Test Sensitivity & Specificity. The positive predictive values displayed above were calculated using a sensitivity of 95.1% and specificity of 98.8%. These test characteristics are from studies of commonly used rapid molecular SARS-CoV-2 tests, such as those found in airports and other point-of-care COVID-19 testing sites: Abbott ID Now, Cephid Xpert Xpress, DNAnudge, DRW SAMBA II, and Mesa Biotech Accula. Because the tests widely available at point-of-care testing sites were only granted FDA Emergency Use Authorization to be used to diagnose symptomatic patients within 7 days of symptom onset, little research is available on performance of these tests when used to screen asymptomatic people.
The systematic review by Dinnes et al. (2021) was referenced for this project because studies of both symptomatic and asymptomatic patients were included. This more accurately reflects the population these tests are being used on, such as people flying to Hawaii getting tested to meet the travel requirement, not because they are ill with symptoms consistent with COVID-19 (the population the manufacturers derived published test sensitivity and specificity from). Unfortunately, only 3 studies included in the systematic review by Dinnes et al. included a combination of symptomatic and asymptomatic patients, 12 were symptomatic patients, and 14 studies did not specify whether patients had symptoms of COVID-19. The lack of data on test performance in asymptomatic populations likely skews the sensitivity and specificity upward, so the PPVs calculated using these test characteristics likewise may be higher than reality.
Estimating COVID-19 Prevalence. Positive predictive values should be calculated using prevalence, not incidence. Incidence is the number of new cases of COVID-19 diagnosed every day, while prevalence is the number of people with COVID-19 during a specified time period. For example, someone who got diagnosed with COVID-19 4 days ago would be a prevalent case (because they still have the infection) but they would not be an incident case because their diagnosis was already reported and is only counted once. While it is relatively easy to find data on the prevalence of diabetes or HIV, COVID-19 is tricky in part because countries don’t agree on what counts as a case. For more on the challenge of approximating prevalence using available data, read El-Gilany (2020). For a strategy to estimate prevalence, see COVID-19 Prevalence Calculator by Prevent Epidemics (2020).
Why do some countries display a PPV of 0%? If a country reports very few or no cases of COVID-19, dividing the true positive cases by all who tested positive = 0. This can be interpreted to mean: if you live in a country that doesn’t have COVID-19 at all, and you get a positive SARS-CoV-2 test result, there is a 0% chance you actually do have COVID-19. The problem is, just because countries report no cases doesn’t mean no cases exist. For example, China reported so few cases the PPV was near 0%, but can we trust that all infections were reported to the global public health community? The data also get wacky with very small countries. Many of these are so small geographically their PPV isn’t displaying properly above, but for some low population areas (e.g. Antarctica, which I cropped off the map), the PPV is 0% because there probably actually weren’t any cases of COVID-19 there in the past week. Another important point to keep in mind is that for a case to be counted, the patient needs to be tested or at least medically evaluated to make a clinical diagnosis. In areas without money for COVID-19 tests and few medical professionals, a large portion of cases go undiagnosed and therefore unreported. This lack of access to testing and medical care leads to inaccurate positive predictive values because the reported prevalence is far lower than the true COVID-19 prevalence.
Data Sources:
Institute for Health Metrics and Evaluation, University of Washington. (2020). COVID-19 mortality, infection, testing, hospital resource use, and social distancing projections. https://covid19.healthdata.org/global?view=cumulative-deaths&tab=trend
United States Census Bureau. (2021). County population totals, 2010-2019. https://www.census.gov/data/tables/time-series/demo/popest/2010s-counties-total.html
mary.mcquilkin@gmail.com