Skip to main content

Our take —

This preprint, which has not yet been subjected to peer review, used both case counts and deaths in an age-specific model of SARS-CoV-2 transmission in New York City. The authors estimate that the reproduction number of the virus decreased dramatically from the beginning of March to mid-April, 2020 in response to a collection of interventions that reduced mobility (through stay-at-home orders, school closures, business closures, etc.) and mandated mask use. The authors’ attribution of transmission declines to each of these two types of interventions, however, is dependent on many assumptions and subject to much uncertainty.

Study design

Modeling/Simulation, Other

Study population and setting

This study estimated the impacts on SARS-CoV-2 transmission of non-pharmaceutical interventions (NPIs) in New York City (NYC) from March 1 to June 6, 2020. COVID-19 cases included all laboratory-confirmed cases reported to the NYC Department of Health and Mental Hygiene, and deaths combined probable and confirmed deaths associated with COVID-19. Mobility data, used as a proxy for contact rates, were obtained from Safegraph and consisted of anonymized, aggregated counts of visitors (measured by mobile phone location) to locations within each ZIP code. The authors used a neighborhood-specific SEIR network model fit to cases and deaths, stratified by age group, to estimate the effects on transmission of 1) all NPIs, 2) contact-reducing NPIs (such as stay-at-home orders, business closures, etc.), and 3) mask use. Mask use was assumed to explain the reduction in estimated transmission rate that was not accounted for by mobility declines during periods when face coverings were mandated in public places. Model projections beyond the end of the study period were compared to observed cases and deaths.

Summary of Main Findings

Observed, diagnosed COVID-19 cases displayed different age-specific patterns compared to model estimates of underlying infection rates: estimated infection rates were highest for those aged 25-44 years and 45-64 years, and rates for all age groups peaked the week of March 22 or one week later. During the first week of the NYC epidemic (beginning March 1), the estimated time-varying reproduction number (Rt) was 2.99, decreased to 1.37 after the stay-at-home order on March 22, and reached a minimum of 0.56 during the week of April 12. Mobility reductions (a proxy for contact rate reductions arising from stay-at-home mandates, school closures, and other contact-reducing interventions) were estimated to result in a 70.7% (95% CI: 65.0% to 76.4%) decline in Rt by the week of April 12. Assuming that effectiveness of mask use would account for the difference between estimates using a) a linear regression with mobility data alone and b) the SEIR model, the authors estimated that mask use reduced the transmission rate and infectious period by 3.4% (95% CI: -1.9% to 8.6%) over eight weeks, with higher effectiveness during the first month. Estimated mask effectiveness was highest in older age groups and remained stable during the study period (for the first month among those 65-74 years old: 20.8%, 95% CI: -0.1 to 41.6%; 75+ years old: 20.8%, 95% CI: 20.8%, 95% CI: -0.9 to 42.5%). Projections from the week of June 7 through the week of July 26 using parameters based on observed mobility data and estimated mask effectiveness underestimated cumulative cases by 27% and underestimated deaths by 2%.

Study Strengths

The model was fit to both observed cases and deaths, and projections beyond the study period were compared to observed outcomes.

Limitations

This is a preprint, and has not yet been subject to peer review. Aggregated zip-code-level mobility data are an imperfect proxy for actual mobility, which is in turn an imperfect proxy for contact rates. The method used to estimate the effectiveness of mask use relies on strong assumptions and oversimplifications (e.g., all residual reduction in predicted transmission rate after accounting for mobility decline is attributable to mask use; mask use affects both transmission risk per contact and infectious period; dates of mask mandates are a perfect proxy for actual mask use; etc.). Projections did not fit observed data well, which may be an indication that the effect of interventions was overestimated.

Value added

This study provides a useful picture of age-specific patterns of SARS-CoV-2 infection during the spring of 2020 in New York City.

Our take —

This study, available as a preprint and thus not yet peer reviewed, evaluated the impact of 13 different non-pharmaceutical interventions on the time varying reproductive number across 130 countries and territories through June 2020. Impact was assessed using a wide range of models with different time lags (i.e., time frame from implementation of intervention to its impact on transmission) and at varying levels of intensity of the interventions. School closures and internal movement restrictions showed evidence of impact across all models, and thus had strong evidence of effectiveness. However, these results should be interpreted cautiously as multiple interventions were often implemented at the same time, making it difficult to fully disentangle the impact of any single intervention from the others.

