Objective: Trials of potential new therapies in acute lung injury are difficult and expensive to conduct. This article is designed to determine the utility, behavior, and statistical properties of a new primary end point for such trials, ventilator-free days, defined as days alive and free from mechanical ventilation. Describing the nuances of this outcome measure is particularly important because using it, while ignoring mortality, could result in misleading conclusions.
Design: To develop a model for the duration of ventilation and mortality and fit the model by using data from a recently completed clinical trial. To determine the appropriate test statistic for the new measure and derive a formula for power. To determine a formula for the probability that the test statistic will reject the null hypothesis and mortality will simultaneously show improvement. To plot power curves for the test statistic and determine sample sizes for reasonable alternative hypotheses.
Setting: Intensive care units.
Patients: Patients with acute respiratory distress syndrome or acute lung injury as defined by the American-European Consensus Conference.
Main results: The proposed model fit the clinical data. Ventilator-free days were improved by lower tidal volume ventilation, but the improvement was mostly caused by the improved mortality rate, so trials that expected similar effects would only have modest increase in power if they used ventilator-free days as their primary end point rather than 28-day mortality. Similar results were obtained using the model in two groups segregated by low or high Acute Physiology and Chronic Health Evaluation score. On the other hand, if patients are divided into two groups on the basis of the lung injury score, both the duration of ventilation and mortality are lower in the low lung injury score group. A trial of a treatment that had a similar clinical effect would have a large increase in power, allowing for a reduction in the required sample size.
Conclusions: Use of ventilator-free days as a trial end point allows smaller sample sizes if it is assumed that the treatment being tested simultaneously reduces the duration of ventilation and improves mortality. It is unlikely that a treatment that led to higher mortality could lead to a statistically significant improvement in ventilator-free days. This would be especially true if the treatment were also required to produce a nominal improvement in mortality.