Skip to main content

Our take —

This study used comprehensive data from China on population mobility patterns and SARS-CoV-2 case detection to estimate the individual and combined impact of non-pharmaceutical interventions (NPIs) on SARS-CoV-2 spread. Results showed that rapid detection and isolation of cases through contact tracing had the greatest impact on averting cases, but social distancing and contact restrictions had a similar magnitude of effect. The combined impact of all interventions resulted in the most infections averted; travel restrictions were implemented too late to substantially affect the overall case counts. The study shows that even a week’s difference in the timing of NPI implementation can have a substantial impact on the trajectory of an epidemic, highlighting the need to act early and decisively when epidemics emerge. The NPIs assessed here represent rough aggregates of individual measures, and some of their components may not be replicable outside China.

Study design

Modeling/Simulation

Study population and setting

This study of SARS-CoV-2 spread within mainland China from December 1, 2019 to April 30, 2020 used mobile phone (i.e., smartphone) data on inter-city and inner-city population movement from three sources: aggregated inflow and outflow data from 340 cities in China, a historical city-specific relative movement matrix from 2014-2015, and county-level daily population movement data used to estimate within-city travel. Additionally, the interval between illness onset to case confirmation was estimated at the county level. The authors used these data different data sources to assess impact of three groups of non-pharmaceutical interventions (NPIs): 1) prohibition of travel between cities to prevent intercity viral spread, 2) early identification and isolation of cases through contact tracing, and 3) contact restrictions and social distancing measures. A susceptible-exposed-infection-removed (SEIR) model was used to simulate viral spread between and within cities in China, accounting for observed changes in contact rates due to travel restrictions and social distancing (estimated from mobile phone data) and reductions in the infectious period due to contact tracing (estimated from data on delay from illness onset to case detection). The impacts of individual NPIs and their timing were assessed by simulating different scenarios.

Summary of Main Findings

Joint implementation of all three classes of NPIs had the largest estimated effect in terms of case counts. Of the three types of NPIs assessed, early detection and isolation of cases had the greatest individual impact on the reduction in total number of cases. Inter-city travel bans were implemented after the virus had already spread outside of Hubei; these had the smallest impact on the case count of the three classes of NPIs. Simulations showed substantial impacts on case counts by varying the timing of NPI implementation in either direction; even a single week’s delay in implementation would have increased the number of new infections by threefold.

Study Strengths

This study used comprehensive mobile data on population movements (>7 billion position requests per day) at the city level to assess changes in contact patterns following travel restrictions between provinces and the implementation of social distancing measures.

Limitations

In part because of the near-simultaneous implementation of NPI measures in China, the interventions assessed are broad aggregates of multiple individual interventions: for example, “contact restrictions and social distancing” comprise many distinct measures that include school closures, bans on gatherings, etc. Similarly, case detection and isolation comprise a suite of linked public health measures, some of which may be replicable outside China and some which may not. The impact of travel restrictions and social distancing may be different outside of China. The study assumes once cases are detected, they are effectively isolated with no subsequent onward transmission; thus, the study may overestimate the impact of case detection and isolation. The estimated number of cases implies that China detected over half of all cases of SARS-CoV-2, which may considerably underestimate the size of the epidemic.

Value added

This is one of the first studies to disaggregate the impact of individual NPIs and to compare their individual versus combined effectiveness.

Our take —

Findings indicate that a large proportion of COVID-19 infections were undocumented (meaning not identified or reported) prior to the implementation of travel restrictions in China on January 23, and a large proportion of the total force of infection was mediated through these undocumented infections. These findings explain the rapid geographic spread of SARS-CoV2 and indicate containment of the virus will be particularly challenging. Reporting inaccuracies and changes in care-seeking behavior add uncertainty to these estimates.

Study design

Modeling/Simulation

Study population and setting

Authors developed a mathematical model that simulates the spatiotemporal dynamics of infections among 375 Chinese cities during the early stage of the epidemic (January 10–23, 2020) and after travel restrictions were implemented (January 24–February 8, 2020). The model divides infections into two classes: 1) documented infected individuals with symptoms severe enough to be confirmed (i.e., observed), and 2) undocumented infected individuals (i.e., not observed).

