Skip to main content

Our take —

This modeling study estimated that in 45 African countries, the cumulative number of COVID-19 cases will surpass 1,000 by the end of April 2020, and 10,000 by mid-May 2020. Due to model assumptions such as no intervention efforts and a constant proportion between reported and unreported cases, this model should only be used for short-term projections. Intense responses have already slowed epidemic growth in several African countries, limiting this model’s applicability.

Study design

Modeling/Simulation

Study population and setting

Using WHO situation reports (SITREPS), authors simulated the cumulative number of reported COVID-19 cases in 45 African countries that had reported at least one case as of March 24, 2020. Using a branching model (a process that models a population in which each individual in generation X produces a random number of offsprings [i.e., secondary cases] for generation X+1), authors used reported case numbers as a proxy for the actual number of cases to forecast future epidemic trends in each country. Ten thousand epidemics were generated per country, and calendar dates by which 1,000 and 10,000 cumulative cases would occur were identified. Conditional quantiles were used to evaluate prediction accuracy (i.e., if a certain date in a country was the 0.95 quantile, then the probability of reaching the case number threshold before or at that date is 95%, whereas if the quantile for that date is 0.5, then the probability of reaching that case threshold before or at that date is 50%).

Summary of Main Findings

The model estimated that all 45 countries will reach 1,000 cumulative cases by the end of April 2020 (if not earlier), and all will reach 10,000 by mid-May 2020, only a few weeks behind the date of 1,000 cases.

Study Strengths

In a method similar to conducting a “positive control” in a laboratory experiment, authors validated the model by applying it to countries that have already exceeded 1,000 cases. Simulated epidemics started with the first cases in each country reported to WHO SITREPS.

Limitations

Authors assumed the number of reported cases was directly proportional to the number of true cases, and that there were always sufficient numbers of unreported cases to sustain transmission. Projections assumed both failed containment and no interventions to reduce transmission early in the epidemic, so these estimates are likely overestimated. Unlike other simulations in African countries, this model assumes surveillance capacity would not be overburdened, and that this would artificially slow accumulation of cases while the unreported epidemic grows unhindered.

Value added

This study adds to the growing body of projections for the impact of COVID-19 in low- and middle-income countries.

Our take —

This study showed that the combination of interventions implemented since January 23, 2020 may have reduced and maintained the reproductive number below 1 across multiple provinces in China. Control measures must be lifted carefully, and the epidemic closely monitored, to maintain the reproduction number below 1 to prevent a second wave of infections. However, as exact policies of lifting specific interventions such as school or workplace closures were not explored, it is difficult to conclude how the authors’ findings could translate to policy decisions.

Study design

Modeling/Simulation

Study population and setting

Using data on confirmed cases of COVID-19 from January and February 2020, the authors estimated i) how transmissible (defined by the reproductive number) the virus was over time in Beijing, Shanghai, Shenzhen, Wenzhou, and the 10 provinces with the largest number of confirmed cases; and ii) how severe (the proportion of confirmed cases who died due to COVID-19) the virus was in all 31 provinces in China. The authors used data from publicly available sources and detailed linelists to reconstruct the epidemic for these analyses. A simple SIR model (in which a population is stratified into three categories: Susceptible to the virus, Infected with the virus and infectious, or Recovered) was used to explore the potential effects of relaxing non-pharmaceutical and social interventions under different scenarios on the relative increase in case counts and the time required to push the disease prevalence back to pre-relaxation levels.

Summary of Main Findings

The authors found that COVID-19 transmissibility across mainland China declined since January 23, 2020, suggesting that the mass public health interventions including social distancing, travel restrictions, and other behavioural changes were effective. The reproductive number in the cities and provinces studied has remained below 1 until the end of the study period (February 29, 2020), and the number of new local cases continued to decline. Estimates of the proportion of confirmed cases who died due to COVID-19 varied across the ten provinces that reported the largest number of confirmed cases from 0.00% (uncertainty: 0.00–0.58%) in Jiangsu to 1.76% (uncertainty:1.11–2.65%) in Henan. Overall, the proportion of confirmed cases who died outside Hubei (0.98%, uncertainty:0.82–1.16%) was significantly lower than in Hubei province (5.91%, uncertainty: 5.73–6.09%). Transmission models exploring the consequences of relaxing non-pharmaceutical interventions showed that, in the absence of herd immunity, relaxation of measures must maintain the reproductive number below 1 to prevent a potential second wave of infections.

