key: cord-0921386-dp4os8p7
authors: Roux, J.; Massonnaud, C.; Colizza, V.; Cauchemez, S.; Crepey, P.
title: Impact of national and regional lockdowns on COVID-19 epidemic waves: Application to the 2020 spring wave in France
date: 2021-04-25
journal: nan
DOI: 10.1101/2021.04.21.21255876
sha: 7554ec72d07e69db5bf96d4ad64eee7533f8a351
doc_id: 921386
cord_uid: dp4os8p7

Background The efficacy of national lockdowns to control COVID-19 epidemics has been demonstrated. This study aimed at assessing the impact of national and regional lockdowns, in the context of quickly growing pandemic waves, considering the French first wave of the COVID-19 epidemic as a case study. Methods We developed a compartmental epidemic model considering the demographic and age profile of the population of the 13 regions of metropolitan France. We assessed the impact on morbidity, mortality, and hospital resources of simulated national and regional lockdowns starting at different time. Results In a regional lockdown scenario aimed at preventing intensive care units (ICU) saturation in continental France in March 2020, almost all regions would have had to implement a lockdown within 10 days from the actual date of the nationwide implementation. By this date, 97% of ICU capacities would have been used and almost 7000 more lives would have been lost, compared to the March 17 lockdown which limited the mortality burden in hospital to 18 130 deaths. For slowly growing epidemics, with a lower reproduction number, the expected delays between regional lockdowns increases. However, the public health costs associated with these delays tend to grow exponentially with time. Conclusions In a quickly growing pandemic wave, defining the timing of lockdowns at a regional rather than national level delays by a few days the implementation of a nationwide lockdown but leads to substantially higher morbi-mortality and stress on the healthcare system.

After it was first identified in December 2019, the novel severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) that causes the 2019 coronavirus disease quickly spread over the world, challenging health systems. To contain this pandemic, countries implemented control strategies of different intensities for example travel restrictions, closing of schools, shops, stay-at-home orders and lockdowns, at national or regional levels. [1] [2] [3] Combinations of these interventions led to important reductions of the effective reproduction number of the epidemic in a number of countries before the Summer 2020. 1, 2, 4, 5 France was one of the European countries most affected by the first pandemic wave, along with Italy, Spain and the UK. The first cases of COVID-19 due to importations were confirmed on January 24, 2020. 6 They were followed by several clusters mostly located in the eastern regions of France (Grand-Est and Bourgogne-Franche-Comté) and Ile-de-France (the most populated French region including the city of Paris). The seeding of the epidemic in these regions mean that they were ahead of other regions in terms of the number of COVID-19 patients admitted into intensive care units (ICU) (mean of 53 and 31 daily ICU admissions from March 11 to March 17 in Ile-de-France and Grand-Est, respectively, compared to 1 to 12 in other French regions). 5, 7 To avoid saturation of the healthcare system, the French government ordered a national lockdown starting on March 17, 2020.

The efficacy of lockdown strategies to control COVID-19 epidemics has been internationally demonstrated at the national 1, 2, 5, 8, 9 and county levels. 10 However, some have argued that the nationwide application in France of such measures was unnecessary and that lockdowns restricted to the two or three most impacted French regions would have been sufficient to contain the pandemic wave. The present study aimed at retrospectively comparing the impact of the nationwide lockdown on March 17 to regional lockdowns on the number of COVID-19 hospitalizations, occupied ICU beds, life-years gained, and deaths avoided during the first pandemic wave in France.

