Accuracy of Early Prediction of Duration of Mechanical Ventilation by Intensivists

Abstract Rationale Predictions of duration of mechanical ventilation are frequently made by intensivists and influence clinical decisions. Objectives We aimed to measure the accuracy of these clinical early predictions. Methods One hundred fifty-five patients within 48 hours of initiation of mechanical ventilation on a general intensive care unit had clinical data prospectively collected and were followed to determine actual duration of mechanical ventilation. Intensivists, after evaluating patients in the first 2 consecutive days, predicted each duration of mechanical ventilation by selecting between less than 3, 4 to 7, 8 to 14, or more than 14 days. Accuracy of predictions was evaluated by comparisons between predicted and actual durations. Measurements and Main Results Raw agreement (95% confidence interval) between predicted and actual durations, classified into the four categories, was 37% (29–45%). Predictions of duration of more than 7 and more than 14 days showed raw agreements of 59% (51–66%) and 83% (76–88%); sensitivities of 40% (28–54%) and 29% (13–51%); specificities of 71% (61–80%) and 93% (87–97%); positive predictive values of 48% (34–63%) and 44% (20–70%); and negative predictive values of 64% (54–73%) and 87% (81–92%), respectively. Conclusions The accuracy of intensivists’ early clinical predictions of duration of mechanical ventilation is limited, particularly for identifying patients who will require prolonged mechanical ventilation.