key: cord-0839497-sxngqdyr
authors: Yang, W.; Kandula, S.; Huynh, M.; Greene, S. K.; Van Wye, G.; Li, W.; Chan, H. T.; McGibbon, E.; Yeung, A.; Olson, D.; Fine, A.; Shaman, J.
title: Estimating the infection fatality risk of COVID-19 in New York City, March 1-May 16, 2020
date: 2020-06-29
journal: nan
DOI: 10.1101/2020.06.27.20141689
sha: 16838387ab92a4da258a130fc60f76dc5c643081
doc_id: 839497
cord_uid: sxngqdyr

During March 1-May 16, 2020, 191,392 laboratory-confirmed COVID-19 cases were diagnosed and reported and 20,141 confirmed and probable COVID-19 deaths occurred among New York City (NYC) residents. We applied a network model-inference system developed to support the City's pandemic response to estimate underlying SARS-CoV-2 infection rates. Based on these estimates, we further estimated the infection fatality risk (IFR) for 5 age groups (i.e. <25, 25-44, 45-64, 65-74, and 75+ years) and all ages overall, during March 1-May 16, 2020. We estimated an overall IFR of 1.45% (95% Credible Interval: 1.09-1.87%) in NYC. In particular, weekly IFR was estimated as high as 6.1% for 65-74 year-olds and 17.0% for 75+ year-olds. These results are based on more complete ascertainment of COVID-19-related deaths in NYC and thus likely more accurately reflect the true, higher burden of death due to COVID-19 than previously reported elsewhere. It is thus crucial that officials account for and closely monitor the infection rate and population health outcomes and enact prompt public health responses accordingly as the pandemic unfolds.

!" # !$ = −" # ' ( ) *#+, -#( . ( where S i , E i , I i , R i , and N i are the numbers of susceptible, exposed (but not yet infectious), 89 infectious, and removed (either recovered or deceased) individuals and the total population, 90 respectively, from a given age group (described below) in neighborhood i. ) *#+, is the citywide 91 transmission rate, which incorporated seasonal variation as observed for OC43, a beta-92 coronavirus in humans from the same genus as SARS-CoV-2. 11 To allow differential transmission 93 in each neighborhood, we included a multiplicative factor, b i , to scale neighborhood local 94 transmission rates. Z and D are the latency and infectious periods, respectively (Table S1) . 95

The matrix [c ij ] represents changes in contact rates over time and connectivity among 97 neighborhoods and was computed based on mobility data. Briefly, changes in contact rates 98 (either intra or inter neighborhoods) for week-t were computed as a ratio of the number of 99 visitors during week-t to that during the week of March 1, 2020 (the first week of the pandemic 100 in NYC when there were no interventions in place), and further scaled by a multiplicative factor 101 m 1 ; m 1 was estimated along with other parameters. To compute the connectivity among the To account for delays in diagnosis and reporting, we included a time-from-infectious-to-case-110 reporting (i.e., diagnosis) lag, drawn from a gamma distribution with a mean of T m and standard 111 deviation (SD) of T sd days. To account for under-detection, we included a case reporting rate (r), 112

i.e. the fraction of infections (including subclinical or asymptomatic infections) reported as 113 cases. To compute the model-simulated number of new cases per week, we multiplied the 114 model-simulated number of infections per day (including those from the previous weeks) by the 115 reporting rate, and further distributed these simulated cases in time per the distribution of 116 time-from-infectious-to-case-reporting. We then aggregated the daily lagged, reported cases to 117 weekly totals for model inference (see below). Similarly, to compute the model-simulated 118 deaths per week, we multiplied the simulated-infections by the IFR and then distributed these 119 simulated deaths in time per the distribution of time-from-infectious-to-death, and aggregated 120 these daily numbers to weekly totals. For each week, the reporting rate (r), the mean (T m ) and 121 standard deviation (T sd ) of time-from-infectious-case-reporting, and the IFR were estimated 122 based on weekly case and mortality data. The distribution of time-from-diagnosis-to-death was 123 based on observations of n=15,686 COVID-19 confirmed deaths in NYC (gamma distribution 124 with mean = 9.36 days and SD = 9.76 days; Table S1 ). 125 75+). To account for stochasticity in model initiation, we ran the parameter estimation process 139 independently 10 times. Results for each age group were combined from these 10 runs (each 140 with 500 realizations). To combine estimates of reporting rate and IFR for <25 year-olds or all 141 ages overall, we weighted the age-group specific estimates by the fraction of estimated 142 infections from each related group. 143

The model-inference system was able to recreate the case and mortality time series for each 146 age group and all ages overall (Fig. 1) . For most age groups, confirmed cases peaked during the 147

week of March 29 and the mortality rate peaked about one week later than the case rate, due 148 to the time-lag from severe infection to death (Fig. 1) . 149 150 There were, however, substantial under-detection of infections, variations by age group, 151 and fluctuations of case reporting rates over time, in part due to changing testing criteria (e.g., 152

testing was restricted to severely ill patients in the early phase due to material shortages in (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. from testing of 25-64 year-olds 13, 14 ). In addition, the spatial variation estimated by our model-178

inference system 15 was in line with reported measures (i.e., highest in the Bronx and lowest in 179