. CC-BY-NC 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted April 25, 2021. ; https://doi.org/10.1101/2021.04.21.21255876 doi: medRxiv preprint

We used a deterministic, age-structured, compartmental epidemic model based on demographic and age profile of the population of the 13 administrative regions of metropolitan France. Model description, equations, and parameter values are presented in supplementary materials. The transmission model was implemented in C++. Data collection, data management, simulations, results analysis and reporting were performed using R. 11

For each region, we estimated the reproduction number before the lockdown (Rprelockdown), and after its implementation (Rlockdown). Then, in order to simulate a regional lockdown, we used the Rprelockdown up to the starting date of the measure; and the Rlockdown estimated in the region after that date. We assumed that the transmission rate remained constant until July 1. We also assumed that the epidemic dynamic was driven by local behaviour within regions and that interactions between regions had no impact on local reproduction numbers.

We simulated national lockdowns starting on all dates from March 10 to March 31, thus exploring other dates of start beyond the actual date of March 17. We then simulated asynchronous regional lockdowns, in which each region could be locked down independently of other regions, on different dates. We considered the following epidemic thresholds as starting dates of regional lockdowns: either (1) the date on which the estimated incidence of hospital admissions in the region reached the level of the most affected region during the first wave (i.e. Grand-Est, here called GES threshold), or (2) the last date on which a lockdown would allow the region to stay below its ICU capacity limit (here called ICU capacity threshold).

Outcomes . CC-BY-NC 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) For each simulation, we computed at the national and regional levels the cumulative numbers of hospital and ICU admissions, maximum occupied ICU beds, deaths, life years and quality-adjusted life years (QALY) lost between March 1 and July 1 and computed the relative change compared to the outcomes obtained when the real lockdown date was used.

To better understand how the reproduction number affect the estimated impact of the lockdown, we analysed scenarios where Rprelockdown is in the range 1.1 to 1.5. This corresponds to the reproduction numbers observed in France at the start of 2021. In this analysis, we kept the previously estimated Rlockdown for each region during the lockdown. 

At the national level, setting a national lockdown one day later than the observed lockdown, i.e on March 18, would have led to an increase of 1445 ICU admissions, the need for 820 additional ICU beds, and 1839 deaths (i.e. 2 ICU admissions and 3 deaths per 100 000 inhabitants); corresponding to a loss of almost 16 600 QALY (Figure 1 -regional results are presented in Supplementary Tables S1 and S2 and Supplementary Figure S5 ). It would have led to an increase in the maximum number of occupied ICU beds at the peak between 8% and 21% at the regional level ( Figure 2) 

The first scenario, where the start of the lockdown is delayed in each region according to the GES threshold (the date on which the estimated incidence of hospital admissions reached 6.09 per 100 000 . CC-BY-NC 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) Figure S9 ). This scenario would have led to the saturation of ICU capacities in some regions ( Figure 2 ).

In the scenario where each region implements a lockdown no later than the moment allowing them to remain within its ICU capacity (ICU capacity threshold), three regions (Grand-Est, Ile-de-France and Bourgogne-Franche-Comté) would have had to be locked down from the start of the study period, on March 17, and one region (Bretagne) could have been locked down three weeks later on April 2, 2020.

However, all regions but Bretagne would have had to implement a lockdown by March 27, 2020. By this date, a maximum of 96.8% of national ICU capacities would have been used, instead of less than 70% with the observed national lockdown 10 days earlier, and almost 7000 more lives would have been lost ( Figure 3 ).

The sensitivity analysis showed that the lockdown impact depends not only on timing but also on the pre-lockdown reproduction number Rprelockdown observed in the region (Supplementary Figure S10 ).

For example, delaying the national lockdown two weeks later on March 31 would have led to 1218 additional deaths with Rprelockdown=1.1 but 5439 additional deaths with Rprelockdown=1.5. In addition, for any given reproduction number, the cost of delaying the lockdown increases exponentially with time, for all outcomes. As an example, a synchronized lockdown on March 10 with a reproduction number equal to 1.5 would have led to 1896 fewer deaths, whereas on March 24 it would have led to 2421 additional deaths.

. CC-BY-NC 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted April 25, 2021. This study aimed at comparing the impact of lockdown strategies where the timing of the lockdown is defined at the national vs at the regional scale, in the context of a quickly growing pandemic wave.