Study design

Ecological, Modeling/Simulation

Study population and setting

This study used the implementation dates of a large variety of non-pharmaceutical policy interventions and assed their relationship with COVID-19 time varying reproductive number, Rt, defined as the mean number of secondary cases that one index case will infect at time t) across 130 countries and territories between January 1 and June 22, 2020. Data on non-pharmaceutical COVID-19 policies, categories, implementation dates, and a general index of strength of COVID-19 policy response were obtained from the Oxford COVID-19 Government Response Tracker. Estimates of the Rt across regions and time were from EpiForecasts.

There were four primary analyses. The first main analysis used a statistical regression model to characterize how different types of NPIs were rolled out over time, such as when they occurred and in what order. The second model observed how those NPIs were clustered in time (i.e. what policies were more likely to occur together with what other policies, forming a group or a cluster of policies). A third model attempted to determine the time lags between policy implementation and their effect on Rt by assessing 3 different time lags (1, 5, and 10 days) and estimating goodness-of-fit statistics for each model. Finally, the study used panel regression models to attempt to disentangle the impact of different types of NPIs on Rt accounting for different time lags and the effort at which NPI was implemented (any effort and maximum effort).

Summary of Main Findings

First, the study finds a major increase in NPI intensity across the world in mid-March, followed by a slow reduction in the stringency of interventions. Second, the study finds substantial evidence that NPI policies were more likely to be rolled out in particular orders, typically clustered together in time, depending in part on how policy intensity is defined (any effort or maximum effort). Thirdly, the study found support for lag times between policy implementation and Rt impact between 1-10 days. Finally, the study finds evidence that school closure and internal movement restrictions, and high-intensity public events cancellations and restrictions on gatherings. There was some evidence for impact for workplace closure, income support and debt/contract relief. Evidence for impact was inconclusive for stay-at-home requirements, public information campaigns, public transport closure, international travel controls, testing, and contact tracing.

Study Strengths

This study critically examines and demonstrates how NPI policies are related to each other, and assesses lags in their impact. These are issues that are critical both to this analysis, and many other policy analyses which often do not acknowledge that NPI policies are highly correlated with respect to timing of implementation, and that interventions have lagging effects, due to changing compliance over time. The study uses well-vetted data appropriate for policy impact evaluation. The discussion section contains a frank and well-written interpretation of the results and the interpretable limitations thereof. We find that the findings that NPIs overall had substantial impact on Rt to be relatively robust.

Limitations

While the discussion section involves cautious interpretation of impact of individual NPI because they were implemented at similar times, the highest impact section (the abstract) does not heed that caution, and strongly implies that they identified which specific types of NPIs were most effective despite temporal correlation. The key difficulty – one examined and partially established in the paper itself – is that these interventions are temporally related to each other, and also have time lagged effects in similar timescales as the policy rollouts. One major issue is that there appears to be substantial limitations in the way in which lagged effects were measured. Lags were assumed to be at a fixed amount of time and the same for all policies, but in reality policy lag effects are spread out over a wide period of time, and will be different for different policies in different situations. This, in addition to the close clustering in time between policies, makes it difficult to conclude which NPIs had the strongest evidence for impact.

Value added

The study shows strong evidence that NPI policies are related to each other over time and how that creates difficulty in examining their impact.

Our take —

Using previously published models, authors quantified the degree to which COVID-19 transmission could interrupt malaria prevention services across sub-Saharan Africa (SSA), and the impact these interruptions could have on malaria deaths. The estimated effects varied according to the duration of interruptions, timing of disruptions and malaria seasonality, and how recently routine vector-control measures were implemented. Across SSA, authors estimated malaria deaths could increase by 696,000. However, due to large uncertainties in the models concerning COVID-19 transmission and how countries will respond, estimates should be considered illustrative rather than true projections.

