Outcome Predictions at Two Time Points among Surrogates and Physicians of Mechanically Ventilated Patients.

Abstract

Objectives: The decisions surrogates and physicians make for incapacitated critically ill patients depend in part on their expectations for patient recovery. We sought to determine whether the accuracy of surrogate and physician outcome predictions made during the ICU stay improves over time.

Design: Survey study.

Setting: Academic Medical Center.

Subjects: Surrogates and physicians of 100 mechanically ventilated patients from March 2018 to April 2019.

Interventions: At the end of the first week of mechanical ventilation and 1 week later, participants indicated on visual analog scales (0-100%) expectations that the patient would require mechanical ventilation in 1 month, require artificial nutrition in 1 month, be alive in 3 months, and be living at home in 3 months. Patient status was determined at 1 and 3 months.

Measurements and main results: Area under the receiver operating characteristic curves (AUROCs) were determined for each outcome, at each time point. Patients who died within the first month were considered to require mechanical ventilation and artificial nutrition in the primary analysis. AUROCs for initial surrogate predictions were 0.61 (95% CI, 0.50-0.72) for mechanical ventilation, 0.67 (95% CI, 0.56-0.78) for artificial nutrition, 0.66 (95% CI, 0.55-0.7) for survival, and 0.61 (95% CI, 0.50-0.73) for living at home. AUROCs for initial physician predictions were 0.60 (95% CI, 0.49-0.71) for mechanical ventilation, 0.72 (95% CI, 0.61-0.0.83) for artificial nutrition, 0.69 (95% CI, 0.59-0.80) for survival, and 0.76 (95% CI, 0.66-0.85) for living at home. Average expectations among surrogates and physicians were highly stable over time; adjustments made to expectations did not result in more accurate predictions for the measured outcomes (p > 0.05).

Conclusions: Among surrogates and physicians of patients who were mechanically ventilated for 1 week, outcome predictions were better than would be expected by chance and not significantly improved 1 week later.