Burden of Comorbid Medical Conditions and Quality of Diabetes Care

Study participants were enrolled in a Medicare managed care health insurance plan providing coverage in Alabama, Florida, and North Carolina. We included patients who were aged at least 65 years, enrolled with the health plan continuously from 1 January through 31 December 2003, and alive on 31 December 2003. Eligible patients had one of the following: 1) at least one pharmacy claim for diabetes medication or 2) an ICD-9 code for diabetes from an in-person visit with a clinician in the outpatient or inpatient setting. We also merged data from the Center for Medicare and Medicaid Services with the health plan’s claims and pharmacy data, allowing us to ascertain race/ethnicity. Only patients identified as African American or European American were included in our analysis. The Western Institutional Review Board approved this study. All analyses were conducted using SAS (version 9.1; SAS Institute, Cary, NC).

Comorbid conditions were identified using ICD-9 codes from encounters in the health insurance claims data. We used Romano’s modification of the CCI (12). The CCI includes 17 specific illnesses and weights them according to severity (1, 2, 3, or 6). The CCI for each patient is calculated by summing the weighted number of points for each diagnosis carried by the patient. CCI (range 0–15) reflects mortality risk at 1 and 10 years, with low scores representing lowest risk (13).

For our main analyses, we used four separate commonly recommended performance measures for diabetes (14). We used data collected as part of the managed care organization’s reporting activities to the National Committee on Quality Assurance Health Plan Employer Data and Information Set. Based on claims data, we determined for the calendar year 2003 whether patients had A1C testing, dilated eye exam, and urinary microalbumin testing in the past year and lipid testing in the past 2 years. Because screening for microalbumin is not clinically indicated once a patient has renal disease, we excluded the 780 participants with known renal disease from the urinary microalbumin analyses.

Initially, we examined performance for each of these measures at every level of the CCI: 0–1, 2, 3, 4, 5, 6, 7, 8, 9, 10, and 11+. Because only eight patients had a CCI of 0, we constructed a single category for patients with 0 and 1 comorbidity. Likewise, because of small numbers, we combined all patients with comorbidity indexes >11 into a single category. Because the data were not normally distributed, we also classified CCI by quartile, which represented a clinically reasonable approach. For multivariable modeling, the main independent variable was quartile of CCI.

Using the proc logistic command in SAS, we developed separate multivariable logistic regression models for each of the four dependent variables, adjusting for age, sex, race/ethnicity, and diabetes medications. Diabetes medications were defined by presence or absence of pharmacy claims for either oral hypoglycemic medications or insulin. These covariates represented important confounders of the relationship between CCI and the outcomes. All patients were included in the models testing the associations between CCI and A1C and lipid testing. As in the unadjusted analyses, for the microalbumin screening model we excluded the 780 individuals with known renal disease, for whom microalbuminuria screening is no longer indicated. All patients were included in the dilated eye exam model, because the presence or absence of diabetic retinopathy does not change the recommendation for at least annual dilated eye exams. However, to account for differential treatment in patients with recognized diabetic retinopathy, we included a variable that reflected presence or absence of diabetic retinopathy in the dilated eye exam model.

Next, we calculated unadjusted and adjusted predicted probabilities for each model. More specifically, for each of the four models, we calculated the average probability of the outcome having a positive response within each comorbidity quartile (15). For the adjusted predicted probabilities, we entered the mean values of all remaining covariates from each comorbidity quartile. Based on the individual variance estimates for each model parameter, we calculated an overall SD for the predicted logit of the outcome. Finally, we transformed each predicted logit along with the upper and lower bounds for the 95% CIs into probabilities. To test for bias from the possible misidentification of patients with diabetes based on ICD-9 codes, we repeated all analyses on the subset of patients taking hypoglycemic medications (n = 2,472) (16).