Study Strengths

Authors used data from multiple provinces across mainland China to give an overall view of the epidemic outside of Hubei province. Empirical estimates of epidemiological delays were used to adjust estimates of severity and the reproductive number. Authors checked whether their findings were affected by unknown quantities such as the proportion of imported cases from Hubei to the 10 provinces studied.

Limitations

As the symptom onset to report delays were not available for all provinces, authors assumed that this delay was the same as that estimated for Beijing. The severity estimates were adjusted based on a relatively small number of onset to death observations from Wuhan City. The lower severity estimates outside of Hubei province may have been affected by a combination of (i) increased testing and detection of milder infections and (ii) changes in the case definition over time.

Value added

This is one of the first comparisons of transmissibility over time and severity of COVID-19 in different provinces across China, and the impact of relaxation of interventions on future case numbers.

Our take —

Using mathematical modeling, the authors show that containment policies and social distancing measures effectively slowed the growth in confirmed COVID-19 cases in China.

Study design

Ecological; Modeling/Simulation

Study population and setting

The authors used an SIR model variant to assess the impact of containment strategies and social distancing measures on daily confirmed COVID-19 cases in mainland China between January and February 2020. A range of measures were implemented in China including: 1) quarantining the sick and suspected in hospitals, at home or monitored house arrest; 2) social distancing measures including closing businesses, universities and mandatory curfews in some areas; 3) stricter handwashing guidance and use of masks; and 4) contact tracing.

Summary of Main Findings

The confirmed case trajectories in each province initially followed a rapid exponential growth trajectory, but then slowed down, moving towards a flattened curve. The authors found that adding two features to a Susceptible Infectious Recovered (SIR) model — effectively removing individuals from the susceptible pool through social distancing, handwashing, masks and contact tracing; and removing infected individuals through quarantine — was sufficient to reproduce the shapes of the confirmed case curves.

Study Strengths

The model was simple and parsimonious, and fit observed data well. The model was robust when applied to different provinces in China.

Limitations

The analysis was restricted up to February 12, 2020, as changes to the case definition for COVID-19 led to an increase in the number of cases after this date. As with similar analyses, efficacy of specific mitigation measures were not assessed but rather, an overall average effect is described. Changes in testing strategies were not described in this paper and may impact confirmed case numbers.

Value added

The simple model developed in the study can be applied to epidemics in different areas to assess the impacts of quarantine and social distancing. The study adds to the already large evidence base for the effectiveness of quarantine and social distancing measures in reducing the spread of COVID-19.

Our take —

Based on data on critical care capacity across the US, a combination of substantial increases in the number of critical care beds and implementation of significant control measures including travel restrictions, self-quarantine, greater access to rapid testing, and behavior changes are needed to reduce the number of contacts to prevent hospitals from being overwhelmed leading to otherwise preventable deaths.

Study design

Modeling/Simulation

Study population and setting

A mathematical model previously developed by the authors was used to simulate the spatio-temporal dynamics of SARS-CoV-2 infections between February and March 24,2020 across 3108 counties in the US. The impact of control measures, including travel restrictions between areas, self-quarantine, and social distancing, was modeled by assuming 0, 25, and 50% reductions in contact patterns. The impact of testing and changes in health-seeking behavior were also considered. Hospital critical care supply was calculated for all continental US counties by linking and harmonizing data from multiple sources including: i) 2020 Centers for Medicare & Medicaid Services; ii) 2018 American Hospital Association (AHA) Annual Survey; iii) 2020 US DHHS Health Resources and Services Administration; and iv) 2017-2019 CMS Medicare Provider of Services. From a baseline of 30% availability of existing ICU beds, four surge capacity scenarios were considered. Excess deaths (the additional deaths compared to historical averages over the same time-frame) due to potential lack of critical care beds was assessed.