This was done thanks to a compartmental model that reproduced the epidemic dynamics of the Spring 2020 pandemic wave in France and simulated several counterfactual scenarios to that of the nationwide lockdown implemented on March 17.

Using our model, we estimated that, at the time of the first national lockdown, all regions were not at the same epidemic stage. Indeed, the threshold of 6.09 hospital admissions per 100,000 inhabitants This delay would have resulted in a 7-fold increase of deaths in some regions (Nouvelle-Aquitaine) compared to the observed outcome. Moreover, these numbers do not take into account the additional deaths due to healthcare system congestion. Therefore, it is likely that the lockdown would have been implemented before the saturation of all regions. On the other hand, starting the national lockdown one week earlier would have resulted into up to 9194 fewer deaths, corresponding to around 83 000 QALY lost at the national level. The reduction was higher in regions where the reproduction number was high, highlighting the importance of this parameter in the decision to start a lockdown.

We also analysed the impact of regional lockdowns. Using a threshold based on the number of hospital admissions to trigger a regional lockdown failed to prevent regions from exceeding their ICU capacity. This is easily explained by different ICU capacities per inhabitant by region, and because the impact of lockdown timing depends not only on the incidence but also on the epidemic dynamics, described by the reproduction number R. Depending on R, the same level of incidence in two regions on a given day can lead to very different dynamics in the following days, as shown in the scenario with a slower . CC-BY-NC 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted April 25, 2021. ; https://doi.org/10.1101/2021.04.21.21255876 doi: medRxiv preprint epidemic growth. The higher the reproduction number, the higher the cost of delaying the lockdown.

Our simulations showed that regional lockdowns, where each lockdown is started on the last date that allows the region to stay within its ICU capacity, would have led to 7000 additional deaths compared to the observed lockdown. In this scenario, all regions but one would have been locked down within 10 days. Although each region would have stayed within its ICU capacities, 97% of national ICU capacities would have been used at peak. These assessments are performed retrospectively. In realtime, it would be difficult to precisely estimate the date when the lockdown should be implemented to avoid the saturation of local ICUs. As a consequence, there is a real risk that the regional approach would have led to the saturation of ICUs in some locations.

Considering the latest information on SARS-CoV-2 and its characteristics, our estimated values of reproduction number were similar to the ones obtained by other researchers before and after the lockdown. 4, 12, 13 Regarding the computation of outcomes of interest (hospital requirements and deaths), we considered the whole care pathway of patients inside hospital settings ( Figure S13 ). In addition, all these estimations were based on observed parameters (lengths of stay, ICU admission risk, and death risk) in real settings and modulated for each region with specific estimated coefficients to account for regional disparities. However, some differences may remain, mainly due to change in hospitalization strategies during the crisis. Concerning the quantification of the impact of a delayed lockdown, we chose Grand-Est as the reference region, as it was the first impacted region of the first wave of the epidemic in France, and thus the number of hospitalizations in this region at lockdown start was a good threshold for comparisons. 5 The mortality estimates presented in this study do not consider deaths occurring outside hospitals (at home -estimated to be around 5% of all COVID-related deaths 14 -or in nursing homes -10 497 deaths recorded as of July 2, 2020) 15 as we limited our analysis to hospital settings. Moreover, our model also does not account for the excess mortality that would have resulted from hospital saturation. Given the important excess hospital and ICU admissions that would have occurred with a delayed lockdown, it is likely that the saturation of the health care system would have resulted in an important excess . CC-BY-NC 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted April 25, 2021. ; https://doi.org/10.1101/2021.04.21.21255876 doi: medRxiv preprint 11 mortality. This was avoided in France thanks to the transfer of 658 patients from highly affected to less affected regions, which remained less infected thanks to the national lockdown (personal communication). Therefore, our analysis may underestimate the number of deaths, life years and QALY lost in scenarios where delays in the implementation of the lockdown increase substantially the stress on the healthcare system. In addition, QALY lost were only estimated based on the deaths of infected individuals, whereas it would be interesting to take into account the lost QALY due to severe COVID-19 leading to hospitalization and ICU admissions, since there may still be after-effects for these individuals. 16 However, to our knowledge, no study has been done in France to provide utility measures in these cases. Regarding the estimation of ICU occupancy, the maximum ICU capacity in each region corresponded to the maximum number of beds that could be mobilised during the pandemic, which was higher than the normal number of ICU beds at the national level. Without this mobilisation of new ICU beds, ICU saturation would have been reached much earlier. Finally, our study relied on the hypothesis that implementing regionals lockdowns is doable and acceptable by the population. Some European countries such as Italy, 3,17 Germany 3 or United Kingdom 18 have implemented such measures.

