The first time I ran analyses on state-by-state data, from the beginning of the epidemic to May 26, covid diagnostic test-positives were correlated strongly and positively with covid-related deaths (r = +.93). I.e., states that had high case counts also had high body counts. Since then the correlation between cases and deaths has weakened, but it persists and has stabilized over the past month (r = +.60 as of August 4).
The correlations make sense. At first diagnostic tests were administered almost exclusively to those who were very ill, so a sizable proportion of those testing positive eventually died. As testing capacity gradually and continually increased, the catchment area widened beyond the ICU into the community, identifying a greater proportion of individuals with less severe symptoms who were more likely to recover from the virus. But more testing also narrowed the enormous gap between the number of test-positive cases and the actual infection rate — I estimate that there are now around three times as many infections as test-positives. As a result the daily test-positive rate now more closely approximates a community-wide survey of daily infections, spanning the actual range of severity from asymptomatic to fatal.
Lately however, testing rates have been dropping. In the US the demand for testing has far outstripped supply, with people being forced to wait for hours to be tested. Typically results are delayed by several days, rendering the tests meaningless in identifying individuals who should self-quarantine. Instead of enabling population-wide surveillance and public health containment of the epidemic, testing is becoming more of a convenience sample, comprised largely of individuals with insatiable curiosity and of corporations with enough resources to ensure rapid testing for their employees. It seems likely that the correlation between testing and deaths will diminish further.
* * *
First comes contagion; later comes recovery or, for some, death. It seems plausible that the correlations between test-positives and deaths would be stronger when a time lag is factored in. I ran two sets of lagged correlations and compared them with contemporaneous correlations.
Deaths 6/15 – 7/5:
- correlated with test-positives from 6/15 – 7/5 –> r = +.22
- correlated with test-positives from 5/26 – 6/15 –> r = +.61
Deaths from 7/5 – 8/4:
- correlated with test-positives from 7/5 – 8/4 –> r = +.84
- correlated with test-positives from 6/15 – 7/5 –> r = +.90
For the first interval the lagged correlation is stronger; i.e., test-positive rates demonstrated potentially sizable predictive power. States with higher test-positive rates were more likely to experience higher death rates 20 to 30 days later. For the second interval the lagged and contemporaneous correlations were equally strong, as well as being stronger than in the earlier interval.
Based on these statistical associations, here are the states with the highest test-positives per 100K of population, all of which (1) suffered higher than average covid-related deaths during July, and (2) are likely to continue experiencing high death rates in August.
- South Carolina
Again, the strength of this predictive correlation between test-positives and deaths may diminish if testing rates continue to decrease.
Next post I’ll look at state-specific data on percent test-positives.