Performance of army medical department health delivery components, 2001-2003: a multi-model approach

Efficiency analyses of 24 military hospitals for the years 2001-2003 were performed using Data Envelopment Analysis (DEA), Stochastic Frontier Analysis (SFA), and Corrected Ordinary Least Squares (COLS) methodologies. The facilities under analysis represented $4 billion of annual Army Medical Department expenditures. The results generally identified the same high and low performance outliers. Output and input slacks and referent sets were analyzed to determine if patterns of performance existed. Hospital cost models were then analyzed. Investigation of the optimal Box-Cox transformations for all variables resulted in the selection a simple loglinear cost function. Hospital cost was modeled as a function of workload, population, a quality and prevention proxy, an access proxy, efficiency scores (and interactions), medical center status, and the interaction between medical center status and workload. Both cross-sectional and panel series studies were conducted. Traditional, ridge, and robust regression methods were applied to the models and compared with the Stochastic Frontier Analysis estimates of hospital cost. Estimation techniques included least squares, Markov Chain Monte Carlo (MCMC) simulation, and Maximum Likelihood Estimation. The results of the comparison indicated that linear models with DEA efficiency variables provided better estimates than SFA models. The best longitudinal model was unbiased with small variance and exhibited an extremely strong linear relationship (R 2=.98). The models provided evidence to support the following relationship: Hospital Cost = f(Workload, Efficiency, Quality, MEDCEN, MEDCEN*Workload, Time). Using the models that demonstrated the smallest empirical variance and bias, parameter estimates generated at time t were then used to forecast cost at time t+1 under either the assumption of time invariant efficiency (for cross-sectional forecasts) or by employing a moving average efficiency score. These forecasts were then evaluated for efficacy. Next, a method for adjusting the funding of facilities by using management directed efficiency minimums and estimation error was proffered. The recommendations associated with this method were compared against the slack variables from DEA analysis. Additional analysis of military hospital and network efficiency were also provided as the basis for future research.