Jonathan R. Bouchard, MS, RPh

This study evaluated the usefulness of the Diabetes Complications Severity Index (DCSI) in assessing healthcare resource utilization (HRU) and costs among Medicare Advantage plan members diagnosed with type 2 diabetes mellitus (T2DM).

A retrospective cohort study of medical and pharmacy claims of 333,576 Medicare members aged 18 to 89 years with ≥1 medical claim with primary diagnosis or ≥2 medical claims with secondary diagnosis, of T2DM (

International Classification of Diseases, Ninth Revision, Clinical Modification

 code 250.x0 or 250.x2) during the period of January 1, 2010, to December 31, 2011.

DCSI was assessed concurrently with HRU and healthcare costs (total, medical, and pharmacy). The cohort was subdivided into 6 DCSI groups: DCSI = 0 (no complications) through DCSI = 5+ (≥5). Associations of complication severity with HRU and costs of care were summarized using regression models.

A 1-point increase in DCSI was associated with a $2744 increase in total costs (a $2480 increase in medical costs plus a $264 increase in pharmacy costs). Increasing DCSI was associated with greater use of inpatient and emergency department (ED) services. Among the higher complications subgroups, there were greater representations of older patients, men, and cases of depressive disorders and hypoglycemia.

DCSI is useful for identifying Medicare plan members with T2DM who should be targeted for clinical programs. HRU and costs increased with DCSI severity. Increases in high-cost HRU, driven by inpatient and ED visits, suggest that preventing or delaying utilization of these services are essential to driving down costs in the T2DM population. Furthermore, high rates of depression and hypoglycemia warrant early screening and necessary treatment to improve patient outcomes.

The Diabetes Complications Severity Index is useful for predicting direct costs and healthcare resource utilization (HRU) among Medicare Advantage patients with type 2 diabetes mellitus (T2DM) and for identifying at-risk subgroups to which preventive therapies and interventions can be targeted.

T2DM complications pose excess burdens on costs and HRU, particularly for members with more severe complications. Preventing or delaying use of these services can reduce costs.

Studies of associations between pharmacy utilization and medical outcomes can clarify whether small increases in drug costs may result in total cost savings.

Depression and hypoglycemia should be addressed early in the treatment of T2DM.

Adults 65 years and older make up the highest proportion of patients in the United States living with type 2 diabetes mellitus (T2DM).

Healthcare resources used by elderly adults account for nearly 60% of diabetes-related healthcare expenditures, and these costs are largely absorbed by the Medicare program.

 The cost of care for T2DM is expected to increase in the coming decades as the US population continues to age and live longer.

Strategies to decrease T2DM-associated morbidity, mortality, and costs are warranted.

Individuals of all ages who are diagnosed with T2DM are at increased risk for developing a range of medical complications including retinopathy, nephropathy, and cardiovascular disease.

 A significant proportion of healthcare resource utilization (HRU) and costs are for T2DM-related complications.

The Diabetes Complications Severity Index (DCSI)

 measures the severity of diabetes-related complications, and predicts the related healthcare utilization and costs in managed care patient populations.

 A DCSI score ranges from 0 to 13 based on the presence and severity of 7 categories of complications: cardiovascular disease, nephropathy, retinopathy, peripheral vascular disease, stroke, neuropathy, and metabolic complications. Each complication category is scored on a scale from 0 to 2, determined by the presence and severity of the complications (0 = not present, 1 = some abnormality, and 2 = severe abnormality, except for neuropathy which is scored on a 2-point scale [0 = not present, 1 = abnormal]).

Complications are identified as abnormal or severe based on 

) codes from medical records or medical claims, as done in previous studies using the DCSI as a measure.

 Within the retinopathy category, for instance, an 

diagnosis of background retinopathy (362.01) would be scored as “abnormal” (score of 1), whereas a diagnosis of blindness (369.xx.00-.99) would be scored as “severe” (score of 2).

The original DCSI was calculated using patients’ pharmacy, medical, and laboratory data. A more recent, modified version of the DCSI omitting laboratory claims from the nephropathy category has demonstrated similar effectiveness as the original DCSI in estimating diabetes severity and in predicting hospitalizations. In addition, the modified DCSI is a good predictor of healthcare costs.

 Furthermore, this DCSI version is superior at predicting concurrent healthcare costs as compared with predicting future costs.

Chang and colleagues constructed the modified version of the DCSI using administrative claims from 7 health insurance plans,

demonstrating its effectiveness in predicting healthcare costs when the predictor (DCSI) and outcomes (healthcare costs) were measured concurrently.

