Effective sample size estimation for a mechanical ventilation trial through Monte-Carlo simulation: Length of mechanical ventilation and Ventilator Free Days

•Many mechanical ventilation RCTs do not achieve significance.•LoMV data is highly skewed and normal trial design methods are often ineffective.•Monte-Carlo simulation provides a flexible trial design tool.•Exclusion criteria improve trial design and better simulate the real trial.•Ventilator Free Days can be used as an alternative outcome metric to LoMV. Randomised control trials have sought to seek to improve mechanical ventilation treatment. However, few trials to date have shown clinical significance. It is hypothesised that aside from effective treatment, the outcome metrics and sample sizes of the trial also affect the significance, and thus impact trial design. In this study, a Monte-Carlo simulation method was developed and used to investigate several outcome metrics of ventilation treatment, including 1) length of mechanical ventilation (LoMV); 2) Ventilator Free Days (VFD); and 3) LoMV-28, a combination of the other metrics. As these metrics have highly skewed distributions, it also investigated the impact of imposing clinically relevant exclusion criteria on study power to enable better design for significance. Data from invasively ventilated patients from a single intensive care unit were used in this analysis to demonstrate the method. Use of LoMV as an outcome metric required 160 patients/arm to reach 80% power with a clinically expected intervention difference of 25% LoMV if clinically relevant exclusion criteria were applied to the cohort, but 400 patients/arm if they were not. However, only 130 patients/arm would be required for the same statistical significance at the same intervention difference if VFD was used. A Monte-Carlo simulation approach using local cohort data combined with objective patient selection criteria can yield better design of ventilation studies to desired power and significance, with fewer patients per arm than traditional trial design methods, which in turn reduces patient risk. Outcome metrics, such as VFD, should be used when a difference in mortality is also expected between the two cohorts. Finally, the non-parametric approach taken is readily generalisable to a range of trial types where outcome data is similarly skewed.