Potential Short-Term Economic Benefits of Improved Glycemic Control: A managed care perspective

This study was based on a retrospective cohort design and used automated enrollment, medical and pharmacy claims, and clinical laboratory data files from the Fallon Clinic in Worcester, Massachusetts, a multispecialty group clinic with a predominantly managed care patient base. Most of the patients were members of the Fallon Community Health Plan, a health maintenance organization with over 200,000 enrollees.

The study population of interest consisted of adult plan members who had a diagnosis of diabetes between 1 January 1994 and 30 June 1998 and who also had multiple HbA_lc values available. These patients were assigned to study cohorts based on their mean levels of HbA_lc over this time period. Inpatient stays for specific short-term complications (events)occurring after the first HbA_lc test (the index date) through 30 June 1998 were identified and related to the average HbA_lc levels to explore variations in medical charges by the degree of glycemic control.

The principal measures of interest included 1) the proportion of patients with one or more inpatient admissions for short-term complications, 2) the mean number of such complications per patient, and 3)the expected costs per patient for these events. Both unadjusted and adjusted analyses were undertaken for the entire study population and stratified according to whether patients had long-term complications of diabetes.

Only patients who met the following eligibility criteria were included in the study: 1) age 35 years or older, 2) a diagnosis of diabetes based on the International Classification of Diseases, Ninth Revision, Clinical Modification (ICD-9-CM 250.xx) listed on at least two medical claims between 1 January 1994 and 30 June 1998, 3) at least two HbA_lc values measured no more than 18 months apart during this period, and 4) continuous enrollment in the plan from 1 January 1994 to 30 June 1998. Two claims indicating a diagnosis of diabetes were required to ensure that patients who were evaluated for diabetes, but not diagnosed,were not included in the study cohort. Multiple HbA_lc values were required to help ensure that patients were not being screened for diabetes periodically but were likely to have the diagnosis and to require monitoring of glycemic control.

Patients were assigned to study cohorts based on the mean of all HbA_lc values available during the study period as follows: good control (<8%), fair control (8-10%), and poor control (>10%).

The patient enrollment file included age and sex, as well as a history of all the dates of enrollment and disenrollment from the plan. The inpatient claims file included the admission and discharge dates, a primary diagnosis(in ICD-9-CM format), up to two secondary diagnoses, and the procedures performed. Outpatient claims provided details on the date of service, the type of service, and the primary diagnosis (ICD-9-CM). The clinical laboratory file included all HbA_lc tests performed and analyzed by a single contract laboratory (CliniTech). The measurement of HbA_lc was stable throughout the study period.

The duration of follow-up for each patient was defined as the number of days between the first HbA_lc test that occurred on or after 1 January 1994 and the cutoff date for the data files (30 June 1998). The statistical techniques used in this study adjusted for variable follow-up, and our findings were expressed on a 3-year basis.

We determined whether each study patient had at least one inpatient admission related to a short-term complication during the study period and the number of such events. These events were defined by an inpatient stay (hospital or skilled nursing facility) with a discharge diagnosis (primary or secondary) in any of the following four groups: 1) hyperglycemia (ICD-9-CM codes 250.1x-250.3x), 2) hypoglycemia (250.8x and 251.0x), 3)selected infections (septicemia, 038.x; pneumonia, 480.x-486.x; kidney infections, cystitis, and urinary tract infections, 590.xx, 595.xx, and 599.0x; cellulitis, 680.xx-682.xx and 686.xx; and bacteremia, 790.7), and 4) electrolyte imbalance (276.xx).

The total cost of inpatient care for short-term diabetes-related complications was estimated for each study patient by totaling the charges on each inpatient medical claim, as indicated on the billing data supplied by the health plan.

The burden posed by comorbidities may differ among cohorts of patients defined by HbA_lc levels and may, thus, confound the relation between HbA_lc levels and rates of inpatient treatment for short-term complications. To adjust for this potential problem, we examined whether each patient had a long-term diabetes complication during the study period, defined as any inpatient discharge diagnosis (primary or secondary)that included neurological disease; cardiovascular, cerebrovascular, or peripheral vascular disease; amputation or ulceration; or renal and ophthalmic symptoms. In addition, cancer was included as a separate comorbidity, because its treatment may increase the risk of infection and other complications. We also assessed comorbidity using the Deyo et al.(11) adaptation of the Charlson Index for administrative claims data, excluding diabetes, i.e., the diagnosis of interest in our study population(11,12). Because our findings were nearly identical to those based on chronic diabetic complications and cancer, they are not reported herein.

We estimated the proportion of patients with at least one inpatient stay for a short-term complication on both an unadjusted and adjusted basis using individual-level data. For the latter, we estimated the probability of inpatient treatment during follow-up using a binomial regression analysis based on a logistic function, controlling for each patient's follow-up time(expressed in the logarithm of years and censored for the inpatient admissions). The independent variables included the HbA_lc group(<8, 8-10, or >10%), age, sex, presence of a long-term diabetes complication, and presence of a cancer diagnosis. A predicted 3-year probability of inpatient treatment was calculated from the logistic function for each patient in each of the HbA_lc groups. An average 3-year risk was then derived for each HbA_lc group.