They concluded that healthcare plans could reduce costs by employing the DCSI to identify and target patients who were the most in need of disease management programs.8 The authors acknowledged, however, that 80% of participants had a DCSI score of 0 (no diabetes complications); thus, their study sample had limited variability in terms of morbidity. A study using a sample with varying levels of illness could allow for a more precise identification of patient groups to target for intervention based on DCSI scores.

This study was designed to describe the demographic and clinical characteristics of a large cohort of Medicare Advantage with Prescription Drug (MAPD) coverage health plan members diagnosed with T2DM, and to evaluate the association of DCSI with HRU and costs in this cohort. The modified DCSI8 was used in this study to determine complications severity.

In this retrospective study involving MAPD health plan members diagnosed with T2DM, the association of DCSI with HRU and costs was analyzed. The data sources included member enrollment, medical, and pharmacy data generated from the claims database of a large national health plan (Humana). This analysis was part of a larger study, which was approved by an independent Institutional Review Board.

The study sample was composed of Humana members enrolled in an MAPD plan in 2010 and 2011. Members with T2DM were identified via 

 diagnostic codes. Members’ medical and pharmacy claims were assessed during a 24-month study period (January 1, 2010, to December 31, 2011) to determine the HRU and cost of care for members, stratified by their level of T2DM complications severity. The severity of complications was quantified using the modified DCSI,

 which in turn was assessed concurrently with annual HRU and costs over the study period.

Eligible study participants were: 1) continuously enrolled for at least 24 consecutive months; 2) aged 18 to 89 years as of January 1, 2010; and 3) diagnosed with T2DM (defined as having at least 1 medical claim with a primary diagnosis, or at least 2 medical claims with a secondary diagnosis, of T2DM [

 code 250.x0 or 250. x2]). Members were excluded from the study if they had medical claims for gestational diabetes mellitus (

 code 250.x1 or 250.x3), or pregnancy (630.xx-679.xx or v22.x-v24.x) during the study period.

The primary outcome measures were all-cause HRU and costs, based on all medical and pharmacy claims regardless of diagnosis or medication therapeutic class. HRU was the quantity of outpatient, inpatient, and emergency department (ED) services captured from medical claims for the study period. Outpatient utilization included medical claims where the service location was a physician’s office or outpatient facility, and included physician office visits, procedures, and tests. Inpatient hospital utilization included medical claims originating from an inpatient hospital facility. Hospitalization claims with the same discharge and admit dates were considered a single hospitalization, and ED utilization was all medical claims where the service location was an ED. ED claims with the same service date as a hospitalization were counted as a hospitalization, as this is viewed as an admission occurring through ED triage.

All-cause healthcare costs (total, medical, and pharmacy) were measured for both study years. Medical and pharmacy costs were adjusted to 2011 dollars using the US Bureau of Labor Statistics’ Consumer Price Index for Medical Care. Medical costs were the sum of costs associated with medical claims where the place of service was an outpatient, inpatient, or ED facility. Pharmacy costs were the sum of costs associated with pharmacy claims. Total healthcare costs were the sum of medical and pharmacy costs.

The modified DCSI was used in this study to determine complications severity.8 The study sample was divided into 6 subgroups consisting of DCSI scores 0 (no complications) through 5+ (5 or more). The DCSI, which served as the independent variable, was assessed concurrently with HRU and costs.

The following demographic characteristics were controlled for when assessing the association of DCSI with HRU and costs: age (as of January 1, 2010), gender (male or female), race/ethnicity (white, black, Hispanic, or other), and regional assignment in 2010 (northeast, midwest, south, and west, based on the US Census Bureau assignment of states to geographic regions). Low-income subsidy status referred to members with limited resources and income below 150% of the US federal poverty threshold who were eligible for additional premium and cost-share assistance for prescription drugs.

The following clinical characteristics were controlled for in order to best isolate the relationship of DCSI with HRU and costs: chronic disease burden, antidiabetic medication use, hypoglycemic events, and depression. These clinical factors were chosen because they either are known or hypothesized to have a potentially significant impact on HRU and costs, and, with the exception of chronic disease burden, these factors are not captured in the DCSI. Chronic disease burden was measured with the unweighted RxRisk-V Score, a prescription claims-based comorbidity index

 that measures a patient’s level of co-occurring conditions by mapping pharmacy claims for drug classes and individual drugs via Medi-Span generic product identifier codes to specific disease states. It is important to note that the RxRisk-V score may be related to the DCSI itself, given that a patient could have been on medications to treat the respective T2DM complication.

