Journal of Clinical Epidemiology 54 (2001) 694–701 0895-4356/01/$ – see front matter © 2001 Elsevier Science Inc. All rights reserved. PII: S0895-4356(00)00367-X Risk adjustment for older hospitalized persons: A comparison of two methods of data collection for the Charlson index Carol van Doorn a , Sidney T. Bogardus a , Christianna S. Williams b , John Concato a,c , Virginia R. Towle a , Sharon K. Inouye a, * a Yale University School of Medicine, Department of Internal Medicine, 20 York Street, TMP 15, New Haven, CT 06504, USA b Department of Epidemiology and Public Health, 20 York Street, TMP 15, New Haven, CT 06504, USA c West Haven Veterans Affairs Medical Center (VAMC), Clinical Epidemiology Unit, 950 Campbell Avenue, West Haven, CT 06516, USA Received 15 October 1999; received in revised form 8 November 2000; accepted 10 November 2000 Abstract To compare Charlson indices based on chart data and ICD-9 data for agreement overall and on rating specific comorbid conditions, and to compare mortality risks associated with these indices. Prospective cohort study. Six general medicine wards at Yale-New Haven Hospital. 524 consecutive patients who had no clinical evidence of delirium at enrollment, admitted between November 6, 1989 and July 31, 1991, aged 70 years or older. Death within 1 year of the index hospital admission date. Scores using the chart-based data were signifi- cantly higher than those using ICD-9 data. About half of the individual conditions showed fair-to-good agreement between the two scores, whereas the other half showed poor agreement. A comparison of mortality prediction indicated that the weightings assigned to in- dividual comorbidities differed substantially from those used in Charlson’s original index. While mortality prediction of each individual index was comparable, the ICD-9 and chart indices contributed independently to mortality prediction in the presence of the other. Low agreement between Charlson scores based on the two methods of data collection and their cumulative contribution to mortality predic- tion suggest that these indices may include different information. Our results suggest that the original Charlson index may not provide op- timal risk adjustment for elderly general medicine samples. We suggest development of an empirically–derived index of comorbid condi- tions and weights may be warranted for older general medical patients. © 2001 Elsevier Science Inc. All rights reserved. Keywords: Comorbidity index; Charlson; ICD-9; Mortality; Risk adjustment; Aged 1. Introduction Comorbidity, defined as the total burden of illnesses un- related to the patient’s principal diagnosis [1], holds impor- tant implications for clinical outcomes, including mortality, surgical results, complication rates, functional status, and length of hospital stay, as well as economic outcomes, includ- ing resource utilization, discharge destination, and intensity of treatments [1,2]. Rating scales for comorbidity provide an essential strategy of risk adjustment, which is critical for prognostication in medical care, clinical research, and health policy and organizational analysis (e.g., quality assessment in managed care organizations, institutional report cards) [3,4]. The Charlson Comorbidity Index [3] is one of the most widely used comorbidity rating systems. This weighted in- dex, developed and validated to predict 1-year mortality, takes into account the number and seriousness of comorbid diseases rated by medical record review. Many investiga- tors, however, consider the reliance on chart-based data as a potential limitation given the current drive for rapid risk as- sessment in large populations, such as needed by hospitals, insurers and managed care organizations. Administrative data provide large, population-based samples that are appropriate for studying real-world differences in patient outcomes [5]. Thus, the Charlson index has been adapted [6,7] for use with ICD-9-CM (International Classification of Diseases, 9th revi- sion, Clinical Modification) administrative data. Two previous studies have compared the chart-based and ICD-9-based Charlson indices in older populations [8,9]. Both studies found low agreement between the chart and administrative data. Malenka et al. [8] demonstrated that the best mortality prediction was obtained by using both indices together. Newschaffer et al. [9] found that the chart-based index was a better predictor of mortality; however, the dif- ference between the two indices was small and the overall predictive power of each was nearly equal. However, these studies did not specifically examine a general medical pop- * Corresponding author. Tel.: 203-688-7302; fax: 203-688-4209. E-mail address: inouye@ynhh.com (S.K. Inouye)