Utility of prescription-based comorbidity indices for predicting mortality among Australian men with prostate cancer.

ABSTRACT

BACKGROUND: Drug prescription registries has become an alternative data source to hospital admission databases for measuring comorbidities. However, the predictive validity of prescription-based comorbidity measures varies based on the population under investigation and outcome of interest. We aimed to determine which prescription-based index of comorbidity has most utility in Australian men with prostate cancer. METHODS: We studied 25,414 South Australian men diagnosed with prostate cancer between 2003 and 2019 from state-wide administrative linked datasets. The Rx-Risk index, Chronic Disease Score (CDS), Drug Comorbidity Index (DCI) and Pharmaceutical Prescribing Profile (P3) with one year lookback period from prostate cancer diagnosis were evaluated. The predictive ability of each index to determine all-cause deaths within two and five years of prostate cancer diagnosis was compared using the c-statistic from flexible parametric survival models, adjusting for age, socioeconomic status and year of prostate cancer diagnosis. RESULTS: The Rx-Risk index performed better in predicting two-year (c-statistic = 0.818) and five-year (c-statistic = 0.784) all-cause mortality than P3, CDS and DCI. Including comorbidity measures as continuous scores resulted in a better performance than including them as categories. Grouping scores into four categories (≤0, >0 - ≤1, >1 - ≤2, and >2) resulted in better performance and calibration than using fewer categories. CONCLUSION: Rx-Risk was validated in Australia and reflects Australian prescribing patterns. It showed better predictive performance for mortality in our study, with a modest improvement over P3, CDS and DCI. For research with prostate cancer populations, we recommend the use of drug-based comorbidity indices that have been validated in a similar population.