Study design

Modeling/Simulation

Study population and setting

This study quantified interruptions to malaria preventive services due to COVID-19, and the potential impact of these interruptions on malaria morbidity and mortality in Nigeria and across sub-Saharan Africa (SSA). COVID-19 trajectories were constructed using a previously developed age-structured Susceptible-Exposed-Infectious-Susceptible model. Malaria deaths were estimated using a previously published malaria transmission dynamics model. Assuming an R0 (basic reproductive number) of 3, authors assessed the four COVID-19 transmission scenarios: 1. unmitigated (no direct action, but contact rates reduced by 20%); 2. mitigation (isolation and social distancing reduced contact rates by 45% for 6 months); 3. indefinite suppression (strict interventions reduce contact rates by 75% and are maintained indefinitely); and 4. suppression lift (strict interventions reduce contact rates by 75%, but these are only maintained for 2 months, after which time contact rates return to 80%). Authors included varying combinations of interruptions to the following malaria prevention activities: 1. distribution of long-lasting insecticide-treated nets (LLIN) would be delayed for a year or continue as normal, 2. seasonal malaria chemoprevention (SMC) would be interrupted, reduced, or continue as normal, and 3. clinical treatments would be interrupted, reduced, or continue as normal. Using Nigeria as an example, authors also estimated how changes to R0 (reproductive number) of COVID-19 could impact malaria deaths and the effect of broadening target ages for SMC.

Summary of Main Findings

If SMC coverage and case treatment were reduced by 50%, malaria deaths were estimated to increase by 42,000 (95% Uncertainty Interval: 22,000-62,000) in Nigeria and by 200,000 (95% UI: 115,000-285,000) across SSA, even if LLIN campaigns were not interrupted. However, if LLIN campaigns were also interrupted in addition to SMC and treatment reductions, excess malaria mortality was estimated to be as high as 495,000 (95% UI: 296,000-693,000) across SSA. If LLIN campaigns, SMC coverage, and treatment were all interrupted, malaria deaths were estimated to increase across SSA by as much as 696,000 (95% UI: 413,000-978,000). If R0 was 2.5 compared to 3.0, authors estimated the epidemic in Nigeria would increase from 6 to 9 months, thereby increasing malaria deaths by approximately 17%, even if LLIN campaigns continued and some case treatment was maintained. In scenarios where LLIN campaigns were interrupted, increasing the target age for SMC from <5 years to <15 years was estimated to save 22,500 lives in Nigeria.

Study Strengths

Authors conducted sensitivity analyses to account for uncertainty in the models.

Limitations

Overall, results are subject to several sources of uncertainty, the foremost being that it is unknown how COVID-19 will spread across SSA. Countries will likely experience varying levels of transmission and respond with different combinations of mitigation strategies, both of which will affect the degree to which malaria activities are disrupted. Parameters included in the COVID-19 model are based on data from the United Kingdom and China, which may not be appropriate for SSA. Authors used Nigeria as a case example for some estimates, but even these results should not be generalized to other parts of SSA.The model assumed that the impact of COVID-19 on malaria deaths was determined solely by the duration of malaria service interruptions. However, duration is unlikely to be the only contributing factor to the effects of service interruptions on malaria-associated deaths; other factors, such as to what degree services are interrupted (i.e., intensity of interruptions), also likely play a role. It is not clear to what extent routine malaria services have truly been disrupted.

Value added

Dual epidemics of COVID-19 and malaria could quickly overwhelm already vulnerable health systems in sub-Saharan Africa. This study contributes to the growing body of literature estimating the potential indirect effects of COVID-19 in Africa, and illustrates the need to prioritize malaria prevention methods such as routine distribution of long-lasting insecticide-treated nets.

Our take —

This analysis of smartphone geolocation data associated with 6,644 nursing homes across 23 U.S. states, published as a working paper and not yet peer reviewed, showed that facilities are highly connected, with 7% of phones appearing in a given facility appearing in at least one other facility.  This high connectivity is likely due to shared staff, and persisted even after the implementation of national guidelines designed to limit mobility between nursing homes. While these data highlight vulnerabilities to COVID-19 transmission, rather than recreate actual COVID-19 transmission pathways, the observed associations of nursing home centrality and connectivity with higher COVID-19 cases suggests additional regulations, like single-facility staffing, may be needed to curb COVID-19 outbreaks in nursing homes.

Study design

Ecological, Modeling/Simulation

Study population and setting

Investigators aggregated nursing home resident COVID-19 case data reported to 23 state health departments in the United States through May 31, 2020. Geolocation smartphone data from March 13 to April 23, 2020 were mapped onto COVID-19 case reports to identify mobility patterns among staff, contractors, and residents across 6,644 nursing homes. Adjusting for compositional characteristics of nursing homes (e.g., demographic factors, quality measures), the investigators assessed the relationship between COVID-19 cases at nursing homes and measures of nursing home connectivity (to other homes).

