An analysis of the effects on Irish hospital care of the supply of care inside and outside the hospital

Walsh, Brendan et al. Research Report An analysis of the effects on Irish hospital care of the supply of care inside and outside the hospital Research Series, No. 91 Provided in Cooperation with: The Economic and Social Research Institute (ESRI), Dublin Suggested Citation: Walsh, Brendan et al. (2019) : An analysis of the effects on Irish hospital care of the supply of care inside and outside the hospital, Research Series, No. 91, ISBN 978-0-7070-0500-3, The Economic and Social Research Institute (ESRI), Dublin, https://doi.org/10.26504/rs91.pdf This Version is available at: https://hdl.handle.net/10419/230299 Standard-Nutzungsbedingungen: Die Dokumente auf EconStor dürfen zu eigenen wissenschaftlichen Zwecken und zum Privatgebrauch gespeichert und kopiert werden. Sie dürfen die Dokumente nicht für öffentliche oder kommerzielle Zwecke vervielfältigen, öffentlich ausstellen, öffentlich zugänglich machen, vertreiben oder anderweitig nutzen. Sofern die Verfasser die Dokumente unter Open-Content-Lizenzen (insbesondere CC-Lizenzen) zur Verfügung gestellt haben sollten, gelten abweichend von diesen Nutzungsbedingungen die in der dort genannten Lizenz gewährten Nutzungsrechte. Terms of use: Documents in EconStor may be saved and copied for your personal and scholarly purposes. You are not to copy documents for public or commercial purposes, to exhibit the documents publicly, to make them publicly available on the internet, or to distribute or otherwise use the documents in public. If the documents have been made available under an Open Content Licence (especially Creative Commons Licences), you may exercise further usage rights as specified in the indicated licence. https://creativecommons.org/licenses/by/4.0/ AN ANALYSIS OF THE EFFECTS ON IRISH HOSPITAL CARE OF THE SUPPLY OF CARE INSIDE AND OUTSIDE THE HOSPITAL BRENDAN WALSH, MAEV-ANN WREN, SAMANTHA SMITH, SEAN LYONS, JAMES EIGHAN AND EDGAR MORGENROTH RESEARCH SERIES NUMBER 91 September 2019 E V I D E N C E F O R P O L I C Y AN ANALYSIS OF THE EFFECTS ON IRISH HOSPITAL CARE OF THE SUPPLY OF CARE INSIDE AND OUTSIDE THE HOSPITAL Brendan Walsh Maev-Ann Wren Samantha Smith Seán Lyons James Eighan Edgar Morgenroth September 2019 RESEARCH SERIES NUMBER 91 Available to download from www.esri.ie © The Economic and Social Research Institute Whitaker Square, Sir John Rogerson’s Quay, Dublin 2 ISBN: 978-0-7070-0500-3 DOI: https://doi.org/10.26504/rs91.pdf Final report of the project entitled: ‘An inter-sectoral analysis by geographic area of the need for and the supply and utilisation of health services in Ireland’ This Open Access work is licensed under a Creative Commons Attribution 4.0 International License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly credited. ABOUT THE ESRI The mission of the Economic and Social Research Institute is to advance evidencebased policymaking that supports economic sustainability and social progress in Ireland. ESRI researchers apply the highest standards of academic excellence to challenges facing policymakers, focusing on 12 areas of critical importance to 21st Century Ireland. The Institute was founded in 1960 by a group of senior civil servants led by Dr T.K. Whitaker, who identified the need for independent and in-depth research analysis to provide a robust evidence base for policymaking in Ireland. Since then, the Institute has remained committed to independent research and its work is free of any expressed ideology or political position. The Institute publishes all research reaching the appropriate academic standard, irrespective of its findings or who funds the research. The quality of its research output is guaranteed by a rigorous peer review process. ESRI researchers are experts in their fields and are committed to producing work that meets the highest academic standards and practices. The work of the Institute is disseminated widely in books, journal articles and reports. ESRI publications are available to download, free of charge, from its website. Additionally, ESRI staff communicates research findings at regular conferences and seminars. The ESRI is a company limited by guarantee, answerable to its members and governed by a Council, comprising 14 members who represent a cross-section of ESRI members from academia, civil services, state agencies, businesses and civil society. The Institute receives an annual grant-in-aid from the Department of Public Expenditure and Reform to support the scientific and public interest elements of the Institute’s activities; the grant accounted for an average of 30 per cent of the Institute’s income over the lifetime of the last Research Strategy. The remaining funding comes from research programmes supported by government departments and agencies, public bodies and competitive research programmes. Further information is available at www.esri.ie. THE AUTHORS Brendan Walsh is a Research Officer, Maev-Ann Wren is a Senior Research Officer, Samantha Smith is Research Associate and Seán Lyons is an Associate Research Professor at the Economic and Social Research Institute (ESRI). James Eighan was a Research Assistant at the ESRI. Edgar Morgenroth is a full Professor of Economics in DCU Business School, Dublin City University, Dublin. Brendan Walsh, Maev-Ann Wren and Seán Lyons have adjunct research positions at Trinity College Dublin. ACKNOWLEDGEMENTS The authors would like to thank the members of the Steering Group and collaborators on the project team for their expert advice and input. The authors would also like to thank people who provided additional analysis for the study. The authors are very grateful to all those who facilitated the many requests for data for this study, in particular: Alan Cahill, Department of Health; Central Statistics Office; Justin Gleeson, All-Ireland Research Observatory (AIRO), Maynooth University; Dr Howard Johnson and the Health Intelligence Unit, Health Service Executive; Vincent Kennedy, Department of Health; Tom O’Regan, Health Information and Quality Authority; Dr Conor Teljeur and Department of Public Health and Primary Care, Trinity College Dublin; Anne Nolan and Paul Redmond, ESRI; Des Williams, National HR Directorate, Health Service Executive; Austin Warters, Michael Fitzgerald and Eithne McAuliffe, Services for Older People, Health Service Executive; and Bob Hennessy and Margaret Cahill, Health Information Quality Authority. Abbreviations|i ABBREVIATIONS ACHI Australian Classification of Health Interventions AHP Allied health professionals AMAU Acute medical assessment unit AMI Acute myocardial infarction (heart attack) BIU Business and Information Unit (in the HSE) CHO Community healthcare organisation COPD Chronic obstructive pulmonary disease CSO Central Statistics Office DoH Department of Health DRG Diagnosis-related group EAPMC Equitable Access to Primary Medical Care ED Emergency department EU European Union FEMPI Financial Emergency in the Public Interest GMS General Medical Services scheme GP General practitioner HCP Home care package HIPE Hospital In-Patient Enquiry Hippocrates Healthcare in Ireland model of effects of Population Projections, Patterns Of CaRe and Ageing Trends on Expenditure and Demand for Services HPO Healthcare Pricing Office HRB Health Research Board HSE Health Service Executive ICD-10 International Statistical Classification of Diseases and Related Health Problems, 10th edition IHI Individual health identifier IP Inpatient ISA Integrated Service Area LHO Local Health Office LOS Length of stay LSAS Long-stay activity statistics LTRC Long-term residential care MAU Medical assessment unit ii|Effects on Irish Hospital Care of the Supply of Care Inside and Outside the Hospital NHI Nursing Homes Ireland NHS National Health Service (of the United Kingdom) NHSS Nursing Home Support Scheme (‘Fair Deal’) NTPF National Treatment Purchase Fund OECD Organisation for Economic Co-operation and Development PET Patient experience time PHA Private Hospitals Association PHI Private health insurance PHN Public health nurse RICO Regional Integrated Care Organisation SAT Single assessment tool SD Standard deviation TILDA The Irish Longitudinal Study on Ageing UHI Universal health insurance UK United Kingdom UQR Unconditional quantile regression WTE Whole-time equivalent Table of Contents|ix TABLE OF CONTENTS EXECUTIVE SUMMARY ......................................................................................................................... xiii CHAPTER 1: Introduction ........................................................................................................................ 1 1.1 Background .............................................................................................................................. 1 1.2 Irish healthcare policy context ................................................................................................ 2 1.3 Outline of report ...................................................................................................................... 4 CHAPTER 2: Motivation for report and literature review ....................................................................... 5 2.1 Scope of chapter ...................................................................................................................... 5 2.2 Motivation ............................................................................................................................... 5 2.3 Literature review ..................................................................................................................... 7 2.3.1 Acute capacity and length of stay ............................................................................... 8 2.3.2 Long-term care capacity and inpatient length of stay ................................................ 9 2.3.3 Irish evidence ............................................................................................................ 11 2.4 Contribution of this report to the literature ......................................................................... 12 CHAPTER 3: Profile of acute care supply, home care and LTRC supply by geographic area ................ 13 3.1 Scope of the chapter ............................................................................................................. 13 3.2 Public hospitals in Ireland ...................................................................................................... 13 3.3 Private hospitals in Ireland .................................................................................................... 16 3.3.1 Private care in acute public hospitals ....................................................................... 16 3.3.2 National Treatment Purchase Fund .......................................................................... 18 3.4 Hospital In-Patient Enquiry (HIPE) dataset ............................................................................ 18 3.5 Acute public hospital bed supply........................................................................................... 19 3.5.1 Inpatient beds ........................................................................................................... 19 3.5.2 Day patient bed supply ............................................................................................. 23 3.5.3 Private hospital beds ................................................................................................ 25 3.6 International comparisons .................................................................................................... 25 3.61 Inpatient beds per capita .......................................................................................... 25 3.6.2 Inpatient bed occupancy rates ................................................................................. 26 3.6.3 Projected bed capacity requirements ...................................................................... 28 3.7 Geographic patterns in the supply of public hospitals .......................................................... 29 3.7.1 Locations of acute public hospitals in 2015 .............................................................. 29 3.7.2 Hospital reconfiguration ........................................................................................... 31 3.7.3 Acute public hospital catchment areas .................................................................... 34 3.8 Home care ............................................................................................................................. 35 3.8.1 Definition of home care ............................................................................................ 35 3.8.2 Home care data ........................................................................................................ 37 3.8.3 Trends over time in home care hours ...................................................................... 38 3.9 Long-term residential care .................................................................................................... 40 3.9.1 Definition of long-term residential care ................................................................... 40 3.9.2 Long-term residential bed data ................................................................................ 41 3.9.3 Trends over time ....................................................................................................... 43 x|Effects on Irish Hospital Care of the Supply of Care Inside and Outside the Hospital 3.10 Home care and long-term residential care correlations ....................................................... 45 3.11 Conclusion ............................................................................................................................. 48 CHAPTER 4: Statistical models and methodology ................................................................................. 49 4.1 Scope of the chapter ............................................................................................................. 49 4.2 Dependent variable ............................................................................................................... 49 4.2.1 Inpatient length of stay ............................................................................................ 49 4.2.2 Delayed discharges ................................................................................................... 50 4.2.3 Discharge to long-term residential care ................................................................... 51 4.3 Independent variables of interest ......................................................................................... 52 4.3.1 Inpatient bed supply ................................................................................................. 52 4.3.2 Home care hours supply ........................................................................................... 52 4.3.3 Long-term residential care bed supply ..................................................................... 53 4.3.4 Confounding variables .............................................................................................. 53 4.4 Patient sample ....................................................................................................................... 56 4.5 Statistical models ................................................................................................................... 57 4.5.1 Linear regression ...................................................................................................... 58 4.5.2 Negative binomial model .......................................................................................... 59 4.5.3 Unconditional quantile regression ........................................................................... 60 4.5.4 Standard errors ......................................................................................................... 62 4.5.5 Statistical programme............................................................................................... 63 CHAPTER 5: Public hospital inpatient length of stay and inpatient bed supply ................................... 65 5.1 Scope of the chapter ............................................................................................................. 65 5.2 Question ................................................................................................................................ 65 5.3 Background ............................................................................................................................ 66 5.3.1 Causal inference ....................................................................................................... 67 5.4 Results .................................................................................................................................. 68 5.4.1 Descriptive statistics ................................................................................................. 68 5.4.2 Length of stay .......................................................................................................... 69 5.4.3 Regression results .................................................................................................... 70 5.4.4 Sensitivity analyses .................................................................................................. 74 5.5 Conclusions ............................................................................................................................ 74 CHAPTER 6: Does formal home care reduce public hospital inpatient LOS and delayed discharges? . 75 6.1 Scope of the chapter ............................................................................................................. 75 6.2 Question ................................................................................................................................ 75 6.3 Background ............................................................................................................................ 76 6.3.1 Causal inference ....................................................................................................... 77 6.4 Results .................................................................................................................................. 78 6.4.1 Descriptive statistics ................................................................................................. 78 6.4.2 Length of stay and inpatient bed days ..................................................................... 79 6.4.3 Regression analyses .................................................................................................. 80 6.4.4 Placebo tests ............................................................................................................. 92 6.5 Conclusions ............................................................................................................................ 93 Table of Contents|xi CHAPTER 7: Does LTRC reduce public hospital inpatient LOS and delayed discharges? ...................... 97 7.1 Scope of the chapter ............................................................................................................. 97 7.2 Question ................................................................................................................................ 97 7.3 Background ............................................................................................................................ 98 7.4 Results ................................................................................................................................. 100 7.4.1 Descriptive statistics ............................................................................................... 100 7.4.2 Length of stay and inpatient bed days ................................................................... 100 7.4.3 Regression analyses ................................................................................................ 101 7.4.4 Placebo tests ........................................................................................................... 114 7.4.5 Probability of discharge to long-stay centre ........................................................... 115 7.5 Conclusions .......................................................................................................................... 116 CHAPTER 8: Concluding discussion and policy recommendations ..................................................... 119 8.1 Scope of chapter .................................................................................................................. 119 8.2 Introduction ......................................................................................................................... 119 8.3 Geographic profile of healthcare needs and non-acute healthcare supply in Ireland (Smith et al., 2019) .......................................................................................................................... 119 8.4 Analysis of the effects on Irish hospital care of the supply of care inside and outside the hospital (Walsh et al., 2019) ................................................................................................ 122 8.5 Policy recommendations for resource allocation................................................................ 128 8.5.1 Improved health system data ................................................................................. 129 8.5.2 Better understanding of substitution and mechanisms to achieve integration of care ......................................................................................................................... 132 8.5.3 Workforce planning ................................................................................................ 134 8.5.4 An allocation system designed to achieve equity of care supply relative to need across geographic areas ......................................................................................... 135 8.5.5 Regular review of resource allocation in line with regional demographic projections .............................................................................................................. 136 8.6 Conclusions .......................................................................................................................... 137 REFERENCES ........................................................................................................................................ 139 APPENDIX 1: Additional data .............................................................................................................. 151 A.1 Catchment areas and reconfiguration ................................................................................. 