Hemostasis Today

March, 2026
March 2026
M T W T F S S
 1
2345678
9101112131415
16171819202122
23242526272829
3031  
Lu (Maggie) Qian: How Adaptive Trial Design Reveals Treatment Effects in COVID-19
Mar 4, 2026, 15:39

Lu (Maggie) Qian: How Adaptive Trial Design Reveals Treatment Effects in COVID-19

Lu (Maggie) Qian, Founder at Zetyra, Creator and Editor at Evidence in the Wild, shared a post on LinkedIn:

”Four randomized controlled trials of IL-6 receptor antagonists in COVID-19 showed no significant benefit.

Then Remap-cap reported that tocilizumab and sarilumab both worked, with >99.9% posterior probability of superiority and confirmed survival benefit at six months.

The difference wasn’t the drug. It was the design.

Remap-cap used a multifactorial platform design.

Patients were randomized across multiple domains simultaneously, corticosteroids, IL-6 antagonists, anticoagulation, antivirals, convalescent plasma, and a Bayesian model learned which components worked, and for whom, across all domains at once.

Earlier trials enrolled less severely ill patients.
Remap-cap enrolled ICU patients.
It allowed concurrent dexamethasone.
And it had the statistical machinery to isolate the IL-6 effect on top of steroids.

It detected what single-question trials could not.

Then the same platform revealed a different vulnerability.

While the corticosteroid domain was randomizing hydrocortisone vs. no hydrocortisone, the recovery trial announced that dexamethasone reduced mortality.

Overnight, 93% of patients were receiving corticosteroids as background therapy regardless of randomization.

The adaptive engine worked as designed. It just didn’t work fast enough.

A simple, large trial answered the question before the more sophisticated Bayesian platform could.

No amount of pre-specification can fully anticipate the information environment a trial will encounter.

Meanwhile, most Phase III trials use covariate-adaptive randomization, and then analyze the data as though they used simple randomization, leaving power on the table.

Corrected inference methods exist.

The carat R package implements them.

Most practicing biostatisticians haven’t heard of it.

Randomization determines what a trial can learn, not just who gets what treatment.

It should be one of the first design decisions.

Not the last.”

Stay updated with Hemostasis Today.