Costs and Utilization Associated With Pharmaceutical Adherence in a Diabetic Population

, ,
The American Journal of Managed Care, February 2004 - Part 2, Volume 10, Issue 2 Pt 2

Objective: To determine whether adherence with pharmaceutical therapy affects well being and total costs associated with diabetes treatment.

Study Design: Retrospective cohort design using insurance claims in an open access, nonmanaged care setting.

Patients and Methods: Patients with diabetes were under age 65 years, continuously enrolled with medical and drug eligibility, and identified by using methodology based on the Health Employer Data and Information Set. Patients were identified in 1998. The level of adherence to drugs used for diabetes, utilization, and medical costs were measured in 1999. Regression analyses statistically controlled for age, sex, illness severity, and product line.

Results: Of the 57 687 patients identified with diabetes, 55% were male and 90% were age 40 years or older. Study members taking a prescription medicine for diabetes were significantly older and marginally sicker than those not taking a prescription medicine for diabetes. Patients without diabetic drug claims had the lowest medical costs, whereas younger patients and female patients had higher costs and utilization. A threshold effect was observed, where a target level of adherence was needed before medical care costs were reduced. Increased pharmaceutical adherence was associated with fewer emergency department visits and inpatient admissions. Increased medication adherence was associated with decreased medical care costs. Increased medication adherence was not associated with decreased overall healthcare costs because medication costs offset medical care cost savings.

Conclusions: Increased adherence with pharmaceutical therapy was associated with decreased use of medical care services, suggesting improved disease control and well being, but not with lower costs.

(Am J Manag Care. 2004;10(part 2):144-151)

The Centers for Disease Control and Prevention estimates that approximately 11.1 million people in the United States have been diagnosed with diabetes and an additional 5.9 million people have the disease but have not yet been diagnosed.1 Diabetes is the leading cause of new cases of blindness, end-stage renal disease, and lower extremity amputations in working-age adults and the fifth-leading cause of death by disease in the United States.2 In addition, people with diabetes are 2 to 4 times more likely than the general population to have heart disease or experience a stroke.1

Not only are serious health burdens associated with the progression of diabetes, there also are significant economic consequences. On average, the cost of healthcare for a person with diabetes in 2002 was 5 times more than the cost for a person without diabetes.3 The American Diabetes Association estimated that $132 billion in expenditures were attributable to the direct medical costs and indirect costs (premature mortality and disability) associated with the disease in 2002.2

Many of the complications associated with diabetes can be delayed or prevented through improved disease management and self care, including aggressive management of cardiovascular risk factors; early identification and treatment of hypertension, kidney disease, retinopathy, peripheral neuropathy and vascular disease; and increased glycemic control through diet, exercise, and/or taking insulin or oral diabetes medications.

Long-term prospective studies have established that improved blood glucose control reduces the risk of onset and slows the progression of complications in people with type 1 and type 2 diabetes. The Diabetes Control and Complications Trial, results of which were published in 1993, established that intensive therapy of patients with type 1 diabetes delays the onset and slows the progression of retinopathy, nephropathy, and neuropathy by 50% to 70% compared with conventional treatment regimens.4 The United Kingdom Prospective Diabetes Study, results of which were published in 1998, demonstrated an overall 25% decrease in microvascular complications among patients with type 2 diabetes who had intensive glucose-lowering therapy compared with conventional therapy.5 Although adherence to medication regimens is crucial to achieving glycemic control and preventing microvascular complications, a number of studies have indicated many people with type 2 diabetes do not use medications as prescribed.6-9

According to a report by the Task Force for Compliance 10 medication nonadherence is a pervasive problem. The Task Force report cites numerous studies that demonstrate patient failure to fill or refill prescriptions and lapses in the continuity of dosing regimens, and notes that compliance is especially low for patients who have chronic diseases not associated with symptoms or in which the symptoms occur erratically. The annual cost of medication noncompliance, including hospital and nursing home admissions, increased ambulatory costs, lost productivity, and premature deaths, has been estimated at over $100 billion per year in the United States.11

  • Develop a methodology to assess adherence with medications used for diabetes.