Manhattan 13, 14 ). This consistency with independent serology survey data provides some 180 independent validation of our model estimates. 181 for these age groups in China. 3, 4 The average IFR was 4.67% (95% CrI: 3.21-6.66%) for 65-74 197

year-olds and 13.83% (95% CrI: 9.65-17.78%) for 75+ year-olds. In addition, the estimated IFR 198 fluctuated substantially over time for these two elderly groups. For 65-74 year-olds, estimated 199 IFR was 6.10% (95% CrI: 4.90-7.57%) during the week of April 5, 2020 but decreased to 3.79% 200 (95% CrI: 1.68-6.90%) during the week of May 10, 2020 ( Fig 3D) . For 75+ year-olds, IFR was 201 estimated to be 16.99% (95% CrI: 13.15-20.11%) during the week of April 5, 2020 but 202 decreased to 9.77% (95% CrI: 4.53-14.81%) during the week of May 10, 2020 ( Fig 3F) . probable COVID deaths. If only COVID confirmed deaths were included, given that 78.0% of 213 deaths among the total were laboratory-confirmed, the estimated overall IFR would be around 214 1.1%. Both estimates were higher than previously reported for elsewhere (e.g., about 0.7% in 215 both China 3 and France 5 ). Importantly, NYC has nosologists who review all death certificates 216 and record deaths into a unified electronic reporting system rapidly. This mortality surveillance 217 infrastructure and enhanced nosology thus allow more rapid and complete death reporting in 218 NYC. As such, our estimates here likely more accurately reflect the underlying fatality risk of 219 COVID-19 infection. Further, given the likely stronger public health infrastructure and 220 healthcare systems in NYC than many other places, 16 the higher IFR estimated here suggests 221 that mortality risk from COVID-19 may be higher in the United States and likely other countries 222 as well than previously reported. Of note, despite the large surge in cases and hospitalizations, 223 through quick expansion of healthcare systems, most hospitals in NYC were able to meet 224 patient care demand during the two-month period. As many jurisdictions in the United States 225 All rights reserved. No reuse allowed without permission.

(which was not certified by peer review) 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 June 29, 2020. . https://doi.org/10.1101/2020.06.27.20141689 doi: medRxiv preprint are considering re-opening after months of social distancing, it is crucial that officials account 226 for and closely monitor the infection rate and health outcomes including hospitalizations and 227 mortality and take prompt public health responses accordingly. 228 229 While the IFR estimated here was similar to those previously reported elsewhere for 230 younger age groups, we found that IFRs for individuals 65 years and older in NYC were about 231 twice as high as prior reports. 3 These higher IFRs may be in part due to differences in 232 population characteristics, in particular, the prevalence of underlying medical conditions such 233 as diabetes mellitus, chronic lung disease, and cardiovascular disease. 17, 18 Regardless, 234 estimated weekly IFR was as high as 6.1% for 65-74 year-olds and 17.0% for 75+ year-olds. In this study, we incorporated multiple data sources, including age-grouped, spatially 241 resolved case and mortality data as well as mobility data, to calibrate our model-inference 242 system. Of note, the timing of the COVID-19 pandemic varied substantially among NYC 243 neighborhoods. For instance, peak mortality rates occurred up to 4 weeks apart among the 42 244

neighborhoods. Fitting the model-inference system simultaneously to these diverse case and 245 mortality time series thus enabled better constraint of key model parameters (e.g., case 246

reporting rate and IFR). However, we note there remain large uncertainties in model estimates. 247

A full assessment of COVID-19 severity will require comprehensive serology surveys of the 248 population by age group and neighborhood, given the large heterogeneity of infection rates 249 across population segments and space. In addition, we only included deaths that were lab-250 confirmed or explicitly coded as related to COVID-19. A recent study reported that excess 251 deaths in NYC during about the same period could be more than 24,000. 8 Further, recent 252 studies have reported severe sequelae of COVID-19 in children, i.e. Multi-system Inflammatory 253 All rights reserved. No reuse allowed without permission.

(which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) 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 June 29, 2020. Reported Cases per 100,000 COVID−Confirmed/Probable Deaths per 100,000

All rights reserved. No reuse allowed without permission.

(which was not certified by peer review) 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 June 29, 2020. . https://doi.org/10.1101/2020.06.27.20141689 doi: medRxiv preprint (which was not certified by peer review) 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 June 29, 2020. . https://doi.org/10.1101/2020.06.27.20141689 doi: medRxiv preprint (which was not certified by peer review) 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 June 29, 2020. . https://doi.org/10.1101/2020.06.27.20141689 doi: medRxiv preprint Supplementary Material Table S1 . Prior ranges for main model parameters and variables. The spatial, temporal, and age resolution of each parameter or variable, estimated in the model-inference system, is specified in the column "Resolution". Note posterior parameter estimates can extend outside the specified prior ranges.

Prior range Source/rationale 

MMWR Morbidity and mortality weekly report

Eight-week model projections of COVID-19

New York State Department of Health

Amid ongoing COVID-19 pandemic, governor cuomo announces 320 results of completed antibody testing study of 15,000 people showing 12.3 percent of 321 population has COVID-19 antibodies 2020

5 new yorkers may have had covid-19, antibody tests 325 suggest 2020

Black boxes show model estimates of cases per 100,000 population and grey boxes show model estimates of mortality rates; thick horizontal lines and box edges show the median, 25 th , and 75 th percentiles; vertical lines extending from each box show 95% Crl. Blue dots indicate observed incidence rates and red dots show observed mortality rates. and SD of 9

NYC DOHMH population estimates, modified from US census bureau interpolated intercensal population estimates

Social contacts and mixing patterns relevant to the spread of infectious diseases

Nowcasting and forecasting the potential domestic and international spread of the 2019-ncov outbreak originating in Wuhan, China: A modelling study

Substantial undocumented infection facilitates the rapid dissemination of novel coronavirus (SARS-cov-2)

Evolving epidemiology and transmission dynamics of coronavirus disease 2019 outside hubei province, China: A descriptive and modelling study. The Lancet Infectious diseases

A simple modification for improving inference of non-linear dynamical systems

Estimates of the severity of coronavirus disease 2019: A model-based analysis. The Lancet Infectious diseases