Original articlesDevelopment of a comorbidity index using physician claims data
Introduction
Comorbidities are coexisting medical conditions that are distinct from the principal diagnosis or the primary illness for which the patient seeks health care services 1, 2, 3. Comorbidities can be either chronic diseases or acute illnesses, and can increase a patient's total burden of illness 3, 4. Compared to patients who do not have these conditions, patients with comorbid illnesses may be at greater risk of complications or death, less able to tolerate particular medical procedures, and less responsive to therapy 1, 3. Furthermore, physicians may factor the presence of particular comorbidities into decisions concerning the most appropriate medical treatments for patients. For example, studies involving breast and prostate cancer patients have found that patients with more comorbidities receive less aggressive treatment for their tumors, even after controlling for patient age and cancer stage 5, 6. The reluctance to pursue a course of aggressive therapy, particularly surgery, in patients with a high burden of comorbid conditions likely stems from their substantially increased risk of complications and death [7].
The complexity of comorbidity data and its potential for creating unwieldy analyses has led to the development of summary comorbidity measures such as the Charlson index 8, 9. Based on medical record review, Charlson and colleagues developed a weighted index measure of comorbidity that was shown to predict 1-year all-cause mortality in a cohort of 559 hospitalized medical service patients and 10-year non-breast cancer mortality in cohort of 685 breast cancer patients. The index is comprised of 19 conditions, each of which is assigned a weight according to its potential for influencing mortality. Charlson and colleagues used as weights the adjusted hazard ratios (referred to in their article as “relative risks”) from a stepwise proportional hazards model, rounded to the nearest integer. The patient's comorbidity index, the sum of the weighted comorbidities, takes into account both the number and seriousness of the conditions. A higher score on the Charlson index indicates a greater burden of comorbid disease.
Deyo et al.[10] adapted the Charlson index for use with the ICD-9-CM diagnostic and procedure codes available in administrative datasets, and demonstrated the utility of this adapted measure in predicting risk of poor outcomes for patients following lumbar spine surgery. The Deyo adaptation involves searching a patient's hospital claims data for the presence of certain ICD-9-CM diagnosis and procedure codes corresponding to the Charlson comorbid conditions. An important limitation of the Deyo/Charlson comorbidity measure, however, is that it was developed and validated on the basis of inpatient hospital care. To date, studies that have used the measure have included only data from the inpatient setting 11, 12, 13. It is possible that important comorbidities recorded on outpatient claims are missed in analyses when only inpatient care is considered 11, 14, particularly because an estimated 80% of Medicare beneficiaries are not hospitalized in a given year [15]. Because of the ongoing shift toward delivery of health care services exclusively in outpatient settings, consideration of comorbidities recorded in outpatient claims is of particular importance when assessing treatment patterns in more recent years.
This article describes the development of a comorbidity measure using Charlson's conditions, the diagnostic and procedure data contained in Medicare physician (Part B) claims, and a new methodologic approach. Although other validated comorbidity measures such as the Kaplan-Feinstein index [16] and Index of Coexistent Disease [5] are available, we chose to build upon the Charlson index because of its wide use with claims data. The new measure is validated by assessing whether comorbidity information derived from physician claims significantly contributes to models of short-term noncancer mortality in national cohorts of elderly breast and prostate cancer patients. Prior studies 10, 17, 18, 19, 20, 21 also have demonstrated the utility of the Charlson comorbidity index, developed as a prospective means of evaluating risk of death, in predicting such nonmortality outcomes as complications, length of stay, and charges. In this study, we also evaluate whether inclusion of physician claims data significantly contributes to models of type of cancer treatment received in breast and prostate cancer patients. Finally, we propose a method of index construction that avoids arbitrary thresholds for condition exclusion and uses a scale more closely related to the model that produces the condition weights.
Section snippets
Data sources
Data for this study were derived from two sources: 1) the National Cancer Institute's Surveillance, Epidemiology, and End Results (SEER) program, and 2) Medicare claims.
Results
Table 1 displays condition prevalences, estimated coefficients, and hazard ratios for comorbidities appearing in inpatient hospital or only in physician claims for the prostate cancer cohort. To facilitate comparison to Charlson's original weights (see Table 1, Charlson et al. article), this table also includes the study-derived hazard ratios for these comorbid conditions rounded to the nearest integer, according to the method of Charlson et al.[9]. Diabetes, chronic pulmonary disease, and
Discussion
The Charlson comorbidity index, adapted for use with administrative datasets 10, 20, has been shown to predict a variety of patient outcomes, including mortality, postoperative complications, length of stay, and hospital charges 10, 17, 18, 19, 20, 21. The Charlson index, however, was developed and validated on the basis of inpatient hospital care [9]. We developed a new index measure of comorbidity derived from physician (Medicare Part B) claims data in two separate cohorts of elderly cancer
Acknowledgements
The authors are grateful to Nicki Schussler of Information Management Services, Inc., Silver Spring, MD, for expert assistance with dataset construction and to Rachel Ballard-Barbash, M.D., M.P.H., for careful review of and helpful comments on the manuscript.
References (42)
- et al.
A new method of classifying prognostic comorbidity in longitudinal studiesdevelopment 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.
The importance of classifying initial co-morbidity in evaluating the outcome of diabetes mellitus
J Chron Dis
(1974) - et al.
Practical considerations on the use of the Charlson comorbidity index with administrative data bases
J Clin Epidemiol
(1996) - et al.
Searching for an improved clinical comorbidity index for use with ICD-9-CM administrative data
J Clin Epidemiol
(1996) - et al.
Adapting a clinical comorbidity Index for use with ICD-9-CM administrative datadiffering perspectives
J Clin Epidemiol
(1993) - et al.
Comorbidity and breast cancer survivala comparison between black and white women
Ann Epidemiol
(1996) - et al.
Evaluation of two competing methods for calculating Charlson's comorbidity index when analyzing short-term mortality using administrative data
J Clin Epidemiol
(1997) - et al.
A comparison of administrative versus clinical datacoronary artery bypass surgery as an example
J Clin Epidemiol
(1994) - et al.
Accuracy of administrative data for assessing outcomes after knee replacement surgery
J Clin Epidemiol
(1997)