Treatment by antidiabetic therapeutic classes was the proportion of members with at least 1 pharmacy claim during the 24-month period for a medication within 1 of the following respective medication classes: biguanides, sulfonylureas, thiazolidinediones, amylin agonists, meglitinides, alpha-glucosidase inhibitors, glucagon-like peptide-1 receptor agonists, dipeptidyl peptidase-4 inhibitors, bile acid sequestrants, dopamine-2 agonists, insulin, and combination oral antidiabetics. A hypoglycemic event was at least 1 outpatient claim with a diagnosis of hypoglycemia (

codes 251.0, 251.1, 251.2, and 250.8 [excluding claims with co-diagnoses of 259.8, 272.7, 681.xx, 682.xx, 686.9x, 707.1-707.9, 709.3, 730.0-730.2, and 731.8 in the same visit or hospitalization]) in any position, or at least 1 ED claim with a diagnosis of hypoglycemia in the primary position.

 A diagnosis of depression, a common comorbidity of T2DM,

indicated that a member had at least 1 claim with an 

 diagnosis of depression (296.2x, 296.3x, 300.4, and 311).

Univariate descriptive statistics were used to describe the demographic and clinical characteristics of each DCSI cohort. Medication use was the mean number of antidiabetic classes. Place of service (outpatient, inpatient, and ED) all-cause HRU was the mean (SD) number of healthcare visits calculated on an annual basis. All-cause HRU included counts and percentages. Utilization of a specific type of service would be reported as a dichotomous status variable indicating the proportion of total members within each DCSI cohort using the specific type of service. Chi-square analyses, ANOVA pairwise comparisons, and t tests were used to determine differences in characteristics across DCSI groupings.

Generalized linear models (GLMs) were used to test the association of DCSI with HRU and costs. GLMs were used because HRU and costs had skewed distributions (ie, a small number of members with very high values). Separate models were generated for each outcome, adjusting for demographic and clinical characteristics. HRU measures included numbers of outpatient visits, inpatient stays, and ED visits; cost measures included total healthcare, medical, and pharmacy costs. HRU and cost measures were averaged over the 2-year period.

A GLM with negative binomial distribution and log link was employed for HRU measures, whereas a GLM with gamma distribution and log link function was applied to cost measures. The adjusted annual HRU/costs between DCSI groups were summarized and compared using regression analysis. These analyses were performed using SAS Enterprise Guide 5.1 statistical software (Cary, North Carolina); statistical significance was determined at the .05 level. All analyses were rerun on the subset of MAPD patients 65 years and older exclusively (not shown), which composed approximately 85% of the study cohort; the results were found to be equivalent to those of the entire cohort. As a result, the data for the entire cohort, aged 18 to 89 years, were reported for the current study.

 contains the results of the demographic and clinical characteristics of the cohort, in aggregate and stratified by DCSI score. The 

shows how the sample was selected based on the specified exclusion/inclusion criteria. The final sample comprised 333,576 MAPD members who were an average 70.8 years old, 52.1% female, 80.2% white, resided primarily in the southern United States, and had significant comorbidities including a 19% rate of depression and a 7.1% rate of hypoglycemic events. As DCSI scores increased, the percentage of black and Hispanic members rose relative to white members. Members with higher DCSI scores tended to be older and to have higher RxRisk-V scores, indicating a greater overall chronic disease burden. The highest DCSI groups (4 and 5+) had higher percentages of males. Moreover, there were increased rates of hypoglycemic events and depression observed with increased DCSI scores. Also, hypoglycemic events approximately doubled from group 4 to group 5+, and overall rates of depression were high, even in the “no complications” group (DCSI = 0; 14.7%).

The unadjusted and adjusted all-cause 24-month HRU, stratified by DCSI group, are presented in 

. As Table 2 demonstrates, there was a significant trend toward increases in the mean number of all-cause outpatient, inpatient, and ED visits per member as DCSI scores increased. After controlling for the demographic and clinical covariates (Table 3), increasing DCSI was associated with increases in all HRU, especially inpatient and ED services. The percent change from a DCSI of 0 to 5+ was prominent across all 3 HRU domains, although the inpatient and ED services reached changes greater than 300%. The percent change from a DCSI of 4 to 5+ was greater than 50% for inpatient and ED services.

Similar to the findings for HRU, there was a positive association between DCSI and all-cause healthcare costs. A 1-point increase in DCSI was associated with a $2744 increase in total costs (a $2480 increase in medical costs plus a $264 increase in pharmacy costs). As shown in Table 3, the majority of the costs were medical, and the relative increases in medical costs associated with higher DCSI score were much greater than those seen in pharmacy costs.

The study’s findings suggest that T2DM complications pose an excessive burden on HRU and costs, and that such burden is incremental with increasing severity of complications. In this study population, an increase in DCSI score from 1 to 5+ corresponded to a 3-fold increase in inpatient service use and a nearly 5-fold increase in ED utilization. Similarly, increases in DCSI were parallel to 2-fold increases in total healthcare and medical costs and a smaller increase in pharmacy costs. This study provides evidence of the usefulness of the modified DCSI in estimating concurrent healthcare costs in a large and clinically diverse sample of Medicare Advantage members. Previous studies suggest that this index is effective for assessing complications severity and predicting healthcare costs and resource use.

