Simulation study of frailty as a baseline confounder: the need to improve reporting intensive care trials

Frailty is an independent predictor of worse outcomes after critical illness, yet intensive care trials almost invariably omit reporting frailty distribution in the investigated cohort. This study investigated the magnitude and prevalence of imbalanced baseline distribution in frailty using the ordinal Clinical Frailty Scale that could contribute to clinically significant differences in apparent mortality.

A database of 6968 patients admitted to intensive care unit was used to simulate enrolment of between 100 and up to 1000 patients into each of the control and interventional arms of a hypothetical two-arm trial. Both 1:1 randomisation and block randomisation using permuted blocks of variable sizes were simulated 100 times and referred to as 'trials'. The observed relationship between the Clinical Frailty Scale and observed in-hospital mortality was used to determine thresholds for frailty imbalance sufficient to generate ≥2% and ≥5% differences in mortality. The proportions of 'trials' meeting these criteria are reported as percentages and 95%CI.

Imbalanced frailty generated a ≥2% mortality difference in 36 [95%CI 31-42]% of 'trials' with 100 patients/arm that was reduced to 9 [95%CI 6-13]% of 'trials' with 400 patients/arm. An imbalance in frailty associated with a ≥5% mortality difference occurred in 3 [95%CI 1-5]% of 'trials' with 100 patients/arm and reduced to 0% [95%CI 0-1.3]% in all 'trials' using larger sample sizes. 'Trials' with a binary balance between non-frail vs frail patients still retained an ordinal frailty imbalance to generate a ≥2% mortality difference in 10 [95%CI 7-14]% with 100 patients/arm. Block randomisation only reduced the proportion of 'trials' with mortality differences related to frailty to a minor and variable degree.

Baseline distributional imbalance in frailty may create clinically significant differences in apparent mortality estimates. The hypothesis-generating findings in this study suggest that intensive care trials should report complete frailty distributions and consider frailty in study designs and analyses.