Physical activity as a protective factor for dementia and Alzheimer’s disease : systematic review, meta-analysis and quality assessment of cohort and case-control studies

This is a self-archived version of an original article. This version may differ from the original in pagination and typographic details. Author(s): Title: Year: Version: Copyright: Rights: Rights url: Please cite the original version: CC BY 4.0 https://creativecommons.org/licenses/by/4.0/ Physical activity as a protective factor for dementia and Alzheimer’s disease : systematic review, meta-analysis and quality assessment of cohort and case-control studies © Authors 2022 Published version Iso-Markku, Paula; Kujala, Urho M; Knittle, Keegan; Polet, Juho; Vuoksimaa, Eero; Waller, Katja Iso-Markku, P., Kujala, U. M., Knittle, K., Polet, J., Vuoksimaa, E., & Waller, K. (2022). Physical activity as a protective factor for dementia and Alzheimer’s disease : systematic review, metaanalysis and quality assessment of cohort and case-control studies. British Journal of Sports Medicine, 56(12), 701-709. https://doi.org/10.1136/bjsports-2021-104981 2022 1 IsoMarkkuP, etal. Br J Sports Med 2022;0:1–11. doi:10.1136/bjsports-2021-104981 Physical activity as a protective factor for dementia and Alzheimer’s disease: systematic review, metaanalysis and quality assessment of cohort and case– controlstudies Paula IsoMarkku ,1,2 Urho M Kujala ,3 Keegan Knittle ,3 Juho Polet ,3 Eero Vuoksimaa ,1 Katja Waller 3 Review To cite: IsoMarkkuP, KujalaUM, KnittleK, etal. Br J Sports Med Epub ahead of print: [please include Day Month Year]. doi:10.1136/ bjsports-2021-104981 ►Additional supplemental material is published online only. To view, please visit the journal online (http:// dx. doi. org/ 10. 1136/ bjsports2021104981). 1Institute for Molecular Medicine Finland (FIMM), HiLIFE, University of Helsinki, Helsinki, Finland 2HUS Diagnostic Center, Clinical Physiology and Nuclear Medicine, University of Helsinki and Helsinki University Hospital, Helsinki, Finland 3Faculty of Sport and Health Sciences, University of Jyväskylä, Jyväskylä, Finland Correspondence to Dr Paula IsoMarkku, Institute for Molecular Medicine Finland (FIMM), HiLIFE, University of Helsinki, Helsinki 00014, Finland; paula. isomarkku@ helsinki. fi Accepted 20 February 2022 © Author(s) (or their employer(s)) 2022. Reuse permitted under CC BY. Published by BMJ. ABSTRACT Objective Physical activity (PA) is associated with a decreased incidence of dementia, but much of the evidence comes from short followups prone to reverse causation. This metaanalysis investigates the effect of study length on the association. Design A systematic review and metaanalysis. Pooled effect sizes, dose–response analysis and funnel plots were used to synthesise the results. Data sources CINAHL (last search 19 October 2021), PsycInfo, Scopus, PubMed, Web of Science (21 October 2021) and SPORTDiscus (26 October 2021). Eligibility criteria Studies of adults with a prospective followup of at least 1 year, a valid cognitive measure or cohort in midlife at baseline and an estimate of the association between baseline PA and followup allcause dementia, Alzheimer’s disease or vascular dementia were included (n=58). Results PA was associated with a decreased risk of allcause dementia (pooled relative risk 0.80, 95% CI 0.77 to 0.84, n=257 983), Alzheimer’s disease (0.86, 95% CI 0.80 to 0.93, n=128 261) and vascular dementia (0.79, 95% CI 0.66 to 0.95, n=33 870), even in longer followups (≥20 years) for allcause dementia and Alzheimer’s disease. Neither baseline age, followup length nor study quality significantly moderated the associations. Dose–response metaanalyses revealed significant linear, spline and quadratic trends within estimates for allcause dementia incidence, but only a significant spline trend for Alzheimer’s disease. Funnel plots showed possible publication bias for allcause dementia and Alzheimer’s disease. Conclusion PA was associated with lower incidence of allcause dementia and Alzheimer’s disease, even in longer followups, supporting PA as a modifiable protective lifestyle factor, even after reducing the effects of reverse causation. INTRODUCTION Worldwide, around 50 million people suffer from dementia. This number is projected to triple by 2050, with twothirds of these people living in lowincome and middleincome countries.1 The economic burden of dementia is estimated to be as high as US$818 billion annually,2 thus making dementia prevention a health priority in ageing societies. Physical inactivity is one of 12 potentially modifiable risk factors suggested to account for about 40% of oldage dementias.3 Several pathways through which physical activity (PA) may prevent dementias have been proposed: decreased production of β-amyloid, increased removal of β-amyloid, improved brain vasculature and blood flow, and antioxidative and inflammatory processes in the brain,4 as well as indirect pathways through improvements in sleep, mood and other cardiovascular risk factors. Metaanalyses indicate an association between PA and decreased risk of allcause dementia,5–7 including in a dose–response manner.8 However, many previous metaanalyses lack rigorous quality assessments,7 and the association between PA and dementia appears absent when PA is measured before the age of 657 8 or in followups longer than 10 years.6 9 As the Alzheimer’s disease process starts decades before diagnosis,10 even studies with 10year followups are likely to include participants with preclinical Alzheimer’s disease. Thus, studies assessing midlife PA and oldage dementia diagnoses with a followup of at least 20 years are needed to confirm whether PA is a modifiable protective lifestyle factor of dementia. This systematic review and metaanalysis examines if midlife PA is a protective factor of allcause dementia, Alzheimer’s disease and vascular dementia. We examine separately studies with followups longer than 20 years, highquality studies and studies with younger cohorts to reduce the effect of reverse causality. Because the association of PA and dementia might potentially be modified by the apolipoprotein E (ApoE) genotype,11–13 education,14 PA type,12 15 sample size16 and funding source,17 we additionally examine these factors as possible moderators of PA– dementia associations. METHODS This systematic review and metaanalysis is reported in line with the Preferred Reporting Items for Systematic Reviews and MetaAnalyses statement18 (see online supplemental material part 1) and was registered on PROSPERO (CRD42018083236). However, due to insufficient data, some registered analyses were not conducted and the original registered plan was adapted (online supplemental material part 1). Kirjasto/Kausijulkaisut. Protected by copyright. on March 21, 2022 at Jyvaskylan Yliopostohttp://bjsm.bmj.com/Br J Sports Med: first published as 10.1136/bjsports-2021-104981 on 17 March 2022. Downloaded from Kirjasto/Kausijulkaisut. Protected by copyright. on March 21, 2022 at Jyvaskylan Yliopostohttp://bjsm.bmj.com/Br J Sports Med: first published as 10.1136/bjsports-2021-104981 on 17 March 2022. Downloaded from Kirjasto/Kausijulkaisut. Protected by copyright. on March 21, 2022 at Jyvaskylan Yliopostohttp://bjsm.bmj.com/Br J Sports Med: first published as 10.1136/bjsports-2021-104981 on 17 March 2022. Downloaded from Kirjasto/Kausijulkaisut. Protected by copyright. on March 21, 2022 at Jyvaskylan Yliopostohttp://bjsm.bmj.com/Br J Sports Med: first published as 10.1136/bjsports-2021-104981 on 17 March 2022. Downloaded from Kirjasto/Kausijulkaisut. Protected by copyright. on March 21, 2022 at Jyvaskylan Yliopostohttp://bjsm.bmj.com/Br J Sports Med: first published as 10.1136/bjsports-2021-104981 on 17 March 2022. Downloaded from Kirjasto/Kausijulkaisut. Protected by copyright. on March 21, 2022 at Jyvaskylan Yliopostohttp://bjsm.bmj.com/Br J Sports Med: first published as 10.1136/bjsports-2021-104981 on 17 March 2022. Downloaded from Kirjasto/Kausijulkaisut. Protected by copyright. on March 21, 2022 at Jyvaskylan Yliopostohttp://bjsm.bmj.com/Br J Sports Med: first published as 10.1136/bjsports-2021-104981 on 17 March 2022. Downloaded from Kirjasto/Kausijulkaisut. Protected by copyright. on March 21, 2022 at Jyvaskylan Yliopostohttp://bjsm.bmj.com/Br J Sports Med: first published as 10.1136/bjsports-2021-104981 on 17 March 2022. Downloaded from Kirjasto/Kausijulkaisut. Protected by copyright. on March 21, 2022 at Jyvaskylan Yliopostohttp://bjsm.bmj.com/Br J Sports Med: first published as 10.1136/bjsports-2021-104981 on 17 March 2022. Downloaded from Kirjasto/Kausijulkaisut. Protected by copyright. on March 21, 2022 at Jyvaskylan Yliopostohttp://bjsm.bmj.com/Br J Sports Med: first published as 10.1136/bjsports-2021-104981 on 17 March 2022. Downloaded from Kirjasto/Kausijulkaisut. Protected by copyright. on March 21, 2022 at Jyvaskylan Yliopostohttp://bjsm.bmj.com/Br J Sports Med: first published as 10.1136/bjsports-2021-104981 on 17 March 2022. Downloaded from Kirjasto/Kausijulkaisut. Protected by copyright. on March 21, 2022 at Jyvaskylan Yliopostohttp://bjsm.bmj.com/Br J Sports Med: first published as 10.1136/bjsports-2021-104981 on 17 March 2022. Downloaded from Kirjasto/Kausijulkaisut. Protected by copyright. on March 21, 2022 at Jyvaskylan Yliopostohttp://bjsm.bmj.com/Br J Sports Med: first published as 10.1136/bjsports-2021-104981 on 17 March 2022. Downloaded from Kirjasto/Kausijulkaisut. Protected by copyright. on March 21, 2022 at Jyvaskylan Yliopostohttp://bjsm.bmj.com/Br J Sports Med: first published as 10.1136/bjsports-2021-104981 on 17 March 2022. Downloaded from Kirjasto/Kausijulkaisut. Protected by copyright. on March 21, 2022 at Jyvaskylan Yliopostohttp://bjsm.bmj.com/Br J Sports Med: first published as 10.1136/bjsports-2021-104981 on 17 March 2022. Downloaded from Kirjasto/Kausijulkaisut. Protected by copyright. on March 21, 2022 at Jyvaskylan Yliopostohttp://bjsm.bmj.com/Br J Sports Med: first published as 10.1136/bjsports-2021-104981 on 17 March 2022. Downloaded from Kirjasto/Kausijulkaisut. Protected by copyright. on March 21, 2022 at Jyvaskylan Yliopostohttp://bjsm.bmj.com/Br J Sports Med: first published as 10.1136/bjsports-2021-104981 on 17 March 2022. Downloaded from Kirjasto/Kausijulkaisut. Protected by copyright. on March 21, 2022 at Jyvaskylan Yliopostohttp://bjsm.bmj.com/Br J Sports Med: first published as 10.1136/bjsports-2021-104981 on 17 March 2022. Downloaded from Kirjasto/Kausijulkaisut. Protected by copyright. on March 21, 2022 at Jyvaskylan Yliopostohttp://bjsm.bmj.com/Br J Sports Med: first published as 10.1136/bjsports-2021-104981 on 17 March 2022. Downloaded from Kirjasto/Kausijulkaisut. Protected by copyright. on March 21, 2022 at Jyvaskylan Yliopostohttp://bjsm.bmj.com/Br J Sports Med: first published as 10.1136/bjsports-2021-104981 on 17 March 2022. Downloaded from Kirjasto/Kausijulkaisut. Protected by copyright. on March 21, 2022 at Jyvaskylan Yliopostohttp://bjsm.bmj.com/Br J Sports Med: first published as 10.1136/bjsports-2021-104981 on 17 March 2022. Downloaded from Kirjasto/Kausijulkaisut. Protected by copyright. on March 21, 2022 at Jyvaskylan Yliopostohttp://bjsm.bmj.com/Br J Sports Med: first published as 10.1136/bjsports-2021-104981 on 17 March 2022. Downloaded from Kirjasto/Kausijulkaisut. Protected by copyright. on March 21, 2022 at Jyvaskylan Yliopostohttp://bjsm.bmj.com/Br J Sports Med: first published as 10.1136/bjsports-2021-104981 on 17 March 2022. Downloaded from Kirjasto/Kausijulkaisut. Protected by copyright. on March 21, 2022 at Jyvaskylan Yliopostohttp://bjsm.bmj.com/Br