Estimating attributable fraction of mortality from sepsis to inform clinical trials

Abstract

Purpose: Nearly all sepsis trials report no statistically significant difference in mortality. The attributable fraction of deaths due to sepsis (AFsepsis) may be an important, yet overlooked consideration. We derived AFsepsis and explored the effect of incorporating AFsepsis into sample size calculations.

Materials and methods: We derived AFsepsis with a matched cohort study using consecutive admissions to adult general intensive care units (ICUs) in England (n = 614,509). Cases were ICU patients with sepsis and the two controls were ICU-non-sepsis controls, matched for propensity to have sepsis and age-sex-matched general population. The primary exposure was sepsis. The primary outcome was hospital mortality. We generated sample size graphs, by varying control group mortality (10%-60%), relative risk reduction (0-1), for 80% power and 5% alpha. We then compared AFsepsis derived sample sizes with sample size calculations from published sepsis trials.

Results: AFsepsis was 15% (95% CI: 14%-16%) compared with propensity matched ICU-non-sepsis controls and 93% (95% CI: 92%-93%) compared with age-sex-matched general population controls. When comparing AFsepsis derived sample sizes with sample size calculations from 18 trials meeting our selection criteria, these calculations assumed very high AFsepsis and/or very effective treatments.

Conclusions: Estimating trial specific AFsepsis to inform sample size calculations could be an additional step in sepsis trial design.

Keywords: Attributable fraction; Randomized controlled trial; Sample size; Sepsis.