Objective assessment of changing mortality risks in pediatric intensive care unit patients

Abstract

Objective: To develop and validate a mortality risk predictor based on physiologic data that estimates daily the probability of a patient dying within the next 24 hrs as that probability changes with disease and recovery.

Setting: Nine pediatric ICUs in tertiary care centers.

Patients: Data from 1,401 patients (116 deaths, 5,521 days of care) were used for predictor development, and 1,227 patients (105 deaths, 4,597 days of care) provided data for predictor validation.

Methods: The predictor was developed by logistic regression analysis using the Pediatric Risk of Mortality scores of all previous days as potential predictor variables. Performance was measured by the area under the receiver operating characteristic curve (Az), and by the comparison of the daily predicted vs. observed patient status in five mortality risk groups (less than 0.01, 0.01 to 0.05, 0.05 to 0.15, 0.15 to 0.3, greater than 0.3) using chi-square goodness-of-fit tests.

Measurements and main results: Only the most recent and the admission day Pediatric Risk of Mortality scores (with a weighting ratio of 3:1) contributed significantly (p less than .05) to the prediction. The overall prediction attained an accuracy of Az = 0.904. The daily number and distribution of survivors and nonsurvivors in the five mortality risk groups were well predicted in the total sample (chi 2 [5 degrees of freedom] = 2.51; p greater than .75), and each ICU separately (chi 2 [5 degrees of freedom] range 2.41 to 7.96; all p greater than .15). This dynamic predictor improved (p less than .01) ICU outcome prediction over an admission-day predictor and, in the opinion of the authors, is essential for pediatric ICU efficiency analysis.

Conclusions: The predictor is valid for assessing the 24-hr mortality risk in pediatric ICU patients hospitalized in other tertiary care institutions, different from those used for predictor development. The predicted mortality risks allow prospective patient stratification into risk groups. The ability of this predictor to follow risk changes over time expands its applicability over static predictors by enabling the charting of patient courses, and permitting ICU efficiency analysis.