Abstract
Background: Comorbidity has an important impact on cancer outcomes, but the optimal approach to measuring comorbidity at a population level has not been established.
Aims:
• To review evidence relating to the importance of comorbidity in relation to cancer care and outcomes; and to examine previously developed approaches to measuring comorbidity in this context.
• To assess the usefulness of routinely collected comorbidity data in New Zealand,
• To develop and validate optimised measures of comorbidity using these data for patients with cancer,
• To develop, and compare more simplified measures against the optimised measures.
Methods: Studies describing methods to measure comorbidity in epidemiological studies related to cancer were identified and reviewed. For this study, development and validation cohorts included patients diagnosed with colorectal, breast, gynaecological, upper gastrointestinal, or urological cancers identified from the national Cancer Registry between July 2006 and June 2008 for the development cohort (n=14096) and July 2008 to Dec 2009 for the validation cohort (n=11014). Data on comorbid conditions identified in administrative hospitalisation and pharmaceutical data were compared with data from manual clinical notes review for a subset of patients. Fifty conditions using administrative hospitalisation and twenty conditions using pharmaceutical data were identified prior to cancer diagnosis. Three sets of indices were developed 1) site-specific indices using hospitalisation data (‘C3’ indices), 2) a single all-cancer index using pharmaceutical data (PBCI) and 3) three simplified versions of the hospitalisation indices; an all-cancer version and two versions including a subset of conditions (SI1, 2 and 3 respectively). Conditions were weighted according to their log hazard ratios from age and stage adjusted Cox regression models of non-cancer death; and indices were calculated by summing these weights. Performance of these indices was compared with the Charlson index, a combination of the C3 and pharmacy-based indices, and with each other.
Results: The review of previously developed comorbidity measures for cancer populations identified 21 separate approaches, with none identified as gold standard. Administrative data were found to be adequate for measuring comorbidity in cancer populations. Comorbidity was associated with poorer survival but the impact varied by condition and across cancer site. No single index clearly outperformed all others for all sites. The best performance overall was achieved with the combined hospitalisation and pharmaceutical indices, particularly for all sites combined, colorectal and upper GI cancers. Generally hospitalisation indices outperformed the pharmaceutical-based index, but the converse was true for non-cancer death for breast and gynaecological cancers. Site-specific weights did not add appreciably to the validity of the indices. Among the simpler indices, the SI2, SI3 and PBCI approaches tended to outperform the Charlson index for all sites combined, although there was little difference between these indices in some sites.
Conclusion: Measuring comorbidity in cancer populations is important, and the C3-based and PBCI indices provide a useful and valid cancer-specific approach based on administrative data. Site-specific indices were not found to be necessary. Future work includes validating these indices in other cancer populations, preferably outside New Zealand; and potentially working to extend the use of these indices beyond cancer.