Improving the Reporting of Trials Evaluating Organ-Support Therapies Using Multistate Modeling.

Abstract

Rationale: Managing critically ill patients in the ICU often involves organ-support therapies (OST), such as mechanical ventilation, extracorporeal membrane oxygenation, renal replacement therapy, and various pharmacologic strategies. Clinical trials in this context pursue diverse goals-including improving survival, reducing OST use, facilitating weaning, or comparing timing of OST initiation-which leads to substantial heterogeneity in OST-related endpoints. One commonly used outcome, the number of OST-free days (OFD), has been criticized for its composite nature, which can obscure important clinical differences between patients with similar OFD values. Variability in how weaning success is defined, how intercurrent OST-free periods are handled, and how death is incorporated further complicates comparisons across trials.

Objectives: To illustrate how multistate modeling can offer an intuitive framework for analyzing randomized clinical trials involving OSTs, and how this approach allows to better describe and compare patient conditions during the entire follow-up.

Methods: We describe the core principles of multistate modeling, including its assumptions (like the Markov assumption), advantages, and limitations. We then present two recent randomized controlled trials evaluating OSTs and identify the main statistical challenges encountered in their analysis.

Measurements and main results: Using a multistate modeling approach, we re-analyzed both trials to characterize and compare patient trajectories over time. The multistate framework enabled clearer insight into how interventions impact the timing of transitions between clinical states, providing a richer and more clinically relevant understanding of treatment effects.

Conclusions: Multistate modeling can substantially inform the interpretation and primary analysis of a clinical trial evaluating an organ-support therapy.