Summary of Main Findings

The best-fitting model estimated 13,118 total new COVID-19 infections (both observed and unobserved) occurred during January 10–23, 2020 in Wuhan City. An estimated eighty-six percent of people infected were infected by undocumented cases. Nationwide, 16,829 total new COVID-19 infections occurred during January 10–23, 86% of which were infected by undocumented cases. After travel restrictions were put into place, the model estimated that 65% of all infections were documented, which is up from 14% prior to travel restrictions.

Study Strengths

In a method similar to performing a “positive control” in a laboratory experiment, the authors verified the model’s ability to recuperate parameters related to unidentified infections using synthetic data prior to applying their techniques to real data.

Limitations

Uncertainty around the estimates exist due to changes in travel restrictions, control measures, reporting inaccuracies, and changes in care-seeking. The ability of a system to identify all infections are likely impacted by differences in control measures, viral surveillance and testing, and case definition and reporting, and these findings may not be applicable to countries with different control, surveillance, and reporting practices.

Value added

Study estimates prevalence and contagiousness of unidentified infections before and after implementation of travel restrictions in China.

Our take —

As a result of social distancing measures, total interpersonal contacts dropped significantly in China, and mixing among school-aged children was almost entirely eliminated. The study estimates that school closures by themselves would reduce peak incidence by a considerable amount, though they would not be sufficient to completely halt transmission. In contrast to a previous study, significant differences in susceptibility to infection by age were observed (younger individuals appeared to have lower susceptibility), though these results come from a sample of index cases that may not represent the age structure and symptom profile of all SARS-CoV-2 infections.

Preprint Review

This expert summary is for the peer-reviewed article linked above. We also summarized this paper before it underwent peer-review.  You can find the original review of the preprint by clicking here.

Study design

Cross-sectional; Ecological; Modeling/Simulation

Study population and setting

This study examined contacts reported by 636 participants in Wuhan, China, and 557 participants in Shanghai, China. The surveys were conducted from February 1 to 10, 2020, and participants were asked to report contacts on two days: a typical weekday in the pre-epidemic period of December 24 to 30, 2019, and then on the day before the survey was conducted. A contact was defined as either direct physical contact or a two-way, in-person conversation of three or more words. The authors used the survey data to estimate changes in age-mixing contact patterns. In a separate analysis of age-specific infection susceptibility, the authors used data from the Hunan CDC field contact tracing and testing of 136 index cases in Hunan Province, China and 7,375 of their contacts, identified between January 15 and March 1, 2020. Close contacts of confirmed cases were involuntarily isolated and tested for SARS-CoV-2; authors estimated odds ratios of infection for different age strata. Finally, the authors used estimated contact patterns and age-specific susceptibility in an SIR model to 1) estimate how differences in age-mixing patterns have affected SARS-CoV-2 transmission, and 2) how school closures and social distancing would be likely to affect future transmission.

Summary of Main Findings

Daily contacts dropped from 14.6 to 2.0 in Wuhan and from 18.8 to 2.3 in Shanghai. The highest density of contacts in the baseline period was observed among school-aged children; in the outbreak period, this mixing almost entirely disappeared, leaving only within-household mixing. Estimated susceptibility to SARS-CoV-2 infection increased with age: relative to the reference category of ages 15-64, the odds ratio (OR) of infection for children under 15 was 0.34 (95% CI: 0.24, 0.49); the OR for adults 65 and older was 1.47 (95% CI: 1.12, 1.92). Alterations in mixing patterns were estimated to decrease the reproduction number (R0) considerably. Social distancing measures (the broad aggregate of all measures implemented in China) were sufficient by themselves to extinguish the epidemic. Simulating pre-emptive school closures without any other intervention reduced peak incidence by 42-64%, depending on assumptions about the degree to which school-aged contacts were diminished. School closures were not sufficient to extinguish the epidemic by themselves, however.

Study Strengths

Age-specific contact matrices were estimated from surveys, and age-specific susceptibility to infection was estimated from detailed contact tracing data.