Examine the relationship between adherence with medications used for diabetes and overall costs of care (medical and pharmaceutical).

  • Determine whether adherence with medications used for diabetes affects well being, as reflected in rates of emergency department visits and inpatient admissions.

METHODS

Study Population

The study involved a retrospective cohort design using Blue Cross Blue Shield of Michigan (BCBSM) claims in an open-access Traditional indemnity and preferred provider organization (PPO) insurance setting. The Traditional indemnity plan refers to fee-forservice health insurance in which members have free choice of physicians, whereas the PPO offers members incentives for using a selected network of participating providers instead of other providers.

Patients with diabetes were identified based on healthcare claims for care provided in calendar year 1998. Claims incurred in calendar year 1999 were used to assess pharmaceutical use, medical costs (defined as BCBSM payments agreed to by participating providers and facilities), and utilization rates.

The eligible study population included all non- Medicare-eligible Michigan residents who were continuously enrolled for calendar year 1999 in a Traditional indemnity or PPO product. In addition, to ensure no coverage gaps, study members had facility, professional, and pharmaceutical benefit coverage.

The National Committee for Quality Assurance Health Plan Employer Data and Information Set (HEDIS®) specifications for identifying patients with diabetes, which rely on administrative and medical record data pertaining to HMO members, were modified for use in an open-access product with administrative data only. To be classified as a patient with diabetes, an individual must have met 1 of the following criteria in calendar year 1998:

  • At least 1 inpatient or emergency department claim with a primary or secondary diagnosis of diabetes (International Classification of Diseases, 9th Revision [ICD-9] 250, 352.2, 362.0, 366.41, 648.0).

Two professional or outpatient facility claims (incurred a minimum of 30 days apart) with a diabetes diagnosis.

  • A filled prescription for insulin and/or an oral hypoglycemic agent.

Patients with claims reflecting gestational diabetes or patients who died in the hospital were excluded from the analysis.

Assessment of Pharmaceutical Use

Adherence measurement was based on filling a prescription used for diabetes (insulin or an oral hypoglycemic agent). For each prescription filled for diabetes, the purchase date of the prescription was recorded. Next, an end date for each prescription was calculated by adding the number of days supplied to the purchase date. This was repeated for each prescription filled for diabetes between the fourth quarter of 1998 and the end of calendar year 1999. The fourth quarter of 1998 was used to calculate those prescriptions for which the days supplied extended into calendar year 1999. Then for each day of the year (January 1, 1999 = day 1; January 2, 1999 = day 2; … December 31, 1999 = day 365), it was determined whether the patient possessed any diabetic drug (yes = 1, no = 0). The outcome for each day was summed across the year, resulting in a range of 0 to 365 days. The medication adherence rate was calculated as percentage of days (rounded to the nearest percentage) that the patient possessed any available diabetic drug during the year, using the equation:

Medication adherence rates ranged from 0% (no claims) to 100%. The closer the adherence rate was to 100%, the more adherent the member was ("adherence" was defined in this study as days with filled prescriptions for diabetic medication). Finally, each study member was placed into 1 mutually exclusive adherence category, defined as 0%, 1% to 19%, 20% to 39%, 40% to 59%, 60% to 79%, 80% to 99%, or 100%. Within each adherence category, members were assessed for costs and utilization of medical care.

It is important to recognize that study members in the 0% drug adherence group either (1) controlled their diabetes through diet and exercise alone, without the aid of diabetic medications; (2) filled prescriptions under another health plan; or (3) did not fill their diabetic prescriptions as recommended.

Assessment of Costs and Utilization of Medical Care

Resource assessment for each diabetic study member included the costs and services incurred for inpatient hospitalization, outpatient care, emergency care, clinic visits, laboratory tests, professional services, and pharmaceuticals in calendar year 1999; however, claims from skilled nursing facilities and home healthcare were excluded.

Three analyses were performed on the cost information (Table 1). First, the overall cost of healthcare (medical and pharmaceutical) for study members was analyzed, which included costs related and unrelated to diabetes care. The second and third analyses focused on costs (medical and pharmaceutical) related to care with a diabetes diagnosis.

