Are Low-Income Elderly Patients at Risk for Poor Diabetes Care?

This study used Medicare claims data from Colorado during a 2-year period from January 1997 through December 1998. Both Medicare part A (institutional claims including those of hospitals and skilled nursing facilities) and part B (physician, laboratory, and other outpatient services) claims were obtained for the entire study population.

The study population consisted of Colorado Medicare fee-for-service (FFS) patients not enrolled in managed care Medicare plans, aged 65–75 years, with the diagnosis of diabetes from 1 January 1998 through 31 December 1998. Diabetes diagnosis was determined by having a single hospitalization with one of the ICD-9 codes of 250.XX, 357.2, 362.01, 362.02, or 366.41 or two outpatient visits at least 7 days apart with the same ICD-9 codes as above. Subjects meeting at least one of the following criteria were excluded to ensure maximum likelihood of capturing diabetes care claims: 1) greater than 1 month break in part B coverage from 1 January 1997 until 31 December 1998; 2) death before 31 December 1998; and 3) residency outside of the state of Colorado from 1 January 1998 through 31 December 1998. Patients were considered DE if they had at least 1 month of Medicaid buy-in by the State of Colorado from 1 January 1997 until 31 December 1998. Of the 9,653 patients who met the study criteria, 200 (2%) were excluded due to missing sociodemographic information, leaving 9,453 patients available for analysis.

Study subjects’ age and sex were determined using Medicare demographic data. It has previously been shown that patient race and ethnicity demographic data are often unreliable to determine Latino ethnicity; therefore, a Latino surname algorithm was used to identify patients of Latino ethnicity based on the Passel-Word Spanish Surname List used by the 1990 census (21). Otherwise, patient race and/or ethnicity (categorized as white [non-Latino], African-American, and other/unknown) were determined using the race/ethnicity information present in the Medicare demographic data.

To determine rural or urban residence, the patient’s zip code was linked to the Rural Urban Commuting Area Codes that are based on 1990 U.S. Census commuting data and urban area and urban place definitions. A Rural Urban Commuting Area Code ≥7 was considered rural (22).

The burden of comorbid disease conditions was determined for each patient using the Johns Hopkins Adjusted Clinical Groups risk-adjustment method (23). The Adjusted Clinical Groups system uses ICD-9 diagnosis codes for all inpatient and outpatient encounters and has 12 collapsed ambulatory diagnosis groups (CADGs) based on the likelihood of persistence or recurrence of the diagnosis, severity of the condition, and types of health care services required for management for each individual patient. Three CADGs were excluded, including pregnancy (due to age of study population), eye/dental, and prevention/administration (as these may be an indirect measure of the outcome).

Outpatient primary care (general practice, family practice, internal medicine, and geriatric) and endocrinology physician visits were determined using Medicare part B claims. Emergency department visits were determined using Medicare part A and B claims. Number of hospitalizations and average length of stay (ALOS) were determined using Medicare part A claims. Each patient had his/her ALOS calculated for all of his/her hospitalizations.

Three indicators of diabetes care were measured: 1) annual A1c; 2) biennial ophthalmologic examination; and 3) biennial lipid measurement. These diabetes process-of-care measures are consistent with those of the Diabetes Quality Improvement Project (24).

As the subject population was identified based on evidence of a diagnosis of diabetes during 1998 and two of the outcome variables (biennial eye and lipid evaluations) were determined to have occurred if claims were present during 1997 and 1998, it was possible that individuals newly identified during the study period did not have the entire 2-year measurement period to receive the recommended diabetes care. To control for the potential effect of this study subject selection criteria, we included a variable to account for the time when a subject first met inclusion criteria in each study model.

The proportion of patients receiving the individual diabetes care measures with the variable of interest, DE status, were compared using the χ², t test, and Wilcoxon two-sample test as appropriate. Logistic regression analysis was used to determine if DE status was independently associated with the likelihood of receiving diabetes care. A separate model was created with each of the three diabetes care indicators (annual A1c, biennial ophthalmologic examination, and biennial lipid measurement) as the outcome variable and DE status as the independent variable. Models were adjusted for potential confounding variables (age, sex, race/ethnicity, rural/urban residence, comorbidity, number of primary care visits, endocrinology visits, and emergency department visits). Interaction terms (including patient sociodemographic, comorbidity, and health care use) with DE status were examined.