J Sports Med: first published as 10.1136/bjsports-2021-104981 on 17 March 2022. Downloaded from Kirjasto/Kausijulkaisut. Protected by copyright. on March 21, 2022 at Jyvaskylan Yliopostohttp://bjsm.bmj.com/Br J Sports Med: first published as 10.1136/bjsports-2021-104981 on 17 March 2022. Downloaded from Kirjasto/Kausijulkaisut. Protected by copyright. on March 21, 2022 at Jyvaskylan Yliopostohttp://bjsm.bmj.com/Br J Sports Med: first published as 10.1136/bjsports-2021-104981 on 17 March 2022. Downloaded from Kirjasto/Kausijulkaisut. Protected by copyright. on March 21, 2022 at Jyvaskylan Yliopostohttp://bjsm.bmj.com/Br J Sports Med: first published as 10.1136/bjsports-2021-104981 on 17 March 2022. Downloaded from Kirjasto/Kausijulkaisut. Protected by copyright. on March 21, 2022 at Jyvaskylan Yliopostohttp://bjsm.bmj.com/Br J Sports Med: first published as 10.1136/bjsports-2021-104981 on 17 March 2022. Downloaded from Kirjasto/Kausijulkaisut. Protected by copyright. on March 21, 2022 at Jyvaskylan Yliopostohttp://bjsm.bmj.com/Br J Sports Med: first published as 10.1136/bjsports-2021-104981 on 17 March 2022. Downloaded from Kirjasto/Kausijulkaisut. Protected by copyright. on March 21, 2022 at Jyvaskylan Yliopostohttp://bjsm.bmj.com/Br J Sports Med: first published as 10.1136/bjsports-2021-104981 on 17 March 2022. Downloaded from Kirjasto/Kausijulkaisut. Protected by copyright. on March 21, 2022 at Jyvaskylan Yliopostohttp://bjsm.bmj.com/Br J Sports Med: first published as 10.1136/bjsports-2021-104981 on 17 March 2022. Downloaded from Kirjasto/Kausijulkaisut. Protected by copyright. on March 21, 2022 at Jyvaskylan Yliopostohttp://bjsm.bmj.com/Br J Sports Med: first published as 10.1136/bjsports-2021-104981 on 17 March 2022. Downloaded from Kirjasto/Kausijulkaisut. Protected by copyright. on March 21, 2022 at Jyvaskylan Yliopostohttp://bjsm.bmj.com/Br J Sports Med: first published as 10.1136/bjsports-2021-104981 on 17 March 2022. Downloaded from Kirjasto/Kausijulkaisut. Protected by copyright. on March 21, 2022 at Jyvaskylan Yliopostohttp://bjsm.bmj.com/Br J Sports Med: first published as 10.1136/bjsports-2021-104981 on 17 March 2022. Downloaded from Kirjasto/Kausijulkaisut. Protected by copyright. on March 21, 2022 at Jyvaskylan Yliopostohttp://bjsm.bmj.com/Br J Sports Med: first published as 10.1136/bjsports-2021-104981 on 17 March 2022. Downloaded from Kirjasto/Kausijulkaisut. Protected by copyright. on March 21, 2022 at Jyvaskylan Yliopostohttp://bjsm.bmj.com/Br J Sports Med: first published as 10.1136/bjsports-2021-104981 on 17 March 2022. Downloaded from Kirjasto/Kausijulkaisut. Protected by copyright. on March 21, 2022 at Jyvaskylan Yliopostohttp://bjsm.bmj.com/Br J Sports Med: first published as 10.1136/bjsports-2021-104981 on 17 March 2022. Downloaded from Kirjasto/Kausijulkaisut. Protected by copyright. on March 21, 2022 at Jyvaskylan Yliopostohttp://bjsm.bmj.com/Br J Sports Med: first published as 10.1136/bjsports-2021-104981 on 17 March 2022. Downloaded from Kirjasto/Kausijulkaisut. Protected by copyright. on March 21, 2022 at Jyvaskylan Yliopostohttp://bjsm.bmj.com/Br J Sports Med: first published as 10.1136/bjsports-2021-104981 on 17 March 2022. Downloaded from Kirjasto/Kausijulkaisut. Protected by copyright. on March 21, 2022 at Jyvaskylan Yliopostohttp://bjsm.bmj.com/Br J Sports Med: first published as 10.1136/bjsports-2021-104981 on 17 March 2022. Downloaded from Kirjasto/Kausijulkaisut. Protected by copyright. on March 21, 2022 at Jyvaskylan Yliopostohttp://bjsm.bmj.com/Br J Sports Med: first published as 10.1136/bjsports-2021-104981 on 17 March 2022. Downloaded from Kirjasto/Kausijulkaisut. Protected by copyright. on March 21, 2022 at Jyvaskylan Yliopostohttp://bjsm.bmj.com/Br J Sports Med: first published as 10.1136/bjsports-2021-104981 on 17 March 2022. Downloaded from Kirjasto/Kausijulkaisut. Protected by copyright. on March 21, 2022 at Jyvaskylan Yliopostohttp://bjsm.bmj.com/Br J Sports Med: first published as 10.1136/bjsports-2021-104981 on 17 March 2022. Downloaded from Kirjasto/Kausijulkaisut. Protected by copyright. on March 21, 2022 at Jyvaskylan Yliopostohttp://bjsm.bmj.com/Br J Sports Med: first published as 10.1136/bjsports-2021-104981 on 17 March 2022. Downloaded from Kirjasto/Kausijulkaisut. Protected by copyright. on March 21, 2022 at Jyvaskylan Yliopostohttp://bjsm.bmj.com/Br J Sports Med: first published as 10.1136/bjsports-2021-104981 on 17 March 2022. Downloaded from Kirjasto/Kausijulkaisut. Protected by copyright. on March 21, 2022 at Jyvaskylan Yliopostohttp://bjsm.bmj.com/Br J Sports Med: first published as 10.1136/bjsports-2021-104981 on 17 March 2022. Downloaded from 2IsoMarkkuP, etal. Br J Sports Med 2022;0:1–11. doi:10.1136/bjsports-2021-104981 Review Eligibility criteria Types of studies We included prospective cohort studies and case–control studies with a baseline measure of PA and a followup measure of allcause dementia, Alzheimer’s disease or vascular dementia. Only studies with followups longer than 1 year were included. Types of participants Participants were adults (≥20 years of age at baseline). We excluded studies where participants had some specific disease at baseline or where the cohort had established dementia or mild cognitive impairment at baseline. For populations that were older than midlife (defined as mean or median age <55 years and maximum age <65 years or mean age plus 1 SD <60 years), a valid measure of baseline cognition was required to be reported. This was done to minimise the possibility of including cohorts with a prodromal state of dementia and to account for the long preclinical period of Alzheimer’s disease10 and the typical age of dementia onset.19 Types of exposure We included studies assessing PA with objective measures or questionnaires. We excluded studies examining single bouts of PA, retrospectively reported PA, fitness levels or PA levels measured extending over the followup period. Types of outcomes Studies needed to report the association between PA and allcause dementia, Alzheimer’s disease or vascular dementia. We included studies that diagnosed dementia based on valid measures of cognition or register data, but excluded studies that based dementia diagnosis on cause of death data in more than 50% of the participants. Types of reports Fulltext reports in English were included. The decision rules that supplement these inclusion and exclusion criteria are described in online supplemental material part 1. Search strategy We conducted a systematic literature search in six electronic databases (PubMed, CINAHL, Scopus, PsycInfo, SPORTDiscus and Web of Science). Two reviewers conducted searches in all six databases, with the last search undertaken on 26 October 2021. The keywords of the original search included physical activity, physically active, sport, athletics, athlete, running, walking, physical training, dementia, Alzheimer’s disease, Alzheimer’s, cognition, cognitive, executive function, TELE (telephone assessment of dementia), TICS (Telephone Interview of Cognitive Status), MMSE (MiniMental State Examination), 3MS (the Modified MiniMental State Examination), memory, processing speed, verbal fluency, semantic fluency, reasoning, delayed recall, prospective, longitudinal, followup, follow up, observational and cohort. In addition to the search results, individual studies known to the authors were added to the metaanalysis. Further details and example searches are described in the online supplemental material part 1. Study selection Inclusion was based on the assessments of two independent reviewers (PIM+KW/JP/KK). Disagreements were discussed, and if consensus was not reached, a third independent researcher made the inclusion decision (UMK). Study screening was done in two phases: clearly irrelevant studies were excluded in the title and abstract phase, and thereafter, fulltext manuscripts were reviewed. In cases where multiple studies reported similar outcome data from the same cohort, we only included the study with the best quality score, longest followup or largest sample size (in this order). Two studies that we excluded from the main metaanalyses (due to other reports from the same cohort being of a higher quality) were however included in the ApoE ɛ4 interaction analysis, as the studies included in the main metaanalyses from these same cohorts did not present any ApoE ɛ4 interaction analyses.20 21 Quality assessment We developed a quality assessment tool specifically for this systematic review and metaanalysis to provide high transparency of the assessment and to account for the precise characteristics of the addressed study questions (see online supplemental material part 1). The new quality assessment tool assesses and scores the representativeness of the exposed cohort, PA assessment methods, demonstration that dementia was not present at start of study, methods used to control for confounders, outcome assessment methods, length of followup and loss to followup. We used three existing quality assessment tools to inform the development of our quality assessment tool: the NewcastleOttawa Quality Assessment Form for Cohort Studies,22 the performance bias estimator by Shiri and FalahHassani23 and the quality assessment tool for quantitative studies from the Effective Public Health Practice Project Quality Assessment.24 Two researchers reviewed the studies with the quality assessment tool independently (PIM+KW/JP). Disagreements were resolved with discussion. If the study cited other papers, at maximum three papers were sought for the required information. We used a quality scoring system with three categories based on the assumption that studies of high quality have less possibility of reverse causation, the study cohort is not selected and the measurement of both dementia and PA is valid (good quality: ≥2.5+1+≥2.5 stars, moderate quality: ≥2+≥0.5+≥2 stars, poor quality: not reaching good or moderate quality). Data extraction The following outcomes and moderator data were extracted from the included studies: rates of allcause dementia, Alzheimer’s disease and vascular dementia incidence; PA levels; estimates of the associations between PA levels and allcause dementia, Alzheimer’s disease or vascular dementia; length of followup; sample age and gender makeup; sample size; country of origin; publication year; study design (including a twin study or not); workrelated or leisuretime PA; confounders (age, cognition at baseline, chronic diseases, education, gender, vascular risk factors, ApoE ɛ4); followup and participation rate; gender interaction; stratification of results according to gender; ApoE ɛ4 interaction results; results stratified according to ApoE ɛ4 allele; number of adjusted confounders; study quality and funding (online supplemental material part 2). Two reviewers extracted the estimates of association (PIM+KW/JP) and followup length (PIM+KW). The likeness of the extractions was compared and disagreements were resolved by discussion. The estimates with the best quality assessment scores and the most extensive adjustments were included. For example, if baseline cognition was only measured and controlled for in one subgroup of the study sample, then data for that subgroup were extracted instead of the uncontrolled data from the full sample. Kirjasto/Kausijulkaisut. Protected by copyright. on March 21, 2022 at Jyvaskylan Yliopostohttp://bjsm.bmj.com/Br J Sports Med: first published as 10.1136/bjsports-2021-104981 on 17 March 2022. Downloaded from 3 IsoMarkkuP, etal. Br J Sports Med 2022;0:1–11. doi:10.1136/bjsports-2021-104981 