Gajewski Byron J, Lee Robert, Dunton Nancy
Department of Biostatistics, University of Kansas School of Medicine, Kansas City, KS, USA, 66160 ; University of Kansas School of Nursing, Kansas City, KS, USA 66160.
J Appl Stat. 2012;39(12):2639-2653. doi: 10.1080/02664763.2012.724664. Epub 2012 Sep 18.
Data Envelopment Analysis (DEA) is the most commonly used approach for evaluating healthcare efficiency (Hollingsworth, 2008), but a long-standing concern is that DEA assumes that data are measured without error. This is quite unlikely, and DEA and other efficiency analysis techniques may yield biased efficiency estimates if it is not realized (Gajewski, Lee, Bott, Piamjariyakul and Taunton, 2009; Ruggiero, 2004). We propose to address measurement error systematically using a Bayesian method (Bayesian DEA). We will apply Bayesian DEA to data from the National Database of Nursing Quality Indicators® (NDNQI®) to estimate nursing units' efficiency. Several external reliability studies inform the posterior distribution of the measurement error on the DEA variables. We will discuss the case of generalizing the approach to situations where an external reliability study is not feasible.
数据包络分析（DEA）是评估医疗效率最常用的方法（霍林斯沃思，2008年），但长期以来人们一直担心DEA假设数据测量没有误差。这极不可能，如果没有意识到这一点，DEA和其他效率分析技术可能会产生有偏差的效率估计（加耶夫斯基、李、博特、皮亚姆贾里亚库尔和汤顿，2009年；鲁杰罗，2004年）。我们建议使用贝叶斯方法（贝叶斯DEA）系统地解决测量误差问题。我们将把贝叶斯DEA应用于国家护理质量指标数据库（NDNQI®）的数据，以估计护理单元的效率。几项外部可靠性研究为DEA变量测量误差的后验分布提供了信息。我们将讨论将该方法推广到外部可靠性研究不可行的情况。
