Statistical Methods for Classifying Hospital Quality Using Hierarchical Nonlinear Mixed-Effects Models
Morton, David J.
MetadataShow full item record
This study presents three methods using nonlinear mixed-effects models to evaluate hospitals using composite measures based on hospital performance measures that are defined as proportion measures. A proportion measure is a measure where the numerator is a subset of the denominator and the ratio can be expressed as a rate or proportion as opposed to a continuous measure such as time from entering the emergency room until admission to the hospital. The first approach utilizes a composite performance measure based on yearly as well as longitudinal data and uses a hierarchical mixed-effects logistic regression model. The parameters are estimated by both Empirical and Full Bayes methods. Measure of agreement between the estimates of the parameters obtained by the Full Bayes and Empirical Bayes models uses the concordance correlation coefficient. The second method incorporates a hierarchical mixed-effects Poisson regression model to each type of data, i.e., hospital performance data and hospital patient safety data. Estimates obtained from the model utilizing the hospital performance data will be compared. For each data type, estimates are obtained by using an Empirical Bayes method and a Full Bayesian model and are compared with each other. The final method uses a bivariate mixed-effects hierarchical model that incorporates the correlation between hospital performance measure data and hospital safety data. An overall measure of quality is constructed based on the bivariate latent variable to form a quality of care measure. In conclusion, each of these methods is a reasonable way to measure hospital quality of care by utilizing aggregated publically available data. The arena of hospital performance measures is constantly changing and some measures used in this study are no longer active measures that are being collected. There are also new measures being developed that represent other types of measures, such as outcomes measures, along with measures being developed in other areas of hospitals care. Although the analysis presented here represents a snapshot of the measures currently available, the methods developed in this study are flexible to incorporate new measures and measure sets as they become available.
Classify Hospital Quality
Latent Variable Model, Poisson Regression
Longitudinal Logistic Regression