Return to: All Abstracts
HSR&D 2004 National Meeting Abstracts


1025. Predicting Mortality and Health Care Utilization with a Single Question
Karen B DeSalvo, MD, VA Puget Sound Healthcare System, HSR&D NW Center of Excellence, VS Fan, VA Puget Sound Healthcare System, HSR&D NW Center of Excellence, MB DcDonell, VA Puget Sound Healthcare System, HSR&D NW Center of Excellence, S Fihn, VA Puget Sound Healthcare System, HSR&D NW Center of Excellence

Objectives: To evaluate the ability of a single item global health measure to predict patient mortality and resource utilization.

Methods: We analyzed prospective cohort data on 21,732 patients collected as part of the Veteran’s Affairs’ Ambulatory Care Quality Improvement Project (ACQUIP). All patients completed a baseline general health status measure (SF-36) and an inventory of chronic conditions. The predictive and discriminative ability of a single question rating subjects’ health (Excellent, Very Good, Good, Fair, Poor [EVGFP]) was compared to the SF-36 Physical Component Score (PCS) and the Seattle Index of Comorbidity (SIC). We created age-adjusted, logistic regression model to predict mortality and hospitalizations during a 1-year follow-up period. We compared the discriminative ability of the predictors by comparing the area under receiver operator curves (AUC) for each model.

Results: Total mortality rate was 11%, and 32% were hospitalized during the study period Compared to subjects reporting ‘Excellent’ or ‘Very Good’ health, patients reporting ‘Poor’ health were 7 times more likely to die in the ensuing year (OR 7.2 [5.1, 10.1]). Patients with ‘Poor’ self-rated health were also significantly more likely to be hospitalized (OR 3.94 [3.4, 4.6]). The EVGFP, PCS and SIC had comparable AUC for predicting mortality (AUC 0.74, 0.73, and 0.73, p<0.001) and hospitalization (AUC 0.63, 0.64, and 0.61, p<0.001)

Conclusions: EVGFP response categories discriminate patients with varying risks.

Impact: EVGFP, collectable at point-of-care, is comparable to more extensive, established risk predictors making it a useful routine risk prediction tool.