Summary of Main Findings

The authors estimated that up to 20,000 (with increased levels of control) and 11,000 (with increased critical care capacity) excess deaths could be averted over the period February 21 – March 24, 2020. Counties in the Northeast of the US were most affected by shortages in critical care beds, and thus projected to have the highest number of excess deaths. Counties in New York, Colorado, and Virginia were projected to exceed critical care limits within the 4 week study period even with high levels (50%) of contact reduction measures and high surge capacity. Spatial clustering of counties at risk of exceeding care capacity was observed in the very low surge capacity and 0% contact reduction scenario, highlighting regions that may require additional resources due to limited existing healthcare capacity. Urban areas were projected to exceed care capacity faster in every scenario compared to rural areas. Under some scenarios of low surge capacity, time to exceeding bed limits was as short as 1-2 weeks from the start of the study period. Urban counties in the Northeast US had the highest proportion of excess COVID-19 deaths that could have been averted with access to critical care.

Study Strengths

The study estimates the critical care capacity at county level from well-established data sources and links this explicitly to projected COVID-19 case numbers. By linking this to projected case numbers of COVID-19 over time and by counties, the level of control, the additional critical care capacity required, and the lead time before health services are overwhelmed can be estimated.

Limitations

Critical care capacity was only assessed in terms of physical equipment and hospital beds. Staffing requirements were not accounted for. Details of the model used to simulate COVID-19 cases across the US were not reported within the paper. These models were hypothetical planning scenarios that did not compare model trajectories to real data. In addition, the modeled scenarios were not tied to empirical data on the effectiveness of interventions in a way that facilitates comparison to real-world situations.

Value added

This was one of the first examinations of when US counties might exceed critical care capacity due to surges in COVID-19 cases which may help hospital and public health resourcing and planning.

Our take —

This paper is available as a pre-print and therefore not yet peer-reviewed. Reduced movement was observed across cities in 23 countries across the world after the introduction of social distancing measures. Although authors estimated reductions in epidemic growth and transmissibility associated with reduced mobility patterns, strong assumptions were made about representativeness of the data and delays from infection to case report across 41 cities which may have biased the results.

Study design

Modeling/Simulation

Study population and setting

The Citymapper Mobility Index (CMI) which measures the relative frequency of planned trips (compared to an internal reference point from late 2019 or early 2020) using the Citymapper application was used as a proxy measure of adherence to social distancing measures implemented across 41 cities in 23 countries from March 2, 2020. Using reported cumulative case numbers over time, authors estimated how fast the total case numbers were growing in each of these 41 cities over 3 weeks beginning March 23 to April 6, 2020. Transmissibility as estimated by the instantaneous reproduction number was estimated for the weeks of March 23, March 30, and April 6, 2020 using the incidence of reported cases between March 8 to April 12, 2020. The authors then assumed a 14 day delay between infection and date of report to estimate the association between CMI and the mean daily growth rate and the reproduction number.

Summary of Main Findings

Authors found that declines in mobility (relative frequency of planned trips) corresponded with a decline in epidemic growth in 41 cities. The majority of cities saw a substantial reduction in mobility, as measured by the CMI, from a mean of 97.6% on March 2 to 12.7% on March 29, 2020. Similar patterns of reduced mobility were observed across cities in Europe, the Americas, and Australia corresponding to the implementation of national or subnational social distancing measures and mandatory closures e.g. of non-essential retail. A 10% reduction in mobility was associated with a decrease in the daily growth rate of 14.6% and a 0.061 reduction in the reproduction number 14 days later.

Study Strengths

Authors used automatically collected app-based data to measure changes in mobility due to social distancing measures. Such data may be more accurate than self-reported behavioural changes. Authors checked whether their findings were affected by the timing of the epidemic and their assumption about the average delay between infection and case report.

Limitations

Whilst the mobility data are only available at the city-level, for some countries COVID-19 case counts were only available at national or sub-national level. Therefore the reduction in mobility in certain urban cities will not necessarily be representative of the whole country. The mobility data only applies to Citymapper app users which is not representative of the whole city, nor covers any journeys by car. The mobility metric also only captures the decline in the relative frequency of trips planned, and does not capture other indicators such as changes in the types of trips planned. Authors assumed a crude 14 day delay between infection and case report, which was the same for all countries considered. In reality this delay will differ significantly between settings due to healthcare capacity and testing policies, and may bias estimates. This bias may also apply to the reproduction number estimate which was based on cases by date of report rather than date of symptom onset. Finally, reduction in mobility does not necessarily equate to a reduction in social and physical contact that could lead to decreased transmission.

