Standard rescue medication adjustments produce biased estimates of analgesic efficacy: reducing bias with multiple imputation for rescue adjustment (MIRA).

Abstract

Acute pain trials rely on imputation procedures to handle the confounding impact of rescue medication on pain scores. Despite long-standing concerns about the bias of single-imputation procedures such as last-, baseline-, and worst-observation carried forward (LOCF, BOCF, WOCF), they remain the standard approaches for rescue adjustment in analgesic trials. To evaluate their performance characteristics, we conducted a data-driven simulation study informed by bunionectomy trials submitted to the U.S. Food and Drug Administration. Our study demonstrates that single-imputation procedures systematically overestimate pain scores after rescue medication use, which produces biased pain curves over time and inaccurate estimates of onset of effect. Because rescue use varies by treatment effectiveness, the direction and magnitude of treatment-effect bias depend on the imputation method. In simulated trials, treatment effects on the sum of pain intensity differences were overestimated by 6% to 7% with LOCF and by 31% to 34% with WOCF and underestimated by 2% to 14% with BOCF depending on the effect size. To address this, we developed Multiple Imputation for Rescue Adjustment (MIRA), a reference-based procedure that uses historical efficacy data on an analgesic to adjust pain scores reported after its use as rescue medication. Multiple Imputation for Rescue Adjustment recovered approximately unbiased estimates under correctly specified simulation conditions and was robust to misspecification of the rescue effect magnitude by up to 50% in either direction. Simulation results were corroborated by reanalyses of 3 independent bunionectomy trials. These findings show that single-imputation methods for handling rescue medication introduce bias in acute pain trials and support MIRA as a more accurate alternative.