The relationship between physiologic status and mortality risk should be reevaluated as new treatment protocols, therapeutic interventions, and monitoring strategies are introduced, and as patient populations change. We developed and validated a third-generation pediatric physiology-based score for mortality risk, Pediatric Risk of Mortality III (PRISM III).
There were 32 pediatric intensive care units (ICUs): 16 pediatric ICUs were randomly chosen and 16 volunteered.
Consecutive admissions at each site were included until at least 11 deaths per site occurred.
Physiologic data included the most abnormal values from the first 12 and the second 12 hrs of ICU stay. Outcomes and descriptive data were also collected. Physiologic variables where normal values change with age were stratified by age (neonate, infant, child, adolescent). The database was randomly split into development (90%) and validation (10%) sets. Variables and their ranges were chosen by computing the risk of death (odds ratios) relative to the midrange of survivors for each physiologic variable. Univariate and multivariate statistical procedures, including multiple logistic regression analysis, were used to develop the PRISM III score and mortality risk predictors.
Data were collected on 11,165 admissions (543 deaths). The PRISM III score has 17 physiologic variables subdivided into 26 ranges. The variables most predictive of mortality were minimum systolic blood pressure, abnormal pupillary reflexes, and stupor/coma. Other risk factors, including two acute and two chronic diagnoses, and four additional risk factors, were used in the final predictors. The PRISM III score and the additional risk factors were applied to the first 12 hrs of stay (PRISM III-12) and the first 24 hrs of stay (PRISM III-24). The Hosmer-Lemeshow chi-square goodness-of-fit evaluations demonstrated absence of significant calibration errors (p values: PRISM III-12 development equals .2496; PRISM III-24 development equals .1374; PRISM III-12 validation equals .4168; PRISM III-24 validation equals .5504). The area under the receiver operating curve and Flora's z-statistic indicated excellent discrimination and accuracy (area under the receiver operating curve--PRISM III-12 development 947 plus minus 0.007; PRISM III-24 development 0.958 plus minus 0.006; PRISM III-12 validation 0.941 plus minus 0.021; PRISM III-24 validation 0.944 plus minus 0.021; Flora's z-statistic--PRISM III-12 validation equals .7479; PRISM III-24 validation equals .9225), although generally, the PRISM III-24 performed better than the PRISM III-12 models. Excellent goodness-of-fit was also found for patient groups stratified by age (significance levels: PRISM III-12 equals .1622; PRISM III-24 equals .4137), and by diagnosis (significance levels: PRISM III-12 equals .5992; PRISM III-24 equals .7939).
PRISM III resulted in several improvements over the original PRISM. Reassessment of physiologic variables and their ranges, better age adjustment for selected variables, and additional risk factors resulted in a mortality risk model that is more accurate and discriminates better. The large number of diverse ICUs in the database indicates PRISM III is more likely to be representative of United States units.
(Crit Care Med 1996; 24:743-752)
From the Departments of Anesthesiology (Dr. Pollack) and Pediatrics (Drs. Pollack, Patel, and Ruttimann), George Washington University School of Medicine; Children's National Medical Center (Drs. Pollack and Patel), Center for Health Services and Clinical Research, Children's Research Institute (Drs. Pollack and Patel); National Institute on Alcohol Abuse and Alcoholism (Dr. Ruttimann), the National Institutes of Health, Washington, DC.
Supported, in part, by grant MCH-110584 from the Maternal and Child Health Bureau (Title V, Social Security Act), Health Resources and Services Administration, Department of Health and Human Services.
See page 752 for information regarding PRISM III and updated PRISM algorithms.
PRISM III and updated PRISM algorithms are copyrighted and may be the subject of one or more patents held by Children's Research Institute. The equations are available without charge for research uses including the independent verification of their accuracy and reliability. Children's National Medical Center may receive compensation resulting from nonresearch uses of PRISM III and PRISM algorithms.
Address requests for reprints to: Murray M. Pollack, MD, Children's National Medical Center, 111 Michigan Avenue, NW, Washington, DC 20010.