Beyond the ventilator-free days: review of several estimands.

backgroundMortality is a critical endpoint in clinical research, but identifying meaningful differences necessitates large sample sizes. Consequently, composite outcomes such as ventilator-free days (VFDs) have been developed, combining survival and ventilation duration into a single measure. Different statistical methods used to analyse VFDs lead to different estimands. Traditionally, VFDs are treated as a count; however, some models consider time to death and time to extubation separately. This review explores the applicability of several time-to-event models and innovative approaches. MAIN TEXT: The first model to consider is the competing risks approach using the Fine-Gray model. This approach focuses solely on the initial extubation event and considers death as a competing event. Second, to incorporate all extubation and reintubation events, multistate models can be employed. Specifically, the multiple-event framework, which allows for multiple transitions between intubation and extubation, while the recurrent events framework, focuses on extubation recurrence. However, these models require complete data and a sufficient number of events for analysis. Third, current ventilation-free survival estimates use methods adapted from leukaemia-free survival to evaluate the probability of remaining extubated and alive over time. Finally, the mixture cure model distinguishes between deceased and extubated individuals within the non-deceased population. It models death through logistic regression and extubation timing through survival regression among living patients.

conclusionIn critical care, especially for acute respiratory distress syndrome, three key states are intubation, extubation, and death. We do not advocate a one-size-fits-all model because the choice depends heavily on the specific goals. The key is to decide which estimand the study will target in the statistical plan, before initiating the study, and to ensure the analysis model is the most appropriate for addressing the research question. .