Limitations

Responses to contact surveys may be limited by recall bias, particularly for the pre-epidemic period. Respondents may have under-reported contacts in the epidemic period due to social pressure. If the age profile of the index cases and their transmission chains is not representative of the population of all individuals infected with SARS-CoV-2 (e.g., because younger individuals may experience less severe symptoms or are more commonly asymptomatic), age-specific susceptibility estimates may be biased. Estimates are derived from a small number (136) of index cases and their contacts in one geographic area. As with most studies of NPIs in China, it is difficult to isolate the effects of individual interventions since multiple measures were implemented nearly simultaneously.

Value added

This is one of the few studies to offer estimates of age-specific susceptibility to SARS-CoV-2 infection, and to estimate effects of school closures on the reproductive number.

Our take —

This study evaluated the effect of the Wuhan travel ban on the spread of COVID-19 within Mainland China and internationally. Using airport closures and airline suspensions, results from this model suggest that reducing long-range travel from Wuhan by up to 90% only delayed the epidemic in Mainland China by less than two weeks. Similarly, reductions in international cases imported from Mainland China were only maintained for a few weeks before picking back up. Overall, this study found that without other interventions aimed at reducing transmissibility (such as early case detection and isolation) implemented in concert with travel restrictions, domestic and international case reductions are modest and unsustainable.

Study design

Modeling/Simulation

Study population and setting

Using publicly available, de-identified population mobility data and data on disease importation, study authors modeled the domestic (within-China) and international spread of COVID-19. Authors also estimated the effects of the travel ban implemented in Wuhan (airport shutdowns beginning on January 23, 2020), mobility restrictions in Mainland China, and other travel restrictions adopted by other countries in early February 2020.

Summary of Main Findings

The model showed a reduction of 10% in cumulative cases within Mainland China by January 31, 2020 as a result of the Wuhan travel ban, relative to the counterfactual scenario of no restrictions. Authors assumed a doubling time (the period required for the number of cases in the epidemic to double) of 4-5 days, so this reduction corresponds to a delay in the epidemic trajectory of only 1-6 days. Furthermore, using WHO situational reports, authors estimated that only 1 in 4 infections are detected and confirmed in Mainland China. Authors estimated that 86% of international cases originated from Wuhan before the ban, and immediately after the ban, cases imported from Mainland China to other countries were reduced by 77%. However, this reduction was not sustained, and internationally imported cases rose again in the following weeks (mainly from Shanghai, Beijing, and Shenzhen). Finally, authors estimated case reduction through various combinations of travel restrictions and other interventions aimed at reducing transmissibility. Results indicate that even 90% travel reductions could at best delay the epidemic by no more than two weeks, without significant reductions in transmissibility from other measures.

Study Strengths

This study estimated spread of COVID-19 using real-world mobility data. Authors independently validated the model by comparing model projections of cumulative number of cases across Mainland China with WHO situational reports; model projections were highly correlated with observed data.

Limitations

Information used by the model regarding transmission (e.g., generation time– the time between source and recipient infections) are based on early results of the COVID-19 outbreak and epidemiology of SARS-CoV-1 and MERS; these parameters have changed as knowledge of SARS-CoV-2 has evolved. The models assumed homogeneous susceptibility and contact patterns by age, which may in fact vary considerably. Containment measures implemented prior to travel restrictions (e.g., body temperature screenings for departure in Wuhan International Airport) were not included in the model. Authors assumed long-term enforcement of travel restrictions in Wuhan through June 2020, which as of April 2020, is no longer applicable.

Value added

Effective national and international public health response planning depends on robust assessments of the effects of travel restrictions and other interventions. Results from this study indicate that the Wuhan travel ban delayed the spread of SARS-CoV-2, but was of modest utility as the epidemic continued, suggesting ongoing responses should focus on mitigation rather than containment.

Our take —

Authors estimated that disruptions to HIV services from the COVID-19 epidemic could result in 38,000 excess HIV-related deaths over a five-year period in South Africa. In some scenarios, excess HIV-related deaths exceeded those saved from social distancing measures. Authors assumed no increased risk of COVID-19 among persons living with HIV, and did not include increased transmission risk due to service interruptions or behavior change due to social distancing in the models.

