Perfusion-guided resuscitation in septic shock: a 13% mortality reduction the frequentist analysis will not call significant

Design

systematic review with prespecified dual frequentist and Bayesian meta-analysis, PROSPERO registered, GRADE assessed

Population

2,408 adults with septic shock across 7 randomised trials

Primary outcome

28-day mortality

Effect

risk ratio 0.87 (95% CI 0.76-1.01, p=0.06, I squared 0%, moderate certainty); Bayesian posterior probability of risk ratio below 1, 97.2%

Seven randomised trials, 2,408 adults, all in septic shock, comparing resuscitation guided by peripheral perfusion - principally capillary refill time - against standard care or lactate-guided resuscitation. The analysis was prespecified as a dual framework: a frequentist random-effects model as primary, and a Bayesian hierarchical model with several prior assumptions alongside it.

The frequentist pooled risk ratio for 28-day mortality was 0.87 (95% CI 0.76 to 1.01, p=0.06), with no heterogeneity at all and moderate-certainty evidence. The Bayesian analysis put the posterior probability that the risk ratio is below 1 at 97.2%, consistent across priors. Both statements describe the same data. The frequentist reading is that the trial evidence does not exclude no effect; the Bayesian reading is that, given these data, benefit is far more likely than harm. The authors note that the two largest and least biased trials were individually non-significant, which is the caution against over-reading the pooled direction.

What this supports in practice is modest and worth doing: capillary refill time belongs in the resuscitation assessment alongside lactate rather than instead of it, and a normalising refill time in a patient whose lactate is still falling is reassurance rather than a reason to keep giving fluid. It is free, needs no device, and works in units where serial lactates are not available - which is a substantial argument in Indian district practice, independent of the mortality question.

• Add a standardised capillary refill time to the resuscitation assessment, measured the same way each time
• Do not abandon lactate - the trials compared perfusion-guided against lactate-guided or usual care, not lactate against nothing
• Use a normalising refill time as a reason to stop escalating fluid, which is where the plausible benefit lies
• In units without serial lactate, this is a usable target rather than a second-best one
• Expect the guidelines to describe this as a reasonable option rather than a recommendation, and say so to trainees

Why it matters

It puts a free bedside measurement on the same footing as serial lactate, which changes what resuscitation looks like in units that cannot run lactates.

Don't overread it

The primary frequentist analysis was not significant, and the two largest low-risk trials were individually null - this is a signal, not an established mortality benefit.

The statistics, in plain English

This is the clearest recent example of two statistical frameworks answering different questions about the same numbers. The frequentist p of 0.06 asks how surprising these data would be if the treatment did nothing, and answers not quite surprising enough. The Bayesian 97.2% asks how probable benefit is given the data and a stated prior, and answers very. Neither is wrong. What both agree on is that the size of any benefit is uncertain: the interval reaches 1.01, and the two largest and least biased trials were individually null, which is exactly the pattern that later fails to replicate in a big trial.

Read the rest in the app

You have read your two free briefings this month. The app carries all 27 specialties, every morning, free — and this finding is waiting in it.

Get Daily Doctor — free

Scan to keep reading on your phone. No account needed to start.