Analysis and prediction of COVID-19 for EU-EFTA-UK and other countries

With the financial support of: Daily report 17-07-2020 Analysis and prediction of COVID-19 for EU-EFTA-UK and other countries Situation report 103 Contact: clara.pra[email protected] Foreword The present report aims to provide a comprehensive picture of the pandemic situation of COVID-19 in the EU countries, and to be able to foresee the situation in the next coming days. We employ an empirical model, verified with the evolution of the number of confirmed cases in previous countries where the epidemic is close to conclude, including all provinces of China. The model does not pretend to interpret the causes of the evolution of the cases but to permit the evaluation of the quality of control measures made in each state and a short-term prediction of trends. Note, however, that the effects of the measures’ control that start on a given day are not observed until approximately 7-10 days later. The model and predictions are based on two parameters that are daily fitted to available data:  a: the velocity at which spreading specific rate slows down; the higher the value, the better the control.  K: the final number of expected cumulated cases, which cannot be evaluated at the initial stages because growth is still exponential. We show an individual report with 8 graphs and a table with the short-term predictions for different countries and regions. We are adjusting the model to countries and regions with at least 4 days with more than 100 confirmed cases and a current load over 200 cases. The predicted period of a country depends on the number of datapoints over this 100 cases threshold, and is of 5 days for those that have reported more than 100 cumulated cases for 10 consecutive days or more. For short-term predictions, we assign higher weight to last 3 points in the fittings, so that changes are rapidly captured by the model. The whole methodology employed in the inform is explained in the last pages of this document. In addition to the individual reports, the reader will find an initial dashboard with a brief analysis of the situation in EU-EFTA-UK countries, some summary figures and tables as well as long-term predictions for some of them, when possible. These long-term predictions are evaluated without different weights to datapoints. We also discuss a specific issue every day. Martí Català Pere-Joan Cardona, PhD Comparative Medicine and Bioimage Centre of Catalonia; Institute for Health Science Research Germans Trias i Pujol Clara Prats, PhD Sergio Alonso, PhD Enric Álvarez, PhD Miquel Marchena David Conesa Daniel López, PhD Computational Biology and Complex Systems; Universitat Politècnica de Catalunya - BarcelonaTech With the collaboration of: Guillem Álvarez, Oriol Bertomeu, Laura Dot, Lavínia Hriscu, Helena Kirchner, Daniel Molinuevo, Pablo Palacios, Sergi Pradas, David Rovira, Xavier Simó, Tomás Urdiales PJC and MC received funding from “la Caixa” Foundation (ID 100010434), under agreement LCF/PR/GN17/50300003; CP, DL, SA, MC, received funding from Ministerio de Ciencia, Innovación y Universidades and FEDER, with the project PGC2018-095456-B-I00; Disclaimer: These reports have been written by declared authors, who fully assume their content. They are submitted daily to the European Commission, but this body does not necessarily share their analyses, discussions and conclusions. 11 (0) Executive summary – Dashboard 22 Situation and highlights The virus is still present in EU+EFTA+UK countries. Countries started to ease the control measures a two or three months ago. Now, it is necessary to increase the surveillance again to face local outbreaks and, if necessary, to implement control measures again. New control measures may not need to be as hard as in previous months. Nevertheless, if the new spread of covid-19 is not slowed down, it is not unlikely to have to take major confinement measures again. Currently 24 countries have a ρ7 greater than 1, from Denmark with 1.96 and Latvia 1.93 to Austria with 1.01. Only 5 countries have a ρ7 below 1 (Italy, Czech Republic, Sweden, Slovenia, Finland). Sweden seems to be improving, although the reported data show some days without information, despite still being the country with the highest A14 with 472 active cases per 100,000 inhab. Luxembourg has an A14 of 286 per 100,000 inhabitants, which places the country at the second in the A14 rank, but its ρ7 of 1.32 situates it at high level of risk. A14 EPG ρ7 Cumulative incidence 33 (1) ρ7 is the average of 7 consecutive ρ, but can still fluctuate. (2,3) EPG stands for Effective Growth Potential. EPGREP is the product of attack-rate of last 14 days per 105 inhabitants by ρ7 (empiric reproduction number). EPGEST is the product of estimated real attack-rate of last 14 days per 105 inhabitants and ρ7. Biocom-Cov degree is an epidemiological situation scale