Study design

Modeling/Simulation

Study population and setting

Using three potential scenarios of disruption to HIV services in South Africa due to the COVID-19 pandemic, authors modeled the number of excess HIV-attributable deaths that could occur between 2020 and 2024 as a result of such interruptions. These scenarios were “managed pause” (least severe; expansion of services paused, but current services maintained), “managed disruption” (disruptions, such as reductions to viral load testing, occur but are managed to mitigate worst impacts), and “interruption of supply” (most severe; supply of key medicines interrupted and a proportion of persons living with HIV [PLWH] are forced off antiretroviral treatment [ART]). Scenarios were cumulative, and as they increased in severity, also incorporated all components from less severe scenarios. Authors compared modeled excess deaths to a baseline model in which no interruptions occurred and coverage would have been maintained as otherwise anticipated. In the first simulation, authors assumed disruptions would begin in mid-April 2020 and last for three months. Authors also compared the effect of interruptions to HIV programs to the estimated direct effects of COVID-19 epidemic in South Africa. Different mortality risks for PLWH who are forced off ART, the proportion of PLWH forced off ART, and duration of interruptions (i.e., months) were used for sensitivity analyses and estimate excess mortality in differing scenarios.

Summary of Main Findings

Model results suggest that excess HIV-related deaths in the “managed pause” and “managed disruption” scenarios are less than 1% in the first year. Conversely, in the “interruption of supply” scenario, the model estimates an excess of 23,000 HIV-related deaths in the first year, which represents more than a 30% increase compared to the 71,000 estimated HIV-related deaths in 2018. Over the five-year period, assuming a three-month interruption, results estimated that total HIV-related deaths will be 38,000, which is less than the number of estimated COVID-19 deaths. Assuming the lowest (0.26%) mortality risk per month for PLWH who are forced to drop off ART, in some situations, the excess number of deaths was estimated to exceed the number of COVID-19 deaths prevented through mitigation efforts such as social distancing. Compared to direct effects of COVID-19, excess deaths from HIV are distributed throughout the 2020-2024 time period (whereas COVID-19 deaths all occur in 2020-2021).

Study Strengths

Authors repeated multiple model iterations using different combinations of scenarios and parameters in order to minimize uncertainty. Given the high HIV burden in South Africa, baseline estimates used for HIV-related deaths are likely robust.

Limitations

Authors did not model interactions between COVID-19 and HIV, and assume that PLWH have the same risk of acquiring or dying from COVID-19 than persons without HIV. Risks for PLWH is currently unknown, and this may have important implications for the model results. Interruptions to services and epidemic mitigation strategies may have important implications for transmission risk. Potential increases in mother-to-child transmission, new partner acquisition, and drug resistance to ART regimens may contribute to excess HIV deaths in the long-term, but were not included in the model. Finally, authors note that the risk of death is likely to increase with accrued time off of ART, but do not include this increase in the model. Limitations in the results also arise from uncertainties regarding the scale of the COVID-19 epidemic in South Africa and the extent to which HIV programs will actually be interrupted.

Value added

The maintenance of primary healthcare programs during an acute healthcare crisis is critical to mitigating overall burden. This study estimates, under varying levels of intensity, how interruptions to primary healthcare services (HIV) can worsen the overall impact of the COVID-19 pandemic, and may be useful for planners and decision-makers regarding mitigation efforts and strategies to minimize interruptions to non-COVID-19 healthcare.

Our take —

Case definitions shape our view of the evolving COVID-19 pandemic. In China, as in many parts of the world, case definitions including testing practices have changed since the start of the epidemic, making it hard to disentangle changes in incidence from changes in case definitions. Analyses from this study highlight how the true epidemic size may have been more than four times greater than that reported with the original case definitions, and how analyses not accounting for these changes can bias our assessments of transmission intensity.

Study design

Modeling/Simulation

Study population and setting

Authors reviewed the evolving case definitions for COVID-19 in mainland China, and used an exponential growth model to estimate the effect each change in case definition had on the number of reported confirmed cases from January 15 to February 20, 2020 as reported by the National Health Commission. Authors assumed changes in the case definition would result in an increase in the number of cases detected and reported relative to the total number of infections. Authors allowed the growth rate of the epidemic curve to change with each changing case definition, and adjusted for major control measures implemented in Mainland China within a statistical model.