Review Studies using the WHO PA recommendation25 as the category cutoff were also preferred if many estimates were presented. The data extraction of moderators other than followup length was done by one reviewer (PIM). Two researchers (PIM and KW) independently assessed whether the PA categories and reference categories in each study met the WHO PA recommendation.25 Disagreements were discussed until consensus was reached. Patient involvement This metaanalysis combines data from preexisting data sets. No patients were involved in study design, planning the search strategies, planning the quality assessment or sensitivity analyses, implementation of the study, interpretation of the results or writing up the results. Statistical analyses Summary statistics were relative risks (RRs) with 95% CIs. For studies that did not report RR data, ORs or HRs were converted into RRs. OR data were converted to RRs using the formula RR=OR/(1−p0+p0*OR), when the outcome occurred in less than 10% of the sample, with p0=outcome incidence in the whole study population.26 When the outcome was common (>10%), we used the square root transformation of OR as recommended by VanderWeele.27 We transformed HRs into RRs using the following formula: RR = (1 −e∧(HRxln(1 −r)))/r , where r is the incidence rate of dementia for the reference group.28 A separate RR was calculated for each higher PA category reported in the included study by comparing each higher PA category to the lowest PA level in the study (eg, an inactive or reference category). For the main metaanalyses, we pooled all estimates of the relationship between PA and allcause dementia, Alzheimer’s disease and vascular dementia, combining categorical and continuous measures of PA. We used a randomeffects model with inversevariance as the weighting method and estimated the statistical heterogeneity with DerSimonianLaird method (indexed with the I2 value). We conducted sensitivity analyses to examine the impacts of removing the study with largest sample size and highest weight on the overall result. An additional analysis examined this relationship within highquality studies that had measured PA in midlife and had a followup longer than 20 years. Metaregressions and comparative subgroup analyses examined the effects of baseline age, followup length and meeting the WHO PA recommendation on the association of PA and allcause dementia, Alzheimer’s disease and vascular dementia. Studies for which the reference category exceeded the WHO guidelines were excluded from the analysis of meeting the PA recommendations. Next, we performed planned sensitivity analyses to examine the effects of sample size, PA type (leisure time or both leisure time and work related) and other covariates on the relationships of PA and allcause dementia, Alzheimer’s disease and vascular dementia. Additionally, we examined the effect of funding source (no commercial funding source vs at least one commercial funding source) on the associations. There were too few twin studies and studies addressing gender effects to conduct the prespecified sensitivity analyses of these issues. Finally, we examined only the highest PA level compared with the lowest PA level as has been done in earlier metaanalyses.5 6 29 Dose–response metaanalyses were performed to explore linear, quadratic and restricted cubic spline trends between PA levels and RRs of dementia onset. These were conducted in R with the ‘dosresmeta’ package.30 A full description of the dose– response methods is available in online supplemental material part 3. An additional preplanned sensitivity analysis was performed to examine if the presence of ApoE ε4 allele moderates the associations between PA and allcause dementia, Alzheimer’s disease and vascular dementia. Pooled estimates for the association of PA and allcause dementia, Alzheimer’s disease or vascular dementia were calculated separately for ApoE ε4 carriers and noncarriers, and a significance test compared the results across the two subgroups.31 Funnel plots were used to examine the potential publication bias. Primary analyses were conducted in Stata V.16.0 (StataCorp). RESULTS Database searches identified 16 324 articles, of which 15 658 were excluded based on title and abstract screening (figure 1). We assessed 666 fulltext articles, 58 of which reported a study that fulfilled the inclusion criteria.9 11 12 15 32–85 Overall, studies included 257 983 (range: 67–81 087), 128 261 (range: 300–71 157) and 33 870 (range: 638–20 639) participants for allcause dementia, Alzheimer’s disease and vascular dementia outcomes, respectively. Methodological quality Methodological quality is reported in online supplemental material part 2. The number of studies of high quality was very low (four allcause dementia studies,9 40 55 61 three Alzheimer’s disease studies9 12 40 and one vascular dementia study).40 Selection, study length and followup rate were the most problematic domains of study quality, with 62%, 65% and 41% of studies receiving the lowest rating on these three quality domains, respectively. PA and all-cause dementia The mean incidence of allcause dementia was 10.9% (total n in the analyses=257 983). When compared with the lowest PA category, the pooled RR in higher PA categories showed an association with a reduced risk of allcause dementia (RR 0.80, 95% CI 0.77 to 0.84) (figure 2, table 1). Mean followup length was 12.9 years (SD 9.5), and mean baseline age was 67.0 (SD 12.9) years. There was substantial heterogeneity between the studies (I2=68.7%), but neither baseline age, the length of followup nor study quality modified the association significantly (table 1). The result was similar within the 16 studies with at least 20 years of followup (RR 0.79, 95% CI 0.71 to 0.87, mean baseline age 50.5 (SD 7.8) years, mean followup 27.6 (SD 5.1) years and percentage of participants with dementia at followup 14.6%). In four highquality studies, the pooled RR was 0.82 (95% CI 0.67 to 0.99), with a mean baseline age of 48.2 (SD 3.5) years and mean followup of 23.2 (SD 4.5) years and 7.6% of participants with dementia at followup. This was very similar to the pooled RR of 0.80 in all studies (table 1). Only three studies were of high quality, had a young baseline age of 30–55 years and had a followup longer than 20 years. The pooled RR in these studies was also similar to the pooled RR in all studies, but not significant (pooled RR 0.79, 95% CI 0.62 to 1.01). Omitting the study with the largest sample size or the study with the largest weight did not significantly change the result (online supplemental table S1). Sample size, funding source, adjusting for ApoE ε4 status, baseline cognition or education did not significantly modify the association of PA Kirjasto/Kausijulkaisut. Protected by copyright. on March 21, 2022 at Jyvaskylan Yliopostohttp://bjsm.bmj.com/Br J Sports Med: first published as 10.1136/bjsports-2021-104981 on 17 March 2022. Downloaded from 4IsoMarkkuP, etal. Br J Sports Med 2022;0:1–11. doi:10.1136/bjsports-2021-104981 Review and allcause dementia (online supplemental table S1). The risk of allcause dementia did not significantly differ between PA levels meeting or not meeting the WHO recommendations of PA (table 1, test for heterogeneity between groups: p=0.202). The two studies examining the association of workrelated PA and allcause dementia showed an opposite trend than other PA (RR 1.25, 95% CI 0.98 to 1.59) (online supplemental table S1). Significant linear, quadratic and cubic spline dose–response relationships were observed between increasing PA levels and lower allcause dementia incidence (figure 3 and online supplemental material part 3). The funnel plot for studies of PA and allcause dementia showed some asymmetry suggesting some publication bias (underreporting of studies with no effect, figure 4). PA and Alzheimer’s disease The mean incidence of Alzheimer’s disease was 8.3% among 128 261 participants. Compared with the lowest PA category, the pooled RR in higher PA categories showed an association with lower incidence of Alzheimer’s disease (RR 0.86, 95% CI 0.80 to 0.93; online supplemental figure S1 and table 2). The mean followup length was 11.5 (SD 8.8) years, and mean baseline age was 68.7 (SD 12.4) years. There was moderate heterogeneity between the studies (I2=47.6%), and neither baseline age, the length of followup nor study quality modified the association significantly (table 2). This result was similar among the seven studies with at least 20 years of followup (RR 0.76, 95% CI 0.64 to 0.90, mean baseline age 52.8 (SD 8.9) years, mean followup 26.8 (SD 6.4) years and percentage of participants with Alzheimer’s disease at followup 5.2%). Among the three highquality studies, the pooled RR of 0.71 (95% CI 0.42 to 1.22) was nonsignificant (table 2). Neither sample size, adjustment for ApoE ε4, baseline cognition nor education significantly moderated the association between PA and Alzheimer’s disease incidence (online supplemental table S2). The risk of allcause dementia did not significantly differ between PA levels meeting or not meeting the WHO recommendations of PA (table 1, test for heterogeneity between groups: p=0.202). Dose–response metaanalyses revealed a significant cubic spline trend between PA levels and Alzheimer’s disease incidence, but linear and quadratic trends were nonsignificant (online supplemental material part 3). The funnel plot for studies of PA and Alzheimer’s disease showed some asymmetry suggesting possibly small publication bias (underreporting of results with no effect, online supplemental figure S2). PA and vascular dementia The mean incidence of vascular dementia was 3.8% among 33 870 participants. When compared with the lowest PA category, the pooled RR in higher PA categories showed an association with reduced incidence of vascular dementia (pooled RR 0.79, 95% CI 0.66 to 0.95) (online supplemental figure S3 and table S3). Mean followup length was 10.9 (SD 8.6) years, and mean baseline age was 67.0 (SD 8.5) years. Statistical heterogeneity between the studies was moderate (I2=36.0%). Neither baseline age, length of followup, meeting the WHO PA recommendation, adjusting for baseline cognition nor study quality significantly modified the association (online supplemental tables S3 and S4). There was only one highquality study with followup longer than 20 years and baseline age between 30 and 55 years.40 The association between PA and decreased incidence of vascular dementia was significant in this study. Significant linear, quadratic and cubic spline dose–response relationships between PA and vascular dementia incidence were observed (online supplemental material part 3). The funnel plot for studies of PA and vascular dementia did not suggest publication bias (online supplemental figure S4). ApoE ε4 interaction Most studies that investigated ApoE ε4 interactions found no significant interactions (9 of 11 studies) (supplementary Figure 1 Flow diagram showing the screening process and the search results. Kirjasto/Kausijulkaisut. Protected by copyright. on March 21, 2022 at Jyvaskylan Yliopostohttp://bjsm.bmj.com/Br J Sports Med: first published as 10.1136/bjsports-2021-104981 on 17 March 2022. Downloaded from 5 IsoMarkkuP, etal. Br J Sports Med 2022;0:1–11. doi:10.1136/bjsports-2021-104981 Review sensitivity analyses for ApoE ε4 allele). In the studies that reported stratified results according to ApoE ε4 carrier status, the pooled RR between PA and allcause dementia or Alzheimer’s disease was similar for ApoE ε4 carriers (RR 0.81, 95% CI 0.67 to 0.98) and noncarriers (RR 0.72, 95% CI 0.56 