In France, spatially targeted measures were introduced only after the first wave and for less restrictive measures than a full lockdown. 19 On one hand it allows to limit the implementation of strong control measures in areas where the infection risk is high, and potentially acknowledged by the population.

On the other hand, it raises questions regarding the consequences related to population movements that may occur following or preceding the lockdown of a region. 10 In addition, some studies have shown that localized mitigation measures, less strict than national lockdown (curfew in specific metropolitan area, closing of shops or schools), have led to a reduction in the French population mobility 20 and a slowdown of the epidemic in non-restricted areas. 19 This study highlights that the impact of a delayed regional lockdown greatly depends on the reproduction number in the region, with costs in terms of morbi-mortality growing exponentially with the delay. It also shows that in a quickly growing pandemic wave such as the one observed in Spring 2020 in France, defining the timing of lockdowns at a regional rather than national level delays by only . CC-BY-NC 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted April 25, 2021. ; https://doi.org/10.1101/2021.04.21.21255876 doi: medRxiv preprint 12 a few days the implementation of a nationwide lockdown but leads to substantially higher morbimortality and stress on the healthcare system. 

The data and code underlying this article will be shared on reasonable request to the corresponding author.

None declared.

. CC-BY-NC 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted April 25, 2021. The last date on which a lockdown would allow the region to stay below its ICU capacity limit. . CC-BY-NC 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. . CC-BY-NC 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted April 25, 2021. ; https://doi.org/10.1101/2021.04.21.21255876 doi: medRxiv preprint Figure 3 ICU beds occupancy in case of regional lockdowns to avoid regional ICU saturation. Vertical dashed lines indicate the moment when regions have to be locked down to avoid a saturation of their ICU capacity. Three regions (GES, IDF, BFC) would be locked from the start of the study period. One region (BRE) could have been locked on April 2, 2020, all others would have been locked before March 27, 2020.

We used a deterministic, age-structured, compartmental epidemic model based on demographic and age profile of the population administrative regions of metropolitan France (Supplementary Figure   S11 ). We considered that individuals were susceptible (S), and then potentially exposed to the virus but not infectious (E). As observed in clinical practice, we set children and adolescents to be less susceptible to infection than other age groups. 21 Exposed individuals stay in their compartment for an average of 2.72 days 22 before moving to either asymptomatic (A) or pre-symptomatic (Ips) compartment according to observed risk of being asymptomatic. 21 We considered that asymptomatic individuals were 45% less infectious than pre-symptomatic individuals were 23 and we assumed they stayed in the compartment for an average of 10.91 days before moving to the removed compartment (R) of patients that are cured or dead from COVID-19. Pre-symptomatic individuals will become infected symptomatic (Is) after an average duration of 2.38 days. This choice of parameters gave a mean incubation period of 5.1 days, 1 within around 2 days of pre-symptomatic transmissions. Then they become after an average of 5 days (assumption) either hospitalized (Ih) or remain within the community (Inh) according to age-dependent hospitalization risks. 4 We assumed that hospitalized individuals were no more infectious because of their hospitalization whereas non-hospitalized keeps spreading the disease with the same intensity as symptomatic individuals. We assumed the average duration in Inh compartment was 3.53 days to match the total duration in the asymptomatic compartment. Finally, both hospitalized and non-hospitalized individuals moved to the removed compartment.

