Grouping Hospitals by Quality A PRIDIT Approach with

Grouping Hospitals by Quality - A PRIDIT Approach with Bootstrapping Robert D. Lieberthal The Wharton School, University of Pennsylvania Abstract Methods Conclusions § § § Background. PRIDIT, a combination of PCA and Ridit scoring, has been used to rank hospitals by quality. § Objectives. To show the additional use of § PRIDIT – Principal Components Analysis (PCA) on the Ridit scores. Bootstrapping – 1000 samples with replacement from the population. § Bootstrapping allows us to gain confidence intervals of hospital quality. Standard errors are greatest for the extremes of the quality distribution. Results Table 1: Philadelphia Hospitals PRIDIT score Bootstrapped standard error Decile Quality Level St Joseph Hospital -0. 02311 0. 00035 1 st 1 Chestnut Hill Hospital -0. 00831 0. 00028 3 rd 2 Bootstrapped standard errors are generally small in comparison to the PRIDIT score. Kensington Hospital -0. 00121 0. 00026 6 th 3 Conclusions. Some hospitals are “close” along the Penn Presbyterian Medical Center -0. 00118 0. 00026 6 th 3 Hahnemann University Hospital -0. 00108 0. 00027 6 th 3 Background Thomas Jefferson University Hospital -0. 00093 0. 00030 6 th 3 § § Hospital Of Univ Of Pennsylvania 0. 00021 0. 00038 6 th 4 Temple East Inc 0. 00151 0. 00035 6 th 5 Frankford Hospital 0. 00413 0. 00037 7 th 6 Nazareth Hospital 0. 00415 0. 00037 7 th 6 Wills Eye Hospital 0. 00546 0. 00030 7 th 7 Jeanes Hospital 0. 00633 0. 00044 7 th 8 Pennsylvania Hospital, The 0. 00858 0. 00033 8 th 9 Albert Einstein Medical Center 0. 01330 0. 00039 8 th 10 Temple University Hospital 0. 01419 0. 00046 9 th 10 bootstrapping can be used to segment hospitals by quality. § Methods. Using national data from Hospital Compare, we demonstrate how Philadelphia area hospitals can be differentiated by quality at different levels of confidence. § § Results. Hospitals can be differentiated by quality metric, while others are clearly distinguishable. Some hospitals of similar quality are also geographically close. PRIDIT is an “unsupervised” learning method. PRIDIT was developed by Brockett et. al. (2002) to construct a rank measure of fraud and applied by Lieberthal (forthcoming) to construct national measure of hospital quality. Objectives § § Develop standard errors for hospital scores using bootstrapping. Compare hospitals in terms of quality and physical distance. Hospital Name Figure 1: Philadelphia Hospitals