156 Executive Summary|xiii EXECUTIVE SUMMARY INTRODUCTION This report provides new evidence on key factors that affect patients’ length of stay (LOS) in Irish public acute hospitals. Overall, the report finds that greater supply of home care and long-term residential care (LTRC) could reduce patients’ LOS and thereby reduce delayed discharges, particularly for older people, in Irish hospitals. This is the second report published from the Health Research Board-funded project, An inter-sectoral analysis by geographic area of the need for and the supply and utilisation of health services in Ireland. Findings in this report build on comprehensive evidence from Smith et al. (2019) on the geographic distribution of long-term care and community care in Ireland in 2014. The overall objective of the project is to provide evidence to inform policymakers about the scope to move care from acute hospitals to other care settings in the community or LTRC (whether for longer stays or shorter-term rehabilitation or convalescence). This project is undertaken in the context of the Sláintecare (Houses of the Oireachtas Committee on the Future of Healthcare, 2017) reforms that seek to achieve greater integration in the Irish healthcare system and move delivery of care into the community where appropriate. The aim of this report is to contribute new evidence on key factors that affect patients’ LOS and the extent of delayed discharges in public hospitals. DATA AND METHODS In this report, we examine public hospital inpatients from 2010 to 2015 using the Hospital In-Patient Enquiry (HIPE) database. We examine the impact of both ‘push’ factors – inpatient bed supply – and ‘pull’ factors – home care hours supply and LTRC bed supply. By comparing differences in acute and non-acute supply across regions and hospitals, and within regions and hospitals over time, and by controlling for patients’ individual characteristics, we aim to identify the extent to which such factors influence patients’ LOS. Data on the number of acute inpatient beds for all acute public hospitals in Ireland between 2010 and 2015 were obtained from the Business and Information Unit (BIU) unit in the HSE. Data on publicly financed home care hours within the home help and home care package schemes across the 32 Local Health Offices for each month between 2012 and 2015 were provided by the HSE Social Care Division. Data on LTRC beds were estimated from a combination of Health Information and Quality Authority (HIQA) bed registry data, the Department of Health’s Long-Stay xiv|Effects on Irish Hospital Care of the Supply of Care Inside and Outside the Hospital Activity Statistics surveys, Nursing Home Ireland surveys, and Compliance Monitoring Inspection reports undertaken by HIQA. These data allow us to estimate the number of LTRC beds in each county for the years 2012 to 2015 inclusive. We applied a number of modelling methods in this report to examine the impact of differences in the supply of home care and long-term care on patients’ LOS, distinguishing between groups more or less likely to be classified as delayed discharges. MAIN FINDINGS The report finds that the LOS for older patients is shorter in counties with better supply of home care and LTRC services. The magnitude of the link between greater home care supply and hospital LOS is greater for patients at older ages, patients with long LOS and patients with conditions which lead to particular reliance on home help services. Detailed findings for the effects of home care supply are as follows. • For patients aged 65 years and over, a 10 per cent increase in per capita home care supply (1.5 million hours) was associated with at least a 1 per cent reduction in average LOS and a 1.3 per cent reduction in LOS for delayed discharges; • An increase of 1.5 million hours in home care supply was associated with about 14,700 fewer inpatient bed days per annum, freeing up 40 inpatient beds daily. Among stroke and hip fracture patients, the effect was greater; a 10 per cent increase in per capita home care supply was associated with a 2.7 per cent and 1.6 per cent reduction in average LOS respectively. • There is evidence that effects may be disproportionately stronger when there are large increments to the supply of home care. Analysis from Dublin North, which experienced large increases in home care supply during the period studied, finds that a 10 per cent increase in per capita home care supply was associated with a 2.7 per cent reduction in average LOS and a 6.9 per cent reduction in LOS for delayed discharges. • This would equate to 40,000 fewer inpatient bed days per annum if applied nationally, freeing up 110 inpatient beds daily. The magnitude of the relationship between greater LTRC supply and hospital LOS is greater for patients at older ages, patients with longer LOS and patients who are eventually discharged to LTRC care facilities. Detailed findings for the effects of LTRC supply are as follows. • For patients aged 65 years and over, a 10 per cent increase in per capita LTRC bed supply was associated with a 1.3–2.2 per cent reduction in average LOS. Executive Summary|xv This equates to about 19,000 fewer inpatient bed days per annum or freeing up 53 inpatient beds daily. • A 10 per cent increase in per capita LTRC bed supply was associated with a 5.3 per cent reduction in average LOS for delayed discharges. • For hip fracture patients and patients with Alzheimer’s disease or dementia aged 85 years and over, a 10 per cent increase in per capita LTRC supply was associated with a 2.5 per cent and 5 per cent reduction in average LOS respectively. • Larger substitution effects were estimated for patients ultimately discharged to an LTRC centre. Separately, this report examines the association between inpatient bed capacity and LOS over the period 2010–2015. During the economic recession there was a 13 per cent reduction in inpatient bed supply, with a slight increase since 2013. As inpatient bed supply fell and then recovered over time, average LOS followed a similar pattern, while inpatient bed occupancy rates increased to the highest in the OECD at 95 per cent. This suggests that lower LOS may simply be a consequence of removing beds rather than an attempt to provide care more efficiently. Overall, our analyses suggest that as much as 40–60 per cent of the reduction in inpatient LOS between 2010 and 2012 may have been due to lower inpatient bed capacity, with subsequent increases in LOS due to once more increasing supply. Due to the lack of an individual health identifier (IHI) we cannot estimate readmissions from lower LOS or the impact these reductions had on patient outcomes and population health. xvi|Effects on Irish Hospital Care of the Supply of Care Inside and Outside the Hospital TABLE ES.1 EFFECTS OF HOME CARE AND LTRC SUPPLY ON EMERGENCY INPATIENT DISCHARGES AGED 65+, IN RELATION TO LOS AND DELAYED DISCHARGES Home care hours supply LTRC bed supply Average LOS 10%Δ implies a 1%–1.7%Δ in average LOS. 1.5 million additional home care hours are associated with 14,700 fewer inpatient bed days per annum. *** Dublin North: 10%Δ implies a 2.7%Δ in average LOS. 1.5 million additional home care hours are associated with 40,000 fewer inpatient bed days per annum extrapolated nationally. *** Stroke: 10%Δ implies a 2.7%Δ in average LOS. *** Hip fracture: 10%Δ implies a 1.6%Δ in average LOS. 10%Δ implies a 1.3%–2.2%Δ in average LOS. 2,965 additional LTRC beds are associated with 19,000 fewer inpatient bed days per annum. *** LTRC discharges: 10%Δ implies a 3.3%–3.9%Δ in average LOS. 2,965 additional LTRC beds are associated with 9,720 fewer inpatient bed days per annum. *** Hip fracture: 10%Δ implies a 2.5%Δ in average LOS. *** Alzheimer’s/dementia: 10%Δ implies a 5%Δ in average LOS for those aged 85+. 90th length of stay percentile (delayed discharges) 10%Δ implies a 0.3 days LOS reduction in the 90th percentile, LOS ≥ 21 days. *** Dublin North: 10%Δ implies a 2 days LOS reduction in the 90th percentile, LOS ≥ 29 days. 10%Δ implies a 0.5 days LOS reduction in the 90th percentile, LOS ≥ 22 days. *** LTRC Discharges: 10%Δ implies a 3.3 days LOS reduction in the 90th percentile, LOS ≥ 62 days. Notes: SD = standard deviation; Δ = change. CONCLUSIONS One of the ‘strategic actions’ of the Sláintecare Implementation Strategy involves the expansion of community-based care in Ireland. Bringing care closer to home, where appropriate and feasible, will require increases in workforces providing primary care, community care and long-term care services, as well as increased investment in LTRC beds. This report contributes to the evidence available on how increasing the supply of community-based care will affect demand for acute care in Ireland. The results show that better home care and LTRC provision are associated with shorter hospital LOS, especially among those groups whose care is most amenable to long-term care services, such as patients with stroke or hip fractures, and delayed discharges. Those geographical areas that received the largest increases in home care supply saw the largest reduction in hospital LOS. Executive Summary|xvii Overall, the findings of this project add significantly to the evidence base on health and social care in Ireland. In Smith et al. (2019), we found that there is substantial variation in the supply of non-acute services across regions, which does not reflect relative need. The east of the country fared particularly poorly. Having selected for more detailed examination the cases of home care and LRTC, in this report we have found an association between greater supply of such services and reduced duration of stay in acute hospitals, which is stronger for some patient groups than others. These headline findings from the two reports of this project suggest that hospital performance in Ireland may be driven by factors outside the control of the hospitals as well as within-hospital factors. Consequently, there is a risk that the system of activity-based funding that is being rolled out across acute hospitals may penalise hospitals inappropriately and further add to challenges for areas, which already have relatively inadequate non-acute services. Greater provision of non-acute services such as home care is essential as the population grows and ages, especially as Ireland starts from a point of relative under-capacity in this area. We show that improving long-term care service supply can help to reduce hospital use, but that this is not a panacea for the pressures on the acute hospital system where significant capacity investment is required. Failure to adequately increase long-term care provision will further exacerbate the pressures on an acute system currently struggling to cope. Inequitable supply of non-acute care across Ireland has arisen in an apparently arbitrary, historical manner, and reflects the absence of any system of planning for population health need. Ireland is unusual in not having a transparent and consistent system of resource allocation related to analysis of need. With rapidly increasing demographic pressures on the healthcare system and planned large capacity increases, introducing such a resource allocation method is a priority to optimise care and achieve cost-effective healthcare expenditure. Factors required to facilitate this transition to better resource allocation include: improved data collection; an evidence-based approach to potential substitution between differing services such as home care and hospital care; systematic methods to integrate care across settings; planning for the development and expansion of the healthcare workforce; and regular updating of resource allocation in line with regional population projections. Motivation for Report and Literature Review|5 CHAPTER 2 Motivation for report and literature review 2.1 SCOPE OF CHAPTER This chapter outlines the motivation for this report. Section 2.2 details the background to the analysis in terms of planned changes to the Irish healthcare system. Section 2.3 reviews the existing academic literature on the effects of acute capacity, home care supply and long-term residential care (LTRC) supply on patient outcomes, hospital utilisation and inpatient length of stay (LOS). Section 2.4 outlines the contribution of this report to the literature. 2.2 MOTIVATION Expanding care and service provision in the primary, community and long-term care sectors is a current priority for policymakers in Ireland (Department of Health, 2019; Government of Ireland, 2018b; Houses of the Oireachtas Committee on the Future of Healthcare, 2017). However, as discussed by Smith et al. (2019), there is only limited documented evidence on the patterns of supply of non-acute services across the counties in Ireland. Smith et al. (2019) built upon previous analyses of non-acute services and generated a more comprehensive picture of the geographic distribution of non-acute services in Ireland. This expansion of care in the non-acute sector is envisaged to allow for care to be provided more appropriately outside of hospitals, and to remove much of the pressure on the overburdened hospital sector in Ireland. Expenditure on acute hospital care accounts for the largest proportion of healthcare spending in OECD countries (OECD, 2018). In 2016, hospital care accounted for 60 per cent of total healthcare expenditure in the OECD, with inpatient services alone accounting for 30 per cent of total spend (OECD, 2018). In Ireland, 55 per cent of current healthcare expenditure was accounted for by the hospital sector in 2017 (CSO, 2019). 3 Ireland is often seen as having a ‘hospital-centric’ service delivery model. This overburdening of the acute sector in Ireland has in part resulted in inpatient bed occupancy rates being the highest in OECD countries (OECD, 2018) and some of the longest waiting lists for elective care of developed countries (Siciliani et al., 2014). The overburdened public hospital sector is ill-equipped to meet the pressures on it, with one of the lowest bed-to-population ratios in the OECD (OECD, 2018). The Irish public hospital system has achieved a number of efficiency improvements in recent years, including a high proportion of procedures as day 3 Curative and rehabilitative care services provided by hospitals. 6|Effects on Irish Hospital Care of the Supply of Care Inside and Outside the Hospital cases (Wren et al., 2017) and one of the lower lengths of stay within the OECD (OECD, 2018); however, pressures on the acute sector remain. Delivering care at the most appropriate level is a fundamental principle in the Sláintecare proposals for reform (Houses of the Oireachtas Committee on the Future of Healthcare, 2017). The Sláintecare report acknowledges the importance of developing non-acute care capacity in order to facilitate integrated care, as well as move away from a hospital-centric system. A key ‘strategic action’ of the Sláintecare Implementation Strategy focuses on expanding non-acute care services to provide adequate care supply closer to home (Department of Health, 2019). The Sláintecare report, the Department of Health’s Capacity Review and the National Development Plan (2018–2027) set out a proposed future Irish health and social care system, and each report acknowledges that greater provision of acute services, such as inpatient beds, is required. This is largely due to the current inadequacy of acute service provision, as well as to projected growth and ageing of the population in coming decades (Wren et al., 2017). However, each plan’s acute care projections are heavily dependent upon substantial expansion in nonacute care, especially in sectors that largely provide care for older people. These projections of acute care requirements are therefore dependent upon two very important elements: the ability of patients and health professionals to substitute care into the community; and the system’s ability to expand workforce and capacity of non-acute services quickly. While the latter element is a vital component of healthcare reforms, in this report we focus on the former element – substitutability. We focus on two specific areas when examining substitutability between acute and non-acute care: home care and LTRC. The report focuses on home care and LTRC for a number of reasons. First, these services are two of the largest components of non-acute care in Ireland more generally (Wren et al., 2017). In 2015, there were over 65,000 uses of public and private home care in Ireland. The State financed 10.46 million home help hours, with an additional estimated 3.86 million hours privately purchased. The State also provided 15,300 home care packages. Furthermore, there were an estimated 29,000 LTRC residents in 2015, with the majority (over 21,000) covered by the Nursing Home Support Scheme (NHSS, known as the ‘Fair Deal’ scheme). These residents used 10.6 million LTRC bed days – over twice as many inpatient bed days as in the public and private acute hospitals systems combined. While the ESRI report projected large increases in projected demand across the board, the largest projected increases were seen for these two services, and even accounting for healthy ageing in the older population in the future, demand for home care and LTRC care was projected to increase by 40–54 per cent between 2015 and 2030. In this context, any substitution of care from acute hospitals to home care and LTRC will likely result in even higher projected demand. Motivation for Report and Literature Review|7 Second, there is a dearth of granular data on healthcare utilisation in Ireland. There is a sufficient level of data to examine substitution effects in Ireland in acute public hospital care using the administrative Hospital In-Patient Enquiry (HIPE) data, as well as home care and LTRC, for which administrative data has been collected at regional level over a number of years. Third, while there is limited national and international evidence on substitutability between acute and non-acute services, a small but revealing body of literature has developed around examining substitution effects between hospital care demand and expenditure and long-term care demand and expenditure. This literature is outlined in Section 2.3. In this report, we also examine an oft-overlooked but key aspect: how acute capacity, changes in acute capacity and hospital reconfiguration can impact access to hospitals and hospital utilisation. The relationship between acute care and nonacute care is complicated and impacted by a range of demand-side and supply-side factors. In order to accurately examine the substitution effects of acute and nonacute care, where non-acute care may ‘pull’ patients out of hospitals into more appropriate care settings (such as an LTRC centre), we must also clearly understand that acute capacity constraints can ‘push’ patients out of hospital early, thereby reducing their LOS. It is acknowledged that acute capacity at this moment in Ireland is insufficient; 2,600 additional hospital beds (day patient and inpatient) are now explicitly planned for within the National Development Plan (Government of Ireland, 2018a), with three elective care-only hospitals to be developed in Dublin, Cork and Galway by 2027. In this context, by examining the relationship between bed capacity and hospital use, we also shed light on how the provision of a defined number of acute beds that will be in a system at a given point in time may impact hospitalisations and inpatient LOS. In this report, we provide a detailed overview of the acute system, including the reconfiguration of services seen in recent years, and compare Ireland to international peers with regard to acute capacity. We also discuss the literature that examines how bed capacity impacts hospital use. Chapter 5 is dedicated to highlighting how inpatient bed capacity is also a key determinant of inpatient LOS. 2.3 LITERATURE REVIEW Despite the fact that substitution of some aspects of care away from hospitals is a key policy recommendation of many health systems, there is a paucity of evidence on substitution between hospital care and non-acute care services. Few studies have determined how changes to acute capacity, such as removal of inpatient beds, can impact patient or population outcomes, as well as overall hospital use. In this section, we detail literature on how acute and non-acute care capacity may 8|Effects on Irish Hospital Care of the Supply of Care Inside and Outside the Hospital affect hospital care, in particular the expected effects on our dependent variable of interest: inpatient LOS. 2.3.1 Acute capacity and length of stay In order to reduce the burden on acute hospitals, many health systems have conducted major reforms. Countries have removed care from acute public hospitals by greatly increasing non-acute capacity, replacing inpatient care with day patient care and providing incentives to reduce the length of inpatient stays such as through activity-based funding (ABF). There have been significant reductions in inpatient LOS over time across Europe and in developed countries generally. OECD statistics highlight that between 2000 and 2014, LOS reduced by almost 20 per cent on average in EU28 countries, from 9.9 days in 2000 to 8 days in 2014 (OECD, 2017b). Inpatient LOS reduced from 6.4 days to 5.6 days in Ireland over the same period and now ranks seventh lowest out of the 33 OECD countries examined (OECD, 2017b). These reductions in inpatient LOS across health systems have been a result of a number of different factors. Within acute hospitals, greater use of more efficient surgery (Downing et al., 2009; Jayne et al., 2010; Laudicella et al., 2016), better discharge planning (Dedhia et al., 2009; Siegler et al., 2013), increasing use of palliative care planning (Brody et al., 2010), reductions in delayed discharges (McCoy et al., 2007; Rae et al., 2007) and more efficient payment mechanisms (Besstremyannaya, 2016; Borghans et al., 2008; Échevin and Fortin, 2014; Farrar et al., 2009) have been shown to reduce LOS. For specific patient populations, focused care units such as early supported discharge units and stroke units (Confalonieri et al., 2015; Keegan and Smith, 2013) and greater provision of non-acute follow-up and rehabilitation care have also been shown to allow patients to be discharged more quickly to potentially more appropriate care settings (Dahl et al., 2015; Gozalo et al., 2015; McCoy et al., 2007; Neiterman et al., 2015; Wren et al., 2014). This long, but not exhaustive, list of factors shows that policymakers have many mechanisms with which to reduce inpatient LOS. However, the oft-overlooked factor, which may also drive lower LOS, relates to bed capacity available for patients to use. Lower LOS may be due to more efficient use of care but may also be indicative of a lack of staffing resources, reduced bed capacity, and/or higher bed occupancy rates. In this sense, whether lower LOS is actually valid as a performance measure can be questioned, as beyond a certain point lower LOS may result in negative consequences for the patient (Madsen et al., 2014). Some studies have examined the impact of hospital bed supply, or a proxy of bed supply such as occupancy rates, on patient outcomes. A number of studies have examined the impact of bed supply changes in the Danish healthcare system. In the mid-2000s, Denmark embarked on a significant reorganisation of hospitals, Motivation for Report and Literature Review|9 centralising many services, and the number of acute hospitals reduced from 41 to 20 (Christiansen and Vrangbæk, 2017). A 16 per cent reduction in hospital beds occurred between 2007 and 2014, while inpatient LOS reduced from 3.9 days to 3.1 days (Christiansen and Vrangbæk, 2017). However, unlike the scenario in Ireland where bed cuts were accompanied by cuts to frontline staff numbers, in Denmark doctor and nurse whole-time equivalents (WTEs) actually increased by 19 per cent and 13 per cent respectively between 2007 and 2015 (Christiansen and Vrangbæk, 2017). Other research from Denmark on all admissions to Danish hospitals over a longer period of time has had contrasting results; in an examination of hospital care between 1995 and 2012, evidence of large inpatient bed shortages was found, with the subsequent high bed occupancy rates (greater than 90 per cent in many cases) associated with a 9 per cent increase in rates of inhospital mortality and thirty-day mortality as compared to lower occupancy rates (Madsen et al., 2014). Studies on other health systems have also found a negative relationship between high bed occupancy and patient outcomes. In the NHS, a simulation study found that as occupancy rates exceeded 85 per cent, negative outcomes were seen, which were greatly exacerbated once the rate exceeded 90 per cent occupancy (Bagust et al., 1999). This was also found in a more recent study from the NHS, which showed that both higher occupancy and bed shortages in an NHS hospital were associated with higher patient mortality (Boden et al., 2016). Evidence from the US has found that higher occupancy and lower nursing WTEs were related to increased mortality (Schilling et al., 2010). A systematic review found lower bed capacity is associated with worse health outcomes, including increased mortality (Eriksson et al., 2017). It would be of substantial benefit to undertake similar analyses on the Irish system; however, due to data limitations a comprehensive study is not possible at present. In the context of the relationships that exist between acute care supply, hospital use and patient outcomes, it is clear that prior to examining substitution effects in the Irish system, we first need to examine how inpatient bed capacity and inpatient LOS interact in Ireland. The extent to which lower LOS reduces acute capacity requirements, or vice versa, is a complex question, as outlined in many of the studies above. This issue is examined in Chapter 5, through the use of granular data on inpatient utilisation and inpatient bed supply in Ireland. 2.3.2 Long-term care capacity and inpatient length of stay Despite the fact that many health systems are moving care towards primary, community and long-term care settings, there is a relative paucity of evidence internationally on substitution between acute care and non-acute care. In general, with no quantitative evidence yet available for Ireland on the effects of improving 10|Effects on Irish Hospital Care of the Supply of Care Inside and Outside the Hospital capacity, or increasing the type and amount, of care provided in different settings, this makes the formulation of such policy difficult. Some evidence exists on the interactions between acute and long-term care and social care in the NHS. A number of studies have found that increases in long-term care supply and expenditure reduced hospitalisations and inpatient LOS (Fernandez et al., 2018; Fernandez et al., 2013; Forder, 2009; Forder et al., 2018; Gaughan et al., 2017a; Gaughan et al., 2015; 2017b). However, substitution away from acute care to long-term care may not necessarily be cost saving. Forder (2009) found that increasing long-term care expenditure in England by £1 reduced acute hospital expenditure by £0.35. In this sense, while increasing long-term care supply may not be cost saving, it may allow for patients to be cared for in more appropriate settings and free up valuable inpatient beds for elective care. Where studies try to decompose substitutive long-term care effects into home care and LTRC, most of the substitutive effects are explained by LTRC supply, with only a weak, non-statistically significant relationship between home care supply and inpatient LOS found for those aged 65 years and over (Fernandez and Forder, 2008). Other work from the NHS found that both the number of delayed discharges and emergency inpatient readmissions are impacted by non-acute care provision, especially for older people (Fernandez and Forder, 2008). In this context, Gaughan et al. (2015) found that inpatient LOS and delayed discharges in particular respond to LTRC bed supply. The authors found that that a 10 per cent increase in LTRC beds reduces delayed discharges by between 6 per and 9 per cent. While some studies from the US have argued that increased spending on home care has reduced hospital care expenditure (Lichtenberg, 2012), others have found little evidence of substitution away from post-acute facilities for a Medicare population with stroke, lower extremity joint replacement or hip fractures for patients with better access to home care (Huckfeldt et al., 2014). Home care has been found as a substitute for more skilled nursing home or intermediate care settings (Balentine et al., 2014). For some groups such as those at the end of life, home-based palliative care has also been shown to reduce hospital use and reduce costs (Chitnis et al., 2013). Two studies from Switzerland (Gonçalves and Weaver, 2017) and Spain (Costa-Font et al., 2018) use geographic variation in home care supply (Switzerland) and home care allowances (Spain) across areas over time to try and determine the impact of home care on use of hospital services. As these studies rely on survey-based data, they examine the probability of hospitalisation over a period, number of hospitalisations over a period and LOS for the previous inpatient admission. Diverse results are found. Gonçalves and Weaver (2017) found that among those aged 65 years and older, no statistically significant reduction in LOS was observed Motivation for Report and Literature Review|11 in areas with more home care hours. However, the number of hospitalisations was shown to increase significantly as home care supply increased. While the authors do not test this result, it is possible that increased home care reduced LTRC, and the probability of requiring a hospitalisation may be lower for those in an LTRC centre as these patients can receive some form of acute care in an LTRC centre. Contrastingly, Costa-Font et al. (2018) showed that a policy that involves an increase in the amount of allowances available to informal carers and individuals to pay for formal care results in a reduction in inpatient LOS by up to 30 per cent, as well as reducing hospitalisation rates and hospital costs by 11 per cent (Costa- Font et al., 2018). The allowances paid to informal carers appear to have the greatest effect on reducing hospital use, though the increased allowances provided to older individuals to pay for formal care were also found to reduce hospital use. The substitution effects found by Costa-Font et al. (2018) in terms of reduced LOS are much larger than found previously in the literature and may in part be a result of allowances rather than home care supply per se improving. Many studies examining specific patient groups show that specialised treatment centre supply can greatly reduce acute care use as well as overall LOS for treatment. Evidence on patients receiving hip replacements in England found that patients in specialised public and private treatment centres had 18 per cent to 40 per cent shorter LOS compared to patients receiving care in acute public hospitals (Siciliani et al., 2013). For specific patient populations, focused care units such as early supported discharge units and stroke units (Confalonieri et al., 2015; Keegan and Smith, 2013) and greater provision of non-acute follow-up and rehabilitation care have also been shown to allow patients to be discharge more quickly to potentially more appropriate care settings (Dahl et al., 2015; Gozalo et al., 2015; McCoy et al., 2007; Neiterman et al., 2015). 