Summary of Main Findings

Even after guidelines restricted social visitors at nursing homes beginning March 13, 2020, approximately 7% of mobile phones that were identified in a given nursing home during the study period were found in at least one other home. Nursing homes were also highly connected (mean number of connections: 15), though these estimates varied widely by state. In regression analysis, nursing home connectivity measures – including the number of other homes to which a nursing home was connected, the number of mobile phones identified in multiple nursing homes, and the number of shared contacts between a nursing home and other homes with high connectivity – were significantly associated with higher cumulative COVID-19 case counts. Higher cumulative COVID-19 cases were also associated with nursing homes in urban areas, more beds, and higher proportions of Black residents (>25%) and Medicaid recipients (>50%).

Study Strengths

Investigators leveraged available geolocation data to identify networks of epidemiologically linked nursing homes vulnerable to COVID-19 transmission. Investigators performed sensitivity analyses to confirm that their findings were robust to COVID-19 prevalence and reporting differences across jurisdictions.

Limitations

The study could not determine whether individuals whose phones appeared in multiple nursing homes were the actual source of SARS-CoV-2 transmission, since the outcome measure was COVID-19 cases aggregated to the nursing home level. Additionally, while geolocation services were useful in constructing nursing home network profiles, the data presented do not offer insights into the duration or frequency of individuals’ exposure to multiple facilities. Lastly, analyses were cross-sectional and did not account for temporal shifts in COVID-19 cases at nursing homes, which could have driven variability in mobility during the observation period that was unexamined in the study.

Value added

This study provides valuable evidence that long-term care facilities are highly connected via shared staff, highlighting potential transmission pathways that threaten vulnerable residents.

Our take —

In this peer-reviewed study, the authors analyzed genomes of SARS-CoV-2 and related viruses (from the Sarbecovirus subgenus) to assess the history of recombination in this group and to estimate the timing of SARS-CoV-2 divergence from its ancestors. The results indicate that while recombination is common in sarbecoviruses, the receptor binding region of SARS-CoV-2 does not appear to be a recent recombination with pangolin coronaviruses and likely derives from ancestral viruses in bats. SARS-CoV-2 was estimated to have diverged from its nearest ancestor in bats between 1948 and 1982, indicating that evolutionary ancestors of the virus have been circulating in bats for many years prior to spillover into humans.

Study design

Ecological, Modeling/Simulation, Other

Study population and setting

The study used 68 full coronavirus genomes from the subgenus Sarbecovirus (containing SARS-CoV and SARS-CoV-2) collected from human cases, bats, and other intermediate hosts in northern, central, and southern China since 2002. The goal of the study was to determine the evolutionary history of SARS-CoV-2, specifically to understand the likely source of SARS-CoV-2 in humans (bats, pangolins, or another species), and to identify how long the virus had been circulating in that animal host.

Summary of Main Findings

The authors found that recombination is common among sarbecoviruses, with 67/68 genomes showing evidence of genomic exchange. They find that SARS-CoV-2 and bat-associated RaTG13 are part of a single lineage separate from SARS-CoV and related sarbevoviruses, suggesting that SARS-CoV-2 is the result of a direct (or nearly-direct) zoonotic transmission from bats. Specifically, SARS-CoV-2 did not acquire its variable loop region of the spike protein (containing the receptor binding domain that interfaces with human ACE2) through a recent recombination event with related sarbecoviruses in pangolins. Rather, RaTG13 is the recombinant virus, having acquired its variable loop domain from an as yet unsampled SARS-related coronavirus. The authors also used three different methods to estimate that SARS-CoV-2 appears to have diverged from a common ancestor in bats between 1948 and 1982. The estimated divergence time between the closest pangolin coronavirus to SARS-CoV-2 and the lineage containing SARS-CoV-2 and RaTG13 was between 1851 and 1877, indicating that pangolins likely acquired coronaviruses independently from bats, and were probably not an intermediate host that facilitated adaptation of SARS-CoV-2 to humans. These results indicate that a direct progenitor for SARS-CoV-2 has been circulating in horseshoe bats for decades before spillover into humans.

Study Strengths

The authors used a robust approach to deal with the issue of recombination in phylogenetic inference, which if unaddressed can lead to longer branch lengths and inflated divergence times. The authors also used a robust approach involving multiple prior distributions to estimate evolutionary divergence times, thereby capturing some of the uncertainty that is inherent in time-measured phylogenetic analysis and improving upon previously published results.