to 0.92). Tests for heterogeneity between groups were invalid because of large heterogeneity between studies (online supplemental figures S5–S8). DISCUSSION This metaanalysis showed that higher PA levels were associated with lower incidence of allcause dementia, Alzheimer’s disease and vascular dementia. These associations were present for allcause dementia and Alzheimer’s disease in studies with long followups (>20 years) and in cohorts with baseline age between 30 and 55 years. Neither baseline age nor followup length moderated the associations of PA with allcause dementia or Alzheimer’s disease. Data for vascular dementia were scarce, especially for long followups, but the results supported an inverse association between PA and vascular dementia incidence. Earlier metaanalyses investigating the associations between PA levels and dementia incidence have been based on short followups,5 29 86 and the results have only been significant in studies with short (<10 years) followups6 9 or elderly populations.7 8 These factors can introduce the possibility of reverse causation, whereby PA levels are affected by dementia.9 Participants’ levels of PA may also change during long followups.87 In this study, we did not find evidence to suggest that reverse causation or regression dilution bias88 affected the observed associations between PA and dementias. Our results therefore support the role of PA as a modifiable protective midlife lifestyle factor of dementia. However, funnel plots for allcause dementia and Alzheimer’s disease suggested some publication bias. In highquality studies, a nonsignificant negative association was found between PA and Alzheimer’s disease. However, with only three highquality studies, the statistical power to show Figure 2 Longitudinal observational studies of physical activity (PA) and allcause dementia: forest plot. APOE, apolipoprotein E; MVPA, moderate to vigorous physical activity; RR, relative risk. Kirjasto/Kausijulkaisut. Protected by copyright. on March 21, 2022 at Jyvaskylan Yliopostohttp://bjsm.bmj.com/Br J Sports Med: first published as 10.1136/bjsports-2021-104981 on 17 March 2022. Downloaded from 6IsoMarkkuP, etal. Br J Sports Med 2022;0:1–11. doi:10.1136/bjsports-2021-104981 Review a significant association is low to moderate but the pooled RR estimate in highquality studies (0.71) was similar to that obtained when including all studies (0.86). Additionally, the metaregression estimate did not show significant moderation by study quality. Our inclusion criteria were also strict, as we excluded studies with baseline in old age and without a validated measure of cognition at baseline. This procedure reduces the risk of bias due to reverse causality, but led to fewer studies being included in the metaanalysis. The strict representativeness criterion may have unnecessarily limited the number of highquality studies88 and thereby increased the CIs of pooled risk estimates from highquality studies. Dose–response metaanalyses showed significant linear, quadratic and cubic inverse associations between PA levels and incidence of allcause dementia and vascular dementia. The finding of a linear dose–response is in line with the results from Xu et al8; however, in our analysis, the effect of PA on allcause dementia incidence was greatest when moving from extreme sedentariness to some PA. While a significant cubic spline relationship was observed between PA and Alzheimer’s disease, the model only included two studies that examined the effects of PA levels greater than 1750 MET*min/week. More studies among more physically active cohorts are needed to conclusively determine whether more PAs offer greater protection at the higher end of the spectrum, or whether a moderate level of PA offers similar protective effects. Our results contrast with those from Kivimäki et al9 who examined individual participant data from many study cohorts worldwide (n=404 840). In that study, no associations were found between PA and allcause dementia or Alzheimer’s disease when followups were longer than 10 years. Notably, the incidence of allcause dementia in their metaanalysis was 0.5%. This is an exceptionally low allcause dementia incidence rate, considering the global annual dementia incidence rate of 17.3% in adults over 60 years of age.89 Two factors can explain this. First, the mean age at baseline was 45.5 years in Kivimäki et al.9 As the mean followup length was 14.9 years, the mean age at the end of followup was approximately 60.4 years, but the mean age of allcause dementia diagnosis in the study was 80.6 years. The result was similar in a subanalysis of persons aged 60 years or older.9 The followup length of over 10 years may also contribute to reverse causation, considering the long preclinical period of Alzheimer’s disease.10 Therefore, these earlier results may be susceptible to bias from both earlyonset dementias and reverse causation. The very low allcause dementia incidence rate in the Kivimäki et al’s9 study may also be explained by the source of incidence rate data: the metaanalysis used register data (hospitalisations, medical reimbursements and death registers) and a few very large cohorts had only death register data. We excluded studies with dementia mortality as the outcome because the relatively low sensitivity of death registers to detect dementia cases may underestimate its association with risk or protective factors.90 Further, dementia mortality in younger individuals is likely to reflect earlyonset dementias. On the other hand, there may be a survival bias, whereby those with higher levels of PA live longer and are therefore at an increased risk of developing dementia. Table 1 PA and allcause dementia: main results, main sensitivity analyses with metaregressions and subgroup analyses, and dose–response analysis Pooled RR 95% CI I2 (%) Studies combined (n) Beta estimate* 95% CI All physical activity 0.80 0.77 to 0.84 68.7 49 Baseline age (continuous) 1.00 0.98 to 1.02 Baseline age (categorical) Age group 30–55 years9 33 35 36 39 40 42 55 61 62 66 73 77 80 0.79 0.71 to 0.87 42.9 14 Age group 55–69 years44 45 57 58 81 83 0.82 0.74 to 0.90 70.8 6 Age group ≥70 years11 32 34 37 38 41 43 46–54 56 59 60 63–65 72 74–76 79 82 0.80 0.75 to 0.85 70.4 29 Followup length (continuous) 1.00 0.97 to 1.03 Followup length (categorical) Followup length <5 years38 43 46 51 52 54 63 75 78 79 0.61 0.50 to 0.74 64.8 10 Followup length 5–20 years9 11 32 34 37 41 44 45 47–50 56–59 64 65 72 74 76 81–83 0.86 0.82 to 0.90 64.2 24 Followup length ≥20 years9 33 35 36 39 40 42 53 55 60–62 66 73 77 80 0.79 0.71 to 0.87 44.8 16 Study quality (high vs moderate vs low)† 0.99 0.64 to 1.53 Low quality9 11 32 34 35 38 41 43–47 49–51 56–58 60 62–66 72 74–76 78 79 81 83 0.81 0.77 to 0.85 75.5 32 Moderate quality9 33 36 37 39 42 48 52–54 59 73 77 80 82 0.79 0.72 to 0.86 28.4 15 High quality9 40 55 61 0.82 0.67 to 0.99 58.9 4 Meeting PA guidelines‡ 11 33 36 38 43 44 48 53–55 59–61 63 65 72 73 76 80 83 0.82 0.76 to 0.87 22.0 20 Not meeting PA guidelines‡ 9 11 32 33 36 38 44 45 47–49 51 57 58 61 62 65 66 72–75 77–79 83 0.76 0.69 to 0.83 60.8 25 Highest quality studies only: age group 30–55 years, followup length >20 years and high quality40 55 61 0.79 0.62 to 1.01 67.4 3 *Beta estimate is the regression coefficient from the metaregression examining the relationship of modifier or continuous PA on the log risk ratio of dementia. †Study quality was assessed with a quality assessment tool we developed (see online supplemental material part 1 for details). ‡The test for heterogeneity between groups was nonsignificant (p=0.202). .I2, heterogeneity; PA, physical activity; RR, relative risk. Kirjasto/Kausijulkaisut. Protected by copyright. on March 21, 2022 at Jyvaskylan Yliopostohttp://bjsm.bmj.com/Br J Sports Med: first published as 10.1136/bjsports-2021-104981 on 17 March 2022. Downloaded from 7 IsoMarkkuP, etal. Br J Sports Med 2022;0:1–11. doi:10.1136/bjsports-2021-104981 Review The sensitivity analyses for allcause dementia and Alzheimer’s disease showed similar estimates among studies that controlled for baseline cognition. Studies have primarily assessed baseline cognition using short screening tests, which are known to result in ceiling effects, and most studies of younger cohorts did not assess cognitive ability at baseline. While many studies Figure 3 Dose–response analysis of physical activity (PA) levels and allcause dementia incidence. Linear trend shown with dasheddotted line and 95% CI in blue; quadratic trend shown with dashed line and 95% CI in orange; and cubic spline trend shown with solid line and 95% CI in green. MET, metabolic equivalent of energy expenditure. Figure 4 Funnel plot for the longitudinal observational studies on physical activity and allcause dementia with pseudo95% CIs. RR, relative risk. Kirjasto/Kausijulkaisut. Protected by copyright. on March 21, 2022 at Jyvaskylan Yliopostohttp://bjsm.bmj.com/Br J Sports Med: first published as 10.1136/bjsports-2021-104981 on 17 March 2022. Downloaded from 8IsoMarkkuP, etal. Br J Sports Med 2022;0:1–11. doi:10.1136/bjsports-2021-104981 Review collected information on baseline cognition, few adjusted for it in their analyses. Because baseline cognitive ability may be the most robust predictor of cognition later in life,91 the absence of rigorous assessments of baseline cognition and failure to control for it in analyses should be seen as major limitations of the included studies and this metaanalysis. Physically active individuals may have higher cognitive reserve to start with,92 so studies should examine whether early cognitive ability predicts PA later in life. In many studies investigating PA and other health outcomes, workrelated PA shows an inverse association with leisuretime PA when adjusted for socioeconomic status or education.93 94 This may also indicate that higher cognitive ability or other unmeasured confounding factors, and not leisuretime PA, may drive the association with a decreased incidence of dementia. Almost all studies in this field have adjusted their results for education level, a widely used proxy for higher cognitive reserve. Still, people with the same number of years of formal education may vary greatly in their cognitive abilities.95 Many studies have suggested that ApoE ε4 carrier status modifies the relationship between PA and dementia.11–13 Our ApoE ε4 interaction analyses suggest no such modification for allcause dementia, Alzheimer’s disease or vascular dementia. Strengths This metaanalysis includes extensive data and has examined the association of PA and dementia in longer followups than earlier metaanalyses. Our quality assessment was specifically developed to account for the long preclinical period of dementia, and the quality assessment of PA has been developed in cooperation with sports and exercise medicine experts (KW and UMK). We also studied PA levels meeting a fixed threshold (WHO PA recommendation), and we addressed dose–response relationships between PA with allcause dementia, Alzheimer’s disease and vascular dementia incidence. Limitations Some limitations of this metaanalysis bear mentioning. Many studies used only rough PA measures (eg, a dichotomous yes or no question to describe exercise participation). These rough measures coupled with the midpoint or mean estimation of MET*min/week PA levels imply that the dose–response metaanalyses likely lack precision, especially when the PA levels within a group were wide. Future cohort studies should use objective or finer grain PA assessments and make individual participant data open whenever possible. Of