Value added

Findings are not novel, but similar to previous studies demonstrates the utility of mobility data to assess the potential impact of social distancing interventions.

Our take —

This study used mathematical models to estimate the impact on COVID-19 epidemics of identifying recovered people with protective antibodies to the virus and putting them in essential roles which involve lots of contact with others. Their model results suggested that this strategy could have a substantial beneficial impact on the epidemic. It is not clear how and at what level this strategy could be implemented in real life, and there are ethical and legal implications that would need to be addressed.

Study design

Modeling/Simulation

Study population and setting

The model was of the general population, with no specific setting mentioned. The age distribution of the population was taken from different states in the United States.

Summary of Main Findings

The study used mathematical models to estimate the impact of identifying people who have recovered from COVID-19 and have protective antibodies to SARS-CoV-2 virus, and deploying them preferentially in roles requiring high levels of interaction with others (e.g. in healthcare, caring for the elderly, schools, food supply). The authors called this approach ‘shield immunity’. Their model findings suggested that the shield immunity strategy could substantially reduce the total number of COVID-19 infections and deaths, reduce the duration of the epidemic and reduce the burden on the healthcare system. Their results suggested that shield immunity could be effective in populations with different age profiles. As it was not known what proportion of COVID-19 infections are asymptomatic (have no symptoms), nor how long protective immunity might last for, the authors re-ran their model with different proportions of infections being asymptomatic, and with different durations of protective immunity, and found that the model still suggested that shield immunity would be effective in all cases, but particularly if fewer infections were asymptomatic, or if immunity lasted for at least 4 months. Their results also suggested that shield immunity provided additional benefits when combined with social distancing.

Study Strengths

The model structure and assumptions were clear and transparent, and the authors checked whether their findings were affected by unknown quantities including the proportion of infections that do not have symptoms, and how long protective immunity lasts for.

Limitations

The model was not fitted to data from a specific location, meaning it may not have represented a realistic epidemic. The shield immunity strategy was represented in the model in a non-intuitive way that makes it hard to understand exactly what the strategy would look like in real life, and hard to know whether the strategies they model could be realistically achieved. The processes of identifying people with immunity (e.g. through antibody testing) and of substituting them into high-interaction roles were not explicitly represented in the model so it is not clear what levels of testing or coverage of substitution would be needed to have the impact that they found. Current uncertainties around whether antibodies are good surrogates for protective immunity were not sufficiently stressed. Ethical and legal issues around employment differing by immune status and concerns that people may deliberately try to become infected so that they can gain immunity to improve their employment prospects were not addressed.

Value added

This study is one of the first to assess the potential impact on the COVID-19 epidemic of identifying people with immunity to the virus and preferentially placing them in roles with high levels of contact with others.

Our take —

Using survey data, this study showed that in-person contacts dramatically declined following implementation of strict social distancing measures in the UK, to levels that would substantially slow epidemic growth. Participants in this study were recruited by email and may have different contact patterns than the population at large, potentially biasing study conclusions.

Study design

Modeling/Simulation; Cross-Sectional

Study population and setting

A cross-sectional survey was administered through email to adults in the United Kingdom (UK) aged 18 years and older. The survey assessed contact patterns beginning one day after the UK implemented a national lockdown to control SARS-CoV-2 transmission. The lockdown included (1) a stay-at-home order whereby individuals were only allowed to leave the home for essential activities such to buy groceries or medicines and to exercise once per day; (2) closure of schools; (3) closure of restaurants, bars, gyms, and other leisure/hospitality businesses; and (4) a ban on mass gatherings, including sporting events. Social contact patterns were compared before and after the lockdown using data from an earlier study that similarly measured patterns (the POLYMOD survey). Contact data from the current survey were then used to estimate the reproductive number following implementation of the lockdown.

