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.