Goodness of fit of the models was evaluated by the Akaike’s Information Criterion, −2 log likelihood, and the Hosmer-Lemeshow statistic (25). All analyses were performed using the SAS software version 8.02.

Sociodemographic, comorbidity, and health use characteristics are presented by DE status in Tables 1 and 2. The mean patient age was 71 ± 2.8 years. Of the 9,453 Medicare patients in the study, 2,096 (22%) were DE. DE patients were significantly more likely to be of younger age, female, minority race/ethnicity, and reside in a rural location compared with the non-DE population. Using the CADG risk adjustment, DE patients were more likely to have “psychosocial diagnoses” compared with non-DE (42 vs. 23%; P < 0.0001). DE patients were also more likely to have “acute minor and major,” “likely to recur,” “asthma,” “chronic unstable medical,” and “eye/dental” diagnoses compared with non-DE patients.

DE patients were significantly more likely to visit primary care physicians and access emergency department care compared with non-DE patients. DE patients were also more likely to be hospitalized and had longer mean ALOS days (P < 0.0001) than their non-DE counterparts. DE patients had significantly less visits to endocrinologists (P < 0.0001) during the study period.

Overall, 79% of the 9,453 Medicare patients had received an annual A1c test, 72% had received a biennial ophthalmologic examination, and 54% had received a biennial lipid test. Only 37% of patients, regardless of DE status, had evidence of receiving all three care indicators, whereas 6% did not receive any of the care indicators.

The proportion of Medicare patients receiving diabetes care by DE status is presented in Fig. 1. DE patients were significantly less likely to receive an annual A1c test, biennial ophthalmologic examinations, and biennial lipid testing and all three care indicators compared with non-DE patients (27 vs. 41% respectively; P < 0.0001).

Multivariate logistic regression analysis revealed that the odds ratios of DE patients receiving diabetes care was significantly lower compared with non-DE patients but differed by their geographic residence (Table 3). This significant interaction was present among each of the care measures except for biennial lipid testing.

DE patients residing in urban and rural locations received less frequent lipid testing and eye examinations than their non-DE counterparts. Urban DE patients were also significantly less likely to receive an annual A1c test and all three care measures. These differences were present after controlling for confounders.

In addition to DE status, African-American race/ethnicity was also associated with not receiving an annual A1c test, biennial lipid testing, and all three care measures together compared with white (non-Latino) patients. Latino patients were significantly less likely to receive biennial eye examinations, lipid examinations, and all three care measures.

Increased emergency department use was significantly associated with not receiving each of the three care indicators, whereas patients who visited an endocrinologist were 2.42 (95% CI 2.06–2.85) more likely to received all three indicators than those who did not visit endocrinologists. No other significant interactions between DE status and other variables were identified after controlling for number of outpatient visits, comorbidity, and time of first diabetes related claim.

In this study, we confirmed that DE Medicare patients were less likely to receive routine diabetes care. Our findings of disparate diabetes preventive care among low-income DE Medicare/Medicaid patients are similar to those described by Srinivasan et al. (15) for Medicaid patients <65 years old. Such lack of care is worrisome given that 22% of the Medicare patients in Colorado were DE for Medicaid.

These diabetes care disparities among DE patients persisted despite increased health care use. Prior studies have attributed such differences to increased health care demands and more comprehensive health insurance coverage among DE patients compared with non-DE patients (26). The association between increased emergency department use and not receiving diabetes care may be due to increased fragmentation of health care resources. Other studies have shown that DE patients were less likely to identify a physician’s office as a usual source of care and more likely to identify emergency departments, which are not designed to provide chronic illness care (26). Frequent emergency department use among those with diabetes may serve as useful identifier of a population at risk of not receiving preventive care.

We found that DE patients were more likely to have comorbid conditions, although it is unclear if such increased comorbidity is in part responsible for their Medicaid status. Psychosocial diagnoses such as depression were especially common among DE patients compared with non-DE patients and were significantly associated with not receiving diabetes care. Previous studies have demonstrated an association between depression and poor glycemic control; therefore, special attention should be given to DE patients who are more likely to a have psychosocial diagnosis to ensure they receive the recommended diabetes care (27).

Despite increased comorbidity and health care use, DE patients were significantly less likely to visit endocrinologists. Decreased access to subspecialty care is also troubling given that endocrinology visits increased the likelihood of receiving appropriate care. It is unclear if such access is due to differences in referral patterns or other access issues such as transportation. The identification of barriers to specialty care among DE patients deserves further attention.