Summary of Main Findings

Authors focused on the first five case definitions for COVID-19, each with less stringent clinical, laboratory, and/or contact criteria than the previous version (i.e., each newer version had increased sensitivity). With the exception of one case definition change that only updated definitions for severity classification, each subsequent case definition change, 2.8-7.1 times more cases were identified than would have been if the previous case definition had been used. Had the fifth version of the case definition (final in these analyses) been used throughout the outbreak, authors estimated that the number of confirmed cases by February 20, 2020 would have been over four times the number actually reported. Before January 23, 2020, authors estimated that 92% of cases went undetected by using earlier case definition versions. Overall, authors found that if changing case definitions were not accounted for, then the growth rate of the COVID-19 epidemic would be overestimated and the doubling time (i.e., time it takes for the number of cases to double) would have been underestimated (indicating faster and greater overall growth of the epidemic).

Study Strengths

Authors estimated the epidemic growth rate for Wuhan, Hubei province alone and the rest of mainland China (excluding Wuhan) separately, which allowed them to account for regional differences in parameters such as epidemic timing and transmissibility. Authors attempted to appropriately account for uncertainty in model parameters such as growth rates and doubling time.

Limitations

Authors used a statistical model that relied on exponential growth and decay only, and did not account for more complex epidemiological parameters that may have affected transmission (e.g., social distancing interventions). Authors were also unable to access information on incidence by illness onset after February 20, 2020 and were therefore unable to evaluate the effects of changes in case definition to epidemic growth after version 5.

Value added

This study demonstrates the real-world implications for evolving case definitions during an epidemic, and the importance of accounting for these changes when estimating epidemiological parameters. Results of this study suggest large and rapid epidemic growth may be artifacts of public health interventions, such as expansions in testing.

Our take —

This study estimated that a coordinated, nationwide policy encouraging individuals to stay at home would have considerably decreased the number of COVID-19 cases in the US. Stay-at-home and shelter-in-place ordinances likely achieved peak effectiveness in mitigating COVID-19 transmission nearly one month after initial announcement, suggesting interventions have impacts relatively soon after implementation. These policies, nonetheless, were often implemented alongside other interventions enforcing physical distancing, adding uncertainty about their independent effects on COVID-19 transmission.

Study design

Ecological

Study population and setting

The objective of the study was to determine the impact of stay-at-home and shelter-in-place ordinances on COVID-19 transmission in the United States. The authors estimated the rate of new COVID-19 infections at the county level from March 1 to April 16, 2020, specifying in a model infectivity, susceptibility, and recovery as parameters, and then modeling outcomes as a function of time and the existence of stay-at-home orders. To estimate the effectiveness of stay-at-home or shelter-in-place ordinances on COVID-19 epidemic growth, the authors compared daily county-level trends in COVID-19 incidence before and after these policies were announced in each municipality.

Summary of Main Findings

Implementation of stay-at-home and shelter-in-place policies were estimated to have exponentially reduced COVID-19 infections over time, from a mean incidence reduction of 3.9% one week after implementation (95% CI: 1.2 – 6.6%) to a 22.6% mean reduction 27 days after implementation (95% CI: 14.8 – 30.5%) corresponding to a net decline in new cases. The initial announcement of these policies (as early as March 16, 2020) preceded estimated peak COVID-19 incidence across municipalities. Applying the effect estimates to the entire United States under the counterfactual scenario in which every county issued stay-at-home orders on March 16, 2020 (the day a national emergency was declared), translated into a 63.2% decrease in cases.

Study Strengths

The authors aggregated stay-at-home and shelter-in-place ordinances at various ecological scales (e.g., state and municipal) to estimate impact of these policies on COVID-19 transmission at a national scale. The authors also accounted for heterogeneity in county characteristics and the timing of shelter-in-place ordinance announcements. The authors recognize some of the challenges in doing causal inference in these longitudinal state-level settings with varying treatment start dates, and do reasonable model checks regarding some of those issues, including inspecting trends in the pre-period.

