Analysis of #Colcorona Trial results in #COVID19 outpatients with risk factors.

I will add Bayesian take and focus on the ITT. (cite @ProfHayward here)

Primary endpoint death or Hospitalization -- legit

Link to preprint
https://www.medrxiv.org/content/10.1101/2021.01.26.21250494v1.full.pdf

Here are primary endpoint
Just looking at the CI, it goes from a possible huge 39% benefit to 3% chance of harm.

A wide spread but mostly to the good.
P- value here is 0.08
If the null is true (no benefit) there is an 8% chance of finding this data or something more extreme.

Now let's calculate the Z score and Bayes Factor

BF - ratio of the null to alternate hypothesis

The Z score is 1.74 (that's the distance from mean)

BF is 0.21 or inversely 4.6. Per Goodman et al -- weak evidence

Now comes the priors. This is tough. Not sure we have had anything to convincing for treating outpatients with COVID19

We either have to be 50/50 or 70/30 against.

I can't see being optimistic here.

Recall Bayes Theorem:

Posterior = BF or (Likelihood Ratio) x Prior

Here is the table

If you are 50/50, the prob of the no benefit is 18%- almost 1 in 5

If are skeptical 70/30, the prob is nearly 4 in 10 chance of no benefit.

Keep in mind, colchicine is inexpensive and pretty well tolerated over the short run.

Although the authors put pneumonia in the adverse events, which is a no-no b/c it is a primary endpoint of interest in COVID,

AE were not bad.

Now we want to know the chance of benefit of more than anything.

Two tables below: Non-inform prior and reasonable skeptical of 0.75-1.33 distribution

Even if skeptical -- there is still a > 50% chance of a 10% benefit of colchicine in early #COVID19

Very little chance of sig harm.

I wish we had more data, but I think colchicine seems reasonable for now.

I'd probably take it.

What do you all think? What did I miss?

@kaulcsmc @djc795 @PulmCrit @AndrewFoy82 @DrToddLee @raj_mehta @mikejohansenmd