Comparison of logistic regression and Mantel -Haenszel statistical procedures to predict length of stay of four diagnosis -related groups

The purpose of this research was to test the appropriateness of logistic regression and Mantel-Haenszel statistical procedures in analyzing health care patient data to examine the relationship between the dependent variable, dichotomized length-of-stay (LOS) and factors that influence the behavior of the LOS. Hospital LOS is a traditional cost driver in health care.;Logistic regression defines the relationship between a dichotomously coded dependent variable and one or more independent variables. This statistical technique frequently is used to assess the influence of independent variables on a dependent variable by evaluating odds-ratios. The odds-ratio demonstrates an index of the likelihood of an expected outcome of the two alternatives given values of the independent variables.;The Mantel-Haenszel statistical procedure examines the relationship between two dichotomous variables using information from 2 x 2 contingency tables. Developed for use in epidemiological research, Mantel-Haenszel was later introduced in the detection of bias in educational research. The Mantel-Haenszel Differential Item Functioning (DIF) is one of the most popular procedures for identifying bias in dichotomous variables.;In this study, the author examined the appropriateness of these two techniques in analyzing hospital LOS to identify factors influencing the length of hospitalization for two sets of Diagnostic Related Groups, pneumonia and myocardio infarction. The results demonstrated that both techniques effectively identified factors that influenced LOS.;The Mantel-Haenszel procedure was sensitive to one type of differential item functioning and was not designed to detect DIF that has a nonuniform effect across trait levels. The MH procedure evaluates one item at a time requiring multiple analyses which could increase the likelihood of Type 1 errors. While the MH results may be easier to explain to a nonstatistician, the limitations of MH restrict the use of MH DIF in analyzing healthcare LOS data.;Logistic regression (LR) allows simultaneous analysis of independent variables. When interpreting results of the LR procedure, inferences can be drawn regarding which variables should be included in the prediction model and which can be excluded from further analyses. Consequentially, the logistic regression may be better suited for this type of healthcare data analysis.