For the 6,032 patients included in the study, the average age was 74.5 years; 56.8% were female, and 39.8% were African American (Table 1). The subset of patients without renal failure, which made up the study population for the urinary microalbumin analyses (n = 5,252), was similar demographically to the overall cohort (data not shown). CCI ranged from 0 to 15, with a mean ± SD of 3.2 ± 2.2. The mean age and the proportion of patients with diabetes complications increased as CCI increased. Fig. 1 shows the unadjusted percentage of patients who received each diabetes service by category of CCI. A1C and lipid testing were somewhat lower in patients with higher CCI. Rates of dilated eye exam and urinary microalbumin were similar across CCI.

In the multivariable analysis, results were similar after adjustment for age, sex, race/ethnicity, and receipt of diabetes medications. The models showed little difference in predicted probability of testing as comorbid disease burden increased (Table 2). Repeating all analyses for the 2,472 patients with pharmacy claims for oral diabetes medications or insulin provided results of similar direction and magnitude (data not shown).

Contrary to our expectations, we found few differences in routine testing by comorbidity burden for this elderly population with diabetes enrolled in a Medicare managed care health plan. While rates of A1C and lipid testing were slightly lower in patients with higher comorbidity than in those with the least comorbidity, rates of dilated eye exam and urinary microalbumin testing did not differ by comorbid burden. These results suggest that physicians are not adjusting the provision of routine diabetes services for patients with varying levels of comorbidity, although our study does not inform the appropriateness of this observation.

In acute settings, tight glycemic control has been shown to improve outcomes (17,18). But in chronic care management, tight glycemic or lipid control yields clinically meaningful benefits only after several years of intervention (9–11). Distinct from lipid and glucose control, detection of proliferative diabetic retinopathy and treatment with laser photocoagulation have been shown to significantly reduce severe visual loss over as short a time period as 2 years (19). Therefore, dilated eye exams should be performed in all patients with diabetes; yet only half of patients in this managed care organization, regardless of comorbid burden, received dilated eye exams.

Patients with multiple comorbid conditions have often been excluded from the evidence-generating randomized controlled trials that form the basis for performance measures (4,20–23). Randomized clinical trials, evidence-based guidelines, and quality measures are remarkably silent on the inevitable trade-off decisions that must be made during the routine clinical care of medically complex, older patients. In the absence of evidence regarding the benefit or harm of applying these guidelines to patients with multiple comorbid conditions, we are unable to definitively determine whether our finding of almost constant testing across comorbidity quartiles represents appropriate or inappropriate care.

In the P4P era, physicians may feel pressure to adopt a “one size fits all” approach and order tests to improve their performance on quality indicators developed from trials that excluded patients with multiple comorbid conditions (24). In patients with limited life expectancy, the appropriate clinical course may be to decrease testing associated with delayed benefit and to focus on interventions with high short-term potential for improving quality of life. Until evidence is available that confirms current performance measures as appropriate in patients with multiple comorbidities or advanced age, P4P programs and managed care organizations may wish to limit accountability to only those populations for whom strong evidence of benefit exists.

Currently, there is only limited evidence that the health care–quality industry acknowledges life expectancy. For example, the Veterans Health Administration/Department of Defense Clinical Practice Guideline for the Management of Diabetes includes the recommendation to consider life expectancy in setting treatment targets. Similarly, the American Geriatric Society and the California Healthcare Foundation have endorsed guidelines that recognize the complexity of managing older patients with diabetes and multiple comorbid conditions (8). However, these acknowledgments have not been translated into measures designed to assess quality of care in these individuals. Indeed, the National Committee for Quality Assurance applies an upper age limit (75 years) for its Health Plan Employer Data and Information Set measures, effectively excluding older individuals from quality assessments (14). Rather than exclude such patients, validated measures that assess their quality of care and directly incorporate life expectancy are needed.

With increasing implementation of performance-based reimbursement, the intersection of multiple comorbidity and quality measurement has become a high-profile topic. Two recent articles, one by Boyd et al. (24) and the other by Tinetti et al. (4), used hypothetical patients to consider possible concerns for applying current guidelines to patients with multiple comorbid conditions. Both articles concluded that guideline-concordant care may result in great expense and marginal benefits, and the authors cautioned that enforcing quality measurement for older complex patients may result in unintended harm unless future quality measures consider chronic conditions, life expectancy, and patient preferences.

