Cost as a study outcome: sensitivity of study conclusions to the method of estimating cost.

Abstract

The analyses presented here are intended to provide empirical guidance to two questions faced by researchers performing clinical trials which include a cost component: Which health care services should we track? Should we use facility specific costs or national average costs for individual services in estimating total costs? We reanalyzed cost data from the Department of Veterans Affairs (VA) multisite clinical trial which compared Adult Day Health Care (ADHC) to Customary Care for patients at high risk for nursing home care. The data presented here compares the original analysis (a combination of local and national costs) to an analysis based on purely facility-specific costs and to an analysis based upon purely VA national costs. Costs for hospital, clinic, nursing home, ADHC, hospital based home care, rehabilitation, pharmacy, and laboratory were included. Hospital, nursing home, clinic, and ADHC in combination account for 98% of the variation in total cost per patient. Including only hospital, clinic, nursing home, ADHC, and hospital-based home care in total cost per patient closely replicated the findings for total cost when all services were included. The originally reported analysis and the 2 new analyses, using respectively facility specific costs and national average costs, did differ substantially in the magnitude of the difference between the total cost per patient of ADHC and Customary Care. They did differ with regard to statistical significance as the P values were either slightly above or below 0.05. Ideally all health care costs should be included in the analysis. When this is not feasible, one should determine utilization and cost for the intervention itself, costly services (usually hospital, nursing home, and clinic care), and lower cost services that are likely to be affected by the intervention. Sensitivity analysis should be performed to determine if different methods of costing (eg, facility specific versus national costs) materially affect the conclusions of the study.