Skip to main content

Our take —

This study conducted early on during the US epidemic found that, if a COVID-19 epidemic were to spread in US cities, ICU capacity could be exceeded even with Wuhan-like levels of containment measures. It provides evidence that plans are needed to reduce the burden on the healthcare system. However, as the pandemic has progressed, local data are now more reliable for making these predictions compared to extrapolating from China.

Study design

Modeling/Simulation

Study population and setting

The study’s primary data source was the number of COVID-19 cases requiring critical care in Wuhan during the period of January to February 2020. Also, the mortality risk by age (over/under 65 years) and co-morbidity (whether the patient had hypertension) was obtained from China CDC case reports. The authors used these data to calculate the probability that an individual in each of these risk groups needed critical care at the peak of the epidemic in Wuhan.

The authors then used data on the proportion of the population over 65 and/or with hypertension in the 30 most populous US cities to predict peak ICU bed requirements in these cities. The study also extracted data from situation reports from Chinese health authorities, including the daily number of confirmed, severe, and critical cases, as well as the number of deaths, in Wuhan and Guangzhou, from January to February.

Summary of Main Findings

The data from Wuhan and Guangzhou showed that even with strict lockdown measures, there was a delay between the start of these measures and the peak in hospitalised cases. Therefore, although these measures reduced the burden on healthcare, the burden in Wuhan in particular was still large. The authors predicted that depending on the US city, 2-5 per 10,000 adults would require critical care at the peak of a Wuhan-like epidemic. This number only differed slightly when taking into account differences in age structure (2.1-4.0 requiring critical care) versus differences in co-morbidity (2.6-4.9 requiring critical care). Since there are 2.8 ICU beds per 10,000 adults in the US, the authors concluded that healthcare capacity could be exceeded if a Wuhan-like outbreak were to occur in some of these cities.

Study Strengths

This study was posted on a preprint server on March 13, 2020. At this time, about 500 cases were confirmed in the US daily, and stay-at-home orders were not in place in any states. It was therefore not possible at this early stage to predict ICU bed requirements conditional on lockdown, without extrapolating from other countries. By doing so, this study identified the need to increase healthcare capacity, even with strict physical distancing measures.

This study made one major assumption: that the risk of requiring critical care at the peak of an epidemic in a US city would be the same as in Wuhan, for a given risk group. A more common approach is to estimate many epidemiological quantities, such as the reproductive number, the average number of contacts between people of different ages, and the effectiveness of interventions. These estimates are then used in a model to project the number of cases or hospitalisations. Errors in each of these estimates are large at the start of an epidemic of an emerging infectious disease, and can compound to produce large errors in model predictions. Hence, most mechanistic models do not attempt to predict beyond a few weeks. This study circumvents the need to estimate many unknown quantities.

Limitations

Epidemics in US cities could differ in many ways from the Wuhan epidemic. The major difference is the effectiveness of physical distancing measures. The authors argued that lockdown measures in the US were unlikely to be stricter than those in Wuhan, so their estimates of ICU bed requirements were likely to be conservative. Other potential differences include different interventions, different population densities, and different contact rates between age groups. Also, risk factors other than age and hypertension were not considered. Together, these could make a large difference to ICU bed requirements. The authors also did not consider differences in ICU capacity between US cities.

The pandemic has progressed vastly since this manuscript; we now have data on the lockdown measures introduced in each city, and the number of cases and hospitalizations since then. Researchers are thus in a much better position to use local data to estimate the parameters of transmission and predict healthcare burden, rather than extrapolating from other countries.

Value added

This study sounded an early warning on the potential of COVID-19 to overwhelm healthcare capacity, under conservative assumptions. It accounted for differences in the need for ICU care based on age and hypertension.

Our take —

This study was available as a pre-print, and thus not yet peer reviewed. Using data from two distinct outbreaks, where close contacts of confirmed cases were monitored and quarantined, and low risk contacts were placed under active surveillance in Singapore and Tianjin, China, this modeling study estimated the serial interval to be shorter than the incubation period suggesting pre-symptomatic transmission with infection occurring 2.55 (Singapore) and 2.89 (Tianjin) days before symptom onset in the primary case. This paper provided early evidence of pre-symptomatic transmission of SARS-CoV-2. If corroborated, isolation of detected cases alone is unlikely to curb the spread of the pandemic.

Study design

Modeling/Simulation

Study population and setting

In this study authors estimated the incubation period and serial interval distribution of SARS-CoV-2 using data from transmission clusters in Singapore (93 cases) and Tianjin, China (135 cases), reported in January and February 2020. Using these estimates, authors calculated the basic reproductive number, R0, and the extent of pre-symptomatic transmission.

