The association between early country‐level COVID‐19 testing capacity and later COVID‐19 mortality outcomes

Abstract Background The COVID‐19 pandemic has overrun hospital systems while exacerbating economic hardship and food insecurity on a global scale. In an effort to understand how early action to find and control the virus is associated with cumulative outcomes, we explored how country‐level testing capacity affects later COVID‐19 mortality. Methods We used the Our World in Data database to explore testing and mortality records in 27 countries from December 31, 2019, to September 30, 2020; we applied Cox proportional hazards regression to determine the relationship between early COVID‐19 testing capacity (cumulative tests per case) and later COVID‐19 mortality (time to specified mortality thresholds), adjusting for country‐level confounders, including median age, GDP, hospital bed capacity, population density, and nonpharmaceutical interventions. Results Higher early testing implementation, as indicated by more cumulative tests per case when mortality was still low, was associated with a lower risk for higher per capita deaths. A sample finding indicated that a higher cumulative number of tests administered per case at the time of six deaths per million persons was associated with a lower risk of reaching 15 deaths per million persons, after adjustment for all confounders (HR = 0.909; P = 0.0001). Conclusions Countries that developed stronger COVID‐19 testing capacity at early timepoints, as measured by tests administered per case identified, experienced a slower increase of deaths per capita. Thus, this study operationalizes the value of testing and provides empirical evidence that stronger testing capacity at early timepoints is associated with reduced mortality and improved pandemic control.

units, hospital beds, and personal protective equipment (PPE). 2 The public health crisis has additionally worsened mental health outcomes by increasing the prevalence of stress, anxiety, and depression worldwide. 3,4 Furthermore, COVID-19 has and will produce adverse socioeconomic outcomes globally as the International Labor Organization projects that the pandemic will push 200 million to unemployment by 2022, and the World Food Program projects that 200 million are at risk of or currently facing acute hunger because of the pandemic. 5,6 With a continued rise in cases, there is still an essential need to control the spread of SARS-CoV-2. To do so, nations initially relied on nonpharmaceutical interventions, including testing, contact tracing, and isolation, 7 as no universally accepted therapeutic regimen or vaccine was available for this infectious respiratory virus for most of 2020. 8 Given that the vaccine rollout is currently ongoing throughout the world, measures such as testing, tracing, and isolation still stand as the cornerstones of COVID-19 control policy. 9 Nonetheless, over the course of the pandemic, the capacity for recognizing and verifying cases has varied across countries. 9 Limitations in country-level testing capacity may stem from shortages in testing equipment, such as swabs, reagents, and PPE, in addition to having a limited number of qualified personnel. Potential limitations in country-level testing capacity may contribute to increased COVID-19 case numbers and impact other COVID-19-related outcomes. 10 Currently, few studies have evaluated the effects of early testing capacity strength on future COVID-19 outcomes. 11,12 Pan et al.
demonstrated that symptom survey rates in Wuhan, China, were indicative of a smaller effective reproductive number of SARS-CoV-2 and were associated with reduced daily confirmed cases. 11 In contrast, Chaudhry et al. adopted a more global approach and found that widespread testing, in addition to full lockdowns, was not associated with country-level COVID-19 death rates. 12 Given such inconsistencies in the literature, we aimed to explore whether early country-level testing capacity, measured as tests administered per case identified, has an association with COVID-19 mortality. The rationale is that a higher level of testing per case identified is indicative of more proactive efforts to find infections in the broader community and break chains of transmission. We assessed COVID-19 mortality as opposed to COVID-19 reported infections, given that testing demand and capacity may change in response to the number of reported infections. In quantifying the effects of testing capacity, researchers may be better able to gauge the value of rapidly scaling up and promoting these public health strategies. Thus, we aimed to assess the relationship between the rigor of early testing and future COVID-19 mortality across countries.
In exploring this research question, we used information regarding daily cumulative testing and mortality for 27 countries. 1 We also incorporated data regarding the timing of nonpharmaceutical interventions, including confinement, school/work closures, event cancellations, travel restrictions, and health practices (contact tracing and mask wearing). 13 We hypothesize that higher country-level tests per case in the early stages of the outbreak is associated with a lower hazard of reaching higher mortality thresholds (i.e., Y deaths per million), after adjusting for country-level characteristics, including median age, GDP, hospital bed capacity, population density, and timing of NPI implementation.