Our findings of disparate diabetes care among African-American Medicare patients are consistent with those reported by Chin et al. (14) in 1993, and more recently, by Schneider et al. (28) among Medicare-managed care plans. Such persistent disparities in diabetes care indicate the need for further interventions to narrow these care gaps.

The extent to which the patient’s socioeconomic status contributes to our findings is not clear but is likely to be substantial. DE patients are by definition of low socioeconomic status. A recent study of the effects of such factors on physician care profiles by Franks and Fiscella (29) detailed potential explanations for the association between low socioeconomic status and health care quality including patient factors such as financial means and literacy, which influence adherence to health care recommendations, along with physician factors such as unconscious bias. Because socioeconomic status is likely to explain in large part our findings, the above-described barriers related to low socioeconomic status should be the major focus of interventions aiming at closing the gap between the DE and their non-DE counterparts.

Administrative data have inherent limitations and are not collected for research purposes, therefore the quality of our data must be critically reviewed. Although these measures of diabetes care quality are consistent with those of the Diabetes Quality Improvement Project initiative, they admittedly reflect the minimal accepted care standards and were chosen based on available data sources. Diagnosis and evaluation codes may be subject to variation, incompleteness, a lack of specificity, and may not document other important clinical issues in outpatient settings (30). Such codes are primarily intended for reimbursement calculation and are subject to “up coding,” a systematic increase intended to improve reimbursement (31). The differentiation of disease complications from preexisting comorbidities using diagnostic codes is problematic (32). This study is also limited in that it is based on process-of-care data rather than clinical outcomes. Process-of-care research allows one to study what is done for the patient, although the relationship between process and outcomes is not always clear (30). The cross-sectional design used in this study also makes the sequence of events unclear.

We only included patients who survived the entire study periods, thus selection bias is likely present as the DE group has a 50% higher age-specific mortality rate than non-DE Medicare beneficiaries. The excess mortality among the DE group most likely explains the higher use rates found for the dually entitled (33). This difference in age-specific mortality would likely further increase the care differences between the groups.

Our study did not investigate the potential influence of nursing home residence status on the likelihood of receiving diabetes care. Approximately 25% of DE patients in the U.S. reside in nursing homes (16). Institutionalization may influence provider attitudes toward the delivery of diabetes care. Few studies are currently available that detail the care of diabetes patients in U.S. nursing homes. One study demonstrated that diabetes care protocols in nursing homes were more likely to detail nutritional and nursing care standards than to guide medical care and monitoring for complications (34). The development of guidelines of the care of nursing home patients with diabetes deserves attention, especially given the growth of the elderly population and the increased prevalence of diabetes among these individuals.

As this study was limited to Colorado, our findings may not be generalizable. Also, this study did not address care provided to DE patients enrolled in Medicare+Choice–managed care plans. Health Plan Employer Data and Information Set (HEDIS) 2000 nationwide data indicate that the mean proportion across Medicaid-managed care plans of diabetes patients receiving A1c testing annually is 66%, and the median across plans was 70% (35). These rates were quite similar to our findings in the FFS population. Of note, HEDIS data indicated that 41% received annual eye examinations and 52% were LDL screened among all Medicaid-managed care enrollees. Because we defined our rates biennially, a comparison with HEDIS rates for these two measures was not possible. Furthermore, this study classified those as DE as having at least 1 month buy-in to the Colorado Medicaid program; however, multiple levels of Medicaid coverage exist.

In conclusion, despite increased evidence detailing the effectiveness of diabetes preventive care, many Medicare patients remain at risk of not receiving such care. Previously identified at-risk populations such as racial/ethnic minorities continue to fail to receive such care. Furthermore, Medicare patients with Medicaid coverage are significantly less likely to receive diabetes care. Given the disproportionate amount of health care service use and expenditures attributable to the DE population compared with non-DE patients, specific targeted quality improvement interventions are warranted to address these care disparities.

The analyses on which this article are based were performed under contract 500-96-P11, entitled “Utilization and Quality Control Peer Review Organization for the State of Colorado,” sponsored by the Centers for Medicare and Medicaid Services, Department of Health and Human Services. This article is a direct result of the Health Care Quality Improvement Program initiated by the Centers for Medicare and Medicaid Services, which has encouraged identification of quality improvement projects derived from analysis of patterns of care, and therefore required no special funding on the part of this contractor.

Ideas and contributions to the author concerning experience in engaging with the issues presented are welcomed.