An integer programming model to limit hospital selection in studies with repeated sampling

Health Services Research, June, 1995 by Michael Shwartz, Ronald K. Klimberg, Melinda Karp, Lisa I. Iezzoni, Arlene S. Ash, Janelle Heineke, Susan M.C. Payne, Joseph D. Restuccia

In most cases, the percentages of patients in the different hospital size categories among the hospitals selected by the model are quite similar to the percentages among all major hospitals treating patients from areas in the septile. Further, despite the complexity of our sampling plan, the distribution of patients across hospital size categories in the final sample is reasonably similar to what one might expect from simple random sampling from areas in the septiles. This suggests that generalizability has not been seriously compromised by our method of hospital selection and record sampling, at least along the dimension of hospital volume.

Shortly after recognizing the value of the model in this study, a second application became apparent, as described in the next section.

STUDY II: VALIDATING SCREENS FOR QUALITY OF CARE USING ADMINISTRATIVE DATA

Background

Iezzoni et al. have developed the Complications Screening Program (CSP), a series of computerized algorithms that use discharge abstract data to screen hospitals for complications potentially resulting from substandard care (Iezzoni, Foley, Heeren, et al. 1992; Iezzoni, Daley, Heeren, et al. 1994). The CSP identifies outcomes or events that although they do not raise concern as individual occurrences, may indicate systematic quality of care problems when they occur in a hospital at rates higher than expected. Examples of the complications screens are postoperative pneumonia, aspiration pneumonia, postprocedural hemorrhage, wound infections, sepsis, medication incidents, "reopening" a surgical site following another major surgery, equipment failures, and perforation or laceration of a viscus.

Twenty-seven screens have been developed. Because the individual screens, which are based on up to a dozen ICD-9-CM codes, identify too few cases to allow meaningful statistical analysis at the hospital level, the concept of a "risk pool" was created. The following six risk pools, defined [TABULAR DATA FOR TABLE 1 OMITTED] by DRGs or ICD-9-CM procedure codes, are used: major surgery, minor and miscellaneous surgery, invasive cardiology and radiology procedures, endoscopy, medical patients, and all patients. Each screen applies to only certain of the risk pools. For example, the postoperative pneumonia screen is considered only for patients with invasive procedures, whereas medication incident screens pertain to all risk pools.

To identify hospitals where more in-depth review might be warranted, the observed rate of cases with complications is compared to the expected rate within each risk pool. Expected rates are determined from a multiple regression model that includes the following independent variables: age, sex, admission source, Major Diagnostic Category, and 13 chronic conditions. By considering the ratio of observed complication rates to expected rates (O/E), account is taken of patient characteristics that might result in higher risk of complications due to intrinsic, patient-related clinical attributes rather than faulty quality of care.