A Refined Comorbidity Measurement Algorithm for Claims-Based Studies of Breast, Prostate, Colorectal, and Lung Cancer Patients
Introduction
The Charlson index (1) is a widely used measure of comorbidity, particularly among investigators conducting epidemiologic and outcomes research studies using administrative claims (2). The measure comprises 19 comorbid conditions, each assigned a weight according to its potential for influencing 1-year mortality. The index is the sum of the weighted comorbidities and accounts for the number and seriousness of the conditions. Originally developed to predict mortality among hospitalized patients, the index has been used to predict other patient outcomes, including treatment and costs. It has been adapted for use with administrative claims by mapping International Classification of Diseases, revision 9 (ICD-9) diagnostic and procedure codes to the conditions identified by Charlson as prognostically important 3, 4.
Since the publication of Charlson's original study, the proportion of medical care provided in the outpatient setting has increased considerably (5). Earlier studies that have used Medicare data generally have relied on codes obtained from inpatient claims to calculate a Charlson index (2). These analyses tend to classify a large proportion of cohort members as having no comorbid conditions simply because only a small proportion of the population is hospitalized in a given year.
An earlier study of comorbidity among Medicare beneficiaries with cancer used Medicare physician (Part B) claims and inpatient data to construct an adaptation of the Charlson index (6). This approach, hereafter referred to as the National Cancer Institute (NCI) index, involved calculation of two separate comorbidity indices from inpatient and physician claims. The analysis demonstrated improved prediction of noncancer mortality and treatment choice with the inclusion of physician claims for breast and prostate cancer patients. The authors obtained comorbid condition weights that differed from Charlson's (1) and showed that condition weights differed by cancer site. Although the NCI index (6) has been used in several claims-based analyses of cancer treatment and outcomes 7, 8, 9, 10, 11, 12, 13, 14, it may be somewhat cumbersome in practice because two separate comorbidity scores must be calculated and interpreted. Moreover, weights were derived only for breast and prostate cancer patients. Other studies 15, 16, 17 have also demonstrated the need for comorbidity indices tailored to the specific population of interest.
In this report, we describe a new analysis to improve the measurement of comorbidity in cancer patients using administrative claims databases. Our objectives are to evaluate (i) how combining comorbid conditions identified from Medicare inpatient or physician claims into a single comorbidity index, unlike the two separate indices used in the NCI index (6), compares with three other comorbidity indices and (ii) the need for comorbid condition weights specific to four different cancer sites. We chose to examine large cohorts of breast, prostate, colorectal, and lung cancer patients in our study because—excluding nonmelanoma skin cancer—these are the four most commonly diagnosed cancers in the United States.
Section snippets
Data Sources
We used data from the NCI's Surveillance, Epidemiology, and End Results (SEER) program, and Medicare Part A hospitalization and Part B physician, supplier, and outpatient facility claims. The SEER and Medicare data have been linked as previously described 18, 19. For this study, we excluded Part B clinical laboratory, diagnostic imaging, and durable medical equipment claims, because the reporting of diagnoses on these claims is not required, and the diagnostic codes from these providers are
Description of Study Cohorts
Table 2 displays summary statistics for each cohort by stage at diagnosis, including the number of noncancer deaths and crude mortality rates, which are not age-adjusted. Stage distributions for the breast, prostate, and colorectal cancer cohorts are approximately 80% early and 20% late but are nearly reversed for the lung cancer cohort. The percent of patients with late-stage disease who died from cancer during the study period approached 85% for the lung and colorectal cohorts but was
Discussion
Researchers working with observational data require methods to appropriately adjust their analyses for underlying differences in patients' health status. The Charlson Index (1) is widely used for this purpose in health services and outcomes research studies, including those involving cancer patients (2). Charlson's work has been extended through the development of comorbidity measures that incorporate physician (Part B) claims in addition to inpatient data from Medicare files (6). This
Acknowledgments
The authors thank Nicki Schussler and Neil Rolfes of Information Management Services, Inc., Silver Spring, MD, for expert programming assistance.
References (26)
- et al.
A new method of classifying prognostic comorbidity in longitudinal studies: development and validation
J Chron Dis
(1987) - et al.
Adapting a clinical comorbidity index for use with ICD-9-CM administrative databases
J Clin Epidemiol
(1992) - et al.
A comparison of administrative versus clinical data: coronary artery bypass surgery as an example
J Clin Epidemiol
(1994) - et al.
Development of a comorbidity index using physician claims data
J Clin Epidemiol
(2000) - et al.
Racial disparity in primary and adjuvant treatment for nonmetastatic prostate cancer: SEER-Medicare trends 1991 to 1999
Urology
(2004) - et al.
The development and validation of a comorbidity index for prostate cancer among Black men
J Clin Epidemiol
(2003) - et al.
Assessment of comorbidity using claims data: an overview
Med Care
(2002) - et al.
Trend data on medical encounters: tracking a moving target
Health Aff
(2001) - et al.
Delay of adjuvant chemotherapy initiation following breast cancer surgery among elderly women
Breast Cancer Res Treat
(2006) - et al.
Chemotherapy and cardiotoxicity in older breast cancer patients: a population-based study
J Clin Oncol
(2005)