Limitations

As with other phylogenetic analyses of sarbecoviruses, inferences about the evolutionary origin of SARS-CoV-2 and the diversification of sarbecoviruses generally are limited by the current availability of genomes related to SARS-CoV-2 in bats and potential intermediate hosts. Additional sampling of bats and potential intermediate hosts around Wuhan and other areas of central China could reveal sarbecoviruses that represent a closer ancestor of SARS-CoV-2 that would provide more information on when and how the virus spilled over into humans.

Value added

This study provides a detailed explanation of the recombination history among sarbecoviruses and in the lineage containing SARS-CoV-2 and its closest relative in bats, RaTG13. The authors demonstrate that SARS-CoV-2 is not a recent recombinant of pangolin and bat viruses, and instead shares features with bat-associated sarbecoviruses. This suggests that spillover from bats to humans may have been direct or near-direct (i.e., a brief residence in an intermediate host). Pangolins do not appear to have been intermediate hosts based on the currently available data.

Our take —

This phylogenetic study analyzed SARS-CoV-2 sequences isolated from individuals across Brazil. The study investigated how SARS-CoV-2 became established in Brazil, and determined the impact of social and physical distancing measures on the spread of the virus. Specifically, the researchers showed that there were multiple international introductions into Brazil, but that relaxation of distancing measures led to increased spread within the country. Internationally connected cities in Brazil played an important role in the local and inter-state spread of SARS-CoV-2 and its subsequent establishment in the country.

Study design

Modeling/Simulation, Other

Study population and setting

The purpose of the study was to investigate how SARS-CoV-2 became established in Brazil and to quantify the impact of non-pharmaceutical interventions (NPIs) on virus spatiotemporal spread. Authors generated 427 new SARS-CoV-2 genomes from Brazilian samples collected between March 5 and April 30, 2020 from 85 municipalities spanning all regions of Brazil and analyzed these data in conjunction with 63 additional previously published sequences from Brazil. Authors used a continuous phylogeographic model to infer the origins of each phylogenetic node and human mobility data to estimate daily changes in R (effective reproductive number).

Summary of Main Findings

The study showed that NPIs reduced R from >3 to close to 1 in São Paulo and Rio de Janeiro during the start of the epidemic, but the increased mobility after this initial decline led to a rising R in Sao Paulo. Genome analysis showed that 99% of the Brazilian sequences belonged to SARS-CoV-2 lineage B, with only 5 strains belonging to lineage A. Within the dominant B lineage, the authors found that sequences from Brazil fell into three primary clusters. Analysis of these three clades indicated that community-driven transmission was already established in Brazil by early March, 2020, and international travel restrictions initiated after this period may have had limited impact. Further phylogeographic analysis revealed there were at least 102 international introductions of SARS-CoV-2 into Brazil, primarily into internationally well-connected states such as São Paulo (36% of all imports), Minas Gerais (24%), Ceará (10%) and Rio de Janeiro (8%). Overall, the authors found that SARS-CoV-2 spread was initially driven by local and within-state movement, but was later characterized by long-distance movement, though within-state virus movement was always more frequent than between-state movement.

Study Strengths

A key strength of this study is that the genomic data is representative of the distribution of cases in Brazil. Furthermore, the methodology is clear and well presented, and the results of this study revealed the importance of customizing diagnostic assays with specificity for locally circulating strains.

Limitations

There is limited description of the 24% Brazilian sequences that did not fall in the three main clades.

Value added

The COVID-19 epidemic in Brazil is currently among the largest in the world, and this paper explains how it grew to that level. The paper demonstrates the role that introductions from other countries had in fueling the spread of the virus and examines the relative contributions of within- and between-state travel. The authors also describe how social and physical distancing measures initially curbed spread of the virus before their relaxation led to increased transmission and details the different SARS-CoV-2 lineages and clades that are spreading in the region, which may have an impact on the diagnostic tests used locally.

Our take —

Authors show that rapid case isolation can shorten the duration of onward transmission and thus be an effective non-pharmaceutical intervention (NPI). Furthermore, the serial interval, the time between symptom onset in the infector and infectee should not be considered as a single fixed distribution that is generalizable across different settings. Authors demonstrate that non-pharmaceutical interventions, and specifically case isolation, shortened the serial interval over the course of the epidemic in China, which can substantially affect the estimates of the reproduction number. An “effective” serial interval distribution may therefore help to refine estimates of transmissibility and offer insights into the impact of public health interventions.

