Predicting inpatient rehabilitation length of stay

Using standardized forms and predefined criteria, information was collected on all 1,238 patients admitted to the inpatient rehabilitation facility at our university hospital between August 1, 1980 and December 30, 1986. Data from 96% of these patients were used retrospectively to create a mathematic model, based on multiple linear regression, that predicts the patient's total rehabilitation length of stay (LOS). The model requires only information about the patient's admitting diagnosis, referral source, admission functional status, and date of admission. The model compared favorably with prospective estimates of LOS made independently by attending physicians at admission to rehabilitation. We conclude that such models could be used to facilitate management of rehabilitation units, forecast patient census, schedule unit personnel, set interim goals for LOS, and facilitate discharge planning. The delivery of rehabilitation services, like the delivery of other medical services, can be defined in part by objective, measurable patient characteristics.