2.3.3 Irish evidence Evidence on substitution effects across healthcare sectors in Ireland is sparse. Much of the work examining differences in care between primary care and acute care uses changes in medical card or GP visit card access as the means to examine the impact. Recent studies found that expanding access to GPs for free, through the extension of free GP care to children aged under 6 years, increases GP visits (O’Callaghan et al., 2018), but does not necessarily reduce emergency department (ED) attendance rates (Walsh et al., 2019). Other studies have shown improved GP access does not necessarily reduce hospitalisations overall (Ma and Nolan, 2016), or specifically for ambulatory care sensitive conditions – those conditions most appropriately treated in primary care amongst older people (Nolan, 2011). A clear negative relationship has also been found between primary care supply and hospitalisations for patients with certain chronic conditions such as chronic obstructive pulmonary disease (COPD) (Sexton and Bedford, 2016). 12|Effects on Irish Hospital Care of the Supply of Care Inside and Outside the Hospital 2.4 CONTRIBUTION OF THIS REPORT TO THE LITERATURE Overall, the literature on substitution effects between hospital and long-term care is ambiguous. One finding, however, that does appear to be consistent across studies is that even where a substitution effect between long-term care and acute care is found, the effects are small and LTRC supply dominates any substitution effect. However, results are context-specific; many results may not be as applicable an Irish setting due to the particular characteristics of the Irish health system. One objective of this report is to help fill this gap and inform future policy in Ireland. This study builds upon the literature outlined above and models substitutive effects of home care and LTRC supply on inpatient LOS in public hospitals. In this context, we use differences in supply across areas and changes in supply over time within areas, controlling for a range of area-level, hospital-level and patient-level characteristics. We use this modelling strategy in an effort to isolate the causal impact that supply of care for older people has on reducing LOS. Using area-level data on long-term care supply is a common technique in the literature to examine substitution effects (Costa-Font et al., 2018; Fernandez and Forder, 2015; Forder, 2009; Forder et al., 2019; Gaughan et al., 2017a; Gonçalves and Weaver, 2017). Profile of Acute Care Supply and Long-term Residential Care in Ireland|13 CHAPTER 3 Profile of acute care supply, home care and long-term residential care supply by geographic area 3.1 SCOPE OF THE CHAPTER This chapter presents a profile of acute hospital care, home care and long-term residential care (LTRC) in Ireland. It provides the institutional context and data sources for subsequent analytical chapters. Sections 3.2 and 3.3 provide an overview of the acute hospital sector in Ireland, discussing both public hospitals (Section 3.2) and private hospitals (Section 3.3). In Section 3.4 we outline the Hospital In-Patient Enquiry (HIPE) dataset. Section 3.5 discusses acute inpatient bed supply and day patient bed supply in public hospitals; this section illustrates the changes of supply over time and across catchment areas, compares supply in Ireland to other jurisdictions internationally and discusses projected future bed requirements. Section 3.6 provides international comparisons. Section 3.7 details the supply of hospitals and hospital beds across regions, as well as the geographic catchment areas of public hospitals in 2015, considering how supply and distance to hospitals have changed over time as a result of reconfigurations of the system. Section 3.8 discusses supply of home care across counties over time. Section 3.9 discusses supply of LTRC across counties over time. Section 3.10 compares supply of home care and LTRC across counties. Section 3.11 concludes. 3.2 PUBLIC HOSPITALS IN IRELAND The acute hospital sector in Ireland is a mixture of publicly-owned, voluntaryowned and privately-owned hospitals. Publicly-owned acute hospitals and acute hospitals owned by charity/voluntary organisations such as religious institutions are usually grouped together when the acute sector is discussed in Ireland, because both types are run on a not-for-profit basis, with the State providing almost all of the funding required (Tussing and Wren, 2006). In this context, we refer to all publicly-funded (public and voluntary) hospitals as ‘public’ hospitals in this report. Historically, the location of public hospitals was distributed so that almost all 26 counties had an acute public hospital. In 2005, there were 59 public hospitals providing acute services as listed by the Department of Health (Tussing and Wren, 2006); in 2015, there were 53 acute public hospitals included by the Healthcare Pricing Office (HPO) as providing administrative data to HIPE (Healthcare Pricing Office, 2016a). 4 While the number of hospitals has remained relatively constant, 4 The smaller number of hospitals in 2015 is mainly a result of closures and consolidation of some hospitals. 14|Effects on Irish Hospital Care of the Supply of Care Inside and Outside the Hospital the services provided by them have changed considerably; there now exists large differences in the services provided across hospitals. A subset of 29 public hospitals provides the vast majority of inpatient services (see Table 3.2). These 29 hospitals, across 28 sites, 5 each operate a Tier 1 emergency department (ED). These 29 EDs are open 24 hours a day, seven days a week, all year round and, generally, activity across them is used when calculating annual ED attendance rates in Ireland. 6 Three of these hospitals located in Dublin provide care exclusively to children aged under 16 years of age (Walsh et al., 2019). In addition to the 29 adults’ and children’s hospitals, five maternity hospitals exclusively provide maternity care. A further 11 medium-sized hospitals which have injury units (i.e. less comprehensive ED services) also provide emergency medicine (HSE, 2019). These units are often only open for selected hours and may transfer the more complex patients to a regional Tier 1 ED. Many of these hospitals previously had more comprehensive emergency medicine services, but saw a reconfiguration of their services in recent years (McHugh et al., 2019). Many still provide some emergency inpatient care, and also provide a large proportion of elective inpatient care. Other hospitals, no longer providing acute care, are not included in the list; many were previously district hospitals but have been reconfigured to provide short-term, rehabilitation or convalescent care. Table 3.1 lists the largest public hospitals, including maternity hospitals, by their ED status in 2015. Table A.1 in Appendix 1 provides a more detailed list of hospitals and also provides information on changes to their ED status over time. This is not a comprehensive list of acute public hospitals (see HPO for more details), but the hospitals listed do provide the majority of acute care in the public system and are the main hospitals examined in later chapters of this report. 5 A children’s hospital and an adults’ hospital are co-located on the Tallaght University Hospital site. 6 The HSE includes these 29 hospitals within their patient experience time (PET) administrative dataset used to collect information on EDs in Ireland. Profile of Acute Care Supply and Long-term Residential Care in Ireland|21 FIGURE 3.1 INPATIENT BEDS IN PUBLIC HOSPITALS IN IRELAND, 1994–2015 Source: Department of Health, BIU HSE. While Figure 3.1 illustrates the trend in total numbers of beds, it understates the extent of changes in supply per capita. The reductions in inpatient bed capacity observed in recent years also occurred in the context of an increasing population, with the size of the population increasing by 30 per cent in the preceding two decades (Wren et al., 2017). Figure 3.2 illustrates the total number of available inpatient beds per capita (per 1,000 population) in public hospitals in Ireland between 1994 and 2015. The reduction in per capita supply showed a broadly linear trend over time, reducing from 3.31 beds per 1,000 in 1994 to 2.23 beds per 1,000 in 2015, which is equivalent to a 33 per cent reduction in inpatient beds per 1,000 people over this time period. 22|Effects on Irish Hospital Care of the Supply of Care Inside and Outside the Hospital FIGURE 3.2 INPATIENT BEDS PER 1,000 POPULATION IN PUBLIC HOSPITALS IN IRELAND, 1994–2015 Source: Bed data: Department of Health, BIU HSE. Population data – CSO annual population estimates, ESRI annual population estimates. Figure A.1 in the appendix provides information on inpatient beds per 1,000 population aged 65 years and over between 2010 and 2015. Once more, a linear, but much steeper, decline is observed. Between 2010 and 2015, there was a 20.3 per cent reduction in inpatient beds for per 1,000 population aged 65 years and over. This dramatic reduction is in the context of those in this age group being high users of inpatient care (Wren et al., 2017). Inpatient bed supply across hospitals In this report, we are interested in the impact of changes in inpatient bed supply, across hospitals and over time, on inpatient LOS in Ireland. In order to examine these relationships, we accessed inpatient bed data from the Planning and Business Information Unit (BIU) in the HSE. These data provide the average monthly number of regularly maintained and staffed acute inpatient beds available for all acute public hospitals in Ireland between 2010 and 2015. These beds are a mixture of beds open on all days (7-day beds) and beds closed on some days such as at the weekend (5-day beds). Average capacity (beds available) in each month was calculated by summing all 7-day and 5-day acute beds in each hospital in every month, and dividing this number by days in said month. This ensures 5-day and 7- day beds are not counted as equivalent in terms of the number of days they are available within a given month. These data are used as an average capacity measure by the HSE and the Department of Health and are provided to organisations including the OECD as reliable estimates of available hospital capacity in Irish public hospitals. In this report, we include inpatient bed data between January 2010 and December 2015 only. While the national number of inpatient beds is available for the pre-2010 years, as shown in Figure 3.1 and Figure 3.2, data at the hospital-month level are only available from 2010. Profile of Acute Care Supply and Long-term Residential Care in Ireland|23 In order to compare changes in inpatient bed supply across hospitals of different sizes, we constructed a standardised capacity variable for each hospital (Equation 3.1): 𝐵𝑒𝑑𝑠_𝑆𝑡ℎ𝑚 =𝐵𝑒𝑑𝑠ℎ𝑚 −𝐵𝑒𝑑𝑠        ℎ √1 72∑(𝐵𝑒𝑑𝑠ℎ𝑚 −𝐵𝑒𝑑𝑠        ℎ)2 72 𝑚=1 , 3.1 where the mean number of inpatient beds available in each hospital (h) over the whole period, 𝐵𝑒𝑑𝑠        ℎ, was subtracted from the number of available inpatient beds in each of the 72 months in this period, 𝐵𝑒𝑑𝑠ℎ𝑚, (with m signifying month) and subsequently divided by the standard deviation in bed availability in each hospital over the whole period. This provides us with a standardised variable with mean equal to 0 and a standard deviation equal to 1. Examining inpatient bed supply using the standardised variable allows us to compare supply changes across different hospitals of different sizes and to take advantage of bed changes at the month level. It also assists ease of interpretation of the results in Chapter 5. 3.5.2 Day patient bed supply While inpatient care and inpatient LOS are the focus of this report, day patient care is also an important component of the acute hospital sector. A day patient (day case) is a patient who attends an acute hospital for curative or medical treatment, but does not stay overnight. Often due to the shorter length of a stay per visit, a number of day patients use the same bed on a given day, with a 133 per cent occupancy rate estimated (Keegan et al., 2018a). In contrast to inpatient beds, the number of day patient hospital beds in Ireland has increased over time. Some of this increase has likely been due to the re-designation of inpatient beds to day patient beds in recent years, rather than a change in overall acute bed capacity, as there has been no substantial increase in new hospital infrastructure to explain the increases in day patient beds. Figure 3.3 illustrates the total number of day patient beds in public hospitals in Ireland between 2003 and 2015 (data from before 2003 are not available) using data from the Department of Health (pre-2008) and the HSE. 12 It is clear that, unlike inpatient bed supply, day patient bed numbers more than doubled between 2003 and 2012, before stabilising. 12 For the Department of Health data, see https://health.gov.ie/publications-research/statistics/statistics-by- topic/public-hospital-bed-numbers/. 24|Effects on Irish Hospital Care of the Supply of Care Inside and Outside the Hospital FIGURE 3.3 DAY PATIENT BEDS IN PUBLIC HOSPITALS IN IRELAND, 2003–2015 Source: Department of Health; BIU HSE. Examining this increase in day patient beds on a per capita basis, Figure 3.4 illustrates the total number of day patient beds per 1,000 population in public hospitals in Ireland between 2003 and 2015. An increase in per capita numbers was observed between 2003 and 2012. However, per capita numbers actually fell between 2012 and 2015 as the population grew, though the absolute number of day patient beds remained static. FIGURE 3.4 DAY PATIENT BEDS PER 1,000 POPULATION IN PUBLIC HOSPITALS IN IRELAND, 2003–2015 Source: Bed data – Department of Health, BIU HSE. Population data – CSO annual population estimates, ESRI annual population estimates. Figure A.1 in the appendix provides information on day patient beds per 1,000 population aged 65 years and over between 2010 and 2015. During this time, there Profile of Acute Care Supply and Long-term Residential Care in Ireland|25 was a 6.5 per cent reduction in day patient beds for per 1,000 population aged 65 years and over. 3.5.3 Private hospital beds Private hospital bed data are much more limited than data for public hospitals, partly because private hospitals do not report activity to HIPE. The number of private hospital beds is likely to have increased substantially over time as new private facilities have opened in Ireland, though the size of any increase is unknown. There were an estimated 1,975 private hospital inpatient beds in Ireland in 2015 (Wren et al., 2017), with approximately 16 per cent of all inpatient beds located in private hospitals. In 2015, there were an estimated 946 day patient beds in private hospitals (Keegan et al., 2018a), corresponding to 32 per cent of all day patient beds in the acute hospital sector. These numbers emphasise that the private hospital system is a large component of the acute hospital sector. However, there is little information available about the composition of bed supply, such as number of beds across hospitals. Access to such information in the future would greatly expand the ability of researchers and policymakers to understand the acute hospital system in Ireland. 3.6 INTERNATIONAL COMPARISONS In this section we use OECD data to compare inpatient bed supply in Ireland with that of other countries. As with all international comparisons, care should be taken when interpreting these numbers, as criteria for inclusion in OECD metrics does differ across countries (OECD, 2012). 3.61 Inpatient beds per capita In 2000, inpatient bed supply per capita was already amongst the lowest in the OECD (Department of Health and Children, 2002). Supply in Ireland reduced further in subsequent years. Figure 3.5 illustrates that OECD data for 2015 indicate that Ireland ranked as the sixth lowest country in the OECD for inpatient bed supply, with 2.4 acute beds per 1,000 population. Slight variation in definitions of an inpatient bed explains the differences between these figures and those provided earlier in this report; in particular, OECD estimates include psychiatric care beds. 26|Effects on Irish Hospital Care of the Supply of Care Inside and Outside the Hospital FIGURE 3.5 ACUTE INPATIENT BED CAPACITY PER 1,000 PEOPLE ACROSS OECD COUNTRIES, 2015* Source: OECD (2016). Notes: * = or nearest year. Acute beds include beds used: for obstetrics; to cure non-mental illness or to provide definitive treatment of injury; to perform surgery; to relieve symptoms of non-mental illness or injury; to reduce severity of non-mental illness or injury; to protect against exacerbation and/or complication of non-mental illness and/or injury which could threaten life or normal functions; and perform diagnostic or therapeutic procedures (OECD, 2012). The inpatient bed supply for Ireland in Figure 3.5 does not include a consistent estimate for beds in private hospitals. However, even accounting for inpatient bed supply in private hospitals, overall acute inpatient bed supply (including beds in both public and private hospitals) remains amongst the lowest in the OECD (Keegan et al., 2018a). The OECD does not compare day patient beds per capita across countries. Comparing day patient bed supply across other countries (including the NHS in England) is more difficult due to differences in how acute and day patient beds are characterised. 3.6.2 Inpatient bed occupancy rates Ireland has very high occupancy rates for inpatient beds. Occupancy rates measure, on average, the percentage of beds occupied by an inpatient at a moment in time. Internationally, an 85 per cent occupancy rate is used as a maximum threshold. This figure is often cited as a level above which concerns arise about patient safety (Bagust et al., 1999) and bed shortages (Madsen et al., 2014; 012345678 Mexico UK Isreal Sweden Spain Ireland Turkey Denmark Finland Portugal Latvia Norway Greece Estonia Switzerland France Luxembourg Slovenia Czech Rep. Hungary Slovakia Poland Belgium Austria Germany Lituania Korea Japan Inaptient Beds Per 1,000 Population Profile of Acute Care Supply and Long-term Residential Care in Ireland|27 The King’s Fund, 2015). This maximum threshold has also been adopted by recent government bed capacity projections in Ireland (PA Consulting, 2018). The occupancy rate is also estimated as an average across the year, with large positive and negative fluctuations likely on a given day or within a hospital. OECD data for 2000 indicate that Ireland experienced occupancy rates of less than 85 per cent (OECD, 2017a). However, occupancy rates in Ireland have steadily increased over time. Figure 3.6 presents average inpatient occupancy rates across OECD countries in 2015. The inpatient bed occupancy rates in Ireland was on average 94.7 per cent in 2015, the highest in the OECD, and almost 20 percentage points above the average of OECD countries included, and 10 percentage points about the maximum threshold after which concerns about patient safety arise. In total, four countries have occupancy rates greater than 85 per cent. It is not surprising that three of these countries (Ireland, Israel, and the UK (England)) have amongst the lowest bed capacity rates in the OECD. 13 In addition, as we discuss in Chapters 4 and 5, Ireland, England and Canada also have significant issues with delayed discharges. In this context, high occupancy is largely a consequence of low bed capacity. 13 While Denmark has been shown to have high occupancy rates (Madsen et al., 2014). This information is not provided to the OECD. 28|Effects on Irish Hospital Care of the Supply of Care Inside and Outside the Hospital FIGURE 3.6 INPATIENT BED OCCUPANCY RATES ACROSS OECD COUNTRIES, 2015* Note: * = or nearest year. Source: OECD (2017). 3.6.3 Projected bed capacity requirements The acute public hospital capacity constraints in Ireland have been acknowledged by the Government as a matter of concern, and there is a recognition that increases in inpatient bed supply in public hospitals are required. In 2018, a report authored by PA Consulting and funded by the Department of Health referred to the necessity to increase inpatient bed capacity considerably. The report found that by 2031, compared with bed supply in 2015, an additional 2,600–7,150 beds will be required in public hospitals (PA Consulting, 2018). The large range in projections is largely dependent upon assumptions around future improvements to non-acute supply and accompanying substitution effects between hospital and non-acute care. The largest projected increases were estimated for inpatient care, with an estimated additional 2,100–5,800 inpatient beds required by 2031 (PA Consulting, 2018). Importantly, these estimates assume a maximum 85 per cent occupancy rate threshold. 14 The National Development Plan has now reemphasised the need for additional beds, and it includes plans to include 2,600 additional acute public hospital beds by 2027 (Government of Ireland, 2018a). This equates to the lowest number in the PA Consulting range, which assumes considerable investment in 14 In order to achieve 85 per cent occupancy rates in the base year (2015), ceteris paribus, 933 additional inpatient beds would be required in that year. 50% 55% 60% 65% 70% 75% 80% 85% 90% 95% 100% United States Portugal Korea Estonia Slovak Republic Slovenia Turkey Hungary Latvia Luxembourg Greece Austria Czech Republic Japan France OECD27 Spain Mexico Chile Belgium Italy Germany Norway Switzerland UK (England)* Canada Israel Ireland Inpatient bed occupancy rates Profile of Acute Care Supply and Long-term Residential Care in Ireland|29 non-acute supply and the ability to substitute many services from the acute hospital sector. The National Development Plan also sets out plans to develop three elective-only hospitals in Dublin, Cork and Galway. A recent study found that an additional 3,200–5,600 hospital beds will be required by 2030 based upon projected demand increases in the system (Keegan et al., 2018a). Partitioned by bed type, the study estimated the need for an additional 2,620–4,430 inpatient beds and an additional 610–670 day patient beds by 2030, as compared to 2015 levels. The lower projected bed requirements in the study are again dependent upon improvements being made to non-acute supply and substitution towards non-acute care where appropriate; they assume 85 per cent occupancy rates are achieved by 2030 (Keegan et al., 2018a). These findings also imply that the extra 2,600 total beds outlined in the National Development Plan are unlikely to be sufficient to meet demand for inpatient care, even in the context of large substitution of care towards the non-acute sector. While the projected number of added hospital beds required by 2030 is high in absolute terms, Keegan et al. (2018a) highlight that even for the highest projected level of expansion, public hospital inpatient beds per capita would not exceed 3 per 1,000 population and Ireland would remain close to the bottom of OECD countries in line with Figure 3.5. However, failure to increase capacity would result in public hospital inpatient beds per capita falling below 2 per 1,000 population, which would be by far the lowest in the OECD based upon 2015 figures (Keegan et al., 2018a). 