Should causal modeling be a required part of a NIH grant applications?

data analysis
,
1

I have been discussing the failure modes of critical care trial science in this site for a long time. However the failure of ARDS RCT guided early ventilator protocols (which mandated COVID patients be treated with EARLY mechanical ventilators (based on a threshold PaO2/FIO2) during the pandemic was a turning point. This failed RCT based protocol was abandoned but too late.

So given this clear example of the massive harm which can be caused by trusting non-transportable RCT. This sentinel event demanded that we investigate why such a catastrophic mistake could be rendered by RCT science. Responsive to that sad clarion call and beginning with Fisher/Bradford Hill, I have written this, now published, comprehensive failure-mode analysis review 50 years of failed critical care RCT syndrome science.

It explains:

  • Why syndrome-based RCTs (ARDS, sepsis, CAP) are simply “RCT mimics” incapable of generating transportable or reproducible results.
  • Why syndrome task forces have been simply modifying RCT inclusion criteria that function as cohort-level colliders, not specifying disease-equivalent causal mechanisms.
  • Why cSM, an explicated, clinical-facing form of SCM, is necessary for valid critical care trial design.

This is a rigorous synthesis of clinical science, causal modeling, and do-calculus that makes clear why explicit causal modeling must become an application requirement for @NIH funded randomized and observational studies in some fields of clinical science.

I hope there will be equally rigorous discussion. The public depends on us not to make this same mistake again. There is no backup.

2

I wouldn’t agree that causal modeling should be part of the analysis of a randomized trial in which there are no time-dependent covariates. But causing thinking should be a critical part of how the treatment options and patients are chosen, as you have advocated.

2 Likes