based on the level of last week’s mean daily new cases (https://upcommons.upc.edu/handle/2117/189661, https://upcommons.upc.edu/handle/2117/189808). Situation and trends per country Table of current situation in EU countries. Colour scale is relative except when indicated, this means that it is applied independently to each column, and distinguishes best (green) form worst (red) situations according to each of the variables. Last column (EPGEST) is assessed with estimated real 14-day attack rate (see report from 22/04 for details). EPGREP is calculated with data reported by countries. EPGREP and EPGEST cannot be compared between them because scales are different, but can be independently used for estimating risk of countries according to reported or estimated real situation, respectively. Data from 2nd July. Disclaimer: estimated active cases and estimated 14-day attack rate are assessed by assuming a lethality of 1 % (see report from 20 to 24 April, #37-41). This value can change in countries where suspicious deaths are reported as well (real values would be lower) and in countries where incidence among elderly people was minor (real values would be higher) 44 Analysis: Dynamics of new outbreaks in three Catalan cities (I). European countries are, in general, dealing with a similar situation: most of them have successfully overcome the first wave, and they are now trying to extinguish local outbreaks that are appearing in their regions. The strategy is clear: test and trace while incident cases are low, and new restrictions when certain thresholds are overcome. These thresholds may vary from country to country, as they are mainly determined by the testing and tracing capacities. This means that daily testing level is important (i.e., number of PCR tests that can be performed per day and per 100,000 inhabitants), but that the number of available health workers to carry out the tracing and isolation of index cases’ contacts is important as well. This is the only way to break transmission chains, one by one. The epidemics in Lleida, L’Hospitalet and Barcelona The situation in Catalunya (Spain) has worsen since countrylevel restrictions were fully removed, on 21st June. The deescalation process started on 2nd May and took almost 2 months, during which the restrictions were gradually eased in a heterogeneous manner, depending on the situation of each region. Last weeks, a region in Western Catalunya (Segrià) started showing symptoms of significant growth. The capital of this region is Lleida, with almost 140,000 inhabitants. Last week, two most populated cities have also started showing a change in previous control trend: L’Hospitalet de Llobregat (265,000 inhabitants) and Barcelona (1,640,000 inhabitants). Figure 1 shows the evolution of the three cities since middle-March, in terms of 7-day cumulative incidence. We also indicate the days at which the process of gradual easing of restrictions started and the moment at which it finished. Figure 1: Evolution of the 7-day cumulative incidence in the Catalan cities of Lleida, L’Hospitalet de Llobregat and Barcelona, together with the starting and end days of the de-escalation process. As shown, the three cities had successfully overcome the first wave when the de-escalation process started. During the process, Lleida was allowed to de-escalate faster at an initial stage, but this was slowed down when a new increase was observed. Once in the final phase of the de-escalation, Lleida showed new 55 symptoms of growth that become uncontrolled in a few days. This was discussed in a previous report1. After a couple of weeks, also L’Hospitalet de Llobregat started showing symptoms of a new growth. Last week, Barcelona has started showing an increase in new cases as well. The increase in testing capacity: more and milder cases are diagnosed It is worth to mention that the 7-day cumulative incidences that are currently being achieved cannot be directly compared with those of the first wave. The testing capacity has increased a factor 4, in Catalunya (Figure 2). Therefore, the same number of reported cases indicate a different epidemiological situation. If the diagnostic rate was between 5-10% in March-April, it has raised up to 20-30% currently. This increase in testing ratio could also explain the generalized mild symptoms of current new cases. In March-April, only serious cases were diagnosed. The front line were hospitals, and PCRs were mostly performer there. Now, diagnosis capacity has been mainly transferred to primary care points, and only serious cases are redirected to hospitals. Therefore, current increase in new cases is affecting those primary care facilities. In these towns, many of them are collapsed. The epidemics in these cities through the index Effective Potential Growth (EPG) We have successfully used the EPG index to analyze the epidemiological situation of regions and