ANDROMEDA-SHOCK – A Bayesian Reanalysis – Schrödinger’s Cat in Action
November 20, 2019
Written by Clay Smith
There is a high probability that a protocol using a skin perfusion-based approach (capillary refill time) vs lactate clearance to manage patients with septic shock reduced 28-day mortality.
Why does the matter?
ANDROMEDA-SHOCK compared a protocol driven by either capillary refill or lactate clearance to guide resuscitation. Capillary refill improved SOFA score but did not show a statistically significant improvement in 28-day mortality. Yet, the effect size on 28-day mortality was large: 34.9% in the cap refill group vs. 43.4% in the lactate clearance group. However, the p value was 0.06, which led the authors to conclude that the cap refill protocol, “did not reduce all-cause 28-day mortality.” Sometimes making a dichotomous “yes or no” outcome on the basis of an arbitrary p value doesn’t seem to make sense. We recently discussed embracing uncertainty and Schrödinger’s Cat, with a call to consider Bayesian statistics as a way to more intuitively interpret studies. Briefly, Bayesian statistics are based on an estimate of a prior (or pretest) probability – what we know going into the study, a continuous evidence update based on that knowledge (likelihood ratio), and the posterior (or posttest) probability based on the data. What would a Bayesian reanalysis of this trial show?
Schrödinger’s cat is on the hunt
This was a post-hoc analysis of the ANDROMEDA-SHOCK trial. For the original study, read this. Originally, the sample size of 424 was based on on the usual cutoffs, with power of 80% and alpha 5% (upper limit of a type I error). Since the study had a p value of 0.06, above the arbitrary cutoff p of 0.05, it was considered negative for the primary outcome. Many cried foul. This was a reanalysis using Bayesian statistics and four selected prior probabilities: optimistic, pessimistic, neutral, and null. Optimistic means that going into the study, based on prior research or belief, one thought there would be a mortality benefit to use a cap refill vs lactate clearance protocol (which corresponds to an OR 0.67). Pessimistic means that cap refill vs lactate worsens mortality (and corresponds to an OR of 1.5). Neutral means that cap refill vs lactate has no impact on mortality (with corresponding OR 1.0). Finally, null would indicate there is no prior information on cap refill vs lactate. What they found was the posterior probability that the odds ratio was <1 (meaning a beneficial effect on 28-day mortality for cap refill over lactate clearance) was >90% for all four priors: optimistic, neutral, null, and even pessimistic.
Gentle reader, I will be the first to confess that this is complex. In fact, it took me nigh upon forever to wrap my head around this, and I’m still not sure I have it. Bayesian analysis is just another way to think about this. Based on the original study, with 34.9% vs 43.4% mortality, which group would you want to be in? It makes intuitive sense to me that there is a >90% chance, in even the most pessimistic scenario, that using cap refill to guide resuscitation in septic shock beats lactate clearance. I want us all to be aware that this way of thinking is coming. Hopefully, this will help us get ready.
Effect of a Resuscitation Strategy Targeting Peripheral Perfusion Status vs Serum Lactate Levels on 28-Day Mortality Among Patients with Septic Shock: A Bayesian Reanalysis of the ANDROMEDA-SHOCK Trial. Am J Respir Crit Care Med. 2019 Oct 1. doi: 10.1164/rccm.201905-0968OC. [Epub ahead of print]
Open in Read by QxMD