Over the weekend of 31 Oct – 1 Nov, Slovakia conducted antigen diagnostic tests on 3.65 million people — two-thirds of the national population. Of those tested, 1.06 percent came up positive.
The methodology deployed in this study has come under criticism, but to my knowledge it’s the only such nationwide population survey that’s been conducted anywhere in the world. As such, it provides an opportunity to evaluate my quick-and dirty algorithm for converting death rates into infection rates.
Here we go:
Death is a lagging indicator of infection, each death resulting from an infection that would have been contracted 2 or 3 weeks earlier.
The fatality rate for covid, inferred from various seroprevalence surveys, is around 0.7 percent in the US. The fatality rate is age-dependent: nations with older populations will have a fatality rate higher than 0.7%. The median age of the US population is 38 years; that of Slovakia, 41 years — 3 years older than the US. So, using the data-based conversion ratio I’ve devised based on age-adjusted mortality data, the Slovakian covid fatality rate = 0.7% x 1.13 = 0.9 percent.
In order to estimate the number of people infected over a given two-day testing interval, I’d need to work backward from the number of people who died 2 or 3 weeks later. Assume that, on average, a person who’s been infected with covid will test positive for 14 days. So, those who tested positive over the 2-day weekend would comprise a cohort of people who’d been infected up to 15 days before the tests were administered. Therefore, in using deaths as a lagging indicator of infections, I’d need a 15-day count of covid deaths recorded two to four weeks following the antigen survey.
It’s been only three days since the antigen survey was conducted; therefore I have to project future death rates based on current rates and recent trends, then work backward from there to estimate infections for this past weekend.
In Slovakia, 43 people died of covid during the 15 days between October 18 and November 2. During that same 15-day interval, the number of covid cases increased by about 70 percent from the preceding 15-day interval. Assuming case counts are correlated strongly with infection rates, then we could project a 15-day death count starting 2 weeks from now to be around three times the most recent count: 43 x 3 = 129 deaths.
To estimate infections from deaths, divide deaths by the fatality rate. For Slovakia, it’s 129/.009 = 14,300 infections. That’s the estimated number of people in Slovakia who’d have been infected over the past weekend.
How does my algorithmic estimate compare with the population antigen survey? Poorly: my lagging-indicator estimate of around 14 thousand infections comes nowhere near the antigen survey result of 38 thousand test-positives. That’s kind of surprising, given that the algorithm has worked quite well for both the US and Europe. Is it the algo’s fault, or the antigen survey methods, or the death count reporting accuracy? As they say in the Discussion section of scientific publications, more research is needed…
ADDENDUM: I think the death count is suspect. To date Slovakia has recorded a total of nearly 63.6K cases but only 235 deaths: that’s a cumulative death rate of 0.4 percent. In contrast, the death rate in the US is = 2.4%; in France it’s 2.5%; UK = 4.4%. Applying an algo to an undercounted death rate can’t help but result in an undercounted infection rate.