Diagnostic Testing’s Gradual Shift from Inpatient Medicine to Community Surveillance

I’ve run these analyses three times before: cumulatively from the beginning of the covid outbreak to April 26, and again a month later, then for the 20-day interval from May 26 to June 15. Now I’m updating those same analyses for state-by-state covid testing data collected between June 15 and July 5.

Mostly I’m looking for evidence that American testing is shifting from the hospital to the community, from medical treatment of sick people to public health abatement of the epidemic. The guideline has remained the same: test and treat the symptomatic; bring contagion down to a minimum through shutdown and sheltering in place; then maintain the low levels of infection during a gradual reopening by identifying and quarantining localized outbreaks before they spread. Absent a national initiative to implement the guideline, each individual state has come up with its own approach. While some states have fared better than others at lowering the rate of new infections while increasing diagnostic testing, it’s generally acknowledged that the country has jumped the gun, reopening while the virus continues to flourish and without having built the necessary infrastructure to implement community-wide control of contagion.

Theoretically, a state is further along in its public health abatement initiatives to the extent that its diagnostic screening is decoupled from its medical treatment of sick patients. As a statistical proxy I’ve been looking at the state-by-state correlations among three variables: testing rates, test-positives, and deaths.

Correlation of testing rate with test-positive rate. When diagnosis is conducted as part of medical treatment, that correlation is strongly positive, because tests are administered almost exclusively to individuals experiencing covid-like symptoms. When contagion is reduced and testing shifts into community abatement the correlation drops: lower rates of serious illness presents the opportunity to widen the net, testing people with less severe symptoms as well as those who are pre- or asymptomatic. Over the past two and a half months the correlation between testing rate and test-positives has in fact decreased. As of May 26 the correlation was +0.66; for the 20-day interval between May 26 and June 15 the correlation dropped to +0.23.

Testing rates nationwide have shown a steady increase since the beginning of the epidemic. Test-positive rates, steadily declining since mid-April, began trending upward again mid-June, implying strongly that the country was experiencing a resurgence of the virus. However, for the 20 days between June 15 and July 5 the state-by-state correlation between testing and test-positives dropped again, to +0.11. Two possible interpretations suggest themselves:

  1. Increased variability between states, with some states experiencing resurgences of contagion while others continue their slow decrease. That’s certainly possible: news reports focus on spiking states like California, Florida, and Arizona, without providing much information on the rest of the country. Between June 15 and July 5, only 19 states suffered an increase in test-positive rate (including CA, FL, and AZ), while the rate dropped for 31 states.
  2. Increased success in public health testing procedures, identifying a greater proportion of individuals who are infected but who aren’t severely ill. Exploring that possibility leads to the next set of analyses…

Correlation of test-positive rate with death rate. During the initial national surge of the coronavirus epidemic, diagnostic tests were administered almost exclusively to hospitalized patients suffering severe symptoms. Most of those who die from covid had previously been hospitalized, so a high positive correlation would have been expected between diagnostic test-positives and deaths. As of May 26 the correlation coefficient was a very high +0.93. For the 20-day interval between May 26 and June 15 the correlation decreased to +0.51; between June 15 and July 5 it dropped further, down to +0.22. By inference, state-by-state differences in test-positives bear very little relation to severity of illness, supporting the contention that testing has shifted markedly from medical treatment of patients to community surveillance of contagion.

Implications. In mid-April the IHME projected that the US would experience 60 thousand covid-related deaths, with daily counts dropping into the single digits by early June. So far 134 thousand Americans have died from covid. While daily body counts have been declining steadily since April, they’re still averaging around 500 per day.

It’s widely agreed that the states reopened prematurely, with contagion likely to increase as social distancing lapses. Yesterday nearly a thousand covid deaths were recorded in the US: is it the beginning of a new surge, or part of ordinary day-to-day variation? Either way, the states are operating largely in uncharted territory, widening surveillance and contact tracing even as infection rates remain high. That’s not how it was supposed to work; it’s now how other countries hit early and hard by the virus have managed to bring the epidemic nearly to a standstill. As they say, it is what it is.



Spain’s Covid Serology Survey: Implications

Recently Spain’s National Centre for Epidemiology completed a systematic randomized survey to estimate the spread of the covid virus through the national population. More than 60 thousand individuals, selected through stratified random sampling, completed questionnaires and were tested for antibodies. Conclusion: 5.2 percent of the Spanish population are seropositive, indicating that at some point they’ve been infected by the virus. That’s a long way from herd immunity.

Spain’s population is 47 million; if 5 percent have been infected, that’s 2.4 million people. By contrast, only 300 thousand Spaniards have tested positive on covid diagnostic tests — one-eighth the actual number of infections.

Divide the number of deaths by the number of infections to calculate the mortality rate. So far 28,400 Spaniards have died from coronavirus; divide that tally by 2.4 million and Spain’s covid mortality rate is 1.2 percent.

Implications for US

Based on the small number of rigorous serology surveys previously conducted, I’ve estimated an age-adjusted covid mortality rate of 0.6 percent for the US. “Age-adjusted”: older people are more likely to die of covid than younger people, so a nation with an older average population should experience a relatively higher mortality rate than a younger population. An analysis of the demographics of covid mortality data in the US indicates that each additional year of age increases the likelihood of dying by 10%. The median age in Spain is 45 years; in the US it’s 38 years — a 7-year difference. To calculate Spain’s age-adjusted mortality rate, multiply the estimated US mortality rate of 0.6% by 1.107 = 1.2%: the same as Spain’s serology survey results. The 0.6% mortality estimate for the US is supported by Spain’s data.

How many Americans have been infected by the coronavirus? Divide the number of deaths by the mortality rate to estimate the total number of infections. So far 133 thousand have died; divided by .006 = 22.2 million. The cumulative number of US diagnostic test-positives is 3 million: like Spain, the confirmed diagnoses amount to only about one-eighth of the estimated number of infections.

What percentage of the US population has been infected so far? Divide the 22.2 million infections to date by the national population of 328 million = 6.8 percent. The prevalence in the US is somewhat higher than Spain’s. An early hotspot that has successfully minimized community spread of the virus, Spain has been averaging just 4 deaths per day over the past three weeks.