What is the optimal approach to analyse ventilator-free days? A simulation study

Area of Science:

• Critical care medicine
• Biostatistics
• Clinical trial methodology

Background:

• Ventilator-free days (VFDs) are a key composite outcome in critical care, assessing survival and mechanical ventilation duration.
• Current statistical methods for analyzing VFDs are inconsistent, leading to varied research interpretations.
• Emerging methods like multistate models and win ratio require evaluation for optimal application.

Purpose of the Study:

• To evaluate and compare various statistical models for analyzing ventilator-free days (VFDs).
• To identify the most statistically powerful and reliable method for VFD analysis in critical care research.

Main Methods:

• Simulated 16 datasets (300 individuals each) comparing control and intervention groups with varied survival and ventilation parameters.
• Applied twelve statistical methods, including count-based, time-to-event, multistate, and win ratio models.
• Validated models using four real-world clinical trial datasets and conducted sensitivity analyses.

Main Results:

• Most methods controlled Type I error, but zero-inflated/hurdle Poisson/negative binomial and cause-specific Cox models showed issues.
• Time-to-event approaches, Mann-Whitney test, proportional odds model, and win ratio demonstrated superior power.
• Multistate model, Mann-Whitney test, proportional odds model, and win ratio showed significant associations in real datasets.

Conclusions:

• The multistate model is recommended as the optimal approach for analyzing VFDs.
• It demonstrated superior performance and provided a more interpretable effect size compared to proportional odds and win ratio models.