 The current study supports those findings, but differs from previous work in that it assessed a wider range of DCSI levels, highlighted clinical and demographic trends associated with DCSI, and assessed these outcomes in Medicare patients, who comprise a sizable proportion of individuals with T2DM and whose HRU and costs are substantial.

Further fine-tuning the use of the DCSI among Medicare patients may assist health plans in making the most judicious use of limited resources for disease management and prevention.

A trend observed in this cohort was that members with the highest DCSI scores tended to be older and more often male. Older age may be a proxy for duration of disease, thereby reflecting the progression of diabetes over time and the onset of diabetes-related complications resulting from poor glycemic control.

 Studies indicate that women are at a higher risk for developing T2DM-related complications and mortality.

 The present study’s findings appear to conflict with other research; however, they do indicate that while higher percentages of men were in the most severe complications subgroups (4 and 5+), there were more women in DCSI categories 1 through 3. These trends suggest that while women with T2DM have overall higher rates of complications, greater representation of males in the highest DCSI groups may reflect interactions of gender with age, race/ethnicity, and other demographic factors.

Two additional trends observed were that members with higher DCSI scores were more likely to reside in the southern United States and as DCSI complications increased, the percentage of black and Hispanic members increased relative to white members. A correlation may exist here as recent studies indicate an increasing prevalence of T2DM among ethnic minorities,

 as well as a higher prevalence of the disease in the southern United States, a region which has a higher prevalence of high-fat diets and sedentary lifestyles.

 Barker and colleagues analyzed nationwide, state-level 2007 and 2008 survey data and identified a “diabetes belt,” which refers to a geographic region situated mainly in the southern United States in which diabetes is much more prevalent compared with other parts of the country.

Residents of this belt were more likely to have lower levels of education, be sedentary, have higher obesity rates, and be black.

This study’s results suggested that depression and hypoglycemic events, common comorbidities of T2DM,

 were present in elevated proportions among the higher DCSI patient groups. Given the high prevalence of depression among the study cohort with no complications, it is imperative that depressive symptoms be identified and treated early among patients with T2DM. Further studies are needed to understand the nature of depression and hypoglycemic events that coexist with T2DM, since these 2 morbidities are associated with worse outcomes and higher healthcare costs.

 For example, studies could seek to understand and address diagnoses of depression that are directly related to T2DM-related distress, as opposed to preexisting depression that could itself be a pathway to obesity-related T2DM, possible future poor self-management,

and the consequent hastening of the development of complications. With regard to hypoglycemic events, more research is warranted on identification, prevention, and patient education.

Hypoglycemic events may contribute to the worsening of complications and increase the risk of acute cardiovascular events and death in older patients with T2DM.

The incidence of T2DM has increased significantly in the past 2 decades due to a surge in obesity, poor dietary habits, sedentary lifestyles, and the aging of the population.

 In addition, smoking, depression, and stress are risk factors that may act in synergy with diet and inactivity to increase the incidence of T2DM and its burden. Lifestyle interventions emphasizing weight loss, a healthier diet, and physical activity can result in a substantial decrease to the incidence of T2DM, particularly among patients 60 years and older, who are at greater risk of developing T2DM.

Health plans should implement comprehensive prevention programs to address the increasing incidence of T2DM and its complications. These efforts may include campaigns to increase member awareness of risk factors, screenings to identify at-risk groups, and increasing the reach of prevention programs focused on dietary patterns, physical activity, smoking, and depression.

The current study has several potential limitations that require acknowledgement. Because this study was cross-sectional in nature, study subject identification did not require a baseline period prior to the 24-month period of examination. This limitation may have resulted in possible imbalances in follow-up times post diagnosis, thereby affecting the calculation of the DCSI. Interpretation of the study results should account for this limitation. Typical of research using administrative claims data, this study was also subject to limitations including potential errors in claims coding and a lack of information in the database on health behaviors, indirect costs, and cash purchases of healthcare services. Causal inferences cannot be made from the results as this was an observational investigation of the association of DCSI with clinical and economic outcomes. This study’s data were obtained from a single health insurance company. Although Humana is a large national health plan with members from various geographic regions, the results may not be generalizable to the overall US population or to subpopulations within certain geographic regions of the United States. Moreover, the results may not be generalizable to all Medicare populations due to differences in benefit structure of MAPD and non-MAPD health plans.

T2DM complications pose an excess burden on HRU and costs. