OBJECTIVE: To develop two comorbidity indexes constructed from the Minimum Dataset (MDS) disease information predictive of 18-month mortality and 2 hospitalizations in a prevalence cohort of 1424 white, female nursing-home residents that could be used as part of a risk adjustment method for quality-of-care outcomes.
METHODS: A split-sample approach was taken for development and cross validation of the indexes. Multivariate logistic regression techniques were employed to identify the MDS diseases most likely to predict each outcome after controlling for age in the development sample. Weights equal to the parameter estimates were assigned to each disease retained in the multivariate model to create a single comorbidity index variable for each outcome. Using the validation samples, the predictive validity of the MDS-based comorbidity indexes was determined and compared to other measures of comorbidity (Charlson Index [ChI], Chronic Disease Score [CDS], count of diseases) as well as to Morris’ activities of daily living (ADL) index for each outcome.
RESULTS: The MDS-based mortality index included weighted variables for atherosclerotic heart disease (ASHD), dysrhythmias, congestive heart failure (CHF), respiratory disease and depression and had a c-statistic of 0.62 in the development sample, and 0.60 in both validation samples. Similar c-statistics were found for the comparison of comorbidity and functional status measures. The c-statistics for the MDS-based hospitalization index (with weighted variables for hemiplegia, glaucoma, peripheral vascular disease (PVD), CHF, diabetes and hypertension) were 0.66 and 0.59 for the development sample and validation sample respectively and were similar to those found for the other measures.
CONCLUSIONS: Presence or absence of diseases measured at baseline are weakly predictive of subsequent hospitalization or mortality in this cohort. Additional patient risk factors such as disease severity, or change in functional status should be investigated in future research aimed at defining risk adjustment methods in this population.
