A predictive model for patient length of stay at a teaching hospital

This research attempts to construct mathematical models for estimating length of stay per admission and investigating the effects of patients\u27 characteristics and clinical indicators on the length of stay for the top ten Diagnosis-Related Groups (DRGs) at a teaching hospital. It is also concerned with the development of cost per admission and cost per patient day functions. Further, these functions are used for analysis to determine a value of the length of stay that would minimize cost per patient day. Also, the effects of changing the length of stay (from the actual to the projected levels) on the total cost per year and the cost per patient day are examined. Moreover, the current cost system for the teaching hospital is evaluated and a new cost system (Activity-Based Costing) is proposed. The nuclear medicine unit is selected to implement the new cost system. The results indicate that the patients\u27 characteristics and the clinical indicators explain approximately 64% of the variation in the length of stay; also model prediction is 79% accurate. The effects of the clinical indicators on the length of stay are much stronger than the patients\u27 characteristics. The cost models fit the data as shown by the following indicators: the average of R2 is 0.79 and the mean of MAPE is 15. The cost variation analysis demonstrates that if a hospital can control the length of stay at the projected level, on an average, the cost per admission and the cost per patient day will decrease. Based on the top ten DRGs (6,367 admissions) in the year 1999, the cost per year and the cost per patient day decreased approximately 13% and 11%, respectively using the cost minimization analysis. The research confirms that the Activity-Based Costing can be applied to healthcare industry, and provides more accurate cost information than the current system. It assists management in effective cost reduction by focusing on non-value-added and providing more accurate statistics for pricing. Overall, this research offers a new decision support instrument for healthcare administrators