Predicting cost for new HMO subscribers.

The purpose of the project was to develop a model for predicting costs for potential new HMO subscribers, using available cost data from fiscal year 1985 for current enrollees of a large HMO. Regression analysis of aggregated clinic, referral, and hospital cost data using a log transformation of cost indicated that 20 percent of the variation in cost could be explained by sex and coverage type of the subscriber, compared with 7 percent explainable by a simple comparison of costs for single versus family subscribers. Subscriber age, while by itself a significant and nonlinear predictor of cost, was not significant when controlled for coverage type. Application of the model to 28 large companies yielded predicted costs well correlated with observed costs (r = .75, p less than .01). Prediction was significantly better for companies with low observed mean costs than for companies with high observed mean costs.