Comparison of risk prediction scoring systems for ward patients: a retrospective nested case-control study

INTRODUCTION The rising prevalence of rapid response teams has led to a demand for risk-stratification tools that can estimate a ward patient's risk of clinical deterioration and subsequent need for intensive care unit (ICU) admission. Finding such a risk-stratification tool is crucial for maximizing the utility of rapid response teams. This study compares the ability of nine risk prediction scores in detecting clinical deterioration among non-ICU ward patients. We also measured each score serially to characterize how these scores changed with time. METHODS In a retrospective nested case-control study, we calculated nine well-validated prediction scores for 328 cases and 328 matched controls. Our cohort included non-ICU ward patients admitted to the hospital with a diagnosis of infection, and cases were patients in this cohort who experienced clinical deterioration, defined as requiring a critical care consult, ICU admission, or death. We then compared each prediction score's ability, over the course of 72 hours, to discriminate between cases and controls. RESULTS At 0 to 12 hours before clinical deterioration, seven of the nine scores performed with acceptable discrimination: Sequential Organ Failure Assessment (SOFA) score area under the curve of 0.78, Predisposition/Infection/Response/Organ Dysfunction Score of 0.76, VitalPac Early Warning Score of 0.75, Simple Clinical Score of 0.74, Mortality in Emergency Department Sepsis of 0.74, Modified Early Warning Score of 0.73, Simplified Acute Physiology Score II of 0.73, Acute Physiology and Chronic Health Evaluation II of 0.72, and Rapid Emergency Medicine Score of 0.67. By measuring scores over time, it was found that average SOFA scores of cases increased as early as 24 to 48 hours prior to deterioration (P = 0.01). Finally, a clinical prediction rule which also accounted for the change in SOFA score was constructed and found to perform with a sensitivity of 75% and a specificity of 72%, and this performance is better than that of any SOFA scoring model based on a single set of physiologic variables. CONCLUSIONS ICU- and emergency room-based prediction scores can also be used to prognosticate risk of clinical deterioration for non-ICU ward patients. In addition, scoring models that take advantage of a score's change over time may have increased prognostic value over models that use only a single set of physiologic measurements.