BCBSM professional claims only have a primary diagnosis code available on the claim. Facility claims, however, have up to 5 secondary diagnosis codes available in addition to the primary diagnosis. BCBSM pharmacy claims do not include a diagnosis field; as a result, pharmaceutical costs related to diabetes were restricted to drugs used for diabetes. Therefore, the second analysis labeled "any diagnosis" included professional claims billed with a primary diagnosis of diabetes or facility claims with a primary or any secondary diagnosis of diabetes, as well as claims for diabetic drugs. The third analysis labeled "primary diagnosis" was restricted to professional and facility claims with a primary diagnosis of diabetes only and claims for diabetic drugs. The cost of diabetic drugs was the same for the "any diagnosis" and "primary diagnosis" analyses. Because BCBSM claims with secondary diagnoses were only available on facility claims, facility claims are responsible for differences between the "any diagnosis" and "primary diagnosis" analyses of diabetes-related care.

In addition to the cost assessment, the rates of study members? emergency department visits and inpatient hospitalizations (both for overall health issues and for diabetes-related care) were measured.

Statistical Analyses

Frequency distributions, means, and standard deviations were used to describe the study population. Comparisons between patients who filled a prescription for diabetes and those without a filled prescription for diabetes were made for several characteristics.

Two types of regression models were performed to determine the impact of patient demographics and adherence with medications used for diabetes on the costs and utilization of medical services. First, an ordinary least squares regression model was used to calculate the expected cost of each individual's healthcare, where the dependent variable was total healthcare costs. Next, for each individual, use of emergency department services or having an inpatient admission was analyzed by using a multivariable logistic regression model, where an emergency department visit or an inpatient admission was the dependent variable. The variables available in BCBSM administrative data files that were relevant to this analysis were age, sex, and product line (Traditional indemnity and PPO). Illness severity (diagnosis cost group [DCG] relative risk score) and drug adherence level (0%, 1%-19%, 20%-39%, 40%- 59%, 60%-79%, 80%-99%, or 100%) also were statistically controlled for in each regression model.

DxCG® software (DxCG, Inc, Boston, Mass) was used to categorize patients? illness severity. The DxCG software applies DCG models, which use age and sex demographics and all diagnoses from both inpatient and outpatient clinician encounters, to predict the relative resource use (or expense) at the individual member level. The age/sex cohort and each diagnostic category contribute a cost weight to the final relative risk score for an individual. The DCG relative risk score is a continuous variable, where the closer to 0 the score, the more healthy the member is. It has been shown the DxCG software can predict up to 40% of the potentially explainable variance in healthcare costs, while using demographic data such as age and sex only can explain 4% to 6%.12

All analyses were conducted with SAS statistical software, version 8.0 (SAS Institute, Inc, Cary, NC). Because the study population was large, statistical testing was performed at the 1% significance level.

RESULTS

Patient demographic characteristics and univariate analysis are presented in Table 2 and Table 3. The study population consisted of 57 687 BCBSM members identified as having diabetes, of whom 90% were age 40 years or older (average age 52.6 ± 10.5 years), 55% were male, and 60% were enrolled in the Traditional indemnity product line. Fewer than 5% were fully compliant with filling their diabetic prescriptions across the year, and more than 17% of the study members did not have a diabetic prescription claim.

P

P

P

P

Univariate analysis showed several differences between members with diabetes filling a prescription (adherence level 1%-100%) and members with diabetes who did not have a diabetic prescription claim (adherence level 0%). Those members with diabetes filling a prescription were significantly older (mean age 52.8 years vs 51.7 years; < .001), marginally sicker (mean DCG score 3.86 vs 3.74; = .0997), more likely to be male (55.9% vs 50.6%; < .001), and less likely to be Traditional indemnity members (60.0% vs 62.6%; < .001).

