Non-Discretionary Inputs Are Important Factors In Hospital Efficiency Studies And Policy Evaluation

Abstract

Objectives: Hospital efficiency is the focus of several studies and increasingly, data envelopment analysis (DEA) is used to compute and compare efficiencies of hospitals for quality improvement and policy analysis. DEA is a non-parametric approach to compute efficiency considering multiple inputs and outputs and benchmark hospitals, with the presumption that the inefficient hospitals can reduce inputs and/or increase outputs to improve their efficiency. Hospitals are complex systems and have multiple non-discretionary inputs, such as type of hospital or region of location that are not under the control of administration, and hence cannot be altered. This study emphasizes the need to consider non-discretionary inputs in DEA models. Methods: One year's worth of Agency for Healthcare Research & Quality's Health Care Utilization Project data was used. Variables included full-time equivalent of registered nurses, licensed practical nurses and nurse aides as inputs and total discharges and percent of surgeries as outputs, with bed size, urban/ rural, teaching status, region and ownership as non-discretionary inputs. First a variable-returns-to-scale DEA model was run without the non-discretionary inputs. Next the model was repeated including an environmental harshness (created using regression). Results: When the efficiency scores of the 862 hospitals in the two stages were compared, 755 hospitals had increased efficiency, and mean increased more than three times from 0.11 (P< 0.000) and standard deviation doubled from 0.15. The number of efficient hospitals increased from 12 to 72. Conclusions: DEA can be a sophisticated method to measure hospital efficiency. Not accounting for non-discretionary inputs can radically alter the efficiency scores and bias study results. The first stage score has inefficiency and the effect of non-discretionary inputs. Since both mean and standard deviation increased dramatically, simple normalization cannot be used as a proxy. Appropriate treatment of non-discretionary inputs is necessary to make meaningful comparisons to inform quality and policy interventions.