3.7 GEOGRAPHIC PATTERNS IN THE SUPPLY OF PUBLIC HOSPITALS The previous sections briefly describe the acute hospital sector in Ireland, the supply of inpatient and day patient hospital beds over time and how Ireland compares to international peers in terms of hospital bed supply. This section discusses the distribution of acute hospital care across Ireland and examines geographic catchment areas of hospitals. It follows a similar template to that used in Smith et al. (2019) to examine the geographic variation in non-acute supply in Ireland. Much of the information in this section is derived from the HIPE dataset. In accordance with this project’s data-sharing agreement with the HPO, in order to prevent disclosure of hospitals, the names of specific hospitals and areas of residence of patients are not identified. 3.7.1 Locations of acute public hospitals in 2015 Figure 3.7 presents the location of each of the 29 hospitals in 2015 with a Tier 1 ED, as mapped by the Irish Association of Emergency Medicine (IAEM). The historical pattern of having a hospital located in every county can still be seen, and a large acute public hospital still exists in most counties. Six hospitals are located in Dublin (three in Dublin North and three in Dublin South), while Cork has 2 large 30|Effects on Irish Hospital Care of the Supply of Care Inside and Outside the Hospital hospitals both located in Cork City. Of the 7 counties in 2015 without a Tier 1 ED public hospital, 3 (Monaghan, Clare, Roscommon) had their local Tier 1 ED reconfigured between 2005 and 2015. A further 6 areas (Tipperary North, southern and northern Cork county, northern Louth county, and Wicklow) also saw their ‘local’ ED reconfigured from Tier 1 ED status in recent years. FIGURE 3.7 LOCATION OF ACUTE PUBLIC HOSPITALS WITH TIER 1 EMERGENCY DEPARTMENT IN IRELAND Source: Irish Association for Emergency Medicine; see http://www.iaem.ie/public/irish-emergency-departments/. Table 3.2 presents the proportion of emergency inpatient discharges that occurred in the larger hospitals with a Tier 1 ED and other smaller hospitals, many of whom were reconfigured, between 2005 and 2015. Overall, it is clear that the 29 larger hospitals provide the majority of emergency inpatient care, with the proportion provided there having increased to over 90 per cent in 2015. Profile of Acute Care Supply and Long-term Residential Care in Ireland|37 spent on home care. This equates to less than half of the amount spent on the LTRC Fair Deal scheme, though the numbers using home care are higher. Between 2006 and 2018, two separate public home care schemes existed: the home care package (HCP) scheme and the home help scheme. Historically, the home help scheme provided domestic support including assistance with cleaning, cooking and basic household tasks, while the HCP scheme was introduced in 2006 to provide more intensive care to allow for older people, particular those discharged from hospital or from a rehabilitation facility, to be cared for in their own home. While differences between the two schemes existed in the past, in recent years they have provided similar care and support to aid individuals at home (Care Alliance Ireland (2018)) and in 2018 were merged into the Home Support Scheme. When generating a metric for home care spending in this report, we combine home care hours provided across both the HCP scheme and the home help scheme. This is necessary due to the practical overlap between the schemes, but it does have the effect of smoothing away some of the actual variation in the mix of services provided that might have existed across areas. This is probably less of an issue for recent years, and combining the spending from the two programmes also means the activity covered by our home care variable is very similar to the care provided in the new Home Support Scheme. Note that we do not examine the recently introduced intensive home care packages (IHCPs). These packages were introduced to target specific high healthcare demand patients, 18 such as those with dementia (Keogh et al., 2018), and in general provide a substantial level of support to only a small number (fewer than 300 in 2015) recipients. 3.8.2 Home care data Due to the lack of information on privately-purchased home care at a regional level, we examine only state-funded home care in this project. Home care data on publicly financed home care supply, between 2012 and 2015, were provided by the HSE Social Care Division. These data capture the number of home care hours provided by the home help scheme and the number of recipients covered by the HCP scheme (with less information on hours) across the 32 Local Health Offices (LHOs – the most granular geographic level for which data are available). These data are available at the month level, allowing us to construct an area-month home care supply variable. 18 The level of services provided as part of an IHCP is much higher than in home care examined here with funding provided for each IHCP between €850 and €1,500 per week (Keogh et al. 2018) equivalent to the cost of a bed in a long-term residential centre. 38|Effects on Irish Hospital Care of the Supply of Care Inside and Outside the Hospital We merge information from the HCP and home help schemes to estimate an overall home care hours supply per capita variable for each LHO using the following steps. First, we estimate the number of home care hours provided annually in an average HCP, using HSE home care data from 2017. This analysis yields an average of 300.6 hours per HCP per annum. Second, we combine the estimated home care hours provided as part of a HCP with hours provided as part of the home help scheme. To account for differences in population size across areas, we construct a ‘home care hours per population aged 65 years and over’ measure for each LHO, by dividing the estimated hours by the population aged 65 years and over in each respective LHO and year. The home care supply variable used is outlined in Equation 3.2: 𝐻𝐶 𝐻𝑜𝑢𝑟𝑠𝑙𝑦 =∑(𝐻𝑜𝑚𝑒 𝐻𝑒𝑙𝑝 𝐻𝑜𝑢𝑟𝑠𝑙𝑚 + 12 𝑚=1 (𝐻𝐶𝑃 𝑟𝑒𝑐𝑖𝑝𝑖𝑒𝑛𝑡𝑠𝑙𝑚 ∗300.6 12 )) 𝑃𝑜𝑝(65+)𝑙𝑦 , 3.2 where l represents the LHO, m the month, and y the year. To aid consistency with Smith et al. (2019), we estimate supply at the county level between 2012 and 2015 in the following sections using the apportioning approach in Smith et al. (2019) to express differences in home care supply across counties. Due to the varying definitions of home care used across countries, it is not possible to compare the supply of home care in Ireland to that in other countries. 3.8.3 Trends over time in home care hours Smith et al. (2019) illustrated the distribution of the annual average number of home care hours per person aged 65 years and over in Ireland in 2014, finding that there was significant geographic variation. In Figure 3.10 and Table 3.3, we expand upon the previous report and illustrate supply across counties over time. Once more, large geographic variations are observed. Counties such as Sligo, Leitrim, Kerry, Louth, Meath and Donegal have consistently higher per capita home care hours across all years. Significant differences within counties are also observed over time. Some counties, including Dublin North, Louth, Meath, Clare and Laois, saw large increases in hours relative to the older population between 2012 and 2015. In the same period, other counties, including Longford, Westmeath, Waterford and Cork, saw reductions in home care hours relative to population. The Gini coefficient provides information on the unequal distribution of supply across counties, where a value of 0 equates to perfect equality in supply, a number greater than 0 denotes inequality in supply, and 0.1 used as a rule of thumb for large inequalities. In Table 3.3, the Gini coefficient for the distribution of home care hours is larger than 0.1 in each year examined, indicating an unequal distribution Profile of Acute Care Supply and Long-term Residential Care in Ireland|39 across areas. The Gini coefficient also increased from 0.118 in 2012 to 0.125 in 2015. 19 FIGURE 3.10 AVERAGE ANNUAL NUMBER OF PUBLICLY FINANCED HOME CARE HOURS PER PERSON AGED 65+ BY COUNTY, 2012–2015 19 Gini coefficients estimated using the conindex command in Stata (O'Donnell et al., 2016). 0 5 10 15 20 25 30 35 40 Number of home care hours per person aged 65+ 2012 2013 2014 2015 40|Effects on Irish Hospital Care of the Supply of Care Inside and Outside the Hospital TABLE 3.3 AVERAGE ANNUAL NUMBER OF PUBLICLY FINANCED HOME CARE HOURS PER PERSON AGED 65+ BY COUNTY, 2012–2015 Geographic area 2012 2013 2014 2015 % change 2012–2015 Carlow 22.3 22.6 22.7 21.1 -5.29 Cavan 29.9 28.5 28.2 27.9 -6.48 Clare 16.7 17.0 16.9 19.5 17.00 Cork 29.5 27.0 28.3 25.3 -14.25 Donegal 30.0 26.2 29.9 30.1 0.54 Dublin North 22.6 27.2 29.6 32.5 43.81 Dublin South 15.9 16.0 15.8 16.6 3.79 Galway 25.3 24.4 24.8 24.6 -2.81 Kerry 31.4 29.6 31.1 30.0 -4.55 Kildare 19.2 18.6 17.1 18.5 -3.61 Kilkenny 21.2 20.0 20.3 18.8 -11.18 Laois 19.6 19.2 22.1 25.1 27.88 Leitrim 34.9 31.3 32.8 34.5 -1.14 Limerick 25.2 22.8 23.1 23.3 -7.40 Longford 22.0 19.7 18.1 17.8 -19.26 Louth 22.1 26.5 29.9 30.3 37.03 Mayo 23.8 21.3 21.7 23.2 -2.76 Meath 27.6 27.9 30.0 31.3 13.33 Monaghan 29.4 28.1 27.5 27.6 -6.27 Offaly 17.1 17.0 19.5 21.9 27.94 Roscommon 27.9 27.5 28.5 29.4 5.61 Sligo 37.6 33.6 34.6 36.7 -2.28 Tipperary North 26.6 25.0 25.0 25.9 -2.49 Tipperary South 26.4 24.7 24.6 25.8 -2.37 Waterford 21.2 16.0 17.0 16.3 -22.84 Westmeath 24.3 21.8 20.1 20.0 -17.89 Wexford 22.8 21.3 22.0 21.8 -4.14 Wicklow 20.7 20.8 20.3 20.3 -1.75 Ireland 23.9 23.3 24.1 24.5 2.63 Gini Coefficient (aged 65+) 0.118 0.113 0.101 0.125 - Source: Social Care Division, HSE. 3.9 LONG-TERM RESIDENTIAL CARE This section discusses the data on LTRC care in Ireland, changes in LTRC bed supply in recent years and differences in LTRC bed supply across counties in Ireland. 3.9.1 Definition of long-term residential care LTRC relates to personal or non-acute health care provided to individuals within a residential care setting over a sustained time period. LTRC is required when it is no longer possible for individuals to be cared for in their own home and the care needed does not require a stay in an acute hospital. Care provided in LTRC is often based around convalescence rather than medical treatment. In this report, LTRC Profile of Acute Care Supply and Long-term Residential Care in Ireland|41 bed data encompasses care beds required for varying durations, and therefore includes beds used for respite, rehabilitation, convalescence and palliative care. This is in line with recent analyses of LTRC in Ireland by Wren et al. (2017). 20 Approximately 4 per cent of all those aged 65 and older are LTRC residents (Wren et al., 2017). While approximately 75 per cent of LTRC residents are covered under the statutory Nursing Home Support Scheme (NHSS), some residents are privately financed or funded under HSE legacy schemes (Wren et al., 2017). The definition, use, delivery and financing of LTRC differs across countries. In 2015, the HSE spent (non-capital) €968 million on LTRC (Department of Health, 2017b), with the majority of expenditure spent on Fair Deal. A needs assessment is required to qualify for Fair Deal, with a co-payment required from each resident based upon a financial assessment of assets, such as the resident’s home (for the first three years of their LTRC stay) and income. Up to 80 per cent of income and up to 7.5 per cent per annum of assets valued greater than €36,000 may be contributed by the resident. 21 Similar to home care, LTRC is provided by a mixture of publicly-owned, voluntaryowned and for-profit centres. Under Fair Deal, individuals have an option to choose from all approved centres, regardless of whether the centre is run by the HSE or not. The majority of LTRC centres in Ireland are privately-owned, with the proportion of such centres increasing. In this report, we concentrate on the use of LTRC for older people (those aged 65 and over). In 2015, only 4.5 per cent of LTRC residents in centres designed to predominantly provide LTRC for older people (as opposed to those caring for younger people with disabilities) was provided to those aged under 65 years (Wren et al., 2017). 22 3.9.2 Long-term residential bed data In this report, LTRC beds per capita are estimated for each region in Ireland over time. There is no definitive list of LTRC centres or beds per centre available. In order to capture all LTRC, we use a number of different data sources. The most comprehensive data on LTRC bed supply was provided by the Health Information and Quality Authority (HIQA). The Health Act (2007) states that all LTRC centres 20 Wren et al. (2017) used the definition of LTRC applied by the Department of Health in its annual Long-stay activity statistics reports. The Department of Health distinguishes between long-stay beds and limited-stay beds. Long-stay beds include those for: extended/continuing care for people who have been assessed as being in need of long-term care; psychiatry of old age, for specialised psychiatric services; and ‘young chronic sick’ for young people with a longlasting illness that is usually irreversible and may be progressive. Limited-stay beds include beds for rehabilitation or convalescence after an illness/injury; palliative care for patients at a time ‘when the medical expectation is no longer cure’; and respite, for ‘the planned admission of dependent persons for short periods of time in order to assist carers in their task of caring’. 21 Contribution deferment (‘Nursing Home Loan’) is possible to be collected posthumously from the resident’s estate. 22 The proportion aged <65 years is much higher when all centres providing care for people with physical and intellectual disabilities are included. 42|Effects on Irish Hospital Care of the Supply of Care Inside and Outside the Hospital must register with HIQA and HIQA is responsible for the regulation of designated centres for older people. Since 2010, HIQA have maintained a database of LTRC centres (both public and private) operating in Ireland. HIQA provided these data for the years 2012–2015 inclusive for this report. The HIQA data provide information on the centre’s ID, provider (HSE, voluntary, or private), address of the centre, most recent registration date and the number of beds the centre has registered for occupancy. The HIQA bed registry is not an exhaustive list, however, with a number of centres missing in earlier years. To fill some of the gaps, we also examined the Department of Health’s long-stay activity statistics (LSAS) and the Nursing Homes Ireland (NHI) surveys. The LSAS survey in 2012 listed the name, address, number of beds (at end December 2012), number of current residents and number of admissions and discharges during 2012 of those centres who responded to the survey. In the 2012 survey, 78.1 per cent of nursing homes responded. The NHI is the representative body for private and voluntary LTRC centres in Ireland. The NHI produced a survey on the private and voluntary long-term residential centres sector in a number of years, with surveys for the years 2007, 2009–2010 and 2014 examined for this project. This survey collected information on the centre name, address and number of beds. In cases where the HIQA registry, LSAS and NHI each failed to provide information on beds, finally we used the Compliance monitoring inspection reports undertaken by HIQA, which include information on the number of residents and number of vacancies (added to provide a beds estimate) on the date of inspection. These reports were also used to check for consistencies in bed numbers within the HIQA bed registry. The most fully-populated HIQA data relate to its 2015 bed registry data, with information on 562 residential centres included. However, information on only 480 centres was included for 2012. Using centre ID, name and address, we have included information on missing centres from 2012–2014 using the LSAS and NHI surveys, as well as the HIQA Compliance monitoring inspection reports, dropping any duplicated centres. 23 , 24 A further 17 centres not included in the HIQA bed registry were added based upon these additional information sources. The final number of LTRC centres found to be operating at some point in the 2012–2015 period was 579. This provided us with a four-year panel of centres across counties. One other issue with the data concerned bed numbers per centre. In some years, bed numbers in each centre were also missing from the HIQA bed registry. We 23 A number of centres saw their name change during the period, or had different names included across data sources. Care was taken to account for name changes and therefore avoid duplication of centres in the data. 24 In a small number of cases, where a centre opened or closed in a particular year, we searched local media outlets to confirm date of opening/closure. Bed data for each centre in year of opening/closing was apportioned based upon month of opening/closure. Profile of Acute Care Supply and Long-term Residential Care in Ireland|43 again used a combination of LSAS and NHI surveys, alongside HIQA Compliance monitoring inspection reports to populate the beds numbers for all centres. By taking these steps, we have developed the most comprehensive LTRC database in Ireland to date, allowing us to examine changes in supply across areas over time. The number of LTRC beds included in this report reflects the maximum number of people that a centre is registered to accommodate, and therefore provides the most accurate reflection of capacity in LTRC in Ireland. The lack of granularity in the data prevents the examination of each type of bed separately. Therefore, LTRC supply examined in this report covers all LTRC bed categories in care for older people, but does not include information from acute hospitals, the National Rehabilitation Hospital or the disabilities sector. We assign centres to counties and LHOs based upon their address. This allows us to estimate the total number of LTRC beds within each area for each year. To examine differences in supply across regions and over time in this chapter, we examine supply at the county level, in line with the approach used in the previous report. When carrying out the substitution analyses in Chapter 7, we include LTRC data at the LHO level in order to match LTRC beds with areas of residence in HIPE. For these analyses, we construct an LTRC bed supply per 1,000 population aged 65 years and over measure for each LHO by dividing the total beds by the population aged 65 years and over in each respective LHO and year. This is outlined in Equation 3.3: 𝐿𝑇𝑅𝐶 𝑝𝑒𝑟 𝑐𝑎𝑝𝑖𝑡𝑎𝑙𝑦 =𝐿𝑇𝐶 𝑏𝑒𝑑𝑠𝑙𝑦 𝑃𝑜𝑝(65+)𝑙𝑦 , 3.3 where l represents the LHO and y the year. 3.9.3 Trends over time The first report from this project illustrated the geographic distribution of LTRC beds per 1,000 population aged 65 years and over in Ireland in 2014 using the same data sources as detailed in the previous section. Smith et al. (2019) found significant geographic variation in LTRC beds. 25 In Figure 3.11 and Table 3.4, we expand upon the previous report and illustrate supply across counties over time. Counties such as Kildare, Roscommon, Tipperary North and Westmeath have consistently higher per capita LTRC beds across all years. Most counties, bar Louth and Wicklow, saw reductions in per capita supply over time, with these reductions especially large in Laois and Meath. However, there is much less volatility in bed 25 There are small differences between Smith et al. (2019) and this report as Smith et al. (2019) used data from early 2015, while this report used data from end of year, 2012-2015 inclusive. 44|Effects on Irish Hospital Care of the Supply of Care Inside and Outside the Hospital supply within counties over time for LTRC than for home care. This is because there were very few centre openings, expansions or closures between 2012 and 2015; changes in bed supply within counties largely reflect growth in the older population over time as opposed to changes in beds. The Gini coefficient for the distribution of LTRC beds remains approximately 0.09 (just below the 0.1 rule of thumb of large inequality) each year examined, indicating an unequal distribution across areas, but no increase or decrease in inequalities over time is observed. FIGURE 3.11 NUMBER OF LONG-TERM RESIDENTIAL CARE BEDS PER 1,000 POPULATION AGED 65+ BY COUNTY, 2012–2015 0 10 20 30 40 50 60 70 80 LTRC beds per 1,000 population aged 65+ 2012 2013 2014 2015 Profile of Acute Care Supply and Long-term Residential Care in Ireland|45 TABLE 3.4 NUMBER OF LONG-TERM RESIDENTIAL CARE BEDS PER 1,000 POPULATION AGED 65+ BY COUNTY, 2012–2015 Geographic area 2012 2013 2014 2015 % change 2012–2015 Carlow 53.74 52.10 50.43 51.76 -3.7 Cavan 58.34 56.68 53.92 51.72 -11.3 Clare 56.83 54.60 52.70 50.75 -10.7 Cork 55.73 54.00 52.82 51.09 -8.3 Donegal 41.64 40.33 38.47 37.46 -10.0 Dublin North 38.68 39.03 38.99 37.73 -2.4 Dublin South 51.58 49.52 49.66 47.27 -8.4 Galway 60.50 59.10 57.87 55.43 -8.4 Kerry 46.53 45.09 43.68 41.88 -10.0 Kildare 80.11 75.53 71.57 71.37 -10.9 Kilkenny 51.77 52.23 53.93 51.79 0.0 Laois 41.36 39.55 32.16 31.41 -24.0 Leitrim 54.60 53.15 51.44 49.40 -9.5 Limerick 54.32 52.57 50.83 49.14 -9.5 Longford 57.49 55.91 53.91 51.68 -10.1 Louth 42.50 40.92 39.51 45.08 6.1 Mayo 55.52 54.47 53.83 51.72 -6.8 Meath 57.70 54.79 52.27 48.07 -16.7 Monaghan 51.31 49.96 48.17 47.03 -8.4 Offaly 55.46 53.55 51.80 51.62 -6.9 Roscommon 68.69 66.52 66.27 64.45 -6.2 Sligo 43.50 42.19 41.02 39.52 -9.2 Tipperary North 67.32 66.28 64.04 61.92 -8.0 Tipperary South 50.69 49.01 48.98 46.83 -7.6 Waterford 50.66 49.17 47.73 45.88 -9.4 Westmeath 63.56 61.94 59.95 58.01 -8.7 Wexford 42.68 42.31 41.10 42.84 0.4 Wicklow 56.38 54.20 57.06 54.53 -3.3 Ireland 52.28 50.82 49.80 48.31 -7.6 Gini Coefficient (Aged 65+) 0.091 0.089 0.094 0.090 - Source: 2012–2015 HIQA, NHI, DoH LSAS. Figure A.4 and Table A.3 show the number of LTRC centres and the average size of LTRC centres (beds per centre) across counties and over time. There are differences in sizes across counties and, in general, the average size of a centre is increasing over time. 3.10 HOME CARE AND LONG-TERM RESIDENTIAL CARE CORRELATIONS The previous sections separately detailed the supply of home care hours and LTRC beds across counties in Ireland. Large geographic variations in both services were observed. In Figure 3.12, using data on both home care hours and LTRC together, we plot supply of the services against one another, for the years 2012, 2013, 2014 46|Effects on Irish Hospital Care of the Supply of Care Inside and Outside the Hospital and 2015, to examine whether those areas with a high relative supply of one service have an accompanying high relative supply of the other. The graphs show that there is, in general, a negative correlation in supply – counties with high relative home care supply tend to have lower relative LTRC bed supply, and vice versa. The negative correlation increases over time. Statistical Models and Methodology|53 number of areas (see Chapter 3). Therefore, for many hospitals, inpatients from a number of LHOs with different levels of home care supply will be included in the analyses at hospital level. This means that our statistical analyses can compare LOS for patients within the same hospital but with different supply of home care in their LHOs, controlling for a range of other pertinent patient-level information. Our analyses, therefore, can more appropriately control for unobserved hospital-level effects, which may have an impact on treatment decisions, quality of care and importantly inpatient LOS. Between 2012 and 2015, Dublin North saw a large increase in home care provision. To examine the relationship between home care and inpatient LOS here, we use variation in home care supply at the month level. As population numbers by month in individual regions are not available, we use absolute differences in home care hours for each month in the analyses. 