Summary of Main Findings

1,346 individuals participated and reported 3,849 in-person contacts. The mean number of in-person contacts declined from 10.8 per day before the lockdown to 2.9 per day after the lockdown. Most contacts (~58%) after the lockdown were in the home. The estimated reproductive number following the lockdown was 0.62 (95%: 0.37-0.89), well below the epidemic threshold for growth.

Study Strengths

This study directly measured impact of social distancing measures on contact patterns using highly detailed age and gender-specific contact data measured by personal diary. This type of contact data is rare and useful for epidemiological studies. The survey also had good geographic and demographic representation among those who participated.

Limitations

This study surveyed and recruited participants by email; participation in email surveys may be related to contact patterns, which would bias study results. Adolescents and children were excluded from the survey and their contact patterns were imputed. Additionally, contact patterns were measured immediately after implementation of the lockdown. The nature of contacts may change as people become fatigued by lockdown measures or as the incidence of disease rises and falls in the population.

Value added

This study provides some of the first empirical data from Europe documenting the extent to which age-specific contact patterns changed following implementation of strict social distancing measures in response to SARS-CoV-2 transmission.

Our take —

This study highlights an overlooked cost of school closures: possible increases in COVID-19 mortality associated with health care workers leaving work because of increased child care burden. The model ignores several potentially important dynamics, and is reliant on assumptions with considerable uncertainty, so its estimate of the critical parameter value should be interpreted with caution. It is nonetheless useful in calling attention to real-world tradeoffs.

Study design

Modeling/Simulation

Study population and setting

This modeling study considers the population of health care workers (HCWs) in the US, drawing on data from the US Current Population Survey on family structure to estimate child-care requirements. The study estimates the critical value of a parameter (the decrease in COVID-19 survival probability associated with a reduction in the health care workforce) at which school closures would “break even” in terms of overall mortality.

Summary of Main Findings

An estimated 29% of US HCWs provide care for children aged 3-12 years. Under model assumptions, a 17.6% increase in mortality probability per COVID-19 patient, occasioned by a reduction in the HCW labor force to meet child-care needs, would offset the benefits (in terms of reduced transmission and thus mortality) from nationwide school closures.

Study Strengths

The study draws on detailed data on household structure among HCWs from a large, nationwide, monthly survey.

Limitations

The model assumes that HCWs will find no alternative child care options upon school closures. The model ignores any effects of health care labor on non-COVID-19 mortality, which may be considerable. The model ignores the mortality benefits of removing HCWs from high-risk transmission environments. HCWs are treated homogeneously; the relationship between COVID-19 survival probability and health care labor is likely highly complex and spatially idiosyncratic.

Value added

This study addresses an overlooked aspect of school closures: namely, the fact that they impose a large child care burden on health care workers with real costs in terms of public health.

Our take —

This study showed how control of local epidemics in China outside Hubei province improved over time with physical distancing measures. It added to the growing evidence that substantial transmission could occur before symptom onset.

Study design

Modeling/Simulation

Study population and setting

The study collected data on the age and sex distributions, the proportion of locally transmitted cases, and the proportion of cases associated with exposure to known cases, among laboratory-confirmed cases in China outside of Hubei province, from December 2019 to February 17, 2020. Individual-level information was used to estimate the distributions of time from symptom onset to hospital admission, time from exposure to illness onset (incubation period), and time between symptom onset of primary and secondary cases(serial interval). These quantities were estimated separately for the time periods before and after January 27, 2020. For nine locations with sufficient data, the study estimated the effective reproduction number over time.

Summary of Main Findings

Between the first and second time period, the proportion of cases with exposure to COVID-19 cases or patients with acute respiratory infections increased from 26% to 66%, and the proportion of cases with exposure to Hubei decreased from 73% to 35%. The mean time from symptom onset to quarantine (through hospital admission) improved from 4.4 days to 2.6 days (p < 0.0001). Across the two time periods, the mean incubation period was 5.2 days (95th percentile at 10.5 days), and the mean serial interval was 5.1 days (with 95% of serial intervals between 1.3 and 11.6 days). Because a large proportion of serial intervals are smaller than the mean incubation period, a substantial amount of transmission could occur before symptom onset, potentially making contact tracing, isolation of infectious individuals, and airport screening difficult. In all nine locations, the effective reproduction number had decreased below 1 by February 8, 2020, indicating that interventions had effectively controlled local epidemics.