Limitations

Stay-at-home and shelter-in-place ordinances were often accompanied by other policies designed to enforce physical distancing (e.g., school and business closures, mass gathering prohibitions), limiting inferences about their independent effects. The epidemiological model used to estimate COVID-19 transmission, while derived from empirical estimates of confirmed COVID-19 cases in the United States, may underestimate true transmission patterns (given shortages in testing capacity in the U.S.), resulting in misleading inferences about the impact of these policies on COVID-19 transmission. There is limited discussion of the selection factors involved, i.e., why some counties implemented this measure and how they may differ from those that did not.

Value added

This is among the first studies to apply causal inference methods to estimate impact of stay-at-home and shelter-in-place orders on COVID-19 transmission in the United States.

Our take —

Authors estimated the post-pandemic transmission dynamics of SARS-CoV-2 assuming variations in duration of immunity, seasonality, cross-immunity with currently circulating beta coronaviruses, and timing and duration of social distancing measures. Model results indicated that future transmission dynamics were highly sensitive to the values of these parameters, ranging from sustained biennial peaks to complete elimination after the initial wave. However, the true degree of some parameters, such as duration of immunity and cross-immunity, are not well established and further research is needed to confirm study results.

Study design

Modeling/Simulation

Study population and setting

Authors used data on commonly circulating beta coronaviruses HCoV-OC43 and HCoV-HK1 to estimate the effects of seasonal forcing (i.e., seasonality) and waning immunity on potential future transmission dynamics of the currently circulating novel SARS-CoV-2. Using the weekly percentage of positive HCoV-OC43 and HCoV-HKU1 laboratory tests and the proportion of physician visits for influenza-like illness (ILI) in the United States from the 2014-15 through 2018-19 influenza seasons, authors estimated R0 (reproduction number), duration of immunity, and the degree of cross-immunity between HCoV-OC43 and HCoV-HKU1 using a two-strain Susceptible-Exposed-Infected-Recovered-Susceptible (SEIRS) model. An SEIRS model extends beyond the traditional SEIR model by assuming that post-infection immunity is not lifelong, and that persons will eventually become susceptible to disease again after they have recovered. Then, using the same model, authors incorporated SARS-CoV-2 into the model to estimate the same parameters while accounting for the two commonly circulating strains. Finally, authors used a one-strain model to estimate the qualitative effectiveness of social distancing interventions on the transmission of SARS-CoV-2.

Summary of Main Findings

Assuming immunity from HCoV-OC43 and HCoV-HKU1 lasts approximately 45 weeks and some degree of cross-immunity occurs between the two, authors estimated the R0 for HCoV-OC43 and HCoV-HKU1 would vary seasonally, ranging from 1.7 in summer months to 2.2 in winter months, with a peak R0 observed around mid-January. Assuming no mitigation activities were implemented during the first year of the COVID-19 pandemic, the dynamics of SARS-CoV-2 post-pandemic showed a highly variable number of scenarios, from annual or biennial peaks, to complete elimination or apparent elimination for several years followed by a resurgence that can occur as late as 2025. These different scenarios were highly sensitive to the duration of immunity, degree of cross-immunity with other circulating beta coronaviruses, and the strength of seasonal variation in transmission assumed in the model. Authors also estimated the long-term dynamics of SARS-CoV-2 under one-time or intermittent social-distancing measures. Intermittent social distancing was shown to prevent exceeding critical care capacity. The worst-case scenario for one-time social-distancing (i.e., with no cross-immunity from currently circulating coronaviruses) resulted in an epidemic wave in autumn bigger than that avoided by the measures in spring.

Study Strengths

Using one-strain, two-strain, and three-strain SEIRS models, this study was able to assess the effects of seasonal forcing (i.e., seasonality), cross-immunity, and duration of immunity on transmission dynamics of SARS-CoV-2. Authors developed the models to capture varying levels of severity of COVID-19, including moderate, mild, and asymptomatic cases, infections that lead to hospitalization but not critical care, and those that required critical care.