To match the epidemic dynamics and reproduce the Erlang distributions of durations in each compartment of the transmission model, we subdivided each compartments having a role in the . CC-BY-NC 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted April 25, 2021. ; https://doi.org/10.1101/2021.04.21.21255876 doi: medRxiv preprint infection process into sub-compartments. This subdivision had no impact on the mean duration spent in each compartment. A sensitivity analysis testing several combinations revealed that 10 subcompartments in each compartment permitted to approach the observed dynamics of the epidemic (Supplementary Figure S12) .

The population was divided into 17 age groups: 16 age-band of 5 years from 0 to 80 years, and a last group for people aged 80 years and older. The population structure of each region was inferred from hospital catchment areas from 2016 and 2017 census data provided by the French National Institute of Statistics and Economic Studies (Insee). 24, 25 To simulate age and location-dependent mixing, we used inter-individual contacts matrices for the French population estimated by Prem et al. 26 We retrieved epidemiological regional data related to the COVID-19 epidemic in metropolitan France gathered by the French National Public Health Agency (SpF-'Santé publique France') 7 daily number of hospital admissions (general and intensive care unit (ICU) wards), daily number of ICU admissions, daily number of occupied ICU beds, and daily number of deaths in hospitals (deaths in nursing homes and at home were not considered). All these epidemiological data were corrected for reporting delays following the same procedure as Salje et al. 4 We also obtained data on the maximum ICU beds capacity per French region from the 'Direction de la Recherche, des Études, de l'Évaluation et des Statistiques' Table S3 ).

Based on the estimated number of new infected hospitalized cases per day provided by our epidemiological model, we inferred outcomes related to hospital requirements, namely ICU admissions, ICU occupied beds, and deaths.

. CC-BY-NC 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted April 25, 2021. ; https://doi.org/10.1101/2021.04.21.21255876 doi: medRxiv preprint 20 We divided the hospital settings in two parts: general ward and ICU ward (Supplementary Figure S13) . The epidemiological transmission model estimated the daily number of new hospitalized cases due to COVID-19 infection, regardless of the ward (i.e. ICU and general wards). Once admitted to hospital, infected cases could either remain in the general ward until the end of their stay or go into ICU, if they became severe cases. We assumed that cases admitted in ICU entered ICU ward the same day as they were admitted in hospital (pre-ICU length of stay equal to 0 day). Once in ICU, cases could either die or stay in ICU until their discharge to general ward. Cases in general ward could either die or stay in general ward until their discharge to home.

We used age-dependent ICU admission risks for hospitalized patients estimated by Salje et al. 4 We also used age-specific lengths of stay in ICU to estimate the number of occupied ICU beds. 27 We estimated the number of deaths using hospital and ICU death risks estimated by the Drees on all the hospitalized cases of the first wave of epidemic in France (March-June 2020). 27 Deaths were delayed in time using the time from hospital or ICU admission to death. 27 Lengths of stay in general ward before discharge, lengths of stay in general ward before death and post-ICU lengths of stay in general ward were not used as we did not estimated the total number of hospital beds needed. This had no impact on the results provided by the model.