Study design

Modeling/Simulation

Study population and setting

Authors compiled a database of 1,407 Covid-19 “transmission pairs” where information on who infected whom (infector to infectee) could be constructed from publicly available data on 9,120 confirmed cases in 27 provinces and 264 health commissions in China. Cases from Hubei province were excluded as chains of transmission were harder to reconstruct. Analysis was restricted to the 677 pairs for which dates of symptom onset and their social relationships were available. Infectors reported symptom onset between January 9 to February 13, 2020 during which key interventions were implemented. The authors estimated the time between symptom onset of the infector and symptom onset of the infectee (the serial interval) over three time periods: i) January 9 to 22 or “pre-peak”; ii) January 23 to 29 or “peak-week”; and iii) January 30 to February 13 or “post-peak”. They also estimated the serial interval over time using a rolling average. Authors then explored whether changes in the serial interval could be explained by factors including age, sex, social relationships, or non-pharmaceutical interventions (NPIs) and the potential impact of changing serial intervals over the course of the epidemic on estimates of transmissibility (Rt or the reproduction number).

Summary of Main Findings

Authors estimated that the serial interval shortened over the course of the epidemic from 7.8 days (95% credible interval: 7.0 – 8.6) pre-peak to 5.1 days (95% CrI: 4.6 – 5.7) during the peak, and 2.6 days (95% CrI: 1.9 – 3.2) post-peak. This decline was also observed when the serial interval was estimated over a rolling time window and when stratified by age, sex, and social relationship. Authors found that the delay from SARS-COV-2 confirmation to isolation was the primary driver and could explain 51.5% of the variability in the observed serial interval duration. For each day of early isolation, the serial interval decreased on average by 0.7 days. NPI strategies and population immunity explained an additional 15-20% of variability. Authors found that using a fixed serial interval distribution in analyses led to mis-estimation of Rt and potentially underestimation early in the pandemic.

Study Strengths

Authors used a large number of transmission pairs (677 pairs) for which detailed epidemiological information was available to estimate the serial interval over multiple time points and explicitly excluded data from Hubei province which was deemed less reliable. Authors estimated the serial interval using non-overlapping and rolling average time-windows and found the same declining trend in the serial interval over time.

Limitations

While authors consider a large number of transmission pairs, some epidemiological data may be limited by recall bias e.g. date of symptom onset. Additional unknown factors aside from time to isolation, NPIs or population immunity could have resulted in the reduction of the serial interval. Additionally, the effect of other NPIs and population immunity could not be disentangled.

Value added

Authors demonstrate that rapid case isolation can be an effective NPI. They additionally show that the serial interval is not a fixed interval as is often assumed and can be impacted significantly by non-pharmaceutical interventions. The use of “effective” serial intervals in analyses may provide better estimates of the reproduction number.

Our take —

Authors used a specialized Susceptible-Exposed-Infected-Removed (SEIR) model to estimate the true extent of the COVID-19 epidemic in Wuhan, China over the course of five time periods between January 1, 2020 through March 8, 2020. Results estimated the effective reproductive number decreased from 3.84 during the first period to 0.28 during the final period. Authors also determined an overall ascertainment rate of 0.13, indicating a high proportion of asymptomatic or mild-symptomatic cases. This study used laboratory-confirmed cases from Wuhan and results were robust to extensive sensitivity analyses.

Study design

Modeling/Simulation

Study population and setting

Authors developed a Susceptible (S), Exposed (E), Presymptomatic infectious (P), unAscertained infectious (A), Isolated (H), and Removed (R) (SAPHIRE) model, an extension of the traditional Susceptible-Exposed-Infected-Recovered model, to estimate the true extent of the COVID-19 epidemic in Wuhan, China. Using 32,583 laboratory-confirmed cases from Wuhan, authors ran the model across five time periods in 2020 (January 1 to 9; January 10 to 22; January 23 to February 1; February 2 to 16; and February 17 to March 8), defined by different events (e.g., Chinese New Year) and the implementation of intervention strategies (e.g., the cordons sanitaire). Authors estimated Re (the effective reproductive number, the reproductive number after the implementation of interventions and community transmission has occurred) for each time period and total cumulative cases across all time periods using these data. Authors also projected what the total cumulative cases would have been assuming the incidence trend from each time period continued uninterrupted until March 8, 2020 (i.e., assuming none or with no additional interventions than what occurred in each respective time period). Authors used confirmed cases exported from Wuhan to Singapore to estimate the ascertainment rate (the percentage of total cases that are reported and confirmed).

