Common comorbidity scales were similar in their ability to predict health care costs and mortality

https://doi.org/10.1016/j.jclinepi.2004.03.002Get rights and content

Abstract

Objective

To compare the ability of commonly used measures of medical comorbidity (ambulatory care groups [ACGs], Charlson comorbidity index, chronic disease score, number of prescribed medications, and number of chronic diseases) to predict mortality and health care costs over 1 year.

Study Design and Setting

A prospective cohort study of community-dwelling older adults (n = 3,496) attending a large primary care practice.

Results

For predicting health care charges, the number of medications had the highest predictive validity (R2 = 13.6%) after adjusting for demographics. ACGs (R2 = 16.4%) and the number of medications (15.0%) had the highest predictive validity for predicting ambulatory visits. ACGs and the Charlson comorbidity index (area under the receiver operator characteristic [ROC] curve = 0.695–0.767) performed better than medication-based measures (area under the ROC curve = 0.662–0.679) for predicting mortality. There is relatively little difference, however, in the predictive validity across these scales.

Conclusion

In an outpatient setting, a simple count of medications may be the most efficient comorbidity measure for predicting utilization and health-care charges over the ensuing year. In contrast, diagnosis-based measures have greater predictive validity for 1-year mortality. Current comorbidity measures, however, have only poor to moderate predictive validity for costs or mortality over 1 year.

Introduction

Researchers and others often seek to compare the burden of illness or outcomes of care across large populations. For example, to compare the health care costs or mortality due to a specific condition across two or more groups of patients, researchers often need to test and/or adjust for possible differences in comorbidity between groups. Most current measures of comorbidity rely on an accounting of the subjects' other diagnosed medical conditions and, in some commonly used approaches, an accounting of the severity of those conditions. Because of the large number of subjects and vagaries associated with medical records, it is often impractical to assess the burden of illness through manual review of individual written medical records. One solution to this problem has been to use diagnostic codes (e.g., the International Classification of Disease ICD codes) contained in administrative or clinical databases to describe the burden of illness among large cohorts.

Comorbidity refers to the total burden of illnesses across multiple potential conditions unrelated to the patient's principal or target diagnosis [1]. Severity of illness, by contrast, typically refers to a single target diagnosis. Different approaches have been used to define comorbidity, depending on the outcome measure, the clinical setting, and source of data. For inpatient settings, manual chart review and laboratory tests have been used to calculate comorbidity or severity scores [2], [3], [4], [5], [6], [7], but community-based survey measures have often relied on self-reported symptoms, diagnoses, quality of life, perceived health, or functional status to assess comorbidity and severity of illness [7], [8]. Other strategies rely on administrative databases to calculate comorbidity using diagnostic codes [9], [10], [11], [12], [13], [14], [15], [16] or prescriptions [16], [17]. Measures can be simple (e.g., count of symptoms or diseases) or quite complex (e.g., using different weighting schemes). The choice and validity of the measure depends on the outcome of interest (e.g., mortality, utilization, costs, and treatment- specific variables), the efficiency in calculating the measure, and the availability and reliability of the data used to create the measure.

Although the predictive validity of individual comorbidity measures has been examined in detail [1], [2], [3], [4], [5], [6], [7], [8], [9], [10], [11], [12], [13], [14], [15], [16], [17], [18], [19], [20], [21], [22], [23], [24], [25], few published studies compare different measures within the same target population. Those studies that have compared predictive validity across measures tend to focus on comparisons in limited disease-specific populations (e.g., patients with cancer) [1], [6], [8], [21], [22]. Others have compared different variants of the same index [11], [12], [13], [14] or compared scores for data derived from written medical records as opposed to administrative databases [10], [20]. There are few published studies comparing the diagnosis-based with medication-based measures [18], [24], [25], [26], and only three used the same population for these comparisons [18], [25], [26]. Furthermore, much current research applies the available comorbidity measures, or variations in the scoring of these measures, to patient populations in whom the measure has not been tested. For example, a research team might use a comorbidity measure developed to predict inpatient mortality to control for the impact of comorbidity on health care utilization among an outpatient population.

Our purpose in the present study was to compare the predictive validity of commonly used measures of medical comorbidity among a large cohort of vulnerable older adults cared for in a single primary care practice that is served by a state-of-the-art electronic medical record system. Because we are interested in measures that can be used with electronic medical records or administrative databases, we limited the comparison to those measures that can be obtained by using data from such sources. We were particularly interested in determining if more complex measures predict health care charges and mortality better than simpler measures. We also examined the impact of modifications in the scoring on the performance of these scales.

Section snippets

Overview of common comorbidity measures

Ambulatory care groups (ACGs) were designed to provide a case-mix adjustment measure for the ambulatory care setting [15], [16]. ACGs are based on the premise that a measure of a population's illness burden can help explain variation in health care resource consumption. Using primary and secondary diagnoses (ICD-9 codes) from ambulatory visits over a defined time period, patients are first classified into one of 34 ambulatory diagnostic groups (ADGs) by clustering similar conditions based on

Methods

Community-dwelling patients at least 60 years of age with a scheduled primary care appointment between July 15, 1999, and August 31, 2001, were eligible for the study. During this time period, 4,386 patients had a scheduled visit and 3,496 (80%) completed a screen. Of the 880 people who were not seen, 499 (56%) did not show up for any of their scheduled appointments during the time period. Additionally, 111 (12%) were missed by the recruiter, 105 (12%) refused, and 175 (20%) were ineligible.

Results

Table 2 presents demographic, utilization, and comorbidity measures for the 3,496 patients. The average age was 68.9 years; 2,418 (69%) patients were women and 1,959 (56%) were African American. Approximately 75% of the patients had Medicare and 959 (27%) had Medicaid. Physicians identified 709 (20%) patients as smokers and 1,581 (45%) patients were obese. This group of patients averaged eight ambulatory visits per year and total health care charges of about $6,600 in the year following the

Discussion

We compared the predictive validity of several comorbidity measures that can be calculated using electronic medical records or administrative databases on four different outcomes. In predicting total charges and the number of ambulatory visits over 1 year, the comorbidity indices based on medications or outpatient diagnoses performed better than the Charlson index (which was developed to predict inpatient mortality) or the total number of chronic conditions. For predicting hospitalization and

Conclusion

In an outpatient setting, a simple count of prescribed medications may be the most efficient comorbidity measure for predicting utilization and health care charges over the ensuing year. The count of medications also compares well to other available measures in predicting hospitalization and mortality over the coming year. Diagnosis-based measures had the greatest predictive validity for 1-year mortality. Despite concerns about face validity, the less complex indices such as a simple count of

References (28)

  • S. Beddhu et al.

    A simple comorbidity scale predicts clinical outcomes and costs in dialysis patients

    Am J Med

    (2000)
  • S.M. Kieszak et al.

    A comparison of the Charlson comorbidity index derived from medical record data and administrative billing data

    J Clin Epidemiol

    (1999)
  • D.J. Malenka et al.

    Using administrative data to describe casemix: a comparison with the medical record

    J Clin Epidemiol

    (1994)
  • P.A. Rochon et al.

    Comorbid illness is associated with survival and length of hospital stay in patients with chronic disability: a prospective comparison of three comorbidity indices

    Med Care

    (1996)
  • Cited by (313)

    View all citing articles on Scopus
    View full text