the included studies, few were high quality, few reported on vascular dementia as an outcome and few had any robust measures of cognition at baseline. The stringent criterion of representativeness may have unnecessarily limited the number of highquality studies. Additionally, some publication bias may have affected the results for allcause dementia and Alzheimer’s disease. We searched only studies published in English which is also a possible source of bias.96 In addition, the impacts of PA modalities which are themselves associated with increased risk of dementia (eg, boxing)97 are not accounted for in the results presented here. CONCLUSIONS AND POLICY IMPLICATIONS This metaanalysis found inverse associations between PA levels and incidence rates of allcause dementia and Alzheimer’s disease, even in studies with followups longer than 20 years. This finding supports PA as a modifiable protective lifestyle factor of dementia. Policy makers should continue to promote PA in school and worklife reforms, city planning and health initiatives. However, these conclusions should be tempered slightly, as the metaanalysis was based on observational studies with known limitations compared with intervention studies, as highquality studies were scarce and some publication bias was present. More research with long followups, adjustment for baseline cognitive performance and valid measures of PA and dementia are Table 2 PA and Alzheimer’s disease: main results, main sensitivity analyses with metaregressions and subgroup analyses, and dose–response analysis Pooled RR 95% CI I2 (%) Cohorts combined (n) Beta estimate* 95% CI All PA 0.86 0.80 to 0.93 47.6 24 Baseline age (continuous) 1.00 0.97 to 1.03 Baseline age (categorical) Age group 30–55 years9 12 33 40 77 80 0.81 0.66 to 0.99 37.3 6 Age group 55–69 years† 44 1.09 0.96 to 1.24 0.0 1 Age group ≥70 years11 37 38 47 48 52 54 59 60 63 67–71 84 85 0.84 0.77 to 0.93 48.5 17 Followup length 1.00 0.96 to 1.04 Followup length <5 years38 52 54 63 70 84 0.93 0.79 to 1.08 48.4 6 Followup length 5–20 years9 11 37 44 47 48 59 67–69 71 85 0.87 0.78 to 0.97 41.3 12 Followup length ≥20 years9 12 33 40 60 77 80 0.76 0.64 to 0.90 16.9 7 Study quality (low vs moderate vs high)‡ 1.14 0.59 to 2.22 Low quality9 11 44 47 60 63 67 68 70 71 0.97 0.88 to 1.07 34.1 10 Moderate quality9 33 37 38 48 52 54 59 69 77 80 84 85 0.81 0.74 to 0.90 24.0 13 High quality 9 12 40 0.71 0.42 to 1.22 71.8 3 Meeting PA guidelines§ 11 12 33 38 44 47 48 54 59 60 63 68 69 71 80 84 0.75 0.64 to 0.88 43.4 16 Not meeting PA guidelines§ 9 11 33 38 44 48 67 69 77 80 0.94 0.85 to 1.04 0.0 10 Age group 30–55 years, high quality and followup length >20 years12 40 0.55 0.29 to 1.03 53.9 2 *Beta estimate is the regression coefficient from the metaregression examining the relationship of modifier or continuous PA on the log risk ratio of dementia. †Only one study, not metaanalytical analysis. ‡Study quality was assessed with a quality assessment tool we developed (see online supplemental material part 1 for details). §The test for heterogeneity between groups was nonsignificant (p=0.126). .I2, heterogeneity; PA, physical activity; RR, relative risk. Kirjasto/Kausijulkaisut. Protected by copyright. on March 21, 2022 at Jyvaskylan Yliopostohttp://bjsm.bmj.com/Br J Sports Med: first published as 10.1136/bjsports-2021-104981 on 17 March 2022. Downloaded from The search in CINAHL in 2021 1. physical activity 2. aerobic exercise 3. sport* 4. walking 5. physical training 6. 1 OR 2 OR 3 OR 4 OR 5 7. dementia 8. Alzheimer* 9. 7 OR 8 10. prospective study 11. longitudinal 12. follow-up 13. observational 14. cohort study 15. 10 OR 11 OR 12 OR 13 OR 14 16. 6 AND 9 AND 15 Filters: English, Human, All adult, 12/2017 – 10/2021, Research articles Search field: Abstract The search in Web of Science in 2021 1. ”physical activity” 2. ”aerobic exercise” 3. sport* 4. walking 5. ”physical training” 6. 1 OR 2 OR 3 OR 4 OR 5 7. dementia 8. Alzheimer* 9. 7 OR 8 10. prospective 11. longitudinal 12. follow-up 13. observational 14. cohort* 15. 10 OR 11 OR 12 OR 13 OR 14 16. 6 AND 9 AND 15 Articles (Document Types) and English (Languages) and 27TH ANNUAL MEETING OF THE SOCIETY FOR THE STUDY OF INGESTIVE BEHAVIOR SSIB or ALZHEIMER S ASSOCIATION INTERNATIONAL CONFERENCE (Exclude – Conference Titles) and Articles (Document Types) and Orthopedics or Surgery or Engineering or Medical Informatics or Obstetrics Gynecology or Urology Nephrology or Cell Biology or Dentistry Oral Surgery Medicine or Emergency Medicine or History or Mechanics or Oncology or Ophthalmology or Otorhinolaryngology or Respiratory System or Rheumatology or Social Work (Exclude – Research Areas) Date range: 22.12.2017 – 21.10.2021 Field of search: Abstract BMJ Publishing Group Limited (BMJ) disclaims all liability and responsibility arising from any reliance Supplemental material placed on this supplemental material which has been supplied by the author(s) Br J Sports Med doi: 10.1136/bjsports-2021-104981–11.:10 2022;Br J Sports Med, et al. Iso-Markku P Specifications to the inclusion and exclusion criteria Specifications to the inclusion and exclusion criteria at title and abstract screening phase in cases of disagreements: The studies are included if • The study is a longitudinal follow-up study addressing something else than physical activity as a dependent variable and dementia as an outcome but physical activity is mentioned as a covariate or it is unclear based on the abstract if physical activity is used as a covariate. • The studies that have frailty as an independent variable with low physical activity is used as one of the criteria (or it is not clear what are the criteria according to the title and abstract) and dementia is the outcome variable. • The studies that have healthy aging, survival or successful aging as an outcome measure with being undemented as one of the criteria or the criteria are not specified in the abstract and physical activity is one the dependent variables. The studies are excluded if • The study is a longitudinal follow-up study addressing something else than physical activity as a dependent variable and dementia as an outcome and covariates are listed in the abstract and physical activity is not one of them. • Dementia mortality is the outcome variable. • The study is considered to be cross-sectional because there is no indication in the title or the abstract that the study has a follow-up. • The studies that are solely addressing the effects of scuba-diving or diving. • The studies that are examining the effect of repetitive head trauma resulted from eg. boxing. • The study addresses only a specific cohort of patients with a certain diagnosis or in for example “fallers”. • Long-term physical activity during the follow-up is examined instead of baseline physical activity in only one time point. • They study the effect of physical activity on structural brain changes and there is no indication in the title or abstract that dementia would also be measured. • The study addresses the effect of sedentary hours in a day on cognition. • The study addresses the effect of life-space mobility on dementia instead of the effect of physical activity. • Baseline physical activity has not been measured or the estimate of physical activity is based on a single bout of physical activity. Specifications to the precise inclusion and exclusion criteria at the full-text screening phase in cases of disagreements: The study is included if • Physical activity is only included as an independent variable that is a composite variable (eg. frailty) and the results for the components are mentioned separately, even if the actual numbers are not specified. If the results for the components are mentioned separately but no numbers are provided, the authors are contacted and asked for the numbers. • There is a valid measure of baseline cognition or the population is in midlife. We define midlife as being 55 years old or younger and require that the cohort members are aged 55 years or less (mean age or median < 55 years and maximum age 65 years or +1 SD < 60 years). BMJ Publishing Group Limited (BMJ) disclaims all liability and responsibility arising from any reliance Supplemental material placed on this supplemental material which has been supplied by the author(s) Br J Sports Med doi: 10.1136/bjsports-2021-104981–11.:10 2022;Br J Sports Med, et al. Iso-Markku P The study is excluded if • The study has used only orientation to date, day of week, month and year as a baseline measure of cognition. • Studies in which dementia diagnosis was self-reported, based solely on disability scale or the main source (≥ 50%) of dementia diagnoses were death registers. BMJ Publishing Group Limited (BMJ) disclaims all liability and responsibility arising from any reliance Supplemental material placed on this supplemental material which has been supplied by the author(s) Br J Sports Med doi: 10.1136/bjsports-2021-104981–11.:10 2022;Br J Sports Med, et al. Iso-Markku P A quality assessment tool for the quality assessment of cohort studies addressing the association of physical activity and dementia or cognition Note: A study can be given a maximum of one star for each numbered item within the Selection, Comparability and Outcome categories. Selection 1) Representativeness of the exposed cohort a) Truly representative (represents well the whole age group or the whole age group of any particular race and is not selected in regards to some disease, socio-economic status or for example only inhabitants of a nursing home, participation rate > 70% (of those alive) and sample size at least 1000 (one star) b) Truly or somewhat representative (participation rate 50-70%) and sample size > 1000 (half a star) c) Selected group in regards to some characteristic or participation rate < 50% or sample size < 1000 d) No description of the derivation of the cohort or participation rate lacks 2) Performance quality (adapted from (Br J Sports Med 2017;51:1410-18)) a) Good: PA assessed with a structured questionnaire of the duration, frequency and intensity of PA or the intensity of PA assessed with a structured question. Or PA assessed with an objective measure of PA (eg. accelerometer) (one star) b) Moderate: Participation only in some types of sports assessed but other activities not considered or assessment of intensity lacks. Frequency or duration are assessed. (half a star) c) Low: A “yes” or “no” question used. Frequency and duration not assessed. Or physical activity index on versatility of sports and somewhat physical household chores but not assessing intensity, frequency or duration. Or not described how exercise or physical activity was measured. (no star) 3) Demonstration that outcome of interest was not present at start of study a) Yes. In a study population whose average age > 55 years, valid measure of cognition is used and demented individuals and individuals with mild cognitive impairment at baseline according to baseline cognition screening have been excluded or population is in midlife (mean age or median < 55 years and maximum age 65 years or +1 SD < 60 years) (one star) b) No Comparability 1) Comparability of cohorts on the basis of the design or analysis controlled for confounders a) The study controls for the following four factors: age, sex (or all cohort members represent the same sex), some vascular risk factor† and education or a measure of general cognitive ability at baseline (education criterion is not needed if all cohort BMJ Publishing Group Limited (BMJ) disclaims all liability and responsibility arising from any reliance Supplemental material placed on this supplemental material which has been supplied by the author(s) Br J Sports Med doi: 10.1136/bjsports-2021-104981–11.:10 2022;Br J Sports Med, et al. Iso-Markku P members have the same education level). In addition, the results have been adjusted with baseline cognition in study population whose average age > 55 years (one star) b) The study controls only for three of the factors presented above (age, sex, some vascular risk factor and education or a measure of general intelligence) or/and in study populations whose mean age > 55 years the results have not been adjusted with baseline cognition or the sociodemographic and health behaviors controlled for are not specified further (half a star) c) Cohorts are not comparable on the basis of covariates controlled for (no star) Outcome 1) Assessment of outcome a) A validated measure of dementia (one star) b) Record linkage (half a star) c) Self report or other d) No description 2) Was follow-up long enough for outcomes to occur a) Yes (one star) b) No Indicate the median duration of follow-up and a brief rationale for the assessment above: 10 years in dementia studies. 