Summary of Main Findings

The Re in each time period was 3.54 (95% CrI: 3.40-3.67), 3.32 (3.19-3.44), 1.18 (1.11-1.25), 0.51 (0.47-0.54), and 0.28 (0.23-0.33), respectively from first to fifth/last. Authors credit the significant decrease in Re to the wide-spread and multi-level public health interventions implemented in Wuhan. Overall, authors estimated that a total of 249,187 (95% CrI: 198,412-307,062) cases (including unascertained cases) occurred when they fit the data across all five time periods; this was notably lower than when authors fit the data according to time period trends, which was estimated to be up to 6,302,694 (6,275,508-6,327,520) when the trend from the second time period was assumed. The model projected the number of daily infections (including unascertained cases) peaked on February 2, 2020 at 55,879 (43,582-69,571). Regardless of time period, case ascertainment rates were low: 0.15 (95% CrI: 0.13-0.17), 0.15 (0.13-0.17), 0.14 (0.11-0.17), 0.10 (0.08-0.12), 0.16 (0.13-0.21), respectively, from first to fifth/last, and 0.13 (0.11-0.16) overall. Due to the high proportion of presymptomatic and unascertained cases, authors estimated the probability of case resurgence could be as high as 97% and would occur approximately one month after the removal of control measures, assuming they were lifted 14 days after the first day of zero ascertained cases.

Study Strengths

Authors used laboratory-confirmed cases, ensuring that false clinical diagnoses did not bias results. Authors validated assumed parameters and estimation methods before running the full simulations; authors determined the model was able to make accurate estimations. The decreases observed in Re between time periods and the low proportion of unascertained cases were robust to sensitivity analyses.

Limitations

Delays in laboratory reporting may have resulted in an underestimation of the ascertainment rate, which would also overestimate the proportion of asymptomatic cases and the effect these cases would have on potential resurgence. Authors also excluded clinically confirmed cases without laboratory confirmation, which may have had similar effects on results. Authors assumed homogeneous transmission between heterogeneous populations; rates of transmission have been shown to vary between groups of differing ages, sexes, race/ethnicities, and geographic locations. Control measures were evaluated as a whole, so the effects of individual interventions on Re are not available.

Value added

This study sought to describe the full spectrum of the dynamics of the COVID-19 epidemic in Wuhan using laboratory-confirmed cases from the city. Results indicated control measures significantly reduced transmission and that there was a high proportion of unascertained cases. Understanding these dynamics is critical for surveillance and control measures, and could be used to inform strategies in countries still experiencing active transmission.

Our take —

Using a stochastic transmission model, authors compared the impact of different contact tracing strategies in reducing Re (effective reproductive number) from 1.2, an assumed value that represents SARS-CoV-2 transmission in settings with physical distancing. A traditional contact tracing scenario was generally unable to bring R to below 1 if there were testing delays, whereas decreases in R achieved through contact tracing via mobile apps was more robust to these delays due to their high contact tracing coverage.

Study design

Modeling/Simulation

Study population and setting

Using a stochastic compartmental model originally developed for smallpox vaccination, authors analyzed how delays in testing and contact tracing at the population level impact Re (effective reproductive number) and the proportion of transmissions that could be prevented per index case. To explore trade-offs within different components of a contact tracing program on Re, authors used their model to explore many sets of parameters. Among these, the authors defined a single conventional contact tracing scenario (80% testing coverage, 80% and 50% tracing coverage of close and casual contacts respectively, 4 day testing delay from symptom onset to receiving a test result, 3 day tracing delay from returning a case test result to contact quarantine) in relation to multiple mobile app contact tracing scenarios that had wider ranges in their assumed testing and tracing coverages and zero-day testing and tracing delays.