We also calculated the mean number of admissions for short-term complications on both an unadjusted and adjusted basis. The latter used a Poisson regression, which was well suited for the relatively rare events(13). Each patient's follow-up time (expressed in the logarithm of years but not censored for inpatient admissions) was controlled for in the analysis. The independent variables included the HbA_lc group, age, sex, presence of a long-term diabetes complication, and presence of a cancer diagnosis. A predicted 3-year probability was calculated from the Poisson model for each patient in each of the HbA_lc groups, and an average 3-year risk was then derived for each HbA_lc group. Inpatient stays were evaluated for all patients and stratified by the presence of a long-term diabetic complication.

Finally, we estimated the average cost of inpatient care for short-term complications using least-squares regression, adjusting for the factors listed above. However, no statistical testing was performed on medical charges,because actual costs were unavailable and charges are not easy to generalize from one setting to another. The analyses of data were conducted using SAS Version 7.0 for Windows NT (SAS Institute, Cary, NC).

We identified a total of 9,156 plan members who were diagnosed with diabetes between 1 January 1994 and 30 June 1998. A total of 3,475 of these people were excluded because they were not continuously enrolled for the entire study period. In addition, 3,089 people were excluded because they either did not have any HbA_lc values or did not have at least two values measured no more than 18 months apart. This left a total of 2,394 diabetic plan members who met our study eligibility criteria and were included in the analysis. The average age of these patients was 63 years, and ∼55%were men. Approximately 30% had a long-term diabetes complication, and 2.5%had a cancer diagnosis (Table 1). Over 80% of the population filled prescriptions for diabetes medications during the study.

There were important differences across the three study cohorts defined by average HbA_lc levels. For example, patients with poor glycemic control were more likely to have long-term diabetic complications than those with good or fair control. Similarly, the duration of follow-up in the study was longer among patients in the two groups with the highest HbA_lclevels. Patients experiencing poorer glycemic control may have had diabetes for a longer amount of time and may have been suffering from long-term complications. These patients, therefore, may have been more closely monitored by their physicians. Finally, the likelihood of treatment with diabetes medications was the greatest in the two groups with the highest HbA_lc levels.

Of the 2,394 patients with diabetes, a total of 251 (∼10%) were admitted for the treatment of short-term complications, accounting for a total of 447 inpatient stays. For these admissions, 59% of the qualifying diagnoses were for infections, 21% were for hyperglycemia or hypoglycemia, and 20% were for electrolyte disturbances. There were a total of 4,562 admissions during the study period for any reason.

We found a statistically significant positive relation between the HbA_lc group and the likelihood of inpatient admissions for short-term complications on both an unadjusted and adjusted basis (both P < 0.01, based on a test for trend). The latter analysis, which controlled for the differences among the study cohorts in age, sex, presence of cancer, and presence of long-term diabetes-related complications, showed that the proportion of patients admitted for a short-term complication over 3 years increased from 8.5% among patients with good control to 17.8% among those with poor control (Table 2). This statistically significant increase in the likelihood of inpatient treatment was also found in both patient subgroups (those who did not have long-term diabetic complications and those who did); the absolute increase in risk was larger for those with long-term complications.

The average number of admissions for short-term complications followed a similar pattern. On an adjusted basis, the average number of admissions over 3 years was 2.5 times greater among patients with poor glycemic control versus those with good control (i.e., 13 inpatient stays per 100 patients vs. 31 per 100, respectively). Although the increase in the number of inpatient stays across the three HbA_lc groups is modest for patients without long-term diabetic complications, the number of inpatient stays is substantially higher among those with such complications, ranging from 30 to 74 per 100 persons. Correspondingly, the average adjusted charges for inpatient treatment for patients with good control was approximately $970 over a period of 3 years compared with $1,380 and $3,040 for those with fair or poor glycemic control, respectively (Table 3). Among the patients with chronic diabetic complications, the results were more marked.

To explore the potential economic benefits of improved glycemic control, we conducted a retrospective study using enrollment, medical claims, and clinical laboratory data for ∼2,500 adults with diabetes enrolled in the Fallon Clinic health plan within a recent 4-year period. We found that the likelihood of inpatient admission for selected short-term complications and the average number of such stays increased significantly with higher levels of HbA_lc. On an adjusted basis,patients with poor glycemic control had more than double the number of inpatient admissions over a 3-year period than those with good glycemic control 31 vs. 13 per 100 patients, respectively; P < 0.01). The largest absolute difference in admission rates occurred in patients who had long-term diabetic complications (30 per 100 patients with good control vs. 74 per 100 patients with poor control; P < 0.01). Corresponding average adjusted charges were also lowest among patients with good control and highest among those with poor control.

