Last week the CDC reported that the number of Americans who’ve been infected by the coronavirus exceeds 20 million — an estimate close to my calcs. Now the CDC announces that their “best estimate” for the US covid infection fatality rate (IFR) is around 0.3% — half the rate that I’ve been using.
Assuming that the CDC number is accurate, then the estimated total number of infections in this country would the number of reported deaths divided by the IFR: 129K/.003 = 43 million. That’s twice the CDC’s own infection estimate: what gives?
Maybe the early runaway contagion that hit NY, NJ, and MA has skewed covid mortality estimates toward the high side. Hospitals were severely understaffed and underequipped; doctors weren’t sure how best to treat patients suffering from this new disease; infected nursing home patients were returned from hospitals to nursing homes, exacerbating viral superspread among the most vulnerable demographics. Consider the ratio of total deaths to total test-positive cases: for NY, NJ, and MA combined the ratio is 7.8%; for the rest of the country it’s 3.8%. This discrepancy can’t be explained by differences in testing rates or in population demographics. Those who got hit hardest and earliest have suffered the highest fatality rates.
At least one modeling expert believes that the CDC’s best estimate is overly optimistic, falsely discounting the New York data and painting an overly rosy picture perhaps for political reasons. Still, halving the US fatality rate could explain, at least in part, the anomaly of rising test-positive rates while death rates are on the decline. On the other hand, the nationwide down-sloping death trend began around the second week of May, by which point deaths in NY, NJ, and MA had already dropped considerably from their peaks.
The CDC might be right, but without better data it’s hard to say. It’s shocking that after all this time the CDC still isn’t coordinating systematic well-designed studies for actually measuring the spread of the virus nationwide, rather than having to rely on models extrapolating from localized, sporadic, and in some instances methodologically suspect investigations.