Summary of Main Findings

Assuming an app-based scenario where 80% of symptomatic persons were tested, 80% of their contacts traced, and testing and tracing delays were zero days, R was reduced from 1.2 to 0.8 (0.7-1.0) and 79.9% of transmission events were prevented. When the testing delay was three days or more, it was not possible to reduce R from 1.2 to below 1 (even when there was no tracing delay) . Mobile app-based tracing scenarios were more effective due to a combination of zero testing and tracing delays and greater tracing coverage for close and casual contacts. Sensitivity analysis showed, however, that high testing and tracing coverage were more important in reducing transmission than time delays. Even assuming a two-day testing delay, contact tracing via a mobile app could keep R below 1 as long as 80% symptomatic persons are tested and 80% of their contacts traced.

Study Strengths

This study accounts for all individual components of contact tracing (e.g., timing of index case infection through quarantining of contacts). Model assumptions were also based on real-world experience from public health professionals working in contact tracing.

Limitations

Considering the high testing coverage in the modeled scenarios, the reductions to R described in this study may be unrealistic in settings like the United States. Further, the authors use the terms “conventional” and “mobile app-based” contact tracing to refer to model assumptions. Thus, we urge caution in assuming these results would apply generically to a conventional or app-based contact tracing program, as their real-world performances could vary widely and even overlap.

Value added

Previous work indicates that contact tracing and isolation alone may not be sufficient to control outbreaks and that other interventions may be needed. This study assessed the degree to which each component of a contact tracing program contributes to its effectiveness.

Our take —

This study, covering over 17 million adults in England, confirms previously observed associations with mortality: higher risks were seen with older age, male sex, chronic comorbidities, socioeconomic deprivation, and nonwhite ethnicity. Although causality cannot be inferred for any given risk factor owing to the study design, the size of the analysis provides a striking replication of results observed elsewhere in smaller studies.

Study design

Retrospective Cohort

Study population and setting

This study analyzed electronic medical records from 17,278,392 adults aged 18 years and over from the National Health Service in England, representing approximately 40% of the adult population in England, to estimate risk factors for COVID-19 mortality. Registered active adult patients with >1 year of prior follow-up and with recorded age, sex, and “deprivation” (a measure related to poverty, derived from area of residence) were eligible for inclusion; the study period was from February 1 to May 6, 2020. Risk factors for COVID-19 mortality were assessed with Cox proportional hazards models 1) adjusted for age and sex; and 2) adjusted simultaneously for all included risk factors.

Summary of Main Findings

There were 10,926 deaths (0.06%) attributed to COVID-19 in the study population. In the fully adjusted model, the risk of mortality increased with every decade of age: relative to the reference category aged 50-59 years, the hazard ratio (HR) for death was 20.61 (95% CI: 18.72-22.70) for those aged 80 years and older. Men had a higher hazard of mortality relative to women (HR: 1.59, 95% CI: 1.53-1.65). Mortality risks for Black (HR: 1.48, 1.30 to 1.69), South Asian (HR: 1.44, 1.32-1.58), and mixed ethnicities (HR: 1.43, 1.11-1.85) were higher relative to white ethnicity. The most deprived quintile (from a geographically-defined index of socioeconomic status) had 1.80 times (1.69-1.91) the mortality hazard of the least deprived quintile. Multiple comorbidities were associated with increased mortality risk, including obesity, diabetes, chronic heart disease, chronic liver disease, kidney dysfunction, recently diagnosed cancer, and severe asthma. Although smoking and hypertension were associated with higher mortality in the age- and sex-adjusted model, they were not in the fully adjusted model. Post-hoc analyses of smoking and hypertension indicated that controlling for chronic respiratory disease (for smoking) and for diabetes and obesity (for hypertension) explained much of the changes in hazard ratios.

Study Strengths

The primary strength of this study is its size; it was able to cover a significant proportion of the adult population of England.

Limitations

Estimated hazard ratios for covariates were derived from a model that adjusted simultaneously for all covariates; results for a given risk factor cannot be interpreted causally due to possible mediation, unmeasured confounding, and collider stratification bias. There were high proportions of missing data for ethnicity (26%), obesity (22%), and hypertension (10%). Some deaths may have been missed due to misclassification. Although the dataset was very large, it may not have been representative of the population of England (e.g., only 17% of health care practices in London were included). The study does not address risk factors for non-fatal but nonetheless severe manifestations of COVID-19.

Value added

This extremely large study confirms associations seen in other studies and provides further evidence for the role of structural inequalities in COVID-19 impacts.