countries. This index is the product between empiric reproduction number (𝜌𝜌7), which is a measurement of the rate at which the epidemic is propagating, and 14-day cumulative incidence (𝐴𝐴14), which is a measurement of the number of active cases (contagious people). As discussed in previous reports, we have situated the threshold for high epidemiological risk at an EPG = 100. This level accounts for an expected growth that would overcome the test and trace capacity in most European countries. Figure 3 shows the evolution of this index in the three cities during the whole epidemic. The EPG = 100 level is also indicated, and the intermediate growth in Lleida that required the slowing down of the de-escalation process is shown as well. Figure 3: Evolution of EPG in the Catalan cities of Lleida, L’Hospitalet de Llobregat and Barcelona, together with the starting and end days of the de-escalation process. 1 https://upcommons.upc.edu/handle/2117/192557 Figure 2: Weekly number of PCR tests per 1,000 inhabitants in Catalunya. 66 EPG dynamics and new restriction measures Figure 3 strengthens the validity of the EPG = 100 threshold. It is not an absolute on-off threshold, but it delimits the control-uncontrol zones pretty well. When the EPG reaches this level, the probability of significant growth increases. Lleida’s EPG is less robust because of lower population, which makes this index to be more sensitive to smaller changes. Let us zoom in the last month (Figure 4). The situation in the three cities has required the intervention of authorities for implementing new restrictions, once they realized that community transmission was present. On 7th July, the Catalan government implemented a safe perimeter around Lleida’s county (Segrià) which forbids the movement of people in and out for reason other than work, because some exported cases had been detected in other Catalan regions. This did not stop the worsen of the situation. Therefore, on 14th July the government implemented a set of measures regarding mass gathering prevention, internal mobility and restaurants capacity, among others. Similar measures were applied on 14th in L’Hospitalet de Llobregat, after a few days with EPG>100. These measures have been extended today to the whole metropolitan area around Barcelona, including the capital. Legislation is still not ready, and most of those control measures remain as recommendations, waiting for their approval by justice services. Figure 4: Evolution of EPG in the Catalan cities of Lleida, L’Hospitalet and Barcelona since beginning of June. Grey dotted line indicates the end of the de-escalation process. Colored dashed lines indicate first restriction measures in each city. The threshold EPG = 100 is also indicated. As shown in Figure 4, perimetric confinement of Lleida’s county has not managed to control the growth. In fact, this measure was applied 17 days after the overcoming of EPG = 100, which was probably too late. The next measure, which limits mobility and meetings, was applied 24 days after that point. It will probably still take another week for the effects to be seen in reported data. In L’Hospitalet de Llobregat, measures were taken before: only 9 days after the overcoming of the threshold. Finally, Barcelona have only been 3 days above EPG = 100 before measures have been implemented. Next days we will be able to observe the delay between these measures and their effect, as well as the magnitude of such effect, which will obviously depend on the current situation. It is also interesting to compare the evolution of these cities once the EPG = 100 threshold is overcome. Figure 5 situates time origin at the day at which EPG gets higher than 100 for each city. Then, we can visualize how the EPG starts a significant increase from that point. In fact, the evolution of L’Hospitalet de Llobregat is being similar to that one of Lleida but with a delay of 2 weeks. Nevertheless, the population density in 77 L’Hospitalet is huge, and this could accelerate the propagation of the epidemic the next days, as it is insinuated by last points. The evolution of Barcelona cannot be observed yet, but it could be hopefully modulated by the effect of earlier control measures. Figure 5: Evolution of EPG in the Catalan cities of Lleida, L’Hospitalet de Llobregat and Barcelona before and after the overcoming of EPG = 100 threshold. New outbreaks in the risk diagrams Risk diagrams are a good way to visualize the aforementioned dynamics, as well. Next, we show the risk diagrams of these cities for the last month (Figure 6). Background color is set according to the EPG scale, situating the red zone where EPG > 100. It can be observed how Lleida has spent more than 3 weeks in the red zones and no improvement symptoms are shown yet. L’Hospitalet de Llobregat entered the red zone 10 days ago, while Barcelona has only spent 3 days in the risk zone. It