Abstract
Objective: We aimed to develop and validate a pharmacy-based instrument to measure comorbidity among cancer patients.
Methods: Patients diagnosed with colorectal, breast, gynecologic, stomach/liver, or renal/bladder cancers were identified from the New Zealand Cancer Registry between July 2006 and June 2008 for a development cohort (n=14096) and from July 2008 to December 2009 for a validation cohort (n=11014). Nineteen conditions were identified using community pharmaceutical data collected in the year before cancer diagnosis; 10 conditions were validated against hospital record data. A pharmacy-based comorbidity index (PBCI) was developed with each identified condition weighted according to their log hazard ratios from age-adjusted and stage-adjusted Cox regression models with noncancer death as the outcome. For each individual the weights were summed to give a score. Predictive abilities of PBCI were compared with the Charlson and C3 (hospitalization-based) comorbidity indices.
Results: Kappa coefficients for conditions identified in notes review compared with pharmaceutical data ranged from 0.83 (diabetes) to 0.26 (anxiety/depression). Correlation coefficients with the Charlson ranged from 0.37 to 0.45 across cancers. All comorbidity indices were significant predictors of mortality, and differences between models were small. The PBCI generally performed as well or better than the Charlson index for predicting noncancer death within all cancer sites and slightly outperformed other indices in predicting noncancer mortality for breast cancer.
Conclusions: The PBCI provides a valid alternative to measuring comorbidity in cancer patients. Researchers can use either hospitalization-based or pharmacy-based comorbidity measures for risk adjustment purposes.