4.3.3 Long-term residential care bed supply The independent variable of interest in Chapter 7 is LTRC bed supply. In this report, using data from a combination of the HIQA bed registry, the Department of Health’s Long-stay activity statistics (LSAS) survey, Nursing Homes Ireland (NHI) surveys, and Compliance monitoring inspection reports undertaken by HIQA, we estimate the number of LTRC centres and LTRC beds in each county and LHO for the years 2012 to 2015 inclusive. We describe these data sources in detail in Chapter 3 (see Equation 3.3). This measure provides us with an LTRC bed supply measure that accounts for changes in supply over time, differences in population size across LHOs and differences in both LTRC bed supply and population within LHOs over time. In this report, we use the supply of LTRC beds within an individual’s LHO to examine the impact of LTRC supply on inpatient LOS. However, unlike publicly-provided home care, where access and supply is based upon an individual’s address, individuals have greater choice of an LTRC centre. Individuals may choose a centre most suitable to them, provided there is an available bed, and this choice may not necessarily be within an individual’s county or LHO. In these analyses, due to lack of data on choice of centre, we use supply in each patient’s LHO as the means to measure substitution between LTRC supply and inpatient LOS. 4.3.4 Confounding variables The rich HIPE dataset allows us to control for a number of confounding variables, at the patient-level and hospital-level, within our analyses. The ability to control for confounders, especially patient-level casemix, allows us to better isolate the effect of our variables of interest on LOS. 54|Effects on Irish Hospital Care of the Supply of Care Inside and Outside the Hospital Using information from HIPE, the statistical analyses in each of the subsequent chapters control for the following variables: day of inpatient admission; year of inpatient admission (which equals year of discharge for over 98 per cent of inpatients in HIPE); age of patient (included linearly and as a squared function); sex of patient; and marital status (married or not) of the patient. In most analyses, we control for whether the patient stated they were covered by a medical card. Medical card status is a proxy for whether or not the patient’s inpatient stay was covered by the State. As medical card eligibility is often based upon a means test, this variable is also a proxy for socioeconomic or deprivation status, as those with a medical card are more likely to come from a lower income household. Unfortunately, patient-level socioeconomic status is unavailable within HIPE. In some analyses and robustness checks, in lieu of medical card status, we also examine whether the patient was a publicly or privately treated patient. Within HIPE, a variable called ‘discharge status’ is captured; this ‘refers to the public/private status of the patient on discharge and not to the type of bed occupied’ (Healthcare Pricing Office, 2016a) and is a proxy for whether or not the patient is covered by private health insurance. The diagnosis-related group (DRG) for each patient is also included in the regression analyses. The HPO states that ‘the DRG scheme enables the disaggregation of patients into homogeneous groups, which are expected to undergo similar treatment processes and incur similar levels of resource use. The data required for DRG assignment include principal and additional diagnoses, procedures performed, age, sex, and discharge status’ (Healthcare Pricing Office, 2016a). Where the analyses is examining a specific patient group, such as those who have suffered a stroke or a hip fracture, or patients with Alzheimer’s disease/dementia, a DRG dummy is not included. A weighted Charlson comorbidity score is included linearly and as a squared function for each patient in the analyses. The Charlson comorbidity index is used to predict mortality by classifying or weighting comorbid conditions such as cancer, stroke, liver disease and renal disease, with the score weighted to each comorbid condition based upon the relative risk of mortality within 12 months (Li et al., 2011). The higher the score, the ‘sicker’ the patient. Using the DRG and weighted Charlson comorbidity score in combination allows to more accurately compare patients of differing illness. When undertaking hospital-month level analyses in Chapter 5, it is not straightforward to incorporate patient casemix, which arises at the individual rather than aggregate level. As a proxy for casemix in models collapsed to the hospital-month level, we include the average number of diagnoses per patient. This variable is constructed by simply counting the individual number of diagnoses listed in the HIPE record. Statistical Models and Methodology|55 Where applicable we control for admission source. In Chapter 5, we group admission sources into the following four categories: admitted from home; admitted from a nursing home/convalescent home or other long-stay accommodation; admitted via transfer from another hospital; 27 and other. 28 In chapters 6 and 7, we only include those admitted from home. Where applicable, we control for discharge destination, dependent upon the patient sample examined in each chapter. In Chapter 5, we group discharge destination into the following five categories: discharged home; 29 discharged to a nursing home, convalescent home, long-stay accommodation or external rehabilitation facility; died at discharge; 30 transferred to another hospital; 31 and other. 32 In chapters 6 and 7, we exclude those who died, in line with other studies examining substitution between hospital care and home care and LTRC (Gaughan et al., 2017a; Gaughan et al., 2015). In all analyses, a year trend is included to capture any secular trend in LOS that affected the whole sample over time. Robustness checks using time fixed effects are also undertaken, though they do not affect the results significantly. Seasonlevel fixed effects are also included to account for differences in demand of hospital care and non-acute care supply across the year (for example, winter initiatives). Month-level fixed effects are not included as in some analyses inpatient bed capacity and home care variables are included at the month level, and are therefore a linear combination of the month effects; it is inappropriate to include both measures simultaneously (Tefft, 2011). While we control for some hospital-level characteristics, these will fail to capture unobserved differences across hospitals that may affect LOS. These unobserved differences may be quite large and may affect interpretation of results. A recent study using decomposition analyses found large unexplained differences in caesarean section rates across hospitals in Ireland after accounting for maternallevel, clinical-level and hospital-level characteristics (Brick et al., 2016). Hospital fixed effects are therefore included in all analyses to capture any systematic or unobserved hospital-level heterogeneity that may affect LOS. Furthermore, we cluster standard errors at the hospital level. As there is considerable overlap between hospitals and LHOs, there are relatively few LHOs included in these 27 Includes patients with the admission sources: ‘transfer from acute hospital’; ‘transfer from non-acute hospital not in HIPE hospital listing’; ‘transfer from hospice not in HIPE hospital listing’; and ‘transfer from psychiatric hospital/unit’. 28 Includes patients with the admission sources: ‘new born’; ‘temporary place of residence’; ‘prison’; and ‘other’. 29 Represents the vast majority of discharges. 30 Includes patients with the discharge destinations: ‘died with post mortem’; ‘died no post mortem’. 31 Includes patients with the discharge destinations coded by HIPE: ‘transfer to hospital – emergency’; ‘transfer to hospital – non emergency’; ‘transfer to psychiatric hospital/unit’; ‘transfer to non-acute hospital not in HIPE hospital listing – emergency’; ‘transfer to non-acute hospital not in HIPE hospital listing – non emergency’; and ‘hospice (not in HIPE hospital listing)’. 32 Includes patients with the discharge destinations: ‘self discharge’; ‘prison’; ‘absconded’; ‘other (e.g. foster care)’; and ‘temporary place of residence (e.g. hotel)’. 56|Effects on Irish Hospital Care of the Supply of Care Inside and Outside the Hospital analyses. The non-acute supply variables are estimated at the LHO-year level; we do not include an LHO fixed effect (or a Resid fixed effect in Chapter 5) as the models become oversaturated when they are included and much of the variation in non-acute supply is accounted for by the area-level fixed effect and year trend. 4.4 PATIENT SAMPLE When examining the impact of acute and non-acute care supply on inpatient LOS, health data limitations in Ireland constrain the types of patients we can include. First, we only include patients from public hospitals in Ireland as these are the patients covered by the HIPE dataset. We are unable to examine private hospital patients as no returns are made by private hospitals to HIPE or any other centralised database. Second, we only examine emergency inpatients in public hospitals. Due to the lack of information on private hospitals, it is difficult to examine elective inpatient care as many patients can substitute public hospital care with private hospital care. This is in contrast to emergency admissions where the vast majority of activity, especially complex cases, is undertaken in public hospitals. Elective procedures are also increasingly undertaken as day patient procedures as technology has improved (in public and private facilities). Therefore, it is difficult to examine changes in elective inpatient LOS over time given that the composition of the sample having an elective admission changes over time and the potential for significant bias related to omitted variables. Patients may select which public or private hospitals to receive elective care, and make their choice based upon objective or subjective quality measures, such as lower LOS. Due to the urgent nature of many emergency procedures, choice of public hospital is much less likely to be a cause of concern for emergency inpatient LOS. Furthermore, as approximately 80 per cent of inpatient bed days in acute public hospitals are emergency, rather than elective (Keegan et al., 2018b), curtailing our analysis to just emergency inpatients still results in our analyses examining the majority of inpatient care. The lack of an IHI in Ireland reduces our ability to examine various aspects of hospital resource use, including number of hospitalisations per annum or readmission to hospitals within 30 days. The lack of an IHI also leads to complications when examining LOS for emergency inpatients. Many emergency inpatient discharges in HIPE have been transferred from another (smaller, less specialised) hospital; these patients may therefore have already received some of their treatment and had an inpatient admission in their previous hospital (a previous inpatient stay). For this reason, we do not include those patients where HIPE has stated they were transferred from another hospital. While the majority of emergency inpatient admissions were admitted directly from an ED, a small number were also admitted from other emergency service facilities, such as medical assessment units (MAUs). As patients admitted from an MAU may have been admitted from another hospital, or already begun their treatment during Statistical Models and Methodology|57 their MAU stay, only discharges admitted directly from an ED are included in most analyses. During the period of study (2010–2015), as detailed in Chapter 3, there was a significant restructuring of the public hospital system in Ireland. Many hospitals were reconfigured, and in some cases hospitals lost their Tier 1 ED status. Therefore, to reduce bias caused by reconfiguration, in some analyses we curtail our patient sample to those with an inpatient stay in the 26 Tier 1 ED hospitals serving adults. 33 In addition, the three Dublin-based children’s hospitals that only cater to patients aged 15 years or under are not examined here. The 26 Tier 1 ED hospitals accounted for 91 per cent of emergency discharges in HIPE between 2010 and 2015 (see Table 4.2). Due to variations across models in the outcome variable being examined and in the data available, the sample of patients differs slightly across the analytical chapters. Chapter 5 includes inpatients, of all ages, discharged between 2010 and 2015 inclusive. The variable of interest in Chapter 5 is inpatient bed capacity and the data provided to us include maternity beds for those hospitals who provide both maternity and non-maternity care, as well as beds in wards that may be used predominantly by children. Therefore, we include emergency inpatients of all ages in our analyses (subject to the inclusion criteria set out above), though exclude patients coded as maternity. We exclude the six Dublin-based hospitals who primarily cater to maternity care and care for children aged under 15 years. Due to the lack of data on home care and LTRC in 2010 and 2011, the analyses in chapters 6 and 7 are limited to the 2012–2015 period. As the variable of interest represents care for older people, we include only emergency inpatients aged 65 years and over. As in Chapter 5, we exclude the six Dublin-based hospitals who primarily cater to maternity care and care for children aged under 15 years. We exclude patients whose area of residence is recorded in HIPE as ‘of no fixed abode’ or outside of Ireland. We also exclude a small number of patients with an LOS greater than 365 days. 4.5 STATISTICAL MODELS We consider four complementary modelling frameworks to examine emergency inpatient LOS: a linear discharge-level model; linear hospital-level models; negative binomial models; and unconditional quantile regression (UQR) models. Within each model, inpatient LOS (or the natural logarithmic transformation of inpatient LOS) is included as the dependent variable. The linear and negative binomial 33 These 26 hospitals had Tier 1 ED status over the full 2010–2015 period, with one hospital briefly losing Tier 1 ED status in 2011. 58|Effects on Irish Hospital Care of the Supply of Care Inside and Outside the Hospital models examine the impact of differences in acute or non-acute supply on average LOS. The UQR model allows us to examine the impact that differences in non-acute supply have on LOS across the LOS distribution. This latter model therefore allows us to examine the different impact that non-acute care supply may have on groups of patients more likely to be experiencing delayed discharge. 4.5.1 Linear regression We use two variants of the linear regression ordinary least squares (OLS) model in the analytical chapters, with LOS examined at the discharge level and hospitalmonth level. Across both models, we examine the impact of changes in acute and non-acute supply on average inpatient LOS. The discharge-level models allow us to control for patient casemix. The appropriate functional form for regression analysis of LOS is an empirical question. The first model we try in each analysis is a pooled cross-sectional linear model. Due to the heavy right-tailed skewness of LOS data, and as we do not have zero counts (for example, as inpatients have an LOS of at least 1 day), we transform LOS in the linear regression models and examine LOS in its natural logarithmic form. This transformation brings the dependent variable closer to a normal distribution. An added benefit of logged LOS is that by transforming the independent variables of interest, we can interpret the coefficients as elasticities. These elasticities represent the predicted percentage change in LOS for the same percentage change in the independent variables of interest. Measuring these elasticities helps our understanding of the relationship between LOS and acute and/or non-acute care supply and aids interpretation of results across independent variables with different scales. The second type of model used in the analyses is a hospital-level fixed effects linear model. The model collapses data on groups of individual discharges to measure the dependent and independent variables at the hospital-month level. This model is more restrictive than the cross-sectional linear model, and a smaller number of observations is used. A disadvantage of this hospital-month-level analyses is that it makes it much more difficult to control for patient casemix. Following previous work (Bloom et al., 2015; Propper and Van Reenen, 2010), we use the age and gender composition of discharges in each hospital-month period as controls for patient casemix. 34 Additionally, the mean proportion of medical card discharges, mean number of diagnoses per discharge, and mean weighted Charlson score are also included to help control for patient casemix, case severity and socioeconomic status of patients. Once more, a log-linear model, with the natural logarithm of mean hospital-month LOS, is estimated with coefficients interpreted as elasticities. 34 Dummies included for: females aged 0–14 years, males aged 0–14 years, females aged 15–24 years, … females aged 85+ years, males aged 85+ years. Statistical Models and Methodology|59 The benefit of the hospital-level fixed effects linear model is that it allows for many of the unobserved differences across hospitals to be stripped out. We can thereby assess the net effect of the predictors on the outcome variable. Interpretation of linear regression Linear model coefficients are straightforward to interpret. As our dependent variable of interest, LOS, is included in its natural logarithmic form, changes in the independent variables are associated with a percentage change in LOS. Where the independent variables of interest are included as standardised variables, changes of one unit are interpreted in terms of their standard deviation. Therefore, coefficients can be interpreted as follows: • A 1 unit (standard deviation) increase in inpatient beds would change LOS by 𝛽∗100 per cent. Where the independent variables of interest are included in their natural logarithmic form, the coefficients can be interpreted as elasticities as follows: • A 10 per cent increase in non-acute supply is associated with a 𝛽∗0.1 per cent change in LOS. 4.5.2 Negative binomial model As LOS is a count variable, many studies in the literature use models designed explicitly to model count data such as the Poisson model, negative binomial model or the zero-inflated negative binomial model. As there are no zero counts with our analyses (all patients have an LOS of at least one day), more flexible models are not required to account for the issue of modelling zero counts in the analysis. In this report, negative binomial models are preferred to more general count models such as Poisson. Previous studies have found Poisson to perform poorly in modelling the overdispersion in the distribution of LOS (the standard deviation of LOS is much greater than the mean) in comparison to the negative binomial model (Allison and Waterman, 2002; Blundell et al., 2002; Cameron and Trivedi, 2013; Hilbe, 2011). Furthermore, as we have clustered data at the hospital level, this has been shown to exacerbate overdispersion in count data (Hilbe, 2011), thereby indicating that a negative binomial would be more appropriate. To determine whether the negative binomial was the correct model to ‘fit’ the LOS variable, we examined the ability to model the LOS variable accurately. Figure A.5 in the appendix plots the residuals from both negative binomial and Poisson models, to determine best fit of each model to the data. Models with lines close to 60|Effects on Irish Hospital Care of the Supply of Care Inside and Outside the Hospital 0 indicate a better fit of the data. 35 It is clear that the negative binomial model is a better fit of the LOS data than Poisson. An additional advantage of the negative binomial model is that the coefficients can be easily compared to elasticities from the log-linear regression. The coefficient in a negative binomial model expresses the absolute change in LOS for a one unitchange in the independent variable of interest. In general in the analysis, results are very similar across OLS and negative binomial models. Therefore, for the sake of parsimony, we discuss only results from the OLS regressions in each analytical chapter, though negative binomial results are also presented in the relevant tables. Interpretation of the negative binomial model Coefficients within a negative binomial model care are interpreted in terms of the expected log of the count of the dependent variable, or as semi-elasticities. However, interpretation differs across functional forms of the independent variables of interest. Where the independent variables of interest are included as standardised variables, changes of 1 unit are interpreted in terms of their standard deviation. Therefore, coefficients can be interpreted as follows: • A 1 unit (standard deviation) increase in inpatient beds would change the logs of expected LOS by 𝛽, or by 𝛽∗100 per cent. Where the independent variables of interest are included in their natural logarithmic form, the coefficients can be interpreted as elasticities as follows: • A 10 per cent increase in non-acute supply is associated with a 𝛽∗0.1 per cent change in LOS. 4.5.3 Unconditional quantile regression It is possible that non-acute care supply affects the LOS of patients that are likely to have short LOS differently from those with long LOS (given their underlying condition and severity). To examine whether the association between non-acute care supply and LOS varies across the distribution of LOS, we estimate unconditional quantile regression (UQR) on inpatient LOS. The family of quantile regressions offer the ability to examine the impact of specific covariates across distributions of healthcare use. Therefore, they are potentially a powerful modelling technique to use when examining variables such as inpatient LOS, when the impact of covariates may differ for those who with lower healthcare resource use (inpatient bed days) use a little versus those with higher healthcare resource use. Quantile regression analyses are commonly used across the areas of labour (Blau and Kahn, 2017; Redmond and McGuinness, 2019), energy economics (Harold et 35 Fit of data modelled using the countfit command in the statistical programme Stata. Statistical Models and Methodology|61 al., 2017) and education (Cullinan et al., 2018; Öckert et al., 2012). Within the health literature, quantile regressions are used to examine issues such as healthcare expenditure (Olsen et al., 2017). However, these models are seldom used to examine LOS, with most studies using linear or count data models. A study from the US used a ‘conditional’ quantile regression approach to examine the impact prospective payments have on inpatient LOS (Norton et al., 2002). This study found differing effects at the bottom and the top of the LOS distribution. A recent paper examining NHS hospitals by Longo et al. (2019) incorporated an ‘unconditional’ quantile regression to examine the impact hospital competition has on a range of efficiency indictors including LOS. Once more, this study found differing effects at the bottom and the top of the LOS distribution. Both these studies highlight the benefits of using UQR in the study of inpatient LOS. It is increasingly clear that the unconditional variant of quantile regression is a more appropriate model than conditional quantile regression in the measurement of healthcare use (Borah and Basu, 2013; Longo et al., 2019). The main benefit of UQR is that it allows coefficients to be compared across the distribution of LOS. In this context, within UQR models the coefficient for those with short LOS can be compared with the coefficient for those with long LOS. Comparisons of coefficients within the UQR is possible as the technique marginalises the effect of the variable of interest over the distributions of all other independent variables in the model (Borah and Basu, 2013). This is not possible within conditional quantile regressions and this inability to accurately compare coefficients increases as the number of covariates included increase. Therefore, as we control for a large number of independent variables (for example, DRG dummies), the benefits of using UQRs over more traditional quantile regression are large in these analyses. Following Firpo et al. (2009) and Longo et al. (2019), we estimate whether nonacute care supply has different effects across unconditional quantiles of the LOS distribution using a two-stage approach. First, rather than regressing LOS on our covariates, we estimate a recentered influence function (RIF). Influence functions are commonly used within statistics; 36 they test the sensitivity of an estimator to the removal of an observation (patient) with different characteristics and at different parts of the outcome distribution. Here, the influence function is recentred (to 0) to allow for comparability of results across LOS quantiles (Equation 4.1): 𝑅𝐼𝐹(𝐿𝑂𝑆𝑖;𝑞𝜏)=𝑞𝜏+𝜏−1[𝐿𝑂𝑆𝑖≤𝑞𝜏] 𝑓𝐿𝑂𝑆(𝑞𝜏) 4.1 36 For example, in the estimation of the Atkinson Index (Cowell and Flachaire, 2007), or the decomposition of the concentration index (Heckley et al., 2016). 62|Effects on Irish Hospital Care of the Supply of Care Inside and Outside the Hospital Here, RIF captures where patients are in the LOS quantile. 𝑞𝜏 is the 𝜏𝑡ℎ quantile of LOS, [𝐿𝑂𝑆𝑖≤𝑞𝜏] is included only when the a patient’s LOS is less than or equal to quantile 𝑞𝜏, with 𝑓𝐿𝑂𝑆(𝑞𝜏) representing the density function within each 𝑞𝜏. In the second step, the estimated RIF is regressed on the same independent variables as the linear and negative binomial models above (Equation 4.2): 𝑅𝐼𝐹(𝐿𝑂𝑆𝑖;𝑞𝜏)=𝛼𝜏𝑋𝑖+𝛽𝜏𝑁𝐴𝑆𝑙𝑦 +𝜀𝑖 4.2 Using this approach, the other independent variables (denoted by the vector 𝑋𝑖) are controlled for, and the 𝛽 coefficient on our non-acute supply (NAS) variable of interest can be interpreted in a similar way as coefficients from a linear or negative binomial. Interpretation of the unconditional quantile regression model Coefficients within a UQR should be interpreted in a similar way to those in a linear regression. As our independent variables of interest are included in their natural logarithmic form, the coefficients can be interpreted as follows. • A 10 per cent increase in non-acute supply is associated with a 𝛽∗[log(1.1)] change in Y for each quantile of Y. • In chapters 6 and 7, we estimate coefficients for the 20th, 30th, 40th, 50th, 60th, 70th, 80, 90th and 95th unconditional quantiles. To aid interpretation of results, LOS for each unconditional quantile is provided. We also discuss the elasticities in our results. 