🧵 Part 3 — Why ICU RCTs fail (beyond colliders) 1. The puzzle Decades of critical care RCTs. Huge effort. Tens of thousands of patients. Very few reproducible breakthroughs. This is Part 3 of my series on why ICU trials fail — and why physiology must guide us.
2
35
102
Replies
2. Heterogeneity (noise, even in “real” diseases) Even when the trial is valid, patients vary hugely: – Baseline physiology, comorbidities, genetics – Different illness phases (early vs late, compensated vs exhausted) – Co-interventions (ventilation, sedation, antibiotics) That
1
0
10
3. Timing & trajectory Critical illness evolves dynamically. The same drug may help early but harm late. RCTs with broad enrolment windows average across very different biological states. Example: steroids in ARDS → benefit if early/prolonged (DEXA-ARDS), harm if late (older
1
0
7
4. Complex interventions ICU therapies are rarely single pills. They’re bundles: – Ventilation strategies – Fluid protocols – Sedation regimens Delivery fidelity varies by clinician, unit culture, resource availability. Noise increases, reproducibility drops.
1
0
8
5. Non-linear physiology ICU physiology isn’t linear. It has tipping points, thresholds (the real kind), and feedback loops. RCTs flatten this into average effects. But the patients at the edge — where autoregulation fails, or where CCP is crossed — are the ones where treatment
1
1
7
6. Biases in what we see – Publication bias: positives get published, negatives vanish → literature looks rosier than reality. – Survivor bias: we only study patients who live long enough for enrolment → case mix is skewed. Together they distort how interventions really work
1
0
10
7. Plausibility vs complexity We test lots of interventions that sound biologically plausible: antioxidants, statins, nitric oxide. But critical illness is biology at its most complex. “Simple” ideas often collapse under rigorous testing.
1
0
8
8. Add colliders (from Part 2) And in syndromes like sepsis & ARDS, threshold definitions act as colliders. So those trials aren’t just noisy — they’re invalid.
1
2
9
9. Why this matters If we can’t rely on bundles and thresholds from RCTs to guide every patient, what do we fall back on? The only thing that travels with us to the bedside: Physiology.
1
1
8
10. The counterbalance Saying “physiology over thresholds” doesn’t mean improvising. It means hypothesis testing at the bedside: – Anchor decisions in physiology – Form a hypothesis (e.g. fluids will raise Pms and CO) – Intervene and measure – Accept or reject the hypothesis,
1
1
7
11. Closing ICU RCTs fail for many reasons: heterogeneity, timing, complex interventions, non-linear physiology — and sometimes collider bias. The message across all 3 threads is the same: Physiology is the compass that helps us navigate uncertainty, trial failures, and patient
3
1
10
Part 1 here ⬇️ https://t.co/qTZ1GwBO17
🧵“Thresholds, consensus & physiology” Decades of critical care research have produced few reproducible breakthroughs. Maybe the problem isn’t our interventions — it’s that we reduce patients to syndromes, instead of treating them as individuals with distinct physiology.
2
1
6