| Study design and sample
We used the Our World in Data, a publicly available scientific database, to quantify country-level testing capacity and health outcomes. 1 The study inclusion criteria broadly identified countries whose first documented testing count was less than 100 tests, and whose case count (at that particular timepoint) was less than the test count. These countries included Bangladesh, Bolivia, Czech Republic,

| Confounders
In assessing the association between early testing capacity and mortality outcomes, we accounted for potential country-level confounders, including median age, gross domestic product (GDP), hospital bed capacity (i.e., hospital beds per 1000 people), population density, and nonpharmaceutical intervention (NPI) implementation.
NPI categories included (1) mandatory/advised confinement, (2) school/work closures, (3) event cancellations (restrictions on public gatherings, entertainment/cultural sector closures, public services closures), (4) travel restrictions (international travel restrictions, restricted freedom of movement), and (5) health practices (contact tracing, mask wearing). If an NPI category was implemented during the full timespan between a specified date marking early testing capacity and a specified date marking the mortality threshold, then it was assigned a value of 1. The summary NPI variable during this specified time span was expressed as either a sum of the five NPI category values (NPI Sum; range = 0-5) or an average of the five NPI category values (NPI Average; range = 0-1). The summary NPI variable accounted for the time that an NPI was in place and the number of NPIs that were in place. NPI data were retrieved from IBM Research: Worldwide NPI Tracker for COVID-19. 13

| Statistical analysis
We determined the trajectory of cumulative tests per case and deaths per million for each country in our analysis from December 31, 2019, to September 30, 2020. We additionally modeled the daily timeevolving pattern of country-level testing capacity and country-level mortality rate, in addition to creating a log-log plot assessing the crude relationship between cumulative tests per case and cumulative deaths per million, measured on September 30, 2020, and determined the corresponding Pearson correlation coefficient. We then used Cox proportional hazards regression to assess the association between early testing capacity and later COVID-19 mortality outcomes. We implemented four models, each adjusting for a different set of covariates. Model 1 adjusted for median age, and GDP at the country-level; Model 2 adjusted for median age, GDP, and hospital bed capacity at the country-level; Model 3 adjusted for median age, GDP, hospital bed capacity, and population density at the country-level; Models 4 and 5 adjusted for median age, GDP, hospital bed capacity, population density, and NPIs, either summed or averaged, at the country-level. We used R version 4.0.3 to perform all analyses.

| Bootstrap analysis
To approximate the variance of the model effect estimates, we conducted 1000 bootstraps on the model coefficients, randomly selecting all 27 countries in the sample, with replacement ( Figure S1A-E). Time-evolving patterns of country-level testing capacity and countrylevel mortality indicate that countries with higher testing capacity early in the pandemic experienced a slower accrual of deaths per capita ( Figure S2). The major strengths of this study are that we chose to incorporate a time component when measuring mortality outcomes, by operationalizing the outcome as the time interval (in days) to reach a certain country-level mortality threshold. In using a time to threshold measure, we were able to determine whether stronger testing capacity at the onset of the pandemic may have yielded a sustained impact on later mortality outcomes. Furthermore, we adjusted for potential confounders, such as the timing of NPIs, to assess the public health measures that were in place during a particular period. NPIs encompassed important measures such as masking policies, which have been associated with reduced rates of SARS-CoV-2 infection. 15 We additionally assessed a longer time frame of analysis compared with existing studies, by exploring the association of testing capacity and mortality up until September 30, 2020, utilizing 7 months of longitudinal data.

| Cox proportional hazards regression analyses
There were certain limitations in our study design approach. We did not identify the type of diagnostic test used in each country.
Countries may have had differential access to rapid antigen tests (RAT), or reverse transcription polymerase chain reaction (RT-PCR) tests. These differences are important given that RT-PCR yields higher sensitivity and specificity than RAT, although RAT is relatively quick to administer. We additionally did not account for the type of healthcare system implemented in a country: healthcare access may influence an individual's access to COVID-19 testing. We also coarsely categorized the NPI variable, which quantified the number of NPIs in place during a particular time period; this variable did not include information regarding the stringency of a specific NPI measure or corresponding public compliance. In addition, we only included countries with daily testing data; thus, the findings may not be applicable to countries that do not report daily testing data. This study also may have yielded limited generalizability, as findings may only correspond to the countries used in this analysis, during this specific period.
Future directions for research may address using a longer time span to determine if this association between early testing capacity and later mortality outcomes continues to hold at longer time intervals. It may also be helpful to explore whether hospitalization rates per capita play a mediating role between country-level testing capacity and mortality rates in order to assess if lower tests per case rates may increase hospital admissions and overrun the capacity of the healthcare system, which may result in increased COVID-19 mortality.
Future analyses may also consider the effects of new therapeutics and vaccines in preventing COVID-19 deaths, as this may dilute the Worldwide Non-pharmaceutical Interventions Tracker for COVID-19 (url: https://ibm.github.io/wntrac/dataset).