Limitations

This study used a deterministic model, meaning that model outcomes are completely predetermined by the model parameters, and do not account for the randomness inherent in the disease transmission processes. Therefore, these models were unable to account for possibilities such as SARS-CoV-2 or other HCoV fade-outs. Furthermore, a large degree of uncertainty in the results exists due to the lack of knowledge regarding the true degree of cross-immunity between the three virus strains. Models only used five years of historic data, limiting future projections. Authors also assumed seasonal forcing was constant year to year, did not account for the effects of schools opening in the fall, did not differentiate between locations, and did not stratify calculations by age – all of which may have significant implications on model results.

Value added

This study is among the first to assess the effects of seasonal forcing (i.e., seasonality), waning immunity, and cross-immunity with other beta coronaviruses on SARS-CoV-2 transmission dynamics, and highlights the importance of measuring these parameters. The effectiveness of interventions, such as social distancing, could be largely impacted by these factors, and the timing of implementing and lifting such mitigation strategies should account for this. Lifting social-distancing measures by the end of summer could result in a big second wave that could potentially coincide with the flu season, thus resulting in an even more overwhelmed health system.

Our take —

Authors sought to describe the dynamics of COVID-19 within the Allegheny County, PA jail system across multiple age groups and personnel and three distinct transmission populations (community, processing and court, and jail) in this study, available as a preprint and thus not yet peer reviewed. Results indicated that interventions that reduced population mixing within the jail system markedly delayed the COVID-19 outbreak and reduced the magnitude of the epidemic curve. Incarcerated populations are more vulnerable to COVID-19 infection and severe outcomes, as these populations experience high rates of turnover, less access to healthcare resources, and limited ability to practice social distancing. This study highlights the importance of considering reform policies that would impact society’s most vulnerable.

Study design

Modeling/Simulation

Study population and setting

Authors used a Susceptible-Exposed-Infected-Medically treated-Removed model to describe the dynamics of COVID-19 transmission within the Allegheny County, PA corrections system, among incarcerated persons and their contacts. Authors segmented the model into five sub-populations (persons <18 years, low-risk adults, high-risk adults, elderly persons >65 years, and corrections personnel) and three transmission systems (the community, processing and court, and jail). Persons <18 years were only included in the community system, but could interact with sub-populations that were able to move within systems (e.g., a low risk adult moving from the community to processing and court). The model assumed that approximately 100 persons were arrested per day. Authors also included several interventions/policy changes (shelter-in-place orders, reductions in arrests, increases in rates of release, and changes to within-jail conditions) to estimate the impact these adaptations had on transmission dynamics within the three transmission systems.

Summary of Main Findings

The model estimated that an unmitigated outbreak would result in 926,108 infections, 51,497 hospitalizations, and 12,133 deaths in the community over the course of 180 days. The model estimated a more severe outbreak among incarcerated individuals: the 2,500 person-capacity jail was 0.2% the size of the greater community and individuals in this system experienced 4,949 infections, 264 hospitalizations, and 79 deaths. The peak was also much earlier in the jail system than the community system, occurring 31 days after the first infective case compared to 88 days after. Assuming shelter-in-place orders, infections, hospitalizations, and deaths in the community were considerably delayed and reduced (e.g., infections were reduced to 450,621). Conversely, within incarcerated populations, infections during the first half of the simulation (days 0-90) were similar to the unmitigated simulation, and the second half (days 91-180) were notably worse, resulting in a total of 7,421 infections over the course of the entire simulation. Reform within the jail system had greater impacts on incarcerated individuals, and to a lesser extent, jail personnel and the community. For example, discontinuing the arrests of bail-eligible individuals results in 22.9% and 3.9% reductions in infections among incarcerated individuals and the community, respectively. Discontinuing all low-level offense arrests resulted in 74.0%, 10.2%, and 19.7% reductions in infections among incarcerated individuals, jail personnel, and the community, respectively.

Study Strengths

The model accounted for movement of populations between the different systems, and did not restrict movement to one direction. Authors calibrated the model based on the size and mixing patterns of each transmission system’s population.