3) Adequacy of follow-up of cohorts a) Complete follow upall subject accounted for (one star) b) Subjects lost to follow up unlikely to introduce biasnumber lost less than or equal to 20%* (when follow-up less than 10 years) or 30%* (when follow-up at least 10 years) of those alive or description of those lost suggested no different from those followed. (one star) c) Follow up rate less than 80%* (when follow-up less than 10 years) or less than 70%* (when follow-up longer than 10 years) and no description of those lost. d) No statement * Follow-up rates are calculated taking into account only the cohort members who have been alive at the time of the follow-up † Vascular risk factor signifies a cardiovascular disease or information on smoking, body mass index, cholesterol levels, diet, blood pressure, diabetes or blood glucose. Thresholds for converting the Newcastle-Ottawa scales to AHRQ standards (good, moderate, and poor): Good quality: Selection: 2,5 - 3 stars, Comparability: 1 star, Outcome: 2,5 - 3 stars Moderate quality: Selection: 2-3 stars, Comparability: 0,5 -1 star, Outcome: 2-3 stars, Poor quality: studies not reaching Moderate/Good quality BMJ Publishing Group Limited (BMJ) disclaims all liability and responsibility arising from any reliance Supplemental material placed on this supplemental material which has been supplied by the author(s) Br J Sports Med doi: 10.1136/bjsports-2021-104981–11.:10 2022;Br J Sports Med, et al. Iso-Markku P A quality assessment tool for the quality assessment of case-control studies addressing the association of physical activity and dementia or cognition Note: A study can be given a maximum of one star for each numbered item within the Selection, Comparability and Outcome categories. Baseline 1) Selection of controls: a) Community controls (half a star) b) Hospital controls c) No description 2) Definition of controls: a) No history of disease (endpoint) (half a star) b) No description of source 3) Demonstration that outcome of interest was not present at start of study a) Yes. In a study population whose average age > 55 years, valid measure of cognition is used and demented individuals and individuals with mild cognitive impairment at baseline according to baseline cognition screening have been excluded or population is in midlife (mean age or median < 55 years and maximum age 65 years or +1 SD < 60 years) (one star) b) No 4) Performance quality (adapted from (Br J Sports Med 2017;51:1410-18)) a) Good: PA assessed with a structured questionnaire of the duration, frequency and intensity of PA or the intensity of PA assessed with a structured question. Or PA assessed with an objective measure of PA (eg. accelerometer). Additionally same method is used for cases and controls and non-response rate is the same for cases and controls. (one star) b) Moderate: Participation only in some types of sports assessed but other activities not considered or assessment of intensity lacks. Frequency or duration are assessed. (half a star) c) Low: A “yes” or “no” question used. Frequency and duration not assessed. Or physical activity index on versatility of sports and somewhat physical household chores but not assessing intensity, frequency or duration. (no star) Comparability 1) Comparability of cohorts on the basis of the design or analysis controlled for confounders BMJ Publishing Group Limited (BMJ) disclaims all liability and responsibility arising from any reliance Supplemental material placed on this supplemental material which has been supplied by the author(s) Br J Sports Med doi: 10.1136/bjsports-2021-104981–11.:10 2022;Br J Sports Med, et al. Iso-Markku P a) The study controls for the following four factors: age, sex (or all cohort members represent same sex), some vascular risk factor† and education or a measure of general cognitive ability at baseline. In addition, the results have been adjusted with baseline cognition in study population whose average age > 55 years (one star) b) The study controls only for three of the factors presented above (age, sex, some vascular risk factor and education or a measure of general intelligence) or in study populations whose mean age > 55 years the results have not been adjusted with baseline cognition (half a star) c) Cohorts are not comparable on the basis of covariates controlled for (no star) Outcome 1) Assessment of outcome a) A validated measure of dementia (if many cognitive tests used, most validated) (one star) b) Record linkage (half a star) c) Self report or other d) No description 2) Was follow-up long enough for outcomes to occur a) Yes (one star) b) No Indicate the median duration of follow-up and a brief rationale for the assessment above: 10 years in dementia studies. 3) Is the case definition adequate?: a) Yes, with a valid measure of dementia (half a star) b) No, does not fulfil the criteria defined above 4) Representativeness of the cases: a) Consecutive or obviously representative series of cases (half a star) b) Potential for selection biases or not stated † Vascular risk factor signifies a cardiovascular disease or information on smoking, body mass index, cholesterol levels, diet, blood pressure, diabetes or blood glucose. Thresholds for converting the Newcastle-Ottawa scales to AHRQ standards (good, moderate, and poor): Good quality: Baseline: 2,5 - 3 stars, Comparability: 1 star, Outcome: 2,5 - 3 stars Moderate quality: Baseline: 2-3 stars, Comparability: 0,5 -1 star, Outcome: 2-3 stars, Poor quality: studies not reaching Moderate/Good quality BMJ Publishing Group Limited (BMJ) disclaims all liability and responsibility arising from any reliance Supplemental material placed on this supplemental material which has been supplied by the author(s) Br J Sports Med doi: 10.1136/bjsports-2021-104981–11.:10 2022;Br J Sports Med, et al. Iso-Markku P Deviations from the original study plan 1. We report the results in two parts o The numbers of subanalyses grew so large that we decided to split the systematic review and meta-analysis into two parts for clarity (one part for dementia and one part for cognition). 2. Our aim was to study what kind of physical activity is associated with decreased incidence of dementia o One aim was to compare the effect of the volume of physical activity and the intensity of physical activity but there weren’t enough studies reporting a relation between the intensity of physical activity and dementia. Another aim was to examine the separately the effect of a rough measure of physical activity (yes or no question, number of different physical activities) or walking distance to more specific measures of physical activity (taking into account the volume, intensity and frequency of physical activity). During the examination of the physical activity measures of the included studies, we decided that it would be better and more clear way to categorize them by WHO physical activity recommendation threshold. This way, we can pinpoint a specific volume of physical activity and take into account whether the study actually is able to measure physical activity reliably enough to tell whether physical activity WHO recommendation is met. At this point, we also glanced the earlier meta-analyses more thoroughly and decided to do an analysis similar as in the most earlier meta-analyses (highest physical activity group vs lowest physical activity group) for better comparison with the earlier studies. 3. In addition to dementia and Alzheimer’s disease, we collected the results separately for vascular dementia. 4. The method of measuring education was extracted with precision but was not examined with a separate sub-analysis because of vast heterogeneity in the classifications. 5. Data extraction of other modifiers than the length of the follow-up was done only by one researcher due to time and resource constraints. 6. The quality assessment tool was developed only after research plan was ready and adjusting for chronic diseases at baseline was not included because adjusting for education and some vascular risk factor was deemed more important. 7. We did not use Review Manager 5 but Covidence for handling data. 8. We conducted the prespecified moderator analyses but the classifications varied a little according to the data found (eg. follow-up length was grouped in a different way because we found many studies with follow-up length over 20 years). o Separate analyses were not done for according to the validity of physical activity measurement because we did a separate analysis according to meeting physical activity WHO recommendations as discussed earlier in point 2. o We did not perform separate analyses for different genders because there was not enough data: almost all studies presented results only for men and women jointly and very few studies reported the results for gender interaction test. o There was not enough data for a separate analysis for twin studies. 