Summary of Main Findings

The authors estimated the mean incubation period as 7.1 days (95% CI: 6.1, 8.3) in Singapore and 9 days (95% CI: 7.9, 10.2) in Tianjin. The mean incubation period was shorter for cases occurring earlier in the epidemic in both locations. Using the first 4 cases in each cluster, the mean serial interval was estimated to be 4.6 days (95%CI: 2.7, 6.4) in Singapore and 4.2 days (3.4, 5.0) in Tianjin. Estimates of the serial interval (the time between symptom onset of the primary case to symptom onset in the secondary case), and incubation period (the time between infection and symptom onset) suggest there is pre-symptomatic transmission, with shorter serial intervals compared to incubation period estimates. Infection occurred, on average, 2.9 days (Tianjin) and 2.6 days (Singapore) before symptom onset. The estimated R0‘s in both settings were estimated to be 2.

Study Strengths

Serial interval and incubation period directly estimated using information from COVID-19 transmission clusters from two distinct locations (Singapore and Tianjin, China). Parameter estimates were robust in sensitivity analyses and inferences were congruent in both locations.

Limitations

The timing of exposure and presumed infectors are uncertain, and the analysis does not does not account for uncertainty in the date of symptom onset within the clusters. This impacts our confidence in both the serial interval estimates and the timing of pre-symptomatic transmission because the serial interval is defined by symptom onset in case pairs. Any changes to estimates in the serial interval would influence estimates of how much transmission occurs before an individual showed symptoms.

Value added

Early studies estimate COVID-19 transmission parameters using prior information from SARS. In this study, the incubation period and serial interval are directly estimated using information from COVID-19 transmission clusters.

Our take —

In an analysis of available SARS-CoV-2 genomes, two genomic types were identified (L and S). However, the “aggressive” nature of the L strain should not be interpreted as more virulent or pathogenic. The higher frequency of this strain in the early outbreak may have been due to higher transmission rates but may also be explained by neutral effects and sampling biases and should thus be interpreted with caution.

Study design

Ecological; Modeling/Simulation; Other

Study population and setting

The data used in the study include genetic information from SARS-CoV-2 infecting humans and related coronaviruses in animals: the reference genome for SARS-CoV-2 (NC_045512), human SARS-CoV, four bat SARS-related coronaviruses (SARSr-CoV: RaTG13, ZXC21, ZC45, and BM48-31), one pangolin SARSr-CoV from Guangdong (GD), and six pangolin SARSr-CoV genomes from Guangxi (GX). Additional analysis focused on 103 SARS-CoV-2 genomes publicly available (on GISAID) from patients inside and outside Wuhan.

Summary of Main Findings

The authors find that the SARS-CoV-2 virus is most closely related to the bat SARS-related coronavirus RaTG13 based on its genetic code (differing ~4% on average across the whole genome), but that there was a larger difference when looking at genetic sequences that are not experiencing natural selection and cause no changes in the proteins that are produced from the genetic code. The authors also find that changes in critical amino acids in the spike protein of SARS-CoV-2 are more similar to a pangolin coronavirus versus bat SARSr-CoV RaTG13, but based on differences in adjacent amino acids near the key residues it is likely that this similarity arose from adaptation of viruses separately within pangolins and humans. Finally, their analysis of 103 SARS-CoV-2 genomes revealed two distinct types (L and S) that differ at two positions in the genetic code. The L type was more common at the early stage of the outbreak in Wuhan (prior to January 7, 2020) compared to after this date and in other locations outside of Wuhan.

Study Strengths

Provides an in-depth analysis of patterns of natural selection on the genomes of SARS-CoV-2 and related coronaviruses.

Limitations

There is evidence from this study that there are multiple genetic types of SARS-CoV-2 circulating. However, their inference that the L type has a higher transmission rate or is more “aggressive” than the S type cannot be concluded from the data presented. Furthermore, “aggressiveness” of a virus can mislead readers into thinking the L type is more virulent, rather than the intended meaning of higher transmissibility. Without access to patient data to assess transmission rates, the authors rely on prevalence rates alone to infer fitness differences between types, yet their inference does not account for other neutral effects (genetic drift, founder effects as types are seeded to different countries). Additionally, insufficient and biased sampling in the early outbreak is likely to be a significant confounding factor in prevalence estimates and any related inference about transmissibility.

Value added

The study indicates that SARS-CoV-2 is evolving, however predominately in a neutral manner, resulting in identifiable genomic types.