Costs of Medical Care

The stacked bar charts in Figure 1 depict average expenditures (medical care and drug) for overall healthcare costs and diabetes-related costs by drug adherence category. Study members categorized in the 0% adherence level had the lowest overall costs and diabetes- related costs.

Although total (medical care and drug) costs increased as the medication adherence level increased, medical care costs showed a slow increase up to the 20% to 39% adherence level for overall costs and diabetesrelated costs with any diabetes diagnosis; after this level was reached, there was a steady decrease. Thus, a threshold effect was observed at the 20% to 39% adherence level; this level of drug adherence was needed before medical care costs were reduced. For the diabetes- related costs with a primary diabetes diagnosis, the threshold effect was not seen until the 40% to 59% adherence level was reached.

As expected, each bar grouping showed that as the pharmaceutical adherence level increased, pharmaceutical costs increased. However, when pharmaceutical costs were added to medical care costs, an overall decrease in total cost as the adherence level increased was not apparent.

After controlling for patient characteristics and drug adherence level, the ordinary least squares regression model determined that age, sex, illness severity, product line, and adherence level were related to healthcare expenditures. (Tables detailing the results from all 3 regression analyses are available from the authors on request.)

  • In the first model, where total medical and pharmacy costs were calculated for all diabetic and nondiabetic-related care, patient characteristics and drug adherence level explained 55% of the cost variation (R2 = 0.5465). Analysis showed increased age, female sex, Traditional indemnity coverage, higher illness severity (DCG score), and higher drug adherence level (>40%) were significant predictors of higher overall costs.

In the second model, where medical claims with a primary or any secondary diagnosis of diabetes and diabetic pharmacy claims were analyzed, only 32% of cost variation (R2 = 0.3220) was explained by the regression variables. Significant predictors of high diabetes-related (any diabetes diagnosis) costs included younger patients, Traditional indemnity coverage, higher illness severity, and higher drug adherence level.

  • In the third model, which pertained to diabetes care with a primary diabetes diagnosis only and diabetic pharmacy claims, 12% of the cost variation (R2 = 0.1213) was explained by the regression variables. Younger patients, female sex, higher illness severity, and higher drug adherence level were significant predictors of high diabetes-related costs with a primary diabetes diagnosis.

Utilization of Medical Care

More than 17% of the 57 687 study members had at least 1 emergency department visit in the measurement year. Approximately 7% of these emergency department visits had a primary diagnosis of diabetes on the claim, 20% had a diabetes diagnosis listed as any secondary diagnosis on the claim, and the remaining 73% of emergency department visits did not have a diabetes diagnosis on the claim. Similarly, approximately 14% of the study population had at least 1 inpatient hospitalization. Of these inpatient admissions, 10% had a primary diagnosis of diabetes on the claim, an additional 66% had a secondary diagnosis of diabetes on the claim, and the remaining 24% did not have a diabetes diagnoses on the claim.

Figure 2 depicts emergency department visits and inpatient hospitalizations for overall healthcare and diabetes-related care by drug adherence category.

Each bar grouping shows that increased patient adherence with diabetic medication was associated with fewer emergency department visits and inpatient hospitalizations. Study members categorized in the 0%, 80% to 99%, and 100% adherence levels had the lowest overall and diabetes-related utilization of emergency department and inpatient services. There was a similar threshold effect at the 20% to 39% adherence category (1%- 19% for overall emergency department visits). Here, a target level of 20% to 39% adherence was needed before decreased utilization of emergency department and inpatient services were realized when all services were considered and when services for diabetes as a primary or secondary diagnosis were evaluated.

Emergency Department Utilization

For the emergency department utilization analysis, a logistic regression model (1 = at least 1 emergency department visit, 0 = no emergency department visits) that controlled for patient characteristics and drug adherence level was used. The logistic regression model determined that age, illness severity, and adherence level were related to emergency department use. Sex and product line were not significant predictors of emergency department utilization. Included below are the resulting odds ratios.

  • Analysis of overall emergency department visits showed that age was not a significant predictor of overall emergency department visits. However, for diabetes-related emergency department visits, older patients were 0.98 times as likely to have an emergency department visit with any diabetes diagnosis and 0.97 times as likely to have an emergency department visit with a primary diabetes diagnosis. These findings were statistically significant, but small in magnitude.

