I think Dr. Farkas is thinking at a deeper level than you perceive. I can’t speak for him, but experienced critical care physicians inevitably apply Bayesian reasoning to almost everything.they analyze, not only to the oseltamivir trial itself, but to the meta-level reliability of the critical-care RCT literature itself
That matters when interpreting the REMAP-CAP oseltamivir raw result and the output derived from adjustment based on so many diverse covariates. A finding suggesting that a long-standing guideline treatment may actually be harmful naturally raises an extraordinary implications including potentially substantial iatrogenic mortality from years of recommended treatment.
An experienced intensivist is unlikely to accept an implication of that magnitude from a single RCT without considering the prior clinical trial history of the field.
And that history is abysmally sobering. Critical care physicians have lived through decades of RCT-based guideline reversals, many involving harm. Importantly, many of those trials were cause-agnostic RCTs where covariate adjustments comprise mathematical window dressing. Since statisticians generally do not distinguish them from cause-integrity RCTs, to clinicians they have all simply been presented as decades of “gold standard RCTs…reversed for harm”.
REMAP-CAP itself provides a striking example. Its recent corticosteroid domain tested another guideline-standard treatment and reported approximately an 89% posterior probability of harm. The results have been largely swept under the rug and the guidelines remain unchanged. This recently prior REMAP CAP trial (which also included influenza) and its covariate structure are discussed here.
https://pubmed.ncbi.nlm.nih.gov/42464342/
Of course the relevant prior is not that “Critical care RCTs are wrong.”, It is that the probability that any single critical-care RCT treatment effect, especially one from the REMAP CAP platform will prove clinically durable and transportable into guidelines may be relatively low.
That prior warrants caution before concluding from the oseltamivir trial that previous guidelines caused deaths on a massive scale AND it also makes it difficult to defend those previous guidelines simply because earlier evidence supported them especially since that “evidence” was transported from a different state of the disease. However failure mode analysis is not a part of clinical RCT based guideline culture.
There is, of course, a deeper structural explanation for the history of so many reversals in critical care. . Cause-agnostic trials can estimate different mixture-weighted effects while appearing to test the same clinical question. So the intensivist’s skeptical prior is not merely experiential, part of it has a structural mathematical basis. Previous broad defense by statisticians of widely discretionary covariate adjustment (as long as pre specified) have been made as if critical care RCTs are structurally similar. They are not. This may be true for cause integrity trials such as the oseltamivir trial but it is not true for much of the standard critical care RCT structures.
So when an expert critical care physician hesitates before accepting the oseltamivir result into a claim of massive historical iatrogenic harm, that hesitation is entirely compatible with deep, thoughtful and informed Bayesian reasoning. She is applying Bayes one level above the individual trial.