4.5.4 Standard errors While the analyses are undertaken at the discharge level, patients’ LOS is affected by a host of non-patient-level characteristics, such as the consultant or hospital decision to discharge them. Therefore, within these analyses, we allow for clustering of standard errors at the level of the hospital throughout. As patients are grouped within hospitals, standard errors across patients within hospitals are therefore correlated. Failure to account for within-hospital correlation will lead to spuriously small errors. This failure to account for within-group (hospital) correlation has been shown to result in incorrect statistical significance as a result of these spuriously small standard errors (Rokicki et al., 2018). In line with asymptotic assumptions (related to the law of large numbers), a larger number of observations in a model results in more accurate, and smaller, standard errors. However, clustering standard errors can be problematic. As standard errors are estimated at the hospital (cluster) level rather than at the patient level, the asymptotic assumptions often relied upon when estimating standard errors in general may not hold when the number of clusters is small. There is evidence that a small number of clusters (fewer than 50) may be insufficient to estimate accurate standard errors (Cameron and Miller, 2015). In this instance, the standard errors Public Hospital Inpatient Length of Stay and Inpatient Bed Supply| 69 TABLE 5.1 DESCRIPTIVE STATISTICS OF EMERGENCY INPATIENT DISCHARGES, 2010–2015 Number of discharges Number of emergency inpatient discharges 2,237,026 Year of discharge 2010 339,994 2011 343,294 2012 375,414 2013 383,411 2014 394,963 2015 399,950 Mean SD Length of stay 6.42 15.56 Tier 1 ED hospital discharges 0.91 0.29 Age 49.64 27.60 Medical card 0.586 – Number of diagnoses 3.93 3.19 Weighted Charlson score 0.78 1.63 Mode of emergency admission Emergency department 0.732 – AMAU – admitted as in-patient 0.068 – Other 0.102 – AMAU only 0.098 – Readmission 0.012 0.109 Emergency admissions per hospital per month 1,174 513 Discharge destination Home 0.852 – Long stay 0.054 – Died 0.026 – Transfer 0.055 – Other 0.013 – Admission day Sunday 0.096 – Monday 0.156 – Tuesday 0.166 – Wednesday 0.161 – Thursday 0.157 – Friday 0.157 – Saturday 0.107 – Marital status Single/widowed/separated/other 0.626 – Married 0.374 – 5.4.2 Length of stay Figure 5.2 illustrates the average LOS for emergency inpatient discharges in all public hospitals and separately in Tier 1 ED hospitals, between 2010 and 2015. For all public hospitals, the estimates include all emergency admissions including those from MAUs, other non-ED units and readmissions. For the Tier 1 ED hospitals, only admissions directly from the ED are included. A U-shape is observed in the average 70|Effects on Irish Hospital Care of the Supply of Care Inside and Outside the Hospital LOS over time, with average LOS reducing between 2010 and 2012, by 6 per cent in the overall sample and 6.2 per cent in the Tier 1 ED sample respectively. However, between 2012 and 2015 an increase was observed. Overall, average LOS was still lower in 2015 than at the beginning of the period. There is evidence that the LOS increase reported in 2015 was not an anomaly, with average inpatient LOS (for those who stay overnight) in 2017 reported by HIPE to be 6.9 days. 38 FIGURE 5.2 AVERAGE LENGTH OF STAY FOR EMERGENCY INPATIENT DISCHARGES, 2010–2015 Source: LOS in Tier 1 ED hospitals include those admitted from the ED and exclude readmissions. 5.4.3 Regression results Table 5.2 presents the determinants of LOS from the pooled linear regression and pooled negative binomial regression models (discharge-level models). Results show a positive and statistically significant relationship between inpatient bed supply and average LOS in all models. To provide some interpretation for these results, in the Tier 1 ED hospital sample (Column III) a 1 standard deviation (SD) reduction in inpatient bed availability implies a drop in average LOS of 1 per cent. In Tier 1 ED hospitals, the bed availability between 2010 and 2012 fell by 2.6 SDs (642 beds), implying a drop in average LOS of 2.7 per cent. This can be compared with the 6.4 per cent drop in LOS actually observed between 2010 and 2012. Results from all public hospitals imply a drop in average LOS of 6.5 per cent, compared to the 10.8 per cent drop actually observed between 2010 and 2012. Other interesting results (from column III) are that patients with a medical card had 5.5 per cent longer LOS, even after controlling for confounders such as age, number 38 See http://www.hpo.ie/latest_hipe_nprs_reports/HIPE_2017/HIPE_Report_2017.pdf. 6.0 6.2 6.4 6.6 6.8 7.0 7.2 All Public Hospitals Full ED Hospitals Average length of stay 2010 2011 2012 2013 2014 2015 Public Hospital Inpatient Length of Stay and Inpatient Bed Supply| 71 of diagnoses and area of residence. Patients who were discharged to long-stay units (nursing homes) had much longer LOS. 39 TABLE 5.2 DETERMINANTS OF LENGTH OF STAY FOR EMERGENCY INPATIENT DISCHARGES (DISCHARGE-LEVEL MODEL), 2010–2015 Ordinary least squares (LN LOS) Negative binomial (LOS) (I) (II) (III) (IV) (V) (VI) Inpatient bed capacity (Standardised) 0.018*** 0.010*** 0.010*** 0.018*** 0.011*** 0.011*** Medical card 0.059** 0.059** 0.055** 0.079*** 0.076*** 0.072*** Discharge destination Home (Ref.) Long stay 0.766*** 0.768*** 0.383*** 0.885*** 0.827*** 0.834*** Died 0.116*** 0.047*** -0.003 0.400*** 0.362*** -0.329*** Transfer 0.082** 0.035 0.110*** 0.270*** 0.233*** 0.297*** Other -0.189*** -0.211*** -0.203*** -0.083 -0.078 -0.086 Tier 1 ED only No No Yes No No Yes Readmissions Yes No No Yes No No Clusters 38 38 26 38 38 26 Observations 2,216,733 1,975,782 1,484,253 2,216,733 1,975,782 1,484,253 R squared 0.431 0.428 0.437 – – – Notes: All models control for age, age squared, sex, weighted Charlson comorbidity index (linear and squared), marital status, day of admission, year of discharge, admission source, discharge destination, DRG, hospital fixed effects, season fixed effects and linear time trend. Standard errors are clustered at the level of the hospital. Hospital/month periods with less than 100 total emergency discharges are excluded. * p < 0.01, ** p < 0.05, *** p < 0.01. Table 5.3 presents the determinants of LOS from the hospital-month level linear regression model. Results once more show a positive and statistically significant relationship between inpatient bed supply and average LOS in all models. Overall, the results are similar to those shown in the previous table for both samples, with the model predicting a 2.7 per cent drop in LOS between 2010 and 2012 in the Tier 1 ED hospital sample (column III), compared to the 6.4 per cent drop observed. Table 5.3 also highlights that hospitals with a greater number of medical card admissions and sicker patients (as measured by the mean weighted Charlson score per discharge) have longer LOS. 39 This may in part reflect waiting lists and capacity constraints in the long-stay sector during this time period, and is examined in greater detail in Chapter 7. 72|Effects on Irish Hospital Care of the Supply of Care Inside and Outside the Hospital TABLE 5.3 DETERMINANTS OF LENGTH OF STAY FOR EMERGENCY DISCHARGES (HOSPITAL-LEVEL MODEL), 2010–2015 Ordinary least squares (Ln LOS) (I) (II) (III) Inpatient bed capacity (standardised) 0.019*** 0.013*** 0.013** Mean medical card 0.042 -0.168** -0.148*** Mean weighted Charlson 0.379*** 0.238*** 0.234*** Tier 1 ED only No No Yes Readmissions Yes No No Clusters 38 38 26 Observations 2,505 2,480 1,867 Notes: All models control for the age/gender composition of discharges, total emergency cases by hospital/month, hospital fixed effects and linear time trend. Standard errors are clustered at the level of the hospital. Hospital/month periods with less than 100 total emergency discharges are excluded. Discharges with the longest 1 per cent of LOS are excluded: >60 days. * p < 0.01, ** p < 0.05, *** p < 0.01. Inpatient bed supply, acting as a ‘push’ factor, may have differing effects on LOS across patients with different characteristics. Therefore, in Figure 5.3, we interact inpatient bed supply with sex, marital status, private discharge status and weighted Charlson comorbidity score to test whether reductions in inpatient supply may reduce LOS at a greater rate for females (versus males), married patients, private patients or sicker patients with more comorbidities. 40 Results are based upon the linear regression model from column III in Table 5.2. Overall, we find that there is little heterogeneity in the effect of inpatient bed supply on inpatient LOS in the groups examined. 40 Regarding ‘private discharge status’, public/private discharge status is included in the regression instead of medical card status to estimate findings for this figure. FIGURE 5.3 EMERGENCY INPATIENT DISCHARGES, 2010–2015 Notes: Ordinary least squares regressions based on Table 5.2, column III. Covariates in models: year, married, medical card (or public/private status), weighted Charlson score, hospital FEs, DRG, hospitallevel standardised bed capacity, age and age squared, season. Sample: patients admitted from home through an ED, not transferred to another hospital at discharge, alive at discharge. Clustered standard errors at hospital level. 11.2 1.4 1.6 1.8 Log Length of Stay -2.5 -2 -1.5 -1 -.5 0 .5 1 1.5 2 2.5 Standardised Inpatient Bed Supply Weighted Charlson = 0 Weighted Charlson = 1 Weighted Charlson = 2 Weighted Charlson = 3+ 74|Effects on Irish Hospital Care of the Supply of Care Inside and Outside the Hospital 5.4.4 Sensitivity analyses A number of sensitivity analyses were undertaken to determine if the relationship between inpatient bed availability and LOS was more pronounced in some groups. The interaction coefficients between bed supply and whether the patient had an operation/surgery or not as part of their emergency inpatient stay, discharge destination and hospital size were found to be insignificant. The relationship between bed supply and LOS was found to be larger amongst older patients. 5.5 CONCLUSIONS This chapter finds that the changes in inpatient LOS in Ireland between 2010 and 2015 were closely related to changes in bed capacity that occurred during those years. Descriptively, there was a clear U-shaped pattern in average LOS in these years, corresponding to a similar pattern in inpatient bed supply. In each of the regression analyses, controlling for a range of pertinent patient-level characteristics and discharge casemix, we find a positive, statistically significant relationship between LOS and bed capacity; a higher level of bed capacity is associated with longer LOS. In order to interpret the coefficients and help illustrate the magnitude of the effects, we use the changes in average LOS between 2010 and 2012 as points of reference. These years also equate to the period where the most severe public healthcare expenditure cuts were experienced. Overall, the analysis predicts that approximately 40–60 per cent of the reduction in LOS for emergency inpatients observed between 2010 and 2012 may have been a result of bed capacity reductions experienced in those years. Formal Home Care, Public Hospital Inpatient LOS and Delayed Discharges |75 CHAPTER 6 Does formal home care reduce public hospital inpatient length of stay and delayed discharges? 6.1 SCOPE OF THE CHAPTER In this chapter, we examine the relationship between inpatient length of stay (LOS) and public home care supply in Ireland for older people between 2012 and 2015. Section 6.2 details the question examined and the patients included in the study. Section 6.3 summarises the background to how we have tried to estimate casual effects in this study. Section 6.4 presents the findings and Section 6.5 concludes. 6.2 QUESTION In this analysis, we aim to estimate the impact of the supply of public home care on emergency inpatient LOS in public hospitals in Ireland amongst patients aged 65 years and older. We explicitly examine the impact that home care has on those with long LOS, many of whom are likely to be classified as delayed discharges. We use variations in home care supply across areas and over time as a means to identify causal effects. Furthermore, we exploit the fact that availability of public home care is based upon the patient’s address (Local Health Office (LHO) area) rather than the hospital they attend. This allows us to compare patients attending the same hospital but with differing access to home care, which helps us to better control for unobserved hospital-level effects that may affect a patient’s LOS. Home care may affect the use of hospital services by: a) reducing hospital use, in particular inpatient admissions; and b) reducing LOS of an inpatient hospital admission. We focus on the latter channel due to data constraints, because the lack of an individual health identifier (IHI) in Ireland prevents us following individuals across separate inpatient admissions. As discussed in Chapter 4, in this chapter we include emergency inpatient hospital discharges, admitted through an emergency department (ED) in the Hospital In- Patient Enquiry (HIPE) dataset between 2012 and 2015. To reduce bias caused by smaller hospitals who do not provide a large amount of emergency inpatient care to older people, we curtail our analyses to the 26 large acute public hospitals with a Tier 1 ED. Readmissions were also excluded for comparability across hospitals and time. We only include patients aged 65 years and older at discharge as this is the group targeted by home care in Ireland. 41 We only include patients admitted from home, as home care may be most applicable to them. In line with previous 41 Home care is provided in more limited circumstances to individuals aged less than 65 years. 76|Effects on Irish Hospital Care of the Supply of Care Inside and Outside the Hospital analyses of LOS and long-term care supply (Gaughan et al., 2015), we exclude patients who died in hospital. We further exclude patients who are transferred to another hospital or setting. These inclusion criteria focus our analysis on a set of patients that has the potential to use home care when discharged, and it allows for consistency across hospitals, areas and time. The sample includes 304,005 discharges of patients aged 65 years and over between 2012 and 2015. We undertake sub-analyses in this chapter to examine the robustness of results across different patient sub-populations. First, owing to the large increases in home care seen in Dublin North between 2012 and 2015, we undertake analyses on patients with a Dublin North residence ID who were discharged from hospitals within the Dublin North catchment area. Second, we examine two specific patients groups seen in the literature as most amenable to home care supply: stroke and hip fracture patients (Gaughan et al., 2015; Huckfeldt et al., 2014). Patients with a primary or secondary diagnosis of stroke were identified using ICD-10 classification codes I60, I61, I63 and I64, resulting in 8,671 stroke discharges. Patients with a primary diagnosis of hip fracture patients were identified using ICD-10 codes S7200, S7201 and S7220, resulting in 6,136 hip fracture discharges. Third, we examine those aged 75 years and over and 85 years and over separately, because while public home care is targeted at people 65 years and over in Ireland, in practice usage rates of home care are much higher in the oldest age groups (Wren et al., 2017). Heterogeneity in the effect of home care on LOS is examined across specific demographic groups, including gender, marital status, public/private status and comorbidity. We estimate linear regression and negative binomial regression models to examine the effect of home care supply on average LOS. To examine whether the association between home care provision and LOS varies across the distribution of LOS, we estimate unconditional quantile regressions (UQRs). We control for a range of other confounders including: day of admission; year of admission; admission source; discharge destination; age and age squared; weighted Charlson comorbidity score (linear and squared); sex; marital status; specific diagnosisrelated group (DRG); and medical card status. Finally, to control for unobserved hospital-level factors that may affect results, we include hospital fixed effects. Due to the considerable overlap between LHO and hospital, we do not include LHO- level fixed effects. A standardised measure of the number of inpatient beds at the hospital-month level is also included to account for bed capacity changes in each hospital over the period. 6.3 BACKGROUND As discussed in Chapter 3, there is a paucity of evidence internationally on substitution between acute care and care for older people, with no quantitative Formal Home Care, Public Hospital Inpatient LOS and Delayed Discharges |77 evidence yet available for Ireland. Evidence from England shows that while social care supply and expenditure may reduce inpatient LOS (Fernandez et al., 2018; Fernandez et al., 2013; Forder, 2009; Forder et al., 2018; Gaughan et al., 2017a; Gaughan et al., 2015; 2017b), home care explains little of the observed finding. Further analyses from the NHS, using similar data to those used in this chapter, find only a weak, non-statistically significant relationship between home care supply and inpatient LOS for those aged 65 years and over (Fernandez and Forder, 2008). Evidence from Switzerland by Gonçalves and Weaver (2017) shows that among those aged 65 years and over, no statistically significant reduction in LOS was observed in areas with more home care hours. However, the number of hospitalisations was shown to increase significantly as home care supply increased. Contrastingly, Costa-Font et al. (2018) show a policy to increase allowances for formal and informal home care resulted in a reduction in inpatient LOS by up to 30 per cent, reduced hospitalisation rates, and reduced hospital costs by 11 per cent (Costa-Font et al., 2018). Building on the literature, we examine differences in home care supply across areas over time. Our analysis makes several contributions that extend previous research in this area. First, we use a large administrative database on the full population of public hospital inpatient discharges. In this context, we are not reliant on surveybased data using 12-month recall of hospital use. We more accurately measure hospital LOS and control for important patient-level characteristics, such as admission and discharge destination and diagnosis-related information. Second, we examine supply rather than allowances and examine public home care in a system where individuals do not require a co-payment in order to acquire public home care. Finally, in addition to examining average LOS, we estimate UQRs to examine whether home care supply has a bigger impact on reducing LOS for those with the longest LOS, many of whom would be classified as delayed discharges. 6.3.1 Causal inference In this chapter, we try to estimate the causal effect of home care supply on inpatient LOS in Ireland. This is often difficult to do in the absence of a clear, natural experiment. We use differences in home care supply across areas and changes in supply over time within areas, controlling for a range of hospital-level and patientlevel characteristics. We use this modelling strategy to isolate the causal impact that supply of care for older people has on LOS. Using area-level long-term care supply is a common technique used in the literature to examine substitution effects (Costa-Font et al., 2018; Fernandez and Forder, 2015; Forder, 2009; Forder et al., 2019; Gaughan et al., 2017a; Gonçalves and Weaver, 2017). We take advantage of the fact that supply of available home care differs significantly across areas and that potential home care supply available to 78|Effects on Irish Hospital Care of the Supply of Care Inside and Outside the Hospital individuals depends upon their residence in that area rather than the hospital in which they receive their inpatient treatment. These factors mean that there is no scope for patients to gain access to additional home care by choosing to attend a particular hospital. As noted earlier, Ireland has a large number of hospitals relative to the size of its population. Crucially (for our empirical strategy), the catchment areas of hospitals overlap in a number of areas (see Chapter 3). Therefore, for many hospitals, inpatients from a number of LHOs with different levels of home care supply will be included in the analyses at the hospital level. This means that our statistical analyses can compare LOS for patients within the same hospital but with differing levels of supply of home care in their LHOs, controlling for a range of other pertinent patient-level information. Our analyses can thereby more appropriately control for unobserved hospital-level effects, which may have an impact on treatment decisions, quality of care and, importantly, inpatient LOS. One possible cause for concern is that the home care variable may be acting as a proxy for better non-acute service provision in an area more generally. For example, regions that saw increases in home care supply may have also seen increases in primary and community care supply, which may separately reduce inpatient LOS for both the older and younger populations. In order to test whether generalised improvements in non-acute care supply might explain our results, we undertake placebo tests to test the robustness of the link between home care supply and inpatient LOS in a younger population – aged 18–44 and 18–64. These younger populations are not targeted by the public home care schemes in Ireland; therefore, local supply of home care should have no bearing on their inpatient LOS. A lack of an effect on home care reducing LOS for a younger population will suggest that it is home care, rather than general improvements in non-acute care, that underpin our substitution results. 6.4 RESULTS 6.4.1 Descriptive statistics Table 6.1 presents descriptive statistics. Overall mean LOS was 10.16 days, and the average age of discharges was 77.31 years. The majority (79 per cent) had a medical card and therefore free care during their inpatient stay. While all of the sample were admitted from home, not all were discharged home, with 12 per cent discharged to a long-stay facility. In the sample, stroke and hip fracture patients accounted for 2.85 per cent and 2 per cent of discharges respectively. Formal Home Care, Public Hospital Inpatient LOS and Delayed Discharges |85 Home care per capita in Dublin North increased by 50 per cent between 2012 and 2015, a much greater rate than the increase observed in Ireland on average. This increase was likely a policy response to the low relative supply of home care in the area and the long LOS in public hospitals in Dublin North (see Table 6.3). Examining Dublin North will also allow us to test the theory that the relationship between home care and LOS might be non-linear, whereby home care supply in an area above some threshold level leads to significantly greater response in reducing LOS. We have used a categorical representation of home care hours to allow for possible non-linearities at individual discharge level, but there might be threshold effects that operate at area level as well; for example, to do with policies affecting discharge timing in general. Table 6.3 presents results from the linear and negative binomial regressions for Dublin North discharges. Results follow a similar pattern as seen in the countrywide analyses, with a negative and statistically significant relationship between home care supply and LOS observed, though elasticities are larger than seen for the countrywide analyses. For the 65 years and over sample, elasticities of approximately 0.27 are observed; a 10 per cent increase in home care supply is associated with a 2.7 per cent (0.4 days) reduction in LOS. The coefficient is larger for those aged 75+, where a 10 per cent increase in public home care provision is associated with a 4.2 per cent (0.7 days) reduction in LOS. Home care elasticities are large in the 85+ models; however they are not statistically significant. Results on other variables including sex, marital status and weighted Charlson morbidity scores follow a similar pattern to the Ireland analyses. TABLE 6.3 DETERMINANTS OF LENGTH OF STAY FOR EMERGENCY INPATIENT DISCHARGES AGED 65+, DUBLIN NORTH, 2012–2015 65+ 75+ 85+ Ordinary least squares Negative binomial Ordinary least squares Ordinary least squares (Ln LOS) (LOS) (Ln LOS) (Ln LOS) (I) (II) (III) (IV) Year trend 0.010 -0.003 0.024 0.012 Married -0.109*** -0.147*** -0.084*** -0.006 Medical card 0.135*** 0.141*** 0.136*** 0.109*** Weighted Charlson score 0.136*** 0.156*** 0.157*** 0.188*** Female 0.016 -0.006 0.029* 0.084*** Home care Ln home care hours -0.271*** -0.279* -0.420*** -0.338 Clusters (hospitals) – – – – Observations 34,776 34,776 21,195 6,833 Adjusted R Squared 0.303 – 0.266 0.211 Average length of stay 13.92 days 13.92 days 16.21 days 19.28 days Notes: Other covariates in models: DRG, hospital-level standardised bed capacity, age and age squared, season. Sample: Patients admitted from home through an ED in Dublin North and with an address in Dublin North, not transferred to another hospital at discharge, alive at discharge. Clustered standard errors at hospital level. Home care is not included in quintiles as only one area is examined. Formal Home Care, Public Hospital Inpatient LOS and Delayed Discharges |87 Figure 6.4 presents results from an unconditional quantile regression on the relationship between home care and LOS across the LOS distribution for Dublin North. A pattern similar to the Ireland analyses is observed, but the size of the coefficients once more are larger for Dublin North. No statistically significant relationship between LOS and home care supply is observed prior to the 80th percentile. However, at the top of the distribution, a 10 per cent increase in home care provision is associated with a 2-day reduction in LOS for those in the 90th percentile (equating to a 6.9 per cent reduction in LOS) and a 5.2 days reduction for those in the 95th percentile (equating to a 10.5 per cent reduction in LOS). FIGURE 6.4 UNCONDITIONAL QUANTILE REGRESSION ON EMERGENCY INPATIENT DISCHARGES AGED 65+ DUBLIN NORTH, 2012–2015 Differences between Dublin North and Ireland more generally, regarding the impact of home care, are quite large. In Table 6.4, we try to scrutinise whether these results are borne out by changes in LOS over time. This table shows that while LOS in the top quantiles in Ireland increased slightly between 2012 and 2015, large reductions were seen in Dublin North in line with what is predicted in the models. LOS in the 90th and 95th quantile in Dublin North reduced by 3 days (31 days to 28 days) and 11 days (56 days to 45 days) respectively. This suggests that the larger coefficients observed in the Dublin North analyses in this chapter are qualitatively accurate. -0.35 -0.14 -1.72 -2.04 -2.64 -3.52 -8.87 -21.05 -52.49 -100 -90 -80 -70 -60 -50 -40 -30 -20 -10 0 10 0.2 (2 days) 0.3 (3 days) 0.4 (5 days) 0.5 (6 days) 0.6 (8 days) 0.7 (11 days) 0.8 (16 days) 0.9 (29 days) 0.95 (50 days) Coefficient Length of stay quantile 88|Effects on Irish Hospital Care of the Supply of Care Inside and Outside the Hospital TABLE 6.4 EMERGENCY INPATIENT LENGTH OF STAY QUANTILES, IRELAND AND DUBLIN NORTH, 2012 AND 2015 Ireland Dublin North Quantile 2012 (n=77,452) 2015 (n=76,433) 2012 (n=9,117) 2015 (n=9,098) <20th n/a n/a n/a n/a 20th 2 days 2 days 2 days 2 days 30th 3 3 3 3 40th 4 4 5 5 50th 5 6 6 6 60th 7 7 8 8 70th 9 9 11 11 80th 12 13 17 16 90th 21 22 31 28 95th 33 35 56 45 Stroke and hip fracture patients Stroke and hip fracture patients may be more amenable to home care reducing their LOS than many other types of patients. Table 6.5 and Table 6.6 present results from the linear and negative binomial regressions examining the impact of home care supply on LOS. Due to smaller numbers, it was not possible to undertake UQRs for these patient groups. Table 6.5 shows that a negative and statistically significant relationship between home care supply and LOS is observed for stroke patients. Within the 65+ group, a 10 per cent increase in home care supply is associated with a 2.7 per cent (0.67 days) and 3.1 per cent (0.76 days) reduction in LOS. When public home care supply in patient area of residence is included as quintiles, stroke patients in the top home care quintile have LOS 19.6 per cent lower than patients in the bottom home care quintile. The size of the relationship between home care and stroke LOS also increases across age, with the impact of public home care provision having a larger impact at older ages; a 10 per cent increase in public home care provision is related to a 3.3 per cent (1 day) and 3.8 per cent (1.4 days) reduction in LOS for those aged 75 years and over and 85 years and over, respectively. TABLE 6.5 DETERMINANTS OF LENGTH OF STAY FOR EMERGENCY INPATIENT STROKE DISCHARGES AGED 65+, 2012–2015 65+ 75+ 85+ Ordinary least squares Negative binomial Ordinary least squares (Ln LOS) (Ln LOS) (LOS) (Ln LOS) (Ln LOS) (I) (II) (III) (IV) (V) Year trend 0.000 -0.001 -0.009 0.012 0.009 Married -0.128*** -0.128*** -0.164*** -0.139*** -0.093* Medical card 0.028 0.027 0.041 0.009 -0.044 Weighted Charlson score 0.044*** 0.044*** 0.043*** 0.041** 0.009 Female -0.004 -0.004 -0.027 0.018 0.060 Home care Ln home care hours -0.270*** -0.307*** -0.331*** -0.382* Home care hour quintiles Quintile 1 (Lowest) Base Quintile 2 -0.079 Quintile 3 -0.044 Quintile 4 -0.224*** Quintile 5 (Highest) -0.196** Clusters (hospitals) 26 26 26 26 26 Observations 8,533 8,533 8,533 5,537 1,921 Adjusted R Squared 0.281 0.281 0.264 0.232 Average length of stay 24.91 days 24.91 days 24.91 days 27.92 days 31.05 days Notes: Other covariates in models: hospital-level standardised bed capacity, age and age squared, season, stroke severity. Sample: Patients with primary or secondary stroke diagnosis, admitted from home through an ED, not transferred to another hospital at discharge, alive at discharge. Clustered standard errors at hospital level. 