Efficiency Measurement Using Diagnosis Procedure Combination (DPC) Data and the Evaluation of Hospital Functions

Abstract

We used Diagnosis Procedure Combination (DPC) data to capture more detailed outputs and measured hospital efficiency using stochastic frontier analysis (SFA).We weighted the number of patients (outputs) by relative coefficients obtained from DPC data. Since previous studies conducted in Japan simply used the number of patients seen and the number of discharged patients, healthcare organizations that treated patients who required more medical resources than average patients experienced worse efficiency measurements. Weighting makes a more accurate estimation of the production functions possible.The hospital standardized mortality ratio (HSMR: the expected mortality rate to the observed mortality rate) obtained from the DPC data was added as a quality variable into the estimate model to control for quality. Quality variables were not used in previous Japanese studies, and it was assumed that the quality of healthcare services was uniform. If the quality of healthcare services rendered in Japanese hospitals varied greatly, then it is possible that gaps in efficiency are greater than previously understood.To control for unobserved heterogeneity in the sampling hospitals, we used the true fixed effect model (TFEM) proposed by Greene (2004, 2005). The characteristics of a TFEM enabled a more accurate estimation of the efficiency by adding dummy variables for each hospital to the more traditional SFA, and complementing the fixed effects with these dummy variables. These fixed effect values reflect gaps in the production structures (cost structures) among hospitals, and may be useful as functional coefficients of the DPC data.Our analyses indicated that the estimated efficiency was approximately 0.6, which was lower than estimated in previous studies. The results were lower because quality was controlled by quality variables in this study. The mean efficiency value in the model without quality variables was 0.74. The histogram of the fixed effect values for each hospital showed a peak around 0.9. When the same histogram was converted to a stem and leaf plot, this difference in distribution was not apparent.