Mortality predictions in the intensive care unit: comparing physicians with scoring systems.

Area of Science:

• Critical Care Medicine
• Medical Informatics
• Health Outcomes Research

Background:

• Risk-prediction models are increasingly used in intensive care units (ICUs).
• Physician predictions are a common method for assessing patient outcomes.
• Comparing the accuracy of these two methods is crucial for clinical decision-making.

Purpose of the Study:

• To systematically review and compare the accuracy of intensive care unit (ICU) physicians' predictions versus objective scoring systems in predicting mortality for critically ill adults.
• To evaluate the diagnostic performance of physician and scoring system predictions.

Main Methods:

• A systematic review of 12 observational studies comparing physician predictions to scoring systems for ICU or hospital mortality.
• Data abstraction and quality assessment were performed independently and in duplicate.
• Summary receiver operating characteristic curves and diagnostic odds ratios were calculated.

Main Results:

• Physician predictions demonstrated significantly higher accuracy (Area Under the Curve [AUC] = 0.85 ± 0.03) compared to scoring systems (AUC = 0.63 ± 0.06, p=.002).
• Physicians' diagnostic odds ratios (12.43) were significantly higher than those for scoring systems (2.25, p=.001).
• Overall, physicians were found to predict mortality more accurately than scoring systems (ratio of diagnostic odds ratios = 1.92, p=.007).

Conclusions:

• Intensive care unit (ICU) physicians show superior ability in discriminating between survivors and non-survivors within the first 24 hours of admission compared to scoring systems.
• The moderate overall accuracy of both prediction methods suggests limited utility for clinical decision-making in the initial 24 hours of ICU care.