Our take —

Using knowledge gleaned from a decade of studies on receptor binding in coronaviruses, this study finds that SARS-CoV-2 likely uses the ACE2 receptor to infect human cells and has favorable amino acid changes that may facilitate human-to-human viral transmission. The virus also shows affinity for other animal ACE2 proteins, which might act as intermediate hosts. However, experiments will be necessary to establish whether these animal species can be successfully infected.

Study design

Ecological, Modeling/Simulation, Other

Study population and setting

The study examines the molecular interactions between SARS-related coronaviruses and cells of their human and animal hosts. Specifically, the authors look at key changes in a special region of the coronavirus spike protein that binds to a receptor protein on cells of its animal and human hosts, the angiotensin-converting enzyme 2 (ACE2), and assess how changes might lead to an increased ability to bind and infect host cells.

Summary of Main Findings

Based on the genome of SARS-CoV-2 and existing knowledge of the interaction between SARS-related coronaviruses and human ACE2, it is likely that the novel virus is capable of infecting human cells by binding to ACE2. Several amino acid changes in SARS-CoV-2 provide favorable binding to sites on human ACE2, which suggests that there is some latent ability for human-to-human transmission of this novel virus. Comparison of ACE2 proteins in other animals indicates that SARS-CoV-2 may be capable of infecting pigs, ferrets, cats, and nonhuman primates, but not mice or rats, as intermediate hosts.

Study Strengths

Capitalizes on years of research on receptor binding in SARS-related coronaviruses collected since 2002, including artificial selection experiments to find the ideal amino acid residues for binding to human ACE2.

Limitations

Computer models of proteins can assess the affinity of the novel coronavirus to human ACE2 but cannot account for other factors that could affect the affinity of SARS-CoV-2 to human cells; these factors must be assessed with tissue culture or animal infection experiments. Likewise, analysis of ACE2 homologues in other animals does not establish that the novel virus will successfully infect these animals, only the potentiality. Experimental infections will be necessary to establish whether the virus can replicate in these intermediate hosts.

Value added

Assesses the favorable substitutions that increases the binding affinity of SARS-CoV-2 to human cells and identifies potential intermediate reservoir host species that may have been involved in the transition of the novel coronavirus from bats to humans.

Our take —

Early major study of COVID-19 estimating the final size of COVID-19 in Wuhan and further projecting the spread within and outside mainland China. Findings from this study support ongoing, self-sustained outbreaks in other metropolitan areas due to the high basic reproductive number (estimated at 2.68), and the exportation of pre-symptomatic cases, to other large cities, from Wuhan early in the epidemic. Large scale public health interventions will be needed, globally, to curb the epidemic. Due to the paucity of data at the time of publication, estimates of key epidemic parameters are likely to change as data within and outside mainland China are updated.

Study design

Modeling/Simulation

Study population and setting

Using data from December 31, 2019 to January 28, 2020 on the number of cases exported internationally from Wuhan, authors estimated the size of the epidemic in Wuhan and forecasted the national and global spread of COVID-19 accounting for the Greater Wuhan region quarantine and other non-pharmaceutical interventions.

Summary of Main Findings

Using data on monthly flight bookings from the Official Aviation Guide (OAG) and confirmed case counts from reports published by the Chinese Center for Disease Control and Prevention (CDC), authors estimated the number of secondary cases one case would infect in a completely susceptible population, (R0), for COVID-19 was 2.68 (95%CI: 2.47-2.86) as of January 25, 2020, and the time it took for the epidemic to double in size was 6.4 days (95% CI: 5.8-7.1). Authors further estimated 0.40% (95% CI: 0.20-0.69%) of the 19 million people in Greater Wuhan had been infected as of January 25, 2020.

Study Strengths

This study provides early estimates of the R0 of COVID-19 and other important COVID-19 epidemic parameters. The study pools information from multiple sources, including line list data from the Chinese CDC and up to date mobility data from OAG and Tencent.

Limitations

This was one of the early papers published about COVID-19; due to the uncertainty surrounding the pathogen, used serial interval estimates based on SARS which are longer than current estimates for COVID-19 (roughly 8 versus 4 days) . In addition, in the analysis, authors assumed travel behavior was not affected by disease status, and that all infections would eventually be symptomatic.

Value added

One of the first major studies published about the COVID-19 outbreak in Wuhan. The study inferred the outbreak size within Wuhan and forecasted the spread of COVID-19 within and outside mainland China in the absence of a robust and reliable line list data characterizing the epidemiology of the pathogen.