is clear again that, once in the red zone, the situation worsens rapidly. Figure 6: Risk diagrams of Lleida, L’Hospitalet de Llobregat and Barcelona corresponding to last 30 days. Final conclusions The situation in Catalunya is worrying. There was a generalized idea among population about a new outbreak coming in October, but it seems that the necessary material and personal resources to overcome summer’s outbreaks were not ready. Different estimations point to the need for 2,000 health workers and trackers fully devoted to the test and trace strategy, but primary health care centers do not have enough means to face current situation. At present, hospitalizations are still low in Barcelona and L’Hospitalet de Llobregat, but the main hospital in Lleida is already working with 3 floors entirely devoted to Covid-19. It is expected that earlier control measures in Barcelona prevent serious cases to increase significantly. 88 Data obtained from https://www.ecdc.europa.eu/en/geographical-distribution-2019-ncov-cases (1) Analysis and prediction of COVID-19 for EU+EFTA+UK 1515 1616 1717 1818 1919 2020 2121 2222 2323 2424 3131 3232 3333 3434 3535 3636 3737 3838 3939 4040 Data obtained from https://www.ecdc.europa.eu/en/geographical-distribution-2019-ncov-cases (2) Analysis and prediction of COVID-19 for other countries 4747 4848 4949 5050 5151 5252 5353 5454 5555 5656 6363 6464 6565 6666 6767 6868 6969 7070 7171 Data updated on 17th July, data series built with the day of the symptoms’ onset, reliable until 10th July. Data obtained from https://github.com/datadista/datasets/tree/master/COVID%2019 and https://covid19.isciii.es/ (3) Analysis and prediction of COVID-19 for Spain and its autonomous communities 7272 7979 8080 8181 8282 8383 8484 8585 8686 8787 8888 95 96 97 98 99 100 101 102 103 104 111 112 113 114 Methods 115 Methods (1) Data source Data are daily obtained from World Health Organization (WHO) surveillance reports2, from European Centre for Disease Prevention and Control (ECDC)3 and from Ministerio de Sanidad4. These reports are converted into text files that can be processed for subsequent analysis. Daily data comprise, among others: total confirmed cases, total confirmed new cases, total deaths, total new deaths. It must be considered that the report is always providing data from previous day. In the document we use the date at which the datapoint is assumed to belong, i.e., report from 15/03/2020 is giving data from 14/03/2020, the latter being used in the subsequent analysis. (2) Data processing and plotting Data are initially processed with Matlab in order to update timeseries, i.e., last datapoints are added to historical sequences. These timeseries are plotted for EU individual countries and for the UE as a whole: Number of cumulated confirmed cases, in blue dots Number of reported new cases Number of cumulated deaths Then, two indicators are calculated and plotted, too: Number of cumulated deaths divided by the number of cumulated confirmed cases, and reported as a percentage; it is an indirect indicator of the diagnostic level. ρ: this variable is related with the reproduction number, i.e., with the number of new infections caused by a single case. It is evaluated as follows for the day before last report (t-1): 𝜌𝜌(𝑡𝑡−1)=𝑁𝑁𝑛𝑛𝑛𝑛𝑛𝑛(𝑡𝑡)+𝑁𝑁𝑛𝑛𝑛𝑛𝑛𝑛(𝑡𝑡 −1)+𝑁𝑁𝑛𝑛𝑛𝑛𝑛𝑛(𝑡𝑡−2) 𝑁𝑁𝑛𝑛𝑛𝑛𝑛𝑛(𝑡𝑡−5)+𝑁𝑁𝑛𝑛𝑛𝑛𝑛𝑛(𝑡𝑡 −6)+𝑁𝑁𝑛𝑛𝑛𝑛𝑛𝑛(𝑡𝑡 −7) where Nnew(t) is the number of new confirmed cases at day t. (3) Classification of countries according to their status in the epidemic cycle The evolution of confirmed cases shows a biphasic behaviour: (I) an initial period where most of the cases are imported; (II) a subsequent period where most of new cases occur because of local transmission. Once in the stage II, mathematical models can be used to track evolutions and predict tendencies. Focusing on countries that are on stage II, we classify them in three groups: •Group A: countries that have reported more than 100 cumulated cases for 10 consecutive days or more; •Group B: countries that have reported more than 100 cumulated cases for 7 to 9 consecutive days; •Group C: countries that have reported more than 100 cumulated cases for 4 to 6 days. 2 https://www.who.int/emergencies/diseases/novel-coronavirus-2019/situation-reports 3 https://www.ecdc.europa.eu/en/geographical-distribution-2019-ncov-cases 4 https://www.mscbs.gob.es/profesionales/saludPublica/ccayes/alertasActual/nCov-China/situacionActual.htm https://github.com/datadista/datasets/tree/master/COVID%2019 , https://covid19.isciii.es/ 116 (4) Fitting a mathematical model to data Previous studies have shown that Gompertz model5 correctly describes the Covid-19 epidemic in all analysed countries. It is an empirical model that starts with an exponential growth but that gradually decreases its specific growth rate. Therefore, it is adequate for describing an epidemic that is characterized by