Thread by @DiseaseEcology

#COVID19 #SARSCOV2 Small seroprevalence study in Tokyo Japan
Attempt at randomized sampling (yay!) & assessment of sample (double yay!)
Seroprev too low to measure accurately (<1.5%; also yay?)
& poor statistical analysis (sad)
Short thread
medrxiv.org/content/10.110…

Japan has not been hit hard by COVID-19 (yet?). Despite instituting only mild restrictions they initially fared much better than many other countries (although cases are now rising).
time.com/5842139/japan-…

This study attempts to quantify the epidemic size by measuring the fraction of people with antibodies from previous exposure. They looked at a part of greater Tokyo & in contrast to previous serosurveys in Japan attempted to collect a random sample.

This is hugely important as I've written about many times (latest:

https://twitter.com/DiseaseEcology/status/1285696432478052355

). Unfortunately, it's not easy to do. Even if you invite a random sample of households to participate, a substantial fraction can say no making your sample unrepresentative.

That's what happened here. Only 32.4% agreed to participate, so sample size was smallish (742 people) & some groups over represented while others were underrepresented. Authors tried to correct for this but w/ refusal rate this high, results are harder to interpret.

Also, if seroprevalence is low, need much bigger N to estimate seroprevalence accurately. Here only 3/742 people were seropositive. Authors present 0.4% (unweighted) and 1.2% (weighted) seroprevalence estimates but huge CIs (0.08-1.2 & 0.17-2.3% resp) &

even these are a bit too narrow given refusal rate. In addition, test wasn't very specific (96.3%) so all +s could have been false positives & this wasn't reflected in analysis (lower CI should be = 0.0%). Under-ascertainment estimates in paper thus wildly off.

So, given refusal rate & poor test specificity, can't conclude a ton from this study except that seroprevalence is definitely very low (0-1%) in this part of Tokyo.
But wanted to at least note authors for attempt at good study design.