Study Strengths

The availability of individual-level data by location, and data on transmission pairs, enabled estimation of the distributions of times to key events, and changes in the effective reproduction number by location. Stratifying these distributions by time period showed the shift in local epidemics from imported cases to local transmission, but also revealed improved epidemic control over time. The individual data collected for this study were made public, enabling further analyses.

Limitations

The study did not explicitly account for expanding case definitions over time, and so the estimated effective reproduction numbers should be treated as upper bounds. The results of the study are also sensitive to changes in reporting rates over time.

Value added

This study was one of the first to characterize transmission in China outside Hubei. It revealed that key quantities for epidemic control such as the serial interval and mean time to isolation were shorter than those estimated for Hubei for previous studies, suggesting better control of the epidemic outside Hubei.

Our take —

In this study, available as a preprint and thus not yet peer reviewed, using influenza-like (ILI) surveillance systems, authors estimated that large surges observed in ILI beginning in March 2020 were due to COVID-19, but by the end of March, the case detection rate across the United States was only 12.5%. Analyses of morbidity and mortality doubling times suggest that SARS-CoV-2 has rapidly spread throughout the country since its introduction on January 15, 2020. Although ILINet is a robust surveillance system, authors make several assumptions in their study and model that contribute to uncertainty, and these results should not be used for hospital surge projections or other healthcare demand forecasts.

Study design

Modeling/Simulation

Study population and setting

Leveraging existing influenza-like illness (ILI) surveillance systems in the United States (ILINet) and data from the past 10 years, this study modeled the excess number of non-influenza ILI to estimate the true prevalence of SARS-CoV-2 infection. Using excess ILI estimates as a proxy for COVID-19 burden, authors then identified surges in non-influenza ILI, estimated ILI admission rates and prevalence of SARS-CoV-2 from the surge, estimated the rate at which patients who are positive for SARS-CoV-2 and have ILI symptoms are positively identified as having COVID-19, and estimated epidemic growth rates (e.g., doubling time, or the length of time it takes for an epidemic to double in size) using an SEIR model.

Summary of Main Findings

Authors identified surges in non-influenza ILI that exceeded expected levels beginning in early March 2020. Some states reported additional non-influenza ILI in excess of up to 50% higher than any previously reported level. Authors were able to assess care-seeking for ILI in New York City, and found that both care-seeking and admission to hospital emergency departments for ILI increased during the month of March, suggesting that care-seeking for mild ILI actually decreased. The rate at which COVID-19 patients were identified using ILI surveillance varied largely by state and over time, but increased over the month of March (from approximately 1% to 12.5% across the United States). Authors estimated that COVID-19-associated deaths doubled every 3.01 days across the United States during the month of March. Assuming the ILI surge is due entirely to COVID-19, authors estimated the slowest doubling time for cases across the United States, starting January 15, is 4 days.

Study Strengths

ILINet is a robust syndromic surveillance platform, and authors were able to estimate background non-influenza ILI levels using ten years of historic data.

Limitations

Authors assumed the patient population reported by providers to ILINet is representative of their entire state. However, ILINet is limited to approximately 2,600 voluntarily enrolled outpatient providers across the United States and may not be representative of statewide populations, and likely underestimates the number of influenza and/or COVID-19 cases. To account for this latter limitation, authors scaled the data in order to compare ILINet data to confirmed COVID-19 case counts per state. Authors also assumed that SARS-CoV-2 is entirely responsible for the excess non-influenza ILI (i.e., the identified surge). Although this is likely, not accounting for increases in other circulating viruses, such as other coronaviruses or RSV, contributes to uncertainty in the model results. However, this assumption does become more robust as the prevalence of SARS-CoV-2 continues to increase. Asymptomatic infections would not be captured in ILINet. The SEIR models were US-wide, and as such were unable to capture regional variations in transmission or the effects of interventions.

Value added

This study is the first to leverage ILINet, a pre-existing and robust influenza-like illness surveillance system, to quantify the prevalence, detection rate, epidemic growth rates, and clinical rates of SARS-CoV-2 in the United States.