Discussed yesterday at both the highest and lowest levels … https://www.nytimes.com/video/us/politics/100000007046134/trump-fauci.html
… this study  apparently dichotomized a PCR cycle threshold C_t such that C_t \ge 35 was deemed a NEGative result, and then analyzed the POS/NEG dichotomy to produce this graph which was “impressive” to some, but “anecdotal” to others:
I had logged comments yesterday at medRxiv and PubPeer, but also wanted to highlight this matter in this forum, as it presents an opportunity for statisticians to offer timely criticism and do some good.
How robust is this plot to small changes in the C_t cutoff? What if (as I suggested in comments linked above) one were to impute POS to any of the (false-)NEG results [highlighted pink below] that was followed by a subsequent POS?
- Gautret P, Lagier JC, Parola P, et al. Hydroxychloroquine and Azithromycin as a Treatment of COVID-19: Preliminary Results of an Open-Label Non-Randomized Clinical Trial. Infectious Diseases (except HIV/AIDS); 2020. doi:10.1101/2020.03.16.20037135
i can’t see where they indicate how pos/neg is defined (>=35??), it’s not even in the definition of the primary outcome. the stats are very “point-and-click”
The main text does not mention this. The only hint to this effect is the fine print under the table: “NEG: negative PCR (CT value ≥35)”. I hadn’t appreciated this myself, either, until I saw this tweet:
then they should do proportional odds modelling as per Harrell (stat med paper from maybe 2017, but also BBR). I analysed a clinical trial this week using proportional odds modelling and got a different result from the primary analysis, it is more apt, more powerful, but the conventional approach gets the ‘primary’ status (it’s complicated by repeated measures, missing data and multiple outcomes).
I’m a bit shocked to be honest at how much attention this study has had. It’s really not much more than a case series, with some comparisons with an irrelevant control group, without any attempt at controlling statistically for confounding. I know people are trying to do research under difficult circumstances etc etc but that doesn’t give you a pass to make basic errors. Sorry if that seems harsh.