an initial exponential growth but a progressive decrease in spreading velocity provided that appropriate control measures are applied. Gompertz model is described by the equation: 𝑁𝑁(𝑡𝑡) = 𝐾𝐾 𝑒𝑒−𝑙𝑙𝑛𝑛 �𝐾𝐾 𝑁𝑁0�· 𝑛𝑛− 𝑎𝑎·(𝑡𝑡−𝑡𝑡0) where N(t) is the cumulated number of confirmed cases at t (in days), and N0 is the number of cumulated cases the day at day t0. The model has two parameters: a is the velocity at which specific spreading rate is slowing down; K is the expected final number of cumulated cases at the end of the epidemic. This model is fitted to reported cumulated cases of the UE and of countries in stage II that accomplish two criteria: 4 or more consecutive days with more than 100 cumulated cases, and at least one datapoint over 200 cases. Day t0 is chosen as that one at which N(t) overpasses 100 cases. If more than 15 datapoints that accomplish the stated criteria are available, only the last 15 points are used. The fitting is done using Matlab’s Curve Fitting package with Nonlinear Least Squares method, which also provides confidence intervals of fitted parameters (a and K) and the R2 of the fitting. At the initial stages the dynamics is exponential and K cannot be correctly evaluated. In fact, at this stage the most relevant parameter is a. Fitted curves are incorporated to plots of cumulative reported cases with a dashed line. Once a new fitting is done, two plots are added to the country report: Evolution of fitted a with its error bars, i.e., values obtained on the fitting each day that the analysis has been carried out; Evolution of fitted K with its error bars, i.e., values obtained on the fitting each day that the analysis has been carried out; if lower error bar indicates a value that is lower than current number of cases, the error bar is truncated. These plots illustrate the increase in fittings’ confidence, as fitted values progressively stabilize around a certain value and error bars get smaller when the number of datapoints increases. In fact, in the case of countries, they are discarded and set as “Not enough data” if a>0.2 day-1, if K>106 or if the error in K overpasses 106. It is worth to mention that the simplicity of this model and the lack of previous assumptions about the Covid- 19 behaviour make it appropriate for universal use, i.e., it can be fitted to any country independently of its socioeconomic context and control strategy. Then, the model is capable of quantifying the observed dynamics in an objective and standard manner and predicting short-term tendencies. (5) Using the model for predicting short-term tendencies The model is finally used for a short-term prediction of the evolution of the cumulated number of cases. The predictions increase their reliability with the number of datapoints used in the fitting. Therefore, we consider three levels of prediction, depending on the country: 5 Madden LV. Quantification of disease progression. Protection Ecology 1980; 2: 159-176. 117 •Group A: prediction of expected cumulated cases for the following 3-5 days6; •Group B: prediction of expected cumulated cases for the following 2 days; •Group C: prediction of expected cumulated cases for the following day. The confidence interval of predictions is assessed with the Matlab function predint, with a 99% confidence level. These predictions are shown in the plots as red dots with corresponding error bars, and also gathered in the attached table. For series longer than 9 timepoints, last 3 points are weighted in the fitting so that changes in tendencies are well captured by the model. (6) Estimating non-diagnosed cases Lethality of Covid-19 has been estimated at around 1 % for Republic of Korea and the Diamond Princess cruise. Besides, median duration of viral shedding after Covid-19 onset has been estimated at 18.5 days for non-survivors7 in a retrospective study in Wuhan. These data allow for an estimation of total number of cases, considering that the number of deaths at certain moment should be about 1 % of total cases 18.5 days before. This is valid for estimating cases of countries at stage II, since in stage I the deaths would be mostly due to the incidence at the country from which they were imported. We establish a threshold of 50 reported cases before starting this estimation. Reported deaths are passed through a moving average filter of 5 points in order to smooth tendencies. Then, the corresponding number of cases is found assuming the 1 % lethality. Finally, these cases are distributed between 18 and 19 days before each one. 6 At this moment we are testing predictions at 4 days for countries with more than 100 cumulated cases for 13-15 consecutive days, and 5 days for 16 or more days. 7 Zhou et al., 2020. Clinical course and risk factors for mortality of adult inpatients with COVID-19 in Wuhan, China: a retrospective cohort study. The Lancet; March 9, doi: 10.1016/S0140-6736(20)30566-3 118