90|Effects on Irish Hospital Care of the Supply of Care Inside and Outside the Hospital Table 6.6 presents results from linear and negative binomial regressions for hip fracture patients. A negative and statistically significant relationship between home care supply and LOS is observed. Within the 65+ group, a 10 per cent increase in home care supply is associated with a 1.6 per cent (0.37 days) reduction in LOS for hip fracture patients. When public home care supply in patient area of residence is included as quintiles, hip fracture patients in the top home care quintile have a 13 per cent lower LOS than patients in the bottom home care quintile. The size of the relationship between home care and hip fracture LOS is highest in those aged 85 years and over, where a 10 per cent increase in public home care supply is related to a 2.8 per cent (0.7 days) reduction in LOS. TABLE 6.6 DETERMINANTS OF LENGTH OF STAY FOR EMERGENCY INPATIENT HIP FRACTURE DISCHARGE AGED 65+, 2012–2015 65+ 75+ 85+ Ordinary least squares Negative binomial Ordinary least squares (Ln LOS) (Ln LOS) (LOS) (Ln LOS) (Ln LOS) (I) (II) (III) (IV) (V) Year trend -0.003 0.000 0.000 -0.005 -0.003 Married 0.007 0.007 -0.130*** 0.033 0.069* Medical card 0.104*** 0.106*** 0.059 0.122*** 0.113** Weighted Charlson score 0.168*** 0.168*** 0.201*** 0.165*** 0.150*** Female -0.079*** -0.079*** -0.146*** -0.038* 0.014 Home care Ln home care hours -0.158** -0.154* -0.165* -0.283* Home care hour quintiles Base Quintile 1 (Lowest) 0.014 Quintile 2 0.041 Quintile 3 -0.102*** Quintile 4 -0.130*** Quintile 5 (Highest) 0.014 Clusters (hospitals) 26 26 26 26 26 Observations 6,136 6,136 6,136 4,794 2,327 Adjusted R Squared 0.146 0.147 0.125 0.107 Average length of stay 23.26 days 23.26 days 23.26 days 24.74 days 26.82 days Notes: Other covariates in models: Hospital-level standardised bed capacity, age and age squared, season. Sample: Patients with primary or secondary hip fracture diagnosis, admitted from home through an ED, not transferred to another hospital at discharge, alive at discharge. Clustered standard errors at hospital level. 92|Effects on Irish Hospital Care of the Supply of Care Inside and Outside the Hospital 6.4.4 Placebo tests In this chapter, we use differences in supply of home care across areas and over time, controlling for a range of patient-level characteristics, to estimate the relationship between home care and earlier discharge for older inpatients from public hospitals. However, regions with better home care supply may also have better non-acute and social care facilities more generally, and changes in the provision of home care may move in line with other service improvements. Therefore, the home care variable may be acting as a proxy for better non-acute service provision in a region. In order to test whether broader changes in non-acute care supply might explain our results, we test the robustness of the link between home care supply and inpatient LOS in a younger population, a population much less likely to avail of public home care services such as home care packages (HCPs), but likely to benefit from improvements to non-acute care more generally. Table 6.7 presents results from the linear regression analysis of the determinants of LOS for those aged 18–44 years and 18–64 years, using the same sets of confounding variables as in the models reported above. Coefficients on marital status, weighted Charlson score medical card status and sex all show similar results to those reported in Table 6.2 for those aged 65 years and over. In the 0–44 and 18–44 age groups, home care does not have a statistically significant association with LOS. Home care shows a small negative and statistically significant (at the 90 per cent level) association for the 18–64 age group. Further analysis shows that this result is driven by those aged 45 years and older. As approximately 10 per cent of all home care provision is provided to those aged 45-64, this small and marginally significant association could be due to this group’s limited use of home care services. In Dublin North, where home care supply saw the largest increase, no statistically significant coefficient on the home care variable is observed for the 18–44 and 18–64 age groups. Similarly, Figure A.6 in the appendix provides results from unconditional quantile regressions showing no relationship between home care and LOS across the LOS distribution for in those aged 18–44 years. Results from these placebo analyses show that the findings of a negative relationship between home care supply and inpatient LOS in inpatients aged 65 years and over are unlikely to be driven by an underlying, omitted, non-acute care variable. Rather, the coefficients on home care are likely to relate to effects of home care supply. Formal Home Care, Public Hospital Inpatient LOS and Delayed Discharges |93 TABLE 6.7 DETERMINANTS OF LENGTH OF STAY FOR EMERGENCY INPATIENT DISCHARGES AGED UNDER 65 YEARS, IRELAND AND DUBLIN NORTH, 2012–2015 Ireland Dublin North 18–44 years 18–64 years 18–44 years 18–64 years (Ln LOS) (Ln LOS) (Ln LOS) (Ln LOS) (I) (II) (III) (IV) Year trend 0.002 0.004 -0.023*** -0.020*** Married -0.056*** -0.076*** -0.041*** -0.057*** Medical card 0.058*** 0.074*** 0.101*** 0.117*** Weighted Charlson score 0.098*** 0.083*** 0.082*** 0.085*** Female 0.020*** 0.016*** 0.010 0.012 Home care Ln home care hours -0.022 -0.034* -0.002 -0.009 Clusters (hospitals) 26 26 - - Observations 204,755 399,329 23,355 44,202 Adjusted R squared 0.319 0.355 0.382 0.412 Average LOS 3.54 days 4.70 days 4.28 days 5.64 days Notes: Linear regression analyses. Other covariates in models: DRG, hospital-level standardised bed capacity, age and age squared, season. Sample: Patients admitted from home through an ED, not transferred to another hospital at discharge, alive at discharge. Clustered standard errors at hospital level for Ireland sample. 6.5 CONCLUSIONS Findings from this chapter show that home care can reduce inpatient LOS for older patients. When average LOS across all emergency inpatients admitted from home aged 65 years and over, both linear and negative binomial regressions show a small substitution effect. These results find that, ceteris paribus, a 10 per cent increase in home care supply per capita (similar to the increase observed in Ireland between 2012 and 2015) is associated with an approximate 1–1.7 per cent reduction in average inpatient LOS. The findings of a small negative relationship between home care supply and average inpatient LOS is consistent with previous evidence from the NHS (Fernandez and Forder, 2008) and Switzerland (Gonçalves and Weaver, 2017). Another way of illustrating the scale of effects shown in this chapter is to say that a 10 per cent increase in home care per capita (1.5 million hours per annum) is associated with a reduction of up to 14,700 inpatient bed days, which equates with 40 more inpatient beds available to the system daily. Examining those with long LOS (delayed discharges), a larger substitutive effect of home care supply is found. Results from UQR analyses show that for those in the 90th percentile of LOS, a 10 per cent increase in per capita home care supply is associated with 0.3 fewer hospital days. The relationship is stronger in the 95th percentile where a 10 per cent increase in per capita home care supply is associated with a 1.2 days reduction in LOS. 94|Effects on Irish Hospital Care of the Supply of Care Inside and Outside the Hospital The results also differ when we examine an area that received a very large increase in home care supply between 2012 and 2015. In Dublin North, an even larger effect of home care on reducing LOS is observed, suggesting that assuming linearity in substitution effects could be misleading. When average LOS is examined, ceteris paribus, a 10 per cent increase in home care supply per capita is associated with an approximate 2.7 per cent reduction in average inpatient LOS for inpatients aged 65 years and over. 43 Imposing these results on the countrywide sample would result in 40,000 fewer inpatient bed days per annum, and 110 more inpatient beds available to the system daily. Results from UQR analyses in Dublin North show that for those in the 90th percentile of LOS, a 10 per cent increase in per capita home care supply is associated with 2 fewer days spent in hospital. Again, the relationship in Dublin North between LOS and home care is stronger in the 95th percentile, where a 10 per cent increase in per capita home care supply is associated with a 5.2 days reduction in LOS. The large effects observed in Dublin North comply with the large actual reductions in LOS found in Dublin North hospitals in recent years. As we show, LOS in the 90th and 95th quantile in Dublin North reduced by 3 days (31 days to 28 days) and 11 days (56 days to 45 days) between 2012 and 2015. While these reductions may be a consequence of a number of factors, results of this chapter do suggest that the increases in home care supply contributed to much of these reductions. Results for stroke and hip fracture patients, two groups of patients that are more likely to benefit from home care services, also show a much larger substitution effect than in the general patient sample. Linear and negative binomial regressions show that a 10 per cent increase in home care supply is associated with a 0.75 and 0.35 day reduction in average LOS respectively. Reducing inpatient care by this amount would save 3 per cent and 1.5 per cent of bed days for those aged 65 years and over with stroke and hip fracture respectively in 2015. Due to the low number of patients, it was not possible to examine whether home care supply had a stronger relationship with reduced LOS for stroke and hip fracture patients with longer LOS; it is expected the relationship would follow that observed for all patients, whereby the largest substitution effects are seen for delayed discharges. Examining the homogeneity of the substitution effects on average LOS across demographic groups (gender, marital status and Charlson comorbidity illness score), we see some small differences. Female patients aged 65 years and over have longer LOS than male patients in general. Results show a sharper reduction in LOS for females than males as LOS increases. No differences were observed across married and not-married patients, or between public and private patients. Previous evidence shows that widowed or unmarried adults have the highest risks of long-term care admission (Thomeer et al., 2016), likely a proxy of low informal 43 A larger elasticity of 0.42 is estimated for those aged 75 years and over. Does LTRC Reduce Public Hospital Inpatient LOS and Delayed Discharges?| 101 per cent of all inpatient bed days in a given year. This figure highlights that examining the longest stays using UQR may be most appropriate to understand how LTRC bed supply impacts LOS for the small proportion of inpatients who use a disproportionate amount of inpatient care. FIGURE 7.1 CUMULATIVE INPATIENT BED DAYS BY LENGTH OF STAY, EMERGENCY INPATIENT DISCHARGES AGED 65+, 2012–2015 7.4.3 Regression analyses Table 7.2 presents results from the linear and negative binomial regressions for the full sample. Results present the marginal effect of LTRC bed supply per capita on average LOS. Across each model, a negative and statistically significant relationship between LTRC bed supply and LOS is observed. For the 65+ sample, elasticities of 0.13 to 0.21 are observed; a 10 per cent increase in LTRC bed supply per capita is associated with a 1.3 per cent (0.1 days) and 2.2 per cent (0.2 days) reduction in average LOS. When LTRC supply is divided into quintiles there is some evidence of an inverted U-shaped relationship, with LOS lowest for patients with the lowest and the highest supply of LTRC beds. However, there are no statistically significant differences between the lowest quintile group and the others. LTRC bed elasticities are shown to increase in the 75+ and 85+ samples. Within the 85+ sample, an elasticity of 0.23 is observed; a 10 per cent increase in LTRC bed supply is associated with a 2.3 per cent (0.3 days) reduction in LOS. In all of these models, we control for home care supply. In all models, the elasticity on home care supply is negative and statistically significant, with elasticities similar to those observed in the previous chapter. This confirms the expectation that both home care and LTRC may substitute for inpatient LOS amongst older people. 0 500,000 1,000,000 1,500,000 2,000,000 2,500,000 3,000,000 3,500,000 Cumulative inpatient bed days Length of stay 90th LOS percentile 47.6% of inpatient bed days TABLE 7.2 DETERMINANTS OF LENGTH OF STAY FOR EMERGENCY INPATIENT DISCHARGES AGED 65+, 2012–2015 65+ 75+ 85+ Ordinary least squares Negative binomial Ordinary least squares (Ln LOS) (Ln LOS) (LOS) (Ln LOS) (Ln LOS) (I) (II) (III) (IV) (V) Year trend 0.006 0.006 0.005 0.005 -0.002 Married -0.115*** -0.115*** -0.149*** -0.105*** -0.057*** Medical card 0.073*** 0.073*** 0.065*** 0.058** 0.050* Weighted Charlson score 0.098*** 0.098*** 0.106*** 0.109*** 0.122*** Female 0.020*** 0.020*** 0.006 0.022*** 0.039*** Ln home care -0.126** -0.104* -0.215*** -0.149* -0.109 Ln LTRC beds -0.130** -0.210*** -0.151** -0.230*** LTRC beds quintiles Quintile 1 (lowest) Base Quintile 2 0.015 Quintile 3 0.006 Quintile 4 -0.018 Quintile 5 (highest) -0.033 Hospital Fes Y Y Y Y Y Observations 333,928 333,928 333,928 202,587 67,092 Clusters (hospitals) 26 26 26 26 26 Adjusted R Squared 0.272 0.272 - 0.241 0.210 Average length of stay 10.51 days 10.51 days 10.51 days 11.87 days 13.62 days Notes: Other covariates in models: DRG, hospital-level standardised bed capacity, age and age squared, season. Sample includes patients admitted from home through an ED, not transferred to another hospital at discharge, alive at discharge. Clustered standard errors at hospital level. Does LTRC Reduce Public Hospital Inpatient LOS and Delayed Discharges?|103 Table 7.3 presents results from the linear and negative binomial regressions for those patients who are ultimately discharged to a long-stay facility. Results again show the marginal effect of LTRC bed supply per capita on average LOS. Across each model, a negative and statistically significant relationship between LTRC bed supply and LOS is observed, with elasticities larger than in the full sample included in Table 7.2 above; in other words, LTRC bed supply has a stronger association with LOS for those discharged to a long-stay facility. This is of greater consequence, as overall inpatient LOS for those patients discharged to a long-stay facility is almost three times as long as those discharged home, ceteris paribus. For the 65+ sample, elasticities of 0.33 to 0.39 are observed; a 10 per cent higher home care supply is associated with a 3.3 per cent (0.85 days) and 3.9 per cent (1 day) lower LOS. When included as quintiles, however, again there are hints of an inverted U-shaped relationship, with LOS lowest for patients with the lowest and the highest supply of LTRC beds. Again, however, the differences between individual quintile coefficients and the lowest category are not statistically significant. LTRC bed elasticities are similar or slightly smaller in the 75+ and 85+ samples. TABLE 7.3 DETERMINANTS OF LENGTH OF STAY FOR EMERGENCY INPATIENT DISCHARGES AGED 65+DISCHARGED TO LONG-STAY FACILITY, 2012–2015 65+ 75+ 85+ Ordinary least squares Negative binomial Ordinary least squares (Ln LOS) (Ln LOS) (LOS) (Ln LOS) (Ln LOS) (I) (II) (III) (IV) (V) Year trend -0.017 -0.013 -0.032 -0.020 -0.030 Married 0.019 0.020 0.001 0.034*** 0.056*** Medical card 0.039** 0.039** 0.023 0.041** 0.042* Weighted Charlson score 0.091*** 0.091*** 0.089*** 0.098*** 0.103*** Female -0.036*** -0.036*** -0.046*** -0.033*** -0.016 Home care -0.231** -0.159* -0.323*** -0.276** -0.238* Ln LTRC beds -0.333** -0.394*** -0.276** -0.303 LTRC beds quintiles Quintile 1 (lowest) Base Quintile 2 0.028 Quintile 3 0.087 Quintile 4 0.017 Quintile 5 (highest) -0.045 Clusters (hospitals) 26 26 26 26 15,315 Observations 38,125 38,125 38,125 31,677 15,315 Adjusted R Squared 0.209 0.209 - 0.206 0.197 Average length of stay 25.88 days 25.88 days 25.88 days 25.69 days 24.92 days Notes: Other covariates in models: DRG, hospital-level standardised bed capacity, age and age squared, season. Sample includes patients admitted from home through an ED, transferred to long-stay facility at discharge, alive at discharge. Clustered standard errors at hospital level. Does LTRC Reduce Public Hospital Inpatient LOS and Delayed Discharges?| 105 As with home care supply, LTRC bed supply may also have differing associations with LOS across patients with different characteristics. Therefore, in Figure 7.2 we interact LTRC bed supply with sex, marital status, public/private status and weighted Charlson comorbidity score to test whether increased LTRC bed supply may reduce LOS at a greater rate for females (versus males), married patients or sicker patients with more comorbidities. Results are based upon the linear regression model from Column I in Table 7.2. Overall, we find little heterogeneity in the effect of LTRC bed supply on inpatient LOS. LTRC supply has slightly stronger effect sizes for females (versus males) and public (versus private) patients, but no statistically significant differences are observed. However, LTRC bed supply does seem to be associated with lower LOS among sicker inpatients, as measured by higher weighted Charlson comorbidity scores. This is in line with the similar results found for home care supply in the previous chapter. FIGURE 7.2 DETERMINANTS OF LENGTH OF STAY FOR EMERGENCY INPATIENT DISCHARGES AGED 65+ INTERACTION PLOTS, 2012–2015 Notes: Ordinary least squares regressions based on Table 7.2, Column I. Covariates in models: year, married, medical card, weighted Charlson score, hospital fees, DRG, hospital-level standardised bed capacity, home care supply, age and age squared, season. Sample: Patients admitted from home through an ED, not transferred to another hospital at discharge, alive at discharge. Clustered standard errors at hospital level. Does LTRC Reduce Public Hospital Inpatient LOS and Delayed Discharges?| 107 The results presented in Table 7.2 and Table 7.3 show that LTRC is associated with lower LOS on average. In Figure 7.3 and Figure 7.4, we present results from the UQR where the impact of LTRC supply is examined across LOS quantiles, for all emergency inpatient discharges aged 65 years and over, and those ultimately discharged to an LTRC centre. In Figure 7.3, examining all emergency inpatients aged 65 years and over, LTRC bed supply has a stronger negative association with reducing LOS for patients with longer LOS. While a small, negative coefficient is observed for an LOS of 4 days and above, the coefficients are larger for those with an LOS greater – in the 90th (LOS of greater than 22 days) and 95th quantiles (LOS of greater than 35 days). Transforming the coefficients to be interpreted as elasticities, 10 per cent higher LTRC bed supply is associated with a 0.5 days lower LOS for those in the 90th percentile (equating to a 2.2 per cent LOS reduction) and 1.2 days lower LOS for those in the 95th percentile (equating to 3.3 per cent LOS reduction). In this regard, higher LTRC bed supply seems to be associated with a greater reduction in LOS in both marginal and absolute terms for patients who may be characterised as delayed discharges. FIGURE 7.3 UNCONDITIONAL QUANTILE REGRESSION, EMERGENCY INPATIENT DISCHARGES FOR PATIENTS AGED 65+, 2012–2015 In Figure 7.4, examining all those inpatient discharged to an LTRC facility, LTRC bed supply once more has a stronger negative association with LOS among patients with longer LOS. The size of the coefficients is also much larger than that observed in the previous graph. Transforming the coefficients to be interpreted as elasticities, 10 per cent higher LTRC bed supply is associated with a 3.3 days lower -0.05 -0.13 -0.39 -0.58 -0.96 -1.38 -1.94 -5.02 -12.25 -18 -15 -12 -9 -6 -3 0 0.2 (2 days) 0.3 (3 days) 0.4 (4 days) 0.5 (6 days) 0.6 (7 days) 0.7 (9 days) 0.8 (13 days) 0.9 (22 days) 0.95 (35 days) Coefficient Length of stay quantile 108|Effects on Irish Hospital Care of the Supply of Care Inside and Outside the Hospital LOS for those in the 90th percentile (equating to 5.3 per cent LOS reduction) and a 6.8 days lower LOS for those in the 95th percentile (equating to 7 per cent LOS reduction). It is clear that the relationship between LTRC bed supply and LOS is much stronger for those ultimately discharged to an LTRC centre, especially for those with longer LOS, who are more likely to be affected by delayed discharges. FIGURE 7.4 EMERGENCY INPATIENT DISCHARGES DISCHARGED TO LONG-STAY FACILITY FOR PATIENTS AGED 65+, 2012–2015 Stroke, hip fracture, and Alzheimer’s/dementia patients Stroke and hip fracture patients may be more amenable to LTRC, while it is known that a clear correlate of LTRC use is cognitive impairment conditions, specifically Alzheimer’s disease and dementia. Table 7.4, Table 7.5 and Table 7.6 present results from the linear and negative binomial regressions examining the impact of LTRC bed supply on LOS for these patient groups. Due to smaller numbers, it was not possible to undertake UQRs for these patient groups. As shown in Table 7.4, we found no statistically significant relationship between LTRC bed and LOS for stroke patients, though the sign and scale of the coefficients are similar to those in the tables above. -1.62 -1.72 -2.32 -3.06 -5.02 -7.69 -14.43 -34.38 -70.95 -110 -90 -70 -50 -30 -10 10 0.2 (6 days) 0.3 (8 days) 0.4 (10 days) 0.5 (13 days) 0.6 (17 days) 0.7 (23 days) 0.8 (34 days) 0.9 (62 days) 0.95 (97 days) Coefficient Length of stay quantile TABLE 7.4 DETERMINANTS OF LENGTH OF STAY FOR EMERGENCY INPATIENT STROKE DISCHARGES AGED 65+, 2012–2015 65+ 75+ 85+ Ordinary least squares Negative binomial Ordinary least squares (Ln LOS) (Ln LOS) (LOS) (Ln LOS) (Ln LOS) (I) (II) (III) (IV) (V) Year trend -0.010 -0.011 -0.017 -0.008 -0.009 Married -0.140*** -0.139*** -0.168*** -0.131*** -0.076*** Medical card 0.038 0.039 0.033 0.026 -0.015 Weighted Charlson score 0.070*** 0.070*** 0.056*** 0.056*** 0.020 Female -0.023 -0.023 -0.038 -0.006 -0.056 Ln home care -0.234*** -0.201** -0.304*** -0.291*** -0.256* LTRC beds Ln LTRC beds -0.199 -0.275 -0.167 -0.095 LTRC beds quintiles Quintile 1 (lowest) Base Quintile 2 -0.004 Quintile 3 -0.001 Quintile 4 -0.055 Quintile 5 (highest) -0.056 Clusters (hospitals) 26 26 26 26 26 Observations 10,775 10,775 10,775 7,015 2,406 Adjusted R Squared 0.232 0.232 0.229 0.214 Average length of stay 24.22 days 24.22 days 24.22 days 26.68 days 29.43 days Notes: Other covariates in models: hospital-level standardised bed capacity, age and age squared, season, stroke severity. Sample: Patients with primary or secondary stroke diagnosis, admitted from home through an ED, not transferred to another hospital at discharge, alive at discharge. Clustered standard errors at hospital level. 