Johnson ML (Houston HSR&D)
Abraham NS (Houston HSR&D)
Richardson P (Houston HSR&D)
Hartman C (Houston HSR&D)
Tran T (Houston MEDVAMC)
El-Serag HB (Houston HSR&D)
A growing literature has compared the performance of pharmacy-based measures of comorbidity to diagnostic-based measures for the prediction of mortality in hospitalized patients or to predict resource use and costs in outpatients. We sought to develop and test a pharmacy-based comorbidity measure for use as a case-mix adjuster of clinical outcomes in a large VA outpatient cohort.
We identified 724,270 veterans aged 18-99 years who were prescribed a traditional NSAID or a coxib in the VA during calendar year 2002. We updated and extended the Rx-Risk-V (Sloan 2003), a derivation of the pharmacy-based Chronic Disease Score applied to the VA pharmacy system, by adding 20 additional disease categories and mapping them to VA drug class codes. We further derived empirical weights for each disease group from a logistic model of one-year mortality, adjusted for age, race and sex. The method of Charlson and Romano was used to score the weights into one measure. The weighted score was compared to the Deyo comorbidity index and validated in a national cohort of 260,321 outpatients with chronic heart failure (CHF).
One-year mortality among the NSAID cohort was 1.6% (n=11,766). Using a baseline model of age, race and sex (c-index =.716), adding the pharmacy comorbidity score improved prediction of mortality by 9.2% (c-index = .782), compared to 6.8% by adding the Deyo measure (c-index = .765). Using both measures further raised the predictive performance (c-index = .792). Among the CHF cohort, 9.7% of patients (n=25,251) died within one-year. The performance of the baseline model controlling for age, race, and sex (c index = .620) was improved 11% by the pharmacy comorbidity score (c index = .689).
An easily constructed outpatient comorbidity measure was created from VA pharmacy records by mapping VA drug class codes to disease groups, and empirically weighted for use as a case-mix adjuster of clinical outcomes in an outpatient population. Further work will examine additional ways to improve upon diagnostic-based measures.
Case-mix adjustment can be improved by using pharmacy records to measure comorbidity.