We also estimated the number of life years and quality-adjusted life years (QALY) lost for each death using life tables provided by INSEE for 2012-2016 28 and utility measures of each age-group in France. 29 

We estimated region-specific model parameters by maximum likelihood in a two-step process using the bbmle R package. 30 First, on the period stretching from March 14 to May 10, 2020, corresponding to the evolution of the epidemic until the end of the national lockdown, we estimated the value of the transmission parameter β, governing the value of R0, the initial state in each compartment per age group on March 1, 2020 and the effects of the national lockdown. The latter was estimated through a transmission reduction parameter, hereafter called , and multiplied the transmission parameter . CC-BY-NC 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted April 25, 2021. ; https://doi.org/10.1101/2021.04.21.21255876 doi: medRxiv preprint 21 (and thus contact matrices) to reproduce the several mitigation measures implemented and their consequences on the regional propagation. We jointly estimated these parameters by fitting the daily de-seasonalized time series of hospital admissions (hereafter denoted Hosp) using the likelihood defined in Equation 1.

where NBin(.|X) is a negative binomial distribution of mean X and overdispersion , being a parameter specific to each region to be estimated. Confidence intervals of these two parameters were estimated using likelihood profiling methods. 30 In a second step, we jointly estimated three regional coefficients adjusting age-specific risks of ICU where NBin(.|X) is a negative binomial distribution of mean X and overdispersion , being a parameter estimated for each region.

. CC-BY-NC 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) 

Estimating the effects of non-pharmaceutical interventions on COVID-19 in Europe

The temporal association of introducing and lifting nonpharmaceutical interventions with the time-varying reproduction number (R) of SARS-CoV-2: a modelling study across 131 countries

Heuvel ER van den. How did governmental interventions affect the spread of COVID-19 in European countries? BMC Public Health

Estimating the burden of SARS-CoV-2 in France. Science (80-)

Lockdown impact on COVID-19 epidemics in regions across metropolitan France

First cases of coronavirus disease 2019 (COVID-19) in France: surveillance, investigations and control measures

Centre for Disease Prevention and Control

Données hospitalières relatives à l'épidémie de COVID-19

COVID-19: One-month impact of the French lockdown on the epidemic burden. medRxiv

Epidemiological monitoring and control perspectives: application of a parsimonious modelling framework to the COVID-19 dynamics in France. medRxiv

Local measures enable COVID-19 containment with fewer restrictions due to cooperative effects

International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted

R: A Language and Environment for Statistical Computing

Impact of lockdown on COVID-19 epidemic in Île-de-France and possible exit strategies

Multi-level modeling of early COVID-19 epidemic dynamics in French regions and estimation of the lockdown impact on infection rate. medRxiv

COVID 19 -Point épidémiologique hebdomadaire du 2 juillet 2020

Long-term follow-up of recovered patients with COVID-19

Novel Criteria for When and How to Exit a COVID-19 Pandemic Lockdown. Front Big Data

New variant of SARS-CoV-2 in UK causes surge of COVID-19

Early assessment of the impact of mitigation measures to control COVID-19 in 22 French metropolitan areas

Evaluating the effect of demographic factors, socioeconomic factors, and risk aversion on mobility during the COVID-19 epidemic in France under lockdown: a population-based study

Age-dependent effects in the transmission and control of COVID-19 epidemics

Transmission interval estimates suggest presymptomatic spread of COVID-19. medRxiv

Substantial undocumented infection facilitates the rapid dissemination of novel coronavirus (SARS-CoV-2). Science (80-)

Base de données de la structure de la population en France par âge et sexe issues du recensement INSEE

Projecting social contact matrices in 152 countries using contact surveys and demographic data

Parcours hospitalier des patients atteints de la Covid-19 lors de la première vague de l'épidémie -Les dossiers de la Drees

Valuing EQ-5D using Time Trade-Off in France

bbmle: Tools for General Maximum Likelihood

. CC-BY-NC 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. . CC-BY-NC 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. . CC-BY-NC 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. . CC-BY-NC 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review). CC-BY-NC 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)The copyright holder for this preprint this version posted April 25, 2021. ; https://doi.org/10.1101/2021.04.21.21255876 doi: medRxiv preprint Estimation. 2020.. CC-BY-NC 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)The copyright holder for this preprint this version posted April 25, 2021. ; https://doi.org/10.1101/2021.04.21.21255876 doi: medRxiv preprint