9. We aimed to include studies with “a valid physical activity questionnaire”, but this criterion would have tremendously restricted the number of included studies, we ended up including also physical activity questionnaires that are not separately validated. BMJ Publishing Group Limited (BMJ) disclaims all liability and responsibility arising from any reliance Supplemental material placed on this supplemental material which has been supplied by the author(s) Br J Sports Med doi: 10.1136/bjsports-2021-104981–11.:10 2022;Br J Sports Med, et al. Iso-Markku P Supplementary analyses Supplementary Table S1. Physical activity and all-cause dementia, supplementary analyses * Moderate heterogeneity observed (I² up to 70.1%) in one or more subgroups; tests for heterogeneity between subgroups are likely to be invalid † Beta estimate is the regression coefficient from the meta-regression examining the relationship of modifier or continuous physical activity on the log risk ratio of dementia. Subanalyses Pooled RR 95% Confidence interval I2 Number of studies combined b estimate† 95% Confidence intervals Prospective cohort studies [8, 10, 26-30, 33, 36-39, 41-62, 68-79] 0.81 0.77, 0.85 68.4% 46 Case-control studies [34, 35, 40] 0.70 0.52, 0.95 73.4% 3 All physical activity without the study with the largest weight (Palta et al. 2019) 0.80 0.77, 0.84 68.0% 48 All physical activity without the study with the largest sample size (Kivimäki et al. 2019) 0.80 0.76, 0.84 70.1% 48 Work-related physical activity [13, 69] 1.25 0.98, 1.59 29.4% 2 All PA Moderators Sample size 1.00 1.00, 1.00 Baseline cognition *, adjusted 0.79 0.72, 0.88 54.3% 12 not adjusted 0.81 0.77, 0.85 70.1% 37 Education *, adjusted 0.80 0.76, 0.85 69.0% 38 not adjusted 0.78 0.69, 0.88 64.4% 11 Chronic diseases *, adjusted 0.80 0.75, 0.85 71.0% 28 not adjusted 0.81 0.76, 0.87 56.1% 21 APOE ε4 status *, adjusted 0.82 0.76, 0.88 56.6% 15 not adjusted 0.79 0.75, 0.84 69.5% 34 Funding source *, only non-commercial 0.79 0.75, 0.84 72.2% 41 also commercial or not told 0.83 0.78, 0.89 4.1% 9 Number of confounders 1.01 0.95, 1.08 BMJ Publishing Group Limited (BMJ) disclaims all liability and responsibility arising from any reliance Supplemental material placed on this supplemental material which has been supplied by the author(s) Br J Sports Med doi: 10.1136/bjsports-2021-104981–11.:10 2022;Br J Sports Med, et al. Iso-Markku P Supplementary Figure S1. Longitudinal observational studies of physical activity and Alzheimer’s disease: forest plot BMJ Publishing Group Limited (BMJ) disclaims all liability and responsibility arising from any reliance Supplemental material placed on this supplemental material which has been supplied by the author(s) Br J Sports Med doi: 10.1136/bjsports-2021-104981–11.:10 2022;Br J Sports Med, et al. Iso-Markku P Supplementary Figure S5. Physical activity and all-cause dementia, Alzheimer’s disease and vascular dementia in APOE ε4 carriers BMJ Publishing Group Limited (BMJ) disclaims all liability and responsibility arising from any reliance Supplemental material placed on this supplemental material which has been supplied by the author(s) Br J Sports Med doi: 10.1136/bjsports-2021-104981–11.:10 2022;Br J Sports Med, et al. Iso-Markku P Supplementary Figure S6. Physical activity and all-cause dementia, Alzheimer’s disease and vascular dementia in APOE ε4 carriers when study quality is good or moderate BMJ Publishing Group Limited (BMJ) disclaims all liability and responsibility arising from any reliance Supplemental material placed on this supplemental material which has been supplied by the author(s) Br J Sports Med doi: 10.1136/bjsports-2021-104981–11.:10 2022;Br J Sports Med, et al. Iso-Markku P Supplementary Figure S7. Physical activity and all-cause dementia, Alzheimer’s disease and vascular dementia in APOE ε4 non-carriers BMJ Publishing Group Limited (BMJ) disclaims all liability and responsibility arising from any reliance Supplemental material placed on this supplemental material which has been supplied by the author(s) Br J Sports Med doi: 10.1136/bjsports-2021-104981–11.:10 2022;Br J Sports Med, et al. Iso-Markku P Supplementary Figure S8. Physical activity and all-cause dementia, Alzheimer’s disease and vascular dementia in APOE ε4 non-carriers when study quality is good or moderate BMJ Publishing Group Limited (BMJ) disclaims all liability and responsibility arising from any reliance Supplemental material placed on this supplemental material which has been supplied by the author(s) Br J Sports Med doi: 10.1136/bjsports-2021-104981–11.:10 2022;Br J Sports Med, et al. Iso-Markku P Sensitivity analysis for APOE ε4 allele To find out if APOE ε4 allele moderates the association between physical activity and dementia, we gathered information from the individual studies on APOE ε4 allele interaction analysis results and results reported separately for APOE ε4 carriers and noncarriers. 12 studies presented results from APOE ε4 interaction analyses (Hansson et al. 2019, Dupré et al. 2020, Tan et al. 2017, Taaffe et al. 2008, Podewils et al. 2005, Kim et al. 2011, Ravaglia et al. 2008, Lindsay et al. 2002, Paillard-Borg et al. 2009, Luck et al. 2014, Rovio et al. 2005, Shih et al. 2018). Most studies reported only results from interaction test not separating multiplicative and additive interaction, one study reported additive but not multiplicative interaction results and two studies reported both multiplicative and additive interaction results. The results were the following: - 12 studies reported interaction analysis results - 9 studies reported no interaction ((Hansson et al. 2019, Dupré et al. 2020, Tan et al. 2017, Taaffe et al. 2008, Rovio et al. 2005, Podewils et al. 2005, Luck et al. 2014, Kim et al. 2011, Ravaglia et al. 2008, Lindsay et al. 2002, Paillard-Borg et al. 2009) - 2 studies reported no multiplicative interaction but significant or probable additive interaction (Luck et al. 2014, Rovio et al. 2005) - 1 study reported no interaction on additive scale. The results were presented stratified according to APOE ε4 status but the result for multiplicative interaction was not given (Shih 2018). BMJ Publishing Group Limited (BMJ) disclaims all liability and responsibility arising from any reliance Supplemental material placed on this supplemental material which has been supplied by the author(s) Br J Sports Med doi: 10.1136/bjsports-2021-104981–11.:10 2022;Br J Sports Med, et al. Iso-Markku P 1 Supplementary Material Part 3 Dose-response meta-analyses Contents Method description ........................................................................................................................................... 1 Results - All-cause dementia (A-CD) .................................................................................................................. 2 Linear model including all studies (A-CD) ...................................................................................................... 2 Models including studies with 3 or more PA exposure levels (A-CD) ........................................................... 4 Results - Alzheimer’s disease (AD)..................................................................................................................... 7 Linear model including all studies (AD) ......................................................................................................... 7 Models including studies with 3 or more PA exposure levels (AD) ............................................................... 9 Results - Vascular dementia (VD) .................................................................................................................... 13 Linear model including all studies (VD) ....................................................................................................... 13 Models including studies with 3 or more PA exposure levels (VD) ............................................................. 15 References ....................................................................................................................................................... 19 Method description Calculation of PA exposure levels Following the procedure of Blond et al (1), we used the midpoint of the physical activity range (mean or median) from each group as the value for PA exposure. MET values for listed activities were taken from the included articles, or estimated using MET values of 3.5 for walking, 4.5 for moderate physical activity and 8.0 for sports participation or vigorous PA. For studies in which PA was assessed as bouts per week, one bout was estimated to be 30 minutes in duration unless otherwise specified in the included article. When physical activity levels were specified in calories per week, mean body weights reported in the articles were used for calculating MET-minutes using the formula 𝑀𝐸𝑇 ∗ 𝑚𝑖𝑛𝑢𝑡𝑒𝑠 = 60 ∗ 𝑘𝑐𝑎𝑙 𝑘𝑔 Where mean body weight was not reported in an article, continental body weight averages were used in the calculation (2). In studies where it was not possible to directly calculate MET*minutes per week for each group, we imputed PA exposure values using the means from other similar studies. We used a cutoff of a maximum of 21 hours of moderate PA per week (3 hours X 7 days), and this corresponds to a maximum value of 5040 MET-minutes per week. Dose-response meta-analyses Dose-response meta-analyses were performed using the dosresmeta package (3) in R and visualizations were created using the shiny webapp based on this package (4). Two-stage random effects meta-analyses using the restricted maximum likelihood procedure were used to pool RRs. Among studies with at least 3 different PA exposure levels, dose–response meta-analyses explored linear, quadratic and restricted cubic spline trends within the data. Knots in the restricted cubic spline regression models were set at the 20th and 80th percentiles of the overall PA exposure distribution. As it was not possible to explore quadratic or spline trends among studies with only two different PA exposure levels, we only examined linear trends BMJ Publishing Group Limited (BMJ) disclaims all liability and responsibility arising from any reliance Supplemental material placed on this supplemental material which has been supplied by the author(s) Br J Sports Med doi: 10.1136/bjsports-2021-104981–11.:10 2022;Br J Sports Med, et al. Iso-Markku P 2 within this larger dataset that included all studies with at least two discrete PA exposure groups. We chose 200 MET min/week as the reference for the dose–response analyses as this was roughly equivalent to the mean MET*min per week value of reference groups in the included studies. Post-estimations based on the dose–response model were conducted to predict RRs and 95% CIs at specific MET*min per week values (approximately 200, 900, 2000, 3000, 4000, and 5000). Results - All-cause dementia (A-CD) Linear model including all studies (A-CD) Scatter plot of RRs vs original exposure variable Linear Trend Call: dosresmeta(formula = logrr ~ exposure, id = id, type = type, cases = cases, n = n, data = dataset(), se = se, covariance = input$pscorr) Two-stage random-effects meta-analysis Estimation method: REML Covariance approximation: Greenland & Longnecker Chi2 model: X2 = 34.6275 (df = 1), p-value = 0.0000 Fixed-effects coefficients Estimate Std. Error z Pr(>|z|) 95%ci.lb 95%ci.ub (Intercept) -0.0002 0.0000 -5.8845 0.0000 -0.0003 -0.0001 *** --- Signif. codes: 0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1 Between-study random-effects (co)variance components Std. Dev BMJ Publishing Group Limited (BMJ) disclaims all liability and responsibility arising from any reliance Supplemental material placed on this supplemental material which has been supplied by the author(s) Br J Sports Med doi: 10.1136/bjsports-2021-104981–11.:10 2022;Br J Sports Med, et al. Iso-Markku P 3 0.0002 Univariate Cochran Q-test for residual heterogeneity: Q = 123.3919 (df = 42), p-value = 0.0000 I-square statistic = 66.0% 43 studies, 43 values, 1 fixed and 1 random-effects parameters logLik AIC BIC 276.7986 -549.5973 -546.1220 Graphical prediction Analytical predictions PA exposure pred.lin ci.lb.lin ci.ub.lin 203.64 1.00 1.00 1.00 916.36 0.86 0.82 0.91 1934.55 0.70 0.62 0.79 2952.73 0.57 0.47 0.69 3970.91 0.46 0.36 0.60 4989.09 0.37 0.27 0.52 BMJ Publishing Group Limited (BMJ) disclaims all liability and responsibility arising from any reliance Supplemental material placed on this supplemental material which has been supplied by the author(s) Br J Sports Med doi: 10.1136/bjsports-2021-104981–11.:10 2022;Br J Sports Med, et al. Iso-Markku P 4 Models including studies with 3 or more PA exposure levels (A-CD) Scatter plot of RRs vs original exposure variable Linear Trend Call: dosresmeta(formula = logrr ~ exposure, id = id, type = type, cases = cases, n = n, data = dataset(), se = se, covariance = input$pscorr) Two-stage random-effects meta-analysis Estimation method: REML Covariance approximation: Greenland & Longnecker Chi2 model: X2 = 21.4096 (df = 1), p-value = 0.0000 Fixed-effects coefficients Estimate Std. Error z Pr(>|z|) 95%ci.lb 95%ci.ub (Intercept) -0.0002 0.0000 -4.6270 0.0000 -0.0003 -0.0001 *** --- Signif. codes: 0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1 Between-study random-effects (co)variance components Std. Dev 0.0001 Univariate Cochran Q-test for residual heterogeneity: Q = 57.7815 (df = 19), p-value = 0.0000 I-square statistic = 67.1% BMJ Publishing Group Limited (BMJ) disclaims all liability and responsibility arising from any reliance Supplemental material placed on this supplemental material which has been supplied by the author(s) Br J Sports Med doi: 10.1136/bjsports-2021-104981–11.:10 2022;Br J Sports Med, et al. Iso-Markku P 5 20 studies, 20 values, 1 fixed and 1 random-effects parameters logLik AIC BIC 128.2688 -252.5375 -250.6487 Spline model Call: dosresmeta(formula = logrr ~ rcs(exposure, knots), id = id, type = type, cases = cases, n = n, data = dataset(), se = se, covariance = input$pscorr) Two-stage random-effects meta-analysis Estimation method: REML Covariance approximation: Greenland & Longnecker Chi2 model: X2 = 37.6403 (df = 2), p-value = 0.0000 Fixed-effects coefficients Estimate Std. Error z rcs(exposure, knots)exposure.