Predicting rural health care utilization with archival data

Abstract

This study explored the usefulness of archival data in predicting rural health care utilization. A regression model was used to see how well observed utilization for local populations could be predicted by calculating expected values in advance from age- and sex-specific national rates applied to local age and sex profiles. Although the correlation between observed and expected utilization was reasonably high (r = 0.92), an attempt was then made to improve prediction by considering other data that do not require independent collection. These archival data included indicators of historic utilization (local Medicaid payments, the percentage of births to county residents occurring in the mother's county of residence, percentage of children immunized, and infant mortality) and services already available. Observed utilization data were obtained by surveys in eight rural counties, and the predictor was tested on three additional rural communities. A predictor equation that added to the expected utilization only one variable (the percentage of births to county residents occurring in the mother's county of residence) was found to account for approximately 95% of the variance in observed utilization. This predictor is recommended for planners who need convenient, low-cost market feasibility estimates for proposed project sites and a way to establish intermediate goals or incentives during early project development.