110|Effects on Irish Hospital Care of the Supply of Care Inside and Outside the Hospital Table 7.5 presents results from linear and negative binomial regressions for hip fracture patients. A negative and statistically significant relationship between LTRC bed supply and LOS is observed. Within the 65 years and over group, 10 per cent higher LTRC supply is associated with a 2.5 per cent (0.51 days) lower LOS for hip fracture patients. When the supply variable is included in quintile form, it loses statistical significance as it did in the full-population model. The strength of the relationship between LTRC bed supply and hip fracture LOS does not increases with age. Does LTRC Reduce Public Hospital Inpatient LOS and Delayed Discharges?| 117 home care supply is associated with a 3.3–3.9 per cent reduction in average inpatient LOS. As another way of illustrating the scale of the effects, we can express the impact of LTRC beds per capita in terms of inpatient bed days. In these data, average LOS is 25.88 days, and 0.987 million inpatient bed days were used between 2012 and 2015. Reducing LOS by 3.3 per cent (0.1 days) equates to 8,141 fewer inpatient beds per annum. However, examining those with long LOS (delayed discharges), a larger substitutive effect of LTRC beds per capita is found. Results from UQR analyses show that for those in the 90th percentile of LOS, a 10 per cent increase in per capita LTRC bed supply is associated with a 0.5 days less spent in hospital. The relationship is stronger in the 95th percentile, where a 10 per cent increase in per capita LTRC bed supply is associated with a 1.2 days reduction in LOS. Returning again to those patients who ultimately are discharged to an LTRC centre, and are therefore inherently more amenable to LTRC supply, a stronger substitution effect is found for the longest stayers, and this effect is intensified by the long LOS for this group more generally. Results from UQR analyses show that for those in the 90th percentile of LOS, a 10 per cent increase in per capita LTRC bed supply is associated with 3.3 fewer days in hospital. The relationship is stronger in the 95th percentile, where a 10 per cent increase in per capita LTRC bed supply is associated with a 6.8 days reduction in LOS. These substitutive effects are found even after controlling for both inpatient bed supply (‘push’ factor) and home care supply (a potentially alternative ‘pull’ factor). The findings of a negative relationship between LTRC supply and average inpatient LOS is consistent with previous evidence from the NHS (Fernandez and Forder, 2008; Forder, 2009). However, while the elasticities for LTRC are larger than for home care, the differences in effect sizes are small. Furthermore, the results from the UQR analysis are similar to those found in the NHS for delayed discharges. Gaughan et al. (2015) found that that a 10 per cent increase in LTRC beds reduced delayed discharges (where need for social care is given as the reason for the discharge delay) by between 6 per and 9 per cent. In our results, we also find a clear association between LTRC bed supply per capita and reduced LOS for those with longer LOS (potentially delayed discharges). Results for stroke, hip fracture patients and Alzheimer’s/dementia patients – three groups of patients that are more likely to benefit from LTRC – do not show consistent negative substitution effects. However, for Alzheimer’s disease and dementia patients aged 85 years and over, a large negative relationship is seen between LOS and LTRC supply. The coefficients, while large, are often statistically insignificant at the most meaningful levels, which is largely a result of smaller sample size and clustering of standard errors. 118|Effects on Irish Hospital Care of the Supply of Care Inside and Outside the Hospital Examining the homogeneity of results across demographic groups (gender, marital status, public/private status and Charlson comorbidity illness score), little differences are observed across gender, marital status and private discharge status. However, similar to the results in home care, the substitution effects are slightly larger for sicker patients. Caution should be taken in interpreting the results on LTRC supply as causal estimates. Our LTRC bed measure includes a range of long-stay, short-stay and rehabilitation beds, all of which may not be amenable to the inpatients in our sample. In addition, our measure of LTRC supply is based upon a patient’s LHO; however, patients who ultimately end up in an LTRC centre may choose a centre outside of their LHO due to the perceived quality of the chosen centre, ability to access a bed quickly, or to be closer to family. In this context, it is uncertain whether the within-hospital differences we accounted for in the previous chapter also apply here. Additionally, we only include inpatients admitted from home. Therefore, all results should be interpreted as the impact of LTRC supply on the inpatient LOS for those not already residing in an LTRC centre. Notwithstanding these caveats, this is the first finding on the impact of LTRC supply on use of acute hospital services to date in Ireland, and in that sense marks an important contribution to our understanding of adds substitution between the two services. Furthermore, in an exploratory analysis we find that the availability of home care reduces the probability of patients who are admitted to hospital from home being discharged to LRTC; this is also new information on the interaction between services in Ireland, and worthy of further study. Concluding Discussion and Policy Recommendations|119 CHAPTER 8 Concluding discussion and policy recommendations 8.1 SCOPE OF CHAPTER In this chapter, we discuss findings from the project and highlight its key conclusions for policy, alongside other factors that may impact resource allocation decisions for health and social care services in Ireland. Section 8.2 describes the project, while Sections 8.3 and 8.4 provide an overview of its key findings across the two reports. Section 8.5 discusses these findings in the context of resource allocation in Ireland and draws on other key factors that may impact resource allocation decisions. Section 8.6 concludes. 8.2 INTRODUCTION This report is part of a broader project entitled, An inter-sectoral analysis by geographic area of the need for and the supply and utilisation of health services in Ireland. The project’s analyses were undertaken in the context of a healthcare system with significant capacity constraints within the acute hospital sector and current policy priorities to achieve a move towards providing more care, where appropriate, in the community and closer to home. In the initial stage of the project, Smith et al. (2019) provided evidence about the supply of non-acute care across counties in Ireland, to help identify whether adequate capacity exists to meet increased (or even existing) demand, should more care be transferred to non-acute services. The analysis undertaken in this report provides evidence on the way in which acute and non-acute sectors interact in the system, especially the substitution effects between inpatient care and care of older people services and home care and long-term residential care (LTRC). 8.3 GEOGRAPHIC PROFILE OF HEALTHCARE NEEDS AND NON-ACUTE HEALTHCARE SUPPLY IN IRELAND (SMITH ET AL., 2019) Smith et al. (2019) provided evidence on the geographic distribution of non-acute care across Ireland. That analyses relied on per capita supply estimates to allow for accurate comparisons across areas. It also examined whether supply differed when proxies for healthcare need and eligibility were accounted for. The breadth of the services examined was wide, with 10 non-acute services – which constitute the vast majority of non-acute care in Ireland – examined (Wren et al., 2017). These were: general practitioners (GPs); community nurses; physiotherapists; occupational therapists; speech and language therapists; podiatrists and chiropodists; counsellors and psychologists; social workers; LTRC beds and home 120|Effects on Irish Hospital Care of the Supply of Care Inside and Outside the Hospital care hours. Due to data limitations, for many services only public supply could be analysed. However, across GPs (who are private providers), physiotherapists and LTRC beds, private supply data were available. Large geographic inequalities were observed across all services, with the largest geographic inequalities in supply seen for podiatrists and chiropodists, driven largely by many areas having no non-acute podiatrist or chiropodist. Gini coefficients of greater than 0.1, often used as a rule of thumb as the measure of large inequalities, were estimated for all services apart from GPs, who have a Gini of 0.096. This means that for all services examined, a redistribution of over 10 per cent of supply from areas in the upper half of the distribution to areas in the lower half of the distribution would be required to achieve perfect equality of supply (van Doorslaer and Koolman, 2004). However, importantly in healthcare, analysis of relative supply should also take relative need into account. Therefore, a key feature of the analyses was the adjustment of supply for a range of eligibility and need factors. Eligibility was based on medical card and GP visit card rates, and need factors followed from those included in the Andersen’s Behavioural Model of Health Services Use. However, the adjustment factors had little impact and the geographic distribution of each non-acute service remains unequal. While Smith et al. (2019) provide clear evidence that the geographic distribution of non-acute care supply is unequal, there also appears to be notable consistency across the supply types in terms of the areas that have low levels of per capita supply relative to the national value. Figure 8.1 provides a summary overview of Smith et al. (2019) findings on the geographic variation in the per capita supply of non-acute and long-term care services. Counties with supply of a service 10 per cent lower than the national average are denoted by a red circle. Counties with supply of a service 10 per cent higher than the national average are denoted by a green circle. Where per capita supply of a service is within 10 per cent of the national average, this is denoted by an amber circle. Counties are sorted by the number of services, where supply is lower than the national average. The greater Dublin commuter belt and south east counties have lower relative supply of many non-acute primary and community care services than the national average. Kildare and Meath have lower relative supply (at least 10 per cent lower than the national average) for all non-acute community and primary care services. Wexford and Wicklow have lower relative supply (at least 10 per cent lower than the national average) for 7 of the eight non-acute community and primary care services examined. In contrast to the low relative supply on the east coast, three counties on the west coast – Galway, Sligo and Leitrim – as well as Cork, Westmeath and Tipperary South Concluding Discussion and Policy Recommendations|121 have higher relative supply for many services, and do not have lower relative supply for more than one of the services examined. In general in Dublin, the supply of services in Dublin North and Dublin South is similar to the national average. Some variation in supply is seen across some services. Dublin North has low relative supply of GPs, counsellors and psychologists and LTRC. Dublin South has low relative supply of counsellors and psychologists, and publicly funded home care hours. Smith et al. (2019) do not analyse the reasons for these patterns of supply but the findings of low supply in the commuter belt counties with rapidly growing population around Dublin is consistent with earlier studies and suggestive that health service planning does not respond adequately to population patterns (Brick et al., 2010; Layte et al., 2009). 122|Effects on Irish Hospital Care of the Supply of Care Inside and Outside the Hospital FIGURE 8.1 SUMMARY OF RELATIVE SERVICE SUPPLY INDICATORS IN IRELAND, 2014 County has supply at least 10 per cent higher than national average. County has supply at least 10 per cent lower than national average. County has supply approximately equal to the national average. Source: Smith et al. (2019). Notes: 1. The number of GPs is converted to estimated whole-time equivalents (WTEs), based on survey evidence on full- and part-time working practices of GPs in Ireland. See Chapter 3 for more details. 2. Publicly employed WTEs. See Chapter 3 for more details. 3. Publicly employed and privately employed WTE PTs. See Chapter 3 for more details. 4. LTRC beds in public and private LTRC centres (supply of beds per 1,000 population aged 65+). See Chapter 3 for more details. 5. Publicly funded home care hours from the home help and home care package schemes (supply of hours per population aged 65+). See Chapter 3 for more details. 8.4 ANALYSIS OF THE EFFECTS ON IRISH HOSPITAL CARE OF THE SUPPLY OF CARE INSIDE AND OUTSIDE THE HOSPITAL (WALSH ET AL., 2019) This report analyses supply in the acute hospital, home care and LTRC sectors, three sectors that provide a substantial amount of the country’s health and social care (Wren et al., 2017). It finds that access to and supply of these services differ across areas, and that variations are evident over time. The report models the impact of ‘push’ (inpatient bed supply) and ‘pull’ (home care and LTRC supply) GP1 PHN/CN2 PT3OT2SLT2P&C2 CO&PSY2 SW2LTRC4HCH5 Kildare 9 Meath 8 Wexford 8 Wicklow 8 Clare 7 Kilkenny 7 Waterford 7 Offaly 5 Carlow 5 Laois 4 Limerick 3 Mayo 3 Tipperary North 3 Dublin North 3 Kerry 3 Monaghan 3 Longford 3 Roscommon 3 Dublin South 2 Cavan 2 Louth 2 Donegal 2 Tipperary South 1 Westmeath 1 Leitrim 1 Sligo 1 Cork 0 Galway 0 Number of services >10 per cent below national average Long-Term Care Non-Acute Primary and Community Care Concluding Discussion and Policy Recommendations|123 factors on hospital care in Ireland, using measures of inpatient bed supply, home care supply per capita, and LTRC bed supply in the analyses. Chapter 3 shows that the acute public hospital system in Ireland has seen significant reconfiguration of services in recent years. There has been a consolidation of emergency department (ED) services in the public hospital system, with many hospitals having their ED services reduced from ‘24/7 365’ status (24 hours a day, 7 days a week). This has resulted in some local hospitals losing fulltime ED services. During the economic recession from 2008, there was a dramatic reduction in inpatient bed supply and staffing levels in the public hospital system. Inpatient bed supply fell by 13 per cent between 2007 and 2012, together with a similar reduction in doctor and nursing WTEs. This was followed by a slight increase in supply from 2013 to 2015. These changes in bed supply were driven by economic stress and related pressures on government finances rather than lower demand for health services. Table 8.1 highlights the main results from chapters 5–7 of this report. In relation to the impact of inpatient bed supply on emergency inpatient length of stay (LOS) – Chapter 5 – in recent years, inpatient bed supply decreases and subsequent increases follow a similar trajectory to inpatient LOS (reductions in supply associated with reduced LOS and vice versa). Using econometric modelling techniques that exploit differences in bed supply across hospitals and within hospitals over time, we find that a large proportion of the LOS changes observed in recent years are explained by changes in inpatient bed supply. These results imply that a 40 per cent of inpatient LOS reductions between 2010 and 2012 were a result of reduced numbers of inpatient beds. While Chapter 5 highlights the role of acute bed supply on inpatient LOS, Chapters 6 and 7 examine the impact of non-acute care for older people on LOS, using granular data on per capita publicly financed home care supply and per capita LTRC bed supply in public and private LTRC centres. However, the size of the substitution effect differs significantly across models, whether average LOS or delayed discharges is examined, and across groups of patients who differ by diagnosis. Chapter 6 illustrates the substitution effect between home care and inpatient LOS. Findings from this chapter show that when average LOS across all emergency inpatients admitted from home aged 65 years and over are examined, ceteris paribus, increased home care supply has a statistically significant negative impact on LOS. We express results in terms of a 10 per cent increase in home care supply per capita (similar to the increased observed in Ireland between 2012 and 2015), equating to 1.5 million additional home care hours in 2015. This 10 per cent increase implies an approximate 1 per cent reduction in average inpatient LOS for inpatients aged 65 years and over. Illustrated in terms of inpatient bed days, 1.5 124|Effects on Irish Hospital Care of the Supply of Care Inside and Outside the Hospital million additional home care hours are associated with a reduction of 14,700 inpatient bed days in 2015. If results from the Dublin North analyses are imposed on the Irish sample, owing to the substantial increase in home care supply in Dublin North between 2012 and 2015, ceteris paribus, a 10 per cent increase in home care supply per capita is associated with an approximate 2.7 per cent reduction in average LOS, and 1.5 million additional home care hours implies a reduction of 40,000 inpatient bed days per annum and freeing up 110 beds daily. 46 Examining the effect of home care supply on those with long LOS (delayed discharges), results from unconditional quantile regression (UQR) analyses show that for those in the 90th percentile of LOS (those with LOS of 21 days or more) a 10 per cent increase in per capita home care supply is associated with a 0.3 days less spent in hospital. Similarly, if results from the Dublin North UQR analysis were imposed countrywide, then even larger reductions, or 2 days, would be estimated. Chapter 7 illustrates the substitution effect between LTRC bed supply and inpatient LOS. Findings from this chapter show that when average LOS across all emergency inpatients admitted from home who are aged 65 years and over is examined, ceteris paribus, increased LTRC bed supply is found to have a statistically significant negative impact on LOS. We express results in terms of a 10 per cent increase in LTRC bed supply per capita, which equates to 2,965 LTRC beds in 2015. This 10 per cent increase in LTRC bed supply per capita is associated with an approximate 1.3– 2.2 per cent reduction in average inpatient LOS. Illustrated in terms of inpatient bed days, an addition of 2,965 beds in LTRC centres is associated with a reduction of 19,000 inpatient bed days per annum using the upper bound elasticity, or 53 available inpatient beds daily. Examining those emergency inpatients aged 65 years and over ultimately discharged to a long-stay facility, a 10 per cent increase in LTRC care supply per capita is associated with an approximate 3.3–3.9 per cent reduction in average inpatient LOS. Illustrated in terms of inpatient bed days, the addition of 2,965 beds in LTRC centres is associated with a reduction of 9,720 inpatient bed days per annum using the upper bound elasticity, for those patients who end up in an LTRC centre. In 2015, emergency inpatients admitted from home and discharged to an LTRC centre used approximately 250,000 inpatient bed days. Therefore, the 10 per increase in LTRC be supply would reduce this amount by 4 per cent for this group of inpatients. Examining the effect of LTRC bed supply on those with long LOS (delayed discharges), results from UQR analyses show that for those in the 90th percentile of LOS (those with LOS of 22 days or more), a 10 per cent increase in per capita home care supply is associated with 0.5 days less spent in hospital. Furthermore, for those ultimately discharged to a long-stay facility, results from UQR analyses show that for those in the 90th percentile of LOS (those 46 In 2015, there were 1.44 million emergency inpatient bed days in Irish public hospitals used by patients aged 65 years and over. Concluding Discussion and Policy Recommendations|125 with LOS of 62 days or more), a 10 per cent increase in per capita home care supply is associated with 3.3 fewer days in hospital. TABLE 8.1 EFFECTS OF HOME CARE AND LTRC SUPPLY ON EMERGENCY INPATIENT DISCHARGES AGED 65+ IN RELATION TO LENGTH OF STAY AND DELAYED DISCHARGES Home care hours supply LTRC bed supply Average LOS 10%Δ implies a 1%-1.7%Δ in average LOS. 1.5 million additional home care hours are associated with 14,700 fewer inpatient bed days per annum. *** Dublin North: 10%Δ implies a 2.7%Δ in average LOS. 1.5 million additional home care hours are associated with 40,000 fewer inpatient bed days per annum extrapolated nationally. *** Stroke: 10%Δ implies a 2.7%Δ in average LOS. *** Hip fracture: 10%Δ implies a 1.6%Δ in average LOS. 10%Δ implies a 1.3%-2.2%Δ in average LOS. 2,965 additional LTRC beds are associated with 19,000 fewer inpatient bed days per annum. *** LTRC discharges: 10%Δ implies a 3.3%– 3.9%Δ in average LOS. 2,965 additional LTRC beds are associated with 9,720 fewer inpatient bed days per annum. *** Hip fracture: 10%Δ implies a 2.5%Δ in average LOS. *** Alzheimer’s/dementia: 10%Δ implies a 5%Δ in average LOS for those aged 85+. 90th LOS percentile (delayed discharges) 10%Δ implies a 0.3 days LOS reduction in the 90th percentile, LOS ≥ 21 days. *** Dublin North: 10%Δ implies a 2 days LOS reduction in the 90th percentile, LOS ≥ 29 days. 10%Δ implies a 0.5 days LOS reduction in the 90th percentile, LOS ≥ 22 days. *** LTRC discharges: 10%Δ implies a 3.3 days LOS reduction in the 90th percentile, LOS ≥ 62 days. Notes: SD = standard deviation; Δ = change. The results in Chapter 5 on the clear, but often ignored, link between beds and hospitalisation use have broad implications for policymakers in both Ireland and internationally. From an Irish perspective, it is clear that there was, and is, an inadequate inpatient bed supply in public hospitals in Ireland. This can be assessed by a number of outcomes, including the long waiting times for elective care (Siciliani et al., 2014; Wren et al., 2017) and the fact that Ireland has the highest bed occupancy rate in the OECD (OECD, 2018). The Government now acknowledges that a greater number of beds are required (Government of Ireland, 2018a; Keegan et al., 2018a). However, the recent reduction in bed capacity was also followed by sharp reductions in staffing levels. This makes the experience in Ireland different to other countries, such as Denmark (Christiansen and Vrangbæk, 126|Effects on Irish Hospital Care of the Supply of Care Inside and Outside the Hospital 2017), that implemented bed capacity reductions as a result of system change, and where increases in staffing often accompanied bed reductions. There are examples where, at the onset of an economic crisis, governments initially enact a counter-cyclical policy by expanding healthcare expenditure in response to recession (Keegan et al., 2013). In Ireland, initial counter-cyclical policies were subsequently replaced by severe cuts to expenditure. Previous analyses of the Irish healthcare system during the economic crisis have argued that efficiencies were seen in the public hospital system at the beginning of the recessionary period (2008–2012), with hospitals ‘doing more with less’, ‘more’ being reflected in inpatient and day patient activity and ‘less’ in reduced budgets (Burke et al., 2014). However, if ‘doing more’ leads to unsafe occupancy rates, this is of concern, and the authors suggest that the continued lack of staffing resources and capacity resulted in lower activity and increased waiting lists post-2012 (Burke et al., 2014). From an international perspective, the findings in Chapter 5 also highlight why one should be cautious when using LOS as a measure of efficiency. LOS can be a useful measure of efficiency when comparing hospitals and health systems. However, interpretations of LOS reductions are clearly dependent upon the context in which they are observed. In periods where expenditure cuts or system changes occur, lower LOS may reflect efficiency gains, but may also simply be indicative of the lack of resources or available bed capacity, as opposed to reduced demand for care. Other countries such as the UK and Canada also have occupancy rates higher than the 85 per cent recommended threshold. In the English NHS for instance, cuts to acute bed numbers in recent years have increased occupancy rates from 87 per cent in 2010–2011 to over 90 per cent in 2016–2017 (The Kings Fund, 2017). Therefore, in those countries with high occupancy rates, using LOS as a measure of efficiency may require caution, especially as shorter inpatient LOS often results in greater readmission rates (Ambugo and Hagen, 2019; Carey, 2015; Martin et al., 2016) and high occupancy has been shown to cause large negative effects in patient outcomes (Boden et al., 2015; Madsen et al., 2014; Schilling et al., 2010). As explained in this report, the lack of an individual health identifier (IHI) makes it difficult to examine mortality and readmissions as outcomes in Ireland. However, given that Ireland has the highest inpatient bed occupancy rates in the EU, these latter findings in the broader literature are worrying, and worthy of further research. Private hospitals provide approximately 15 per cent of all inpatient care in Ireland (Wren et al., 2017), with the majority of this care being elective or less complex emergency care. Due to the lack of information on private hospital care, it is difficult to ascertain the impact of bed capacity changes on elective LOS in public hospitals. This suggests that in order to understand the impact of non-acute care supply on elective care in public hospitals, information on private hospitals would also be required.