Health Care Expenditure Prediction With a Single Item, Self-Rated Health Measure

Background: Prediction models that identify populations at risk for high health expenditures can guide the management and allocation of financial resources. Objective: To compare the ability for identifying individuals at risk for high health expenditures between the single-item assessment of general self-rated health (GSRH), "In general, would you say your health is Excellent, Very Good, Good, Fair, or Poor?", and 3 more complex measures. Study Design: We used data from a prospective cohort, representative of the US civilian noninstitutionalized Population, to compare the predictive ability of GSRH to: (1) the Short Form-12, (2) the Seattle Index of Comorbidity, and (3) the Diagnostic Cost-Related Groups/Hierarchal Condition Categories Relative-Risk Score. The outcomes were total, pharmacy, and office-based annualized expenditures in the top quintile, decile, and fifth percentile and my inpatient expenditures. Data Source: Medical Expenditure Panel Survey panels 8 (20032004, n = 7948) and 9 (2004-2005 5 n = 792 1). Results: The GSRH model predicted the top quintile of expenditures, as well as the SF-12, Seattle Index of Comorbidity, though not as well as the Diagnostic Cost-Related Groups/Hierarchal Condition Categories Relative-Risk Score: total expenditures [area Under the Curve (AUC): 0.79, 0.80, 0.74, and 0.84, respectively], pharmacy expenditures (AUC: 0.83, 0.83 0.76, and 0.87, respectively), and office-based expenditures (AUC: 0.73, 0.74, 0.68, and 0.78, respectively), as well as any hospital inpatient expenditures (AUC: 0.74, 0.76, 0.727 and 0.78, respectively). Results were similar for the decile and fifth percentile expenditure cut-points. Conclusions: A simple model of GSRH and age robustly stratifies populations and predicts future health expenditures generally as well as more complex models.