Patients with higher DCG scores were more likely to have an emergency department visit, with odds ratios of 1.08 overall, 1.06 for any diagnosis of diabetes, and 1.05 for a primary diagnosis of diabetes.

  • For overall emergency department use, patients with an adherence level of 80% to 99% were approximately 0.88 times as likely to have an emergency department visit. For diabetes- related visits (any diagnosis or primary diagnosis), patients with any level of drug adherence (1%-100%) were 1.58 to 4.21 times as likely to have an emergency department visit than those without claims for diabetic drugs (0% adherence).

Inpatient Utilization

A similar analysis was performed on inpatient hospitalizations. A logistic regression model (1 = at least 1 inpatient admission, 0 = no inpatient admissions) that controlled for patient characteristics and drug adherence level was used. The logistic regression model determined that sex, illness severity, and adherence level were predictive variables of inpatient hospitalizations for overall and diabetes-related care. Presented below are the resulting odds ratios.

Male patients were 0.41 times as likely to have an inpatient hospitalization overall, 0.48 times as likely for any diabetes diagnosis, and 0.44 times as likely for a primary diabetes diagnosis.

  • Patients with higher DCG scores were more likely to have an inpatient admission, with odds ratios of 1.33 overall, 1.20 for any diabetes diagnosis, and 1.09 for a primary diabetes diagnosis.

Compared with patients without diabetic drug claims (0% adherence), patients with an adherence level of 1% to 19% or 20% to 39% were 1.26 and 1.23 times, respectively, more likely to have an inpatient admission overall. Patients with any level of drug adherence (1%-100%) were 1.78 to 4.32 times as likely to have a diabetes-related inpatient admission for any diabetes diagnosis or a primarydiabetes diagnosis than patients in the 0% adherence group.

  • In addition, for the primary diabetes diagnosis analysis, age and product line were predictive variables of inpatient admissions. Older patients were 0.95 times as likely and Traditional indemnity patients were 0.80 times as likely to have an inpatient admission with a primary diabetes diagnosis.

DISCUSSION

The study had several limitations that should be kept in mind while interpreting the results. Measuring filled prescriptions does not equate to actual drug use. Although claims data identify when study members filled prescriptions, they cannot be used to determine whether patients were taking medication properly. Further, administrative data cannot measure the use of free medication samples or whether prescriptions were filled through a different healthcare plan. However, it is unlikely that a high proportion of diabetic members with continuous pharmacy coverage received free medication samples on an ongoing basis or filled prescriptions through more than 1 healthcare plan. We examined coordination-of-benefit issues, payment of claims for members covered by more than one health care plan, and partial pharmacy carve-outs, which could explain no more than 5% of the 0% adherence group.

In addition, we were unable to account for other demographic factors (eg, race, education, income) that were not available on claims data. Benefit design differences may explain variation between product lines, which may indirectly affect the utilization of medical services. For example, higher copayment levels may deter patients from seeking necessary medical care or filling prescriptions. We did not account for the various levels of copayment, deductibles, and coinsurance for prescriptions and other medical services, including physician office care, medical emergency care, and inpatient hospitalization in the analysis. However, we believe these differences were minor in contrast to the relatively comprehensive benefit structure represented by this study population.

It is important to recognize the 0% adherence group is heterogeneous. It includes (1) patients receiving ongoing care for diabetes, who were intentionally being treated solely by diet and exercise; (2) patients who filled prescriptions under another health plan; and (3) patients who were not filling prescriptions despite receiving them during ongoing medical care for diabetes. The relative proportions of these groups, however, cannot be quantified using administrative data. Additionally, people identified as diabetic in 1998 who did not have claims for any services for diabetes (including drugs) in 1999 also were included; however, this segment represented fewer than 2% of the 0% compliant population.

