Return to search

Comorbidity indicators: Validation and application

The objectives of this study were to assess the construct validity and predictive validity of a previously published comorbidity classification scheme designed for use with administrative data. The scheme groups non-primary discharge diagnoses into a set of thirty comorbidity indicators, which may be used to describe and compare populations with respect to burden of comorbid illness. The scheme was developed on a large population of hospitalized patients in California in 1992 (training population) and the predictive effect of the indicators estimated with respect to the outcomes length of stay, hospital charges, and in-hospital death. The current study drew data from the Massachusetts Hospital Case Mix Data Base of 1992 (testing population). The effect of the comorbidity indicators on each outcome was estimated by fitting ordinary least squares regression (OLSR) models of length of stay and hospital charges, as well as logistic regression models of in-hospital mortality, to the testing population. The estimated effect of the comorbidity indicators on each outcome, adjusted for demographics and characteristics of index hospitalization, was compared between the training and testing populations. The characteristics of the testing population were largely similar to those of the training population. The relationship between burden of comorbid illness (as measured by the number of comorbidity indicators per patient) and the outcomes was comparable in the two populations. The estimated adjusted effect of the comorbidity indicators and the predictive ability of the OLSR models were comparable in the training and testing population with respect to the outcomes length of stay and charges. The estimated adjusted effect of the comorbidity indicators on in-hospital death was not comparable in the two populations. The results support construct validity and predictive validity of the comorbidity classification in Massachusetts discharge data in 1992. Other aspects of baseline risk must be accounted for separately. The estimated adjusted effect of the indicators in the training population on the outcomes length of stay and charges, but not in-hospital death, is generalizable to Massachusetts' discharge data and may be further generalizable. Practical application of the comorbidity indicators for comorbidity adjustment in epidemiological research should be further explored.

Identiferoai:union.ndltd.org:UMASS/oai:scholarworks.umass.edu:dissertations-3664
Date01 January 2002
CreatorsHeimisdottir, Maria
PublisherScholarWorks@UMass Amherst
Source SetsUniversity of Massachusetts, Amherst
LanguageEnglish
Detected LanguageEnglish
Typetext
SourceDoctoral Dissertations Available from Proquest

Page generated in 0.0092 seconds