Limitations

Authors assumed that once patients were hospitalized, they did not spread COVID-19 any further. Although PPE worn by hospital staff limits ongoing transmission of COVID-19 in such settings, this may not always be true. Authors also assumed that all individuals leaving jail went straight back to the community, and did not account for other post-jail destinations, such as prison. However, authors argued that this is a small percentage of the population leaving jail and would not have a significant impact on results. The model used parameters specific to Allegheny County, PA (e.g., daily arrests, number of corrections personnel, size of jail population), so results may not be generalizable to locations with different population makeups.

Value added

Despite the United States having one of the highest incarceration rates in the world, most public models have excluded jails and other correctional facilities from their projections, potentially vastly underestimating the impact of COVID-19. This study is among the first to estimate the impact of COVID-19 on incarcerated populations, as well as on those with whom incarcerated individuals interact.

Our take —

This study was available as a preprint and thus not yet peer reviewed. Anomalies in influenza-like illness reports over time, which could be estimated from self-reported syndromic data, could be a lead indicator for COVID-19 transmission. However for such a system to work at-scale, widespread and representative uptake and high self-reporting rates will be needed.

Study design

Modeling/Simulation

Study population and setting

Geo-located and self-reported temperature data from more than 1 million smartphone-connected personal thermometers managed by Kinsa, Inc. were anonymized and aggregated at the US county-level. Temperature readings were used to construct and track influenza-like illness (ILI) signals and generate county-specific ILI-incidence forecasts from the beginning of March (before widespread outbreaks were underway) to mid-May 2020. Forecasts were based on day-of-year reproduction numbers (R) estimated back to August 1, 2016, and forecasts were made from March 1, 2020 using R estimates for that day-of-year. However, it was unclear what time frame of ILI data from thermometer readings were used for this back calculation. Influenza forecasts with lower levels of transmission due to the assumed impact of “shelter-in-place” social distancing measures were also produced to assess levels of potential bias. Authors compared this ILI forecast to the real-time thermometer-based ILI incidence at the city-level to identify “anomalous” ILI incidence which was defined as the difference between the real-time thermometer ILI and the upper uncertainty bound of the ILI forecast. Authors estimated the probability that this anomalous incidence was driven by normal seasonal influenza and assessed whether these detected anomalous incidence were indicative of COVID-19 cases.

Summary of Main Findings

Results show that self-reported temperature or ILI data can be used to produce forecasts of ILI incidence comparable to other influenza models for the first few weeks. ILI anomalies correlated strongly to county- and state-level positive COVID-19 case counts. County-level ILI anomalies also corresponded spatially to areas where major outbreaks have occurred including the Seattle Area, San Francisco Bay Area, New York Metro Area, and Florida. However, anomalous ILI incidence was not observed in Minnesota, Wisconsin, and South Dakota where COVID-19 cases have been confirmed. Authors found that ILI forecasts were likely sensitive to any social distancing measures implemented which could impact influenza transmission and affect whether ILI anomalies would occur.

Study Strengths

Authors re-ran their influenza forecasts assuming that “shelter-in-place” orders and social distancing reduced influenza transmission by different amounts which could bias their estimates of anomaly incidence. The study uses county-level real-time syndromic data which has the potential to reflect changes in case numbers in a more timely manner.

Limitations

As the study relies on self-reported data from smartphone-connected thermometers, it is unclear how representative the real-time ILI incidence used in this study is. For example, users of the Kinsa Inc product may only represent a certain age-group or there may not be any or a very limited number of users in certain counties or states. Only the total COVID-19 case counts have been used to validate the ILI anomalies making it difficult to assess whether these anomalies are indicative of COVID-19 case counts over time. It is not clear from the methods or the supplementary material exactly how, or on what time frame of Kinsa data the ILI forecasts were derived.

Value added

By using temperature data from a network of over 1 million users across the US, the study demonstrates the potential use of self-reported syndromic data to identify potential hotspots of COVID-19 transmission. This may provide an early-warning system to locate areas where an outbreak response is needed. Such syndromic surveillance for emerging pathogens can be valuable in settings with strong surveillance in place for other diseases with similar symptoms (in this case influenza and respiratory symptoms) and settings where a lack of diagnostic capacity may lead to under-reporting.