(Intercept) -0.0005 0.0001 -5.5896 rcs(exposure, knots)exposure'.(Intercept) 0.0004 0.0001 4.5834 Pr(>|z|) 95%ci.lb 95%ci.ub rcs(exposure, knots)exposure.(Intercept) 0.0000 -0.0006 -0.0003 *** rcs(exposure, knots)exposure'.(Intercept) 0.0000 0.0002 0.0006 *** --- Signif. codes: 0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1 Between-study random-effects (co)variance components Std. Dev Corr rcs(exposure, knots)exposure 0.0002 rcs(exposure, knots)exposure rcs(exposure, knots)exposure' 0.0001 -1 Univariate Cochran Q-test for residual heterogeneity: Q = 58.7637 (df = 38), p-value = 0.0169 I-square statistic = 35.3% 20 studies, 40 values, 2 fixed and 3 random-effects parameters logLik AIC BIC 222.9501 -435.9001 -427.7122 Quadratic trend BMJ Publishing Group Limited (BMJ) disclaims all liability and responsibility arising from any reliance Supplemental material placed on this supplemental material which has been supplied by the author(s) Br J Sports Med doi: 10.1136/bjsports-2021-104981–11.:10 2022;Br J Sports Med, et al. Iso-Markku P 12 covariance = input$pscorr) Two-stage random-effects meta-analysis Estimation method: REML Covariance approximation: Greenland & Longnecker Chi2 model: X2 = 2.8700 (df = 2), p-value = 0.2381 Fixed-effects coefficients Estimate Std. Error z Pr(>|z|) 95%ci.lb exposure.(Intercept) -0.0002 0.0001 -1.6085 0.1077 -0.0004 I(exposure^2).(Intercept) 0.0000 0.0000 1.3787 0.1680 -0.0000 95%ci.ub exposure.(Intercept) 0.0000 I(exposure^2).(Intercept) 0.0000 --- Signif. codes: 0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1 Between-study random-effects (co)variance components Std. Dev Corr exposure 0.0002 exposure I(exposure^2) 0.0000 -1 Univariate Cochran Q-test for residual heterogeneity: Q = 19.6762 (df = 16), p-value = 0.2352 I-square statistic = 18.7% 9 studies, 18 values, 2 fixed and 3 random-effects parameters logLik AIC BIC 158.5896 -307.1793 -303.3163 BMJ Publishing Group Limited (BMJ) disclaims all liability and responsibility arising from any reliance Supplemental material placed on this supplemental material which has been supplied by the author(s) Br J Sports Med doi: 10.1136/bjsports-2021-104981–11.:10 2022;Br J Sports Med, et al. Iso-Markku P 13 Graphical prediction Linear Trend (dot-dash); Spline Model with knots at 20 and 80% of distribution (solid); Quadratic Trend (dashed) Analytical predictions PA exposure pred.lin ci.lb.lin ci.ub.lin pred.spl ci.lb.spl ci.ub.spl pred.quadr ci.lb.quadr ci.ub.quadr 203.64 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 916.36 0.95 0.88 1.02 0.84 0.73 0.96 0.89 0.78 1.02 1934.55 0.88 0.74 1.04 0.84 0.73 0.97 0.81 0.63 1.04 2952.73 0.82 0.62 1.07 0.86 0.73 1.02 0.78 0.59 1.04 3970.91 0.76 0.52 1.10 0.89 0.72 1.10 0.82 0.62 1.09 4989.09 0.70 0.44 1.13 0.91 0.69 1.19 0.93 0.64 1.34 Results - Vascular dementia (VD) Linear model including all studies (VD) Scatter plot of RRs vs original exposure variable BMJ Publishing Group Limited (BMJ) disclaims all liability and responsibility arising from any reliance Supplemental material placed on this supplemental material which has been supplied by the author(s) Br J Sports Med doi: 10.1136/bjsports-2021-104981–11.:10 2022;Br J Sports Med, et al. Iso-Markku P 14 Linear Trend Call: dosresmeta(formula = logrr ~ exposure, id = id, type = type, cases = cases, n = n, data = dataset(), se = se, covariance = input$pscorr) Two-stage random-effects meta-analysis Estimation method: REML Covariance approximation: Greenland & Longnecker Chi2 model: X2 = 14.3032 (df = 1), p-value = 0.0002 Fixed-effects coefficients Estimate Std. Error z Pr(>|z|) 95%ci.lb 95%ci.ub (Intercept) -0.0002 0.0000 -3.7820 0.0002 -0.0003 -0.0001 *** --- Signif. codes: 0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1 Between-study random-effects (co)variance components Std. Dev 0.0000 Univariate Cochran Q-test for residual heterogeneity: Q = 8.2459 (df = 7), p-value = 0.3114 I-square statistic = 15.1% 8 studies, 8 values, 1 fixed and 1 random-effects parameters logLik AIC BIC BMJ Publishing Group Limited (BMJ) disclaims all liability and responsibility arising from any reliance Supplemental material placed on this supplemental material which has been supplied by the author(s) Br J Sports Med doi: 10.1136/bjsports-2021-104981–11.:10 2022;Br J Sports Med, et al. Iso-Markku P 15 45.5563 -87.1126 -87.2208 Graphical prediction Analytical predictions PA exposure pred.lin ci.lb.lin ci.ub.lin 203.64 1.00 1.00 1.00 916.36 0.88 0.82 0.94 1934.55 0.73 0.62 0.86 2952.73 0.61 0.47 0.79 3970.91 0.51 0.36 0.72 4989.09 0.42 0.27 0.66 Models including studies with 3 or more PA exposure levels (VD) Scatter plot of RRs vs original exposure variable BMJ Publishing Group Limited (BMJ) disclaims all liability and responsibility arising from any reliance Supplemental material placed on this supplemental material which has been supplied by the author(s) Br J Sports Med doi: 10.1136/bjsports-2021-104981–11.:10 2022;Br J Sports Med, et al. Iso-Markku P 16 Linear Trend Call: dosresmeta(formula = logrr ~ exposure, id = id, type = type, cases = cases, n = n, data = dataset(), se = se, covariance = input$pscorr) Two-stage random-effects meta-analysis Estimation method: REML Covariance approximation: Greenland & Longnecker Chi2 model: X2 = 7.8907 (df = 1), p-value = 0.0050 Fixed-effects coefficients Estimate Std. Error z Pr(>|z|) 95%ci.lb 95%ci.ub (Intercept) -0.0002 0.0001 -2.8090 0.0050 -0.0004 -0.0001 ** --- Signif. codes: 0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1 Between-study random-effects (co)variance components Std. Dev 0.0000 Univariate Cochran Q-test for residual heterogeneity: Q = 1.1021 (df = 2), p-value = 0.5764 I-square statistic = 0.0% 3 studies, 3 values, 1 fixed and 1 random-effects parameters logLik AIC BIC BMJ Publishing Group Limited (BMJ) disclaims all liability and responsibility arising from any reliance Supplemental material placed on this supplemental material which has been supplied by the author(s) Br J Sports Med doi: 10.1136/bjsports-2021-104981–11.:10 2022;Br J Sports Med, et al. Iso-Markku P 17 14.2687 -24.5373 -27.1510 Spline model Call: dosresmeta(formula = logrr ~ rcs(exposure, knots), id = id, type = type, cases = cases, n = n, data = dataset(), se = se, covariance = input$pscorr) Two-stage random-effects meta-analysis Estimation method: REML Covariance approximation: Greenland & Longnecker Chi2 model: X2 = 8.1864 (df = 2), p-value = 0.0167 Fixed-effects coefficients Estimate Std. Error z rcs(exposure, knots)exposure.(Intercept) -0.0001 0.0003 -0.1959 rcs(exposure, knots)exposure'.(Intercept) -0.0001 0.0003 -0.5438 Pr(>|z|) 95%ci.lb 95%ci.ub rcs(exposure, knots)exposure.(Intercept) 0.8447 -0.0006 0.0005 rcs(exposure, knots)exposure'.(Intercept) 0.5866 -0.0007 0.0004 --- Signif. codes: 0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1 Between-study random-effects (co)variance components Std. Dev Corr rcs(exposure, knots)exposure 0.0000 rcs(exposure, knots)exposure rcs(exposure, knots)exposure' 0.0000 -0.9995 Univariate Cochran Q-test for residual heterogeneity: Q = 1.5369 (df = 4), p-value = 0.8201 I-square statistic = 0.0% 3 studies, 6 values, 2 fixed and 3 random-effects parameters logLik AIC BIC 24.6521 -39.3042 -42.3727 Quadratic trend Call: dosresmeta(formula = logrr ~ exposure + I(exposure^2), id = id, type = type, cases = cases, n = n, data = dataset(), se = se, BMJ Publishing Group Limited (BMJ) disclaims all liability and responsibility arising from any reliance Supplemental material placed on this supplemental material which has been supplied by the author(s) Br J Sports Med doi: 10.1136/bjsports-2021-104981–11.:10 2022;Br J Sports Med, et al. Iso-Markku P 18 covariance = input$pscorr) Two-stage random-effects meta-analysis Estimation method: REML Covariance approximation: Greenland & Longnecker Chi2 model: X2 = 8.5347 (df = 2), p-value = 0.0140 Fixed-effects coefficients Estimate Std. Error z Pr(>|z|) 95%ci.lb exposure.(Intercept) -0.0000 0.0003 -0.0541 0.9568 -0.0005 I(exposure^2).(Intercept) -0.0000 0.0000 -0.8025 0.4223 -0.0000 95%ci.ub exposure.(Intercept) 0.0005 I(exposure^2).(Intercept) 0.0000 --- Signif. codes: 0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1 Between-study random-effects (co)variance components Std. Dev Corr exposure 0.0000 exposure I(exposure^2) 0.0000 0.9446 Univariate Cochran Q-test for residual heterogeneity: Q = 1.2812 (df = 4), p-value = 0.8646 I-square statistic = 0.0% 3 studies, 6 values, 2 fixed and 3 random-effects parameters logLik AIC BIC 40.3790 -70.7581 -73.8266 BMJ Publishing Group Limited (BMJ) disclaims all liability and responsibility arising from any reliance Supplemental material placed on this supplemental material which has been supplied by the author(s) Br J Sports Med doi: 10.1136/bjsports-2021-104981–11.:10 2022;Br J Sports Med, et al. Iso-Markku P 19 Graphical prediction Linear Trend (dot-dash); Spline Model with knots at 20 and 80% of distribution (solid); Quadratic Trend (dashed) Analytical predictions PA exposure pred.lin ci.lb.lin ci.ub.lin pred.spl ci.lb.spl ci.ub.spl pred.quadr ci.lb.quadr ci.ub.quadr 188.48 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 424.00 0.95 0.92 0.99 0.98 0.87 1.12 0.99 0.90 1.08 895.11 0.86 0.78 0.96 0.92 0.71 1.20 0.93 0.75 1.16 1366.22 0.78 0.66 0.93 0.83 0.63 1.08 0.86 0.65 1.13 1837.33 0.71 0.56 0.90 0.74 0.56 0.98 0.76 0.57 1.01 2308.44 0.64 0.47 0.88 0.66 0.48 0.91 0.65 0.48 0.88 References 1. Blond K, Brinkløv CF, Ried-Larsen M, Crippa A, Grøntved A. Association of high amounts of physical activity with mortality risk: a systematic review and meta-analysis. Br J Sports Med. 2020 Oct 1;54(20):1195–201. 2. Walpole SC, Prieto-Merino D, Edwards P, Cleland J, Stevens G, Roberts I. The weight of nations: an estimation of adult human biomass. BMC Public Health. 2012 Jun 18;12(1):439. 3. Crippa A, Orsini N. Multivariate Dose-Response Meta-Analysis: The dosresmeta R Package. J Stat Softw. 2016 Aug 16;72:1–15. 4. Crippa A, Orsini N. Multivariate Dose-Response Meta-Analysis [Internet]. Multivariate Dose-Response Meta-Analysis. [cited 2021 Dec 22]. Available from: http://alessiocrippa.com/shiny/dosresmeta/ BMJ Publishing Group Limited (BMJ) disclaims all liability and responsibility arising from any reliance Supplemental material placed on this supplemental material which has been supplied by the author(s) Br J Sports Med doi: 10.1136/bjsports-2021-104981–11.:10 2022;Br J Sports Med, et al. Iso-Markku P