Our finding that HbA_lc levels positively correlate to medical charges in clinical practice was consistent with the study by Gilmer et al.(9), although their data were somewhat older and included all costs regardless of diagnosis. To the best of our knowledge, our study is the first to relate short-term complications to levels of glycemic control. The results from this investigation may be helpful in understanding the potential economic implications of short-term interventions designed to improve glycemic control, such as disease management programs and newer drug therapies.

We acknowledge several limitations of our study. First, although our analyses suggest that a correlation exists between selected short-term complications and the extent of glycemic control, it does not demonstrate that these events could have been avoided. Only intervention studies can establish whether lowering HbA_lc levels reduces the costs for acute diabetic complications. Second, the precise diagnoses that could vary with the level of glycemic control on a short-term basis are subject to judgment. In our study,we chose to focus on hyperglycemia, hypoglycemia, and cellulitis, which were all classified as short-term (acute) complications of diabetes in a recent database study by Selby et al.(10). We also included several additional conditions that diabetic patients experience more frequently than nondiabetic patients, including other infections and electrolyte imbalance(10). Further work is required to identify which diagnoses are correlated to glycemic control on a short-term basis.

Furthermore, short-term complications were identified based on selected ICD-9-CM codes that appeared as primary or secondary diagnoses on inpatient bills. In the database used for this study, it was not possible to distinguish acute hospital stays from admissions to skilled nursing facilities. There are also inherent inaccuracies associated with the use of ICD-9-CM diagnosis codes, which may have led us to misestimate the number of events. To address these issues, we conducted a chart review based on samples of two groups of inpatient stays from two randomly selected patient samples. The first group consisted of 69 inpatient stays (∼15% of the observed number of events)with a diagnosis of a short-term diabetic complication noted on medical claims for 50 study patients. Of these 69 inpatient stays, 58 (85%) involved admission to an acute-care hospital. The diagnosis data from the discharge summaries, which were completed only for hospital stays, showed that a qualifying discharge diagnosis (i.e., for a short-term event) was listed on 64(93%) of these admissions. For the second group of inpatient admissions with discharge summaries (n = 29, drawn from 20 patients), which did not have a short-term complication noted in the billing data, 12 stays (41%) were found on examination to have a qualifying diagnosis. Therefore, the results of the chart review indicate that we may have underestimated the number of inpatient stays for acute diabetic complications, which suggests that our results may be conservative.

It is possible that some of the inpatient stays we identified as events were not the sole cause for admission or that they could, perhaps, have been incidental findings for patients who were admitted principally for other medical problems. Determining the reasons for admission based on ICD-9-CM discharge diagnoses is fraught with difficulties(14,15). Moreover, to the best of our knowledge, there are no accepted methodologies for apportioning costs among multiple diagnoses. The emphasis that we placed on acute complications may help limit but not eliminate this problem.

Because the frequency of inpatient admissions probably increases with the severity and duration of diabetes, each of which may also be correlated with HbA_lc levels, it is possible that our findings are due to residual confounding by these two factors. In this study, we attempted to adjust for the severity of diabetes among the study cohorts by controlling for age, sex,several chronic diabetic complications, and follow-up time. The latter two factors are especially important because they may be a reasonable proxy for the duration of diabetes. Nonetheless, to the extent that we were unsuccessful in these efforts at statistical control, we may have overstated the reduction in costs that may be associated with better glycemic control.

We assessed the occurrence of short-term diabetes-related complications any time after the first observed HbA_lc value, because the precise temporal relation between these two factors is uncertain. Moreover, patients were assigned to groups based on their average HbA_lc level during follow-up because glycemic control was quite variable and a predefined starting date was not available, given that patients were not required to be newly diagnosed. Randomized studies may be required to better understand the temporal relation between glycemic control, as evaluated by HbA_lc,and the occurrence of diabetic complications.

Finally, our findings regarding costs may not be generalizable to all patients with diabetes. Treatment patterns for this disease may differ according to individual physician practice styles, health plan guidelines, and geographical regions. In addition, data on resource costs were not available to us, but rather our analyses relied on charges. It is known that charges often exceed resource costs, but the magnitude of this mark-up differs from one institution to another. However, by reporting our findings in terms of admission rates, other researchers may use our findings in conjunction with their own cost data.

Despite these limitations, this exploratory analysis suggests that in typical practice, better glycemic control is associated with lower inpatient admissions and costs during a 3-year time period. These potential short-term savings in cost are important to consider when making decisions regarding the adoption and use of new interventions for the management of diabetes.

This study was funded in part by Parke-Davis Pharmaceuticals (Morris Plains, NJ).

We thank Ellen Trencher and Barbara Lewis, PhD, for allowing us access to the data used in this study and for commenting on the manuscript, as well as Jan Guilbert for conducting the chart reviews. We also appreciate the helpful comments provided by Mauricio Calero, MD, Phillip Sarocco, and Joseph Gricar.