Higashi et al. (25) and Min et al. (26) examined the association of comorbidity with quality of care in two articles based on three adult cohorts, the Community Quality Index Study, the Assessing Care of Vulnerable Elders Study, and the Veterans Health Administration quality of care project. They examined 236–439 quality indicators covering 22–30 clinical areas and found, “contrary to [their] expectations,” that as the number of comorbid conditions increased, adherence to quality measures increased. Both articles concluded that multiple comorbid conditions result in better quality of care.

In an editorial accompanying the study by Min et al., Ritchie (27) argues that the study raises as many questions as it answers. Although the Assessing Care of Vulnerable Elders Study attempted to account for patient preferences by including desire for hospitalization or surgery, it did not balance the added patient burden of guideline adherence with goals of reduced patient/caregiver burden and symptom control. In fact, current quality measures focus on reducing mortality and mainly ignore functional status and quality of life, which may be more important in older populations. Ritchie also notes that the patient and provider may place different values on symptom relief versus control of asymptomatic risk factors for disease progression. Consistent with our main thesis, Ritchie and others conclude that more research must be done to assist providers in making evidence-based decisions that reflect multiple competing clinical factors for patients with high comorbidity (27–29).

This study has important limitations. First, it is well documented that ICD-9 codes in claims data have variable sensitivity and specificity for the actual presence of disease (16,30). To increase the sensitivity of our algorithm for identifying patients with diabetes, we required only one ICD-9 code from a face-to-face physician encounter. Hebert et al. (30) found that a single diagnosis of diabetes in Medicare claims data from a face-to-face physician encounter had a sensitivity of 57.9% and a specificity of 96.9%. The ICD-9 code used to determine testing for urine protein is specific to urinary microalbumin; thus, we did not capture other methods of proteinuria screening such as 24 h urine collection, which may have contributed to the low observed testing rate. Using this ICD-9 code allowed us to be specific for microalbumin and to not overestimate testing by including routine urinalysis. The analysis of patients only on diabetes medications confirmed our findings in the larger group. Likewise, claims data may not adequately document patient comorbidity. However, in an analysis of older men with diabetes, there was a 38% increase in 1-year mortality for each point increase in the CCI score, suggesting that CCI score could serve as an important factor for tempering guideline adherence (31). We note that many patients did not participate in this health plan’s pharmacy benefit; thus, we lacked complete medication data. Our study population included patients of a Medicare managed care health organization, so we lacked data on other elderly populations. We were unable to examine treatment rates or patient preferences. Finally, unlike recent studies on the topic of the association between multiple comorbid conditions and quality measurement, we focused on quality measures associated with only one disease, diabetes, rather than multiple disease-associated quality measures (4,24–26).

Providers in this Medicare managed care health plan had similar rates of adherence to diabetes performance measures across all quartiles of comorbid illness burden. It is not clear whether equally aggressive care confers equal benefits to patients with varying comorbid illness burden. Empiric research should evaluate the optimal screening intervals for A1C, lipid, and urinary microalbumin testing in patients with high comorbid illness burden and decreased life expectancy, as well as possible unintended consequences of such testing. Finally, patient outcomes in the P4P era should be carefully monitored to assure that health outcomes and quality of life in complex patients are not compromised by pressures to perform well on “required” tests.

This work was supported by National Center on Minority Health and Health Disparities/National Institutes of Health (NIH) Grant P60MD00502-01. J.H.H. is also supported by a Minority Investigator Research Supplement from the National Heart, Lung, and Blood Institute/NIH (N01-HC-48047, modification no. 42).

We thank Nelda Wray, MD, and Catarina Kiefe, MD, PHD, for insightful comments and suggestions on this manuscript.