The study used a short time frame; patients were identified in year 1 and studied in year 2. Diabetes is a chronic illness, and many complications of diabetes become evident after several years of treatment. Improved adherence with diabetic medication regimens is likely to increase pharmaceutical costs in the short term. It may take many years of improved glycemic control to reduce medical costs associated with the treatment of long-term complications. The cost and utilization findings of this analysis may differ with a longer measurement period.

Costs of care for diabetes as a secondary diagnosis primarily reflect inpatient care (with a relatively small dollar amount of outpatient facility care for emergency services and procedures for diabetes), because only facility claims, other than facility laboratory and radiology claims, contain more than 1 diagnosis. Thus, the approximately twofold difference between costs of care for patients with diabetes as a primary or secondary diagnosis, compared with the costs for patients with diabetes as a primary diagnosis only, mostly reflects the cost of inpatient care. For inpatient admissions with diabetes as a secondary diagnosis, diabetes may or may not be an important focus of attention and active management. Diabetes often is coded on facility claims when it is present as a comorbidity, even when its management requires little attention during a hospital stay. Therefore, it is interesting to note that there is an identical threshold effect (ie, medical care costs decrease above the 20%-39% adherence level) for both overall costs and costs of care for patients for whom diabetes was a primary or secondary diagnosis. In addition, the patterns of costs attributable to medical care and pharmacy care, by drug adherence level, were proportionately quite similar for these 2 categories of patients.

In the model based on the costs associated with diabetes as a primary diagnosis, the R2 was substantially lower than that for the other 2 analyses (overall costs and any diabetes diagnosis costs). Most of the costs in this portion of the analysis were attributed to inpatient care where diabetes management was the primary focus of attention. Patients at any medication adherence level are generally only admitted specifically for the treatment of diabetes when the condition is acutely out of control. Since we are only analyzing care associated with a primary diagnosis of diabetes, costs related to the treatment of the long-term complications of poorly managed diabetes are not included and less variation would be expected.

In contrast, we noted a minor difference when examining the proportion of costs for medical care and pharmacy services for diabetes when these costs were limited to services billed with diabetes as a primary diagnosis. In this group, a threshold effect was noted, but only beginning with a decrease in medical care costs in the patients with 60% to 79% medication adherence. Although no conclusions can be drawn from these data, it may be that for patients with diabetes and moderate levels of adherence with diabetes medication, there also were moderate levels of adherence with other recom- mendations for care of comorbid health conditions and lifestyle modifications. This could explain the decreasing levels of overall medical costs and costs for diabetes as a secondary diagnosis at relatively low levels of drug adherence. It also may be that to affect costs for medical care focused primarily on problems directly associated with diabetes, greater medication adherence is required, explaining the threshold effect that was observed at higher adherence levels for this group of patients.

Lastly, these analyses are based on administrative claims from BCBSM, a large insurer in Michigan that may not be representative of other insurance companies. In addition, these analyses focus on diabetes-specific medication adherence, cost, and utilization patterns; results could vary with other disease conditions.

CONCLUSIONS

Overall healthcare costs (medical and pharmaceutical) were not lower among diabetic patients who were compliant with medication regimens because pharmaceutical cost increases offset decreases in medical care costs. Increased pharmaceutical adherence was associated with decreased use of emergency department visits and inpatient admissions, suggesting improved disease control and well being. Similar cost and utilization patterns were evident for both overall healthcare and diabetes-related care. The impact of medication adherence on decreasing the use and cost of nonpharmacy care was observed at a threshold of 20% to 39% adherence, except for services for diabetes only as a primary diagnosis, in which case the threshold of 40% to 59% of adherence was associated with a subsequent decrease in cost and use of nonpharmacy care.

Investments in disease management programs to promote adherence with medication regimens may substantially increase well being without a subsequent increase in the overall cost for patients with diabetes. Adherence to pharmacologic regimens is key to improving glycemic levels and reducing the complications of diabetes. This study demonstrated that the consistent use of prescribed diabetic medication is associated with a reduced likelihood of future emergency department and inpatient admissions. These findings provide further support for the need to encourage and monitor long-term adherence with medications to ensure that patients with diabetes receive appropriate treatment on an ongoing basis. Administrative data can be used to help monitor the effectiveness of disease management, care management, and self-management in achieving increased adherence with medications used to improve glycemic control.

Acknowledgments

This paper was prepared as part of a research project by The Center for Health Care Quality and Evaluative Studies, Blue Cross Blue Shield of Michigan. The initial concept and design were based on a report prepared by Merck Medco for a large customer group. The authors would like to thank Kevin Hawkins, PhD, for his contributions to the concept, design, and data analyses; Vicki Wilson, BS, and Sheryl Spath, BS, for their programming assistance; and Susan Rubin, MPH, Leaden Hickman, PhD, Huda Fadel, PhD, MPH, and Kathy Laing, RN, MBA, CPHQ, for their assistance in preparing the manuscript.

From Blue Cross Blue Shield of Michigan, Detroit, Mich.

Presented as a poster at the Academy for Health Services Research and Health Policy Annual Research Meeting, June 25, 2002, Washington, DC.

Address correspondence to: Mary T. Martus, RN, BSN, 600 Lafayette East, MC J426, Detroit, MI 48226. E-mail: mmartus@bcbsm.com.

1. Centers for Disease Control and Prevention, National Center for Chronic Disease Prevention and Health Promotion. National Diabetes Fact Sheet—National Estimates on Diabetes. Available at: http://www.cdc.gov/diabetes/pubs/estimates.htm. Accessed October 30, 2002.

Diabetes Care.

2. American Diabetes Association. Economic consequences of diabetes in the US in 2002. 2003;26:917-932.

3. Centers for Disease Control and Prevention, National Center for Chronic Disease Prevention and Health Promotion. Diabetes: Disabling, Deadly, and on the Rise. At a Glance 2002. Available at: http://www.cdc.gov/nccdphp/aag/aag_ddt.htm. Accessed October 30, 2002.

N Engl J Med

4. The Diabetes Control and Complications Trial Research Group. The effect of intensive treatment of diabetes on the development and progression of long-term complications of insulin-dependent diabetes mellitus. . 1993;329:977-986.

Lancet.

5. UK Prospective Diabetes Study (UKPDS) Group. Intensive blood-glucose control with sulphonylureas or insulin compared with conventional treatment and risk of complications in patients with type 2 diabetes (UKPDS 33). 1998:352:837-853.

Diabet Med.

6. Donnan PT, MacDonald TM, Morris AD. Adherence to prescribed oral hypoglycaemic medication in a population of patients with type 2 diabetes: a retrospective cohort study. 2002;19:279-284.

Diabetes Care.

7. Boccuzzi SJ, Wogen J, Fox J, Sung JC, Shah AB, Kim J. Utilization of oral hypoglycemic agents in a drug-insured US population. 2001;24:1411-1415.

Clin Ther.

8. Brown JB, Nichols GA, Glauber HS, Bakst A. Ten-year followup of antidiabetic drug use, nonadherence, and mortality in a defined population with type 2 diabetes mellitus. 1999;21:1045-1057.

J Clin Outcomes Manage.

9. Stephens JM, Blak BT, Gold KF, Botteman MF, Pashos CL, Walkinshaw CS. Persistence patterns with oral hypoglycemic agents in type 2 diabetes. 2002;9: 491-499.

Noncompliance With Medications: An Economic Tragedy With Important Implications for Health Care Reform.

10. The Task Force for Compliance. Revised April 1994. Available from: Administrator, The Task Force for Compliance, 2300 North Charles Street, Suite 200, Baltimore, MD 21218.

11. National Pharmaceutical Council. Emerging Issues in Pharmaceuical Cost Containment. Levy R, ed. Reston, Va: National Pharmaceutical Council; December 1992;2(3):1-16. Available through: National Pharmaceutical Council (703-620-6390).

Society of Actuaries.

12. Cummings R, Cameron B. A comparative analysis of claimsbased methods of health risk assessment for commercial populations. May 24, 2002. Available at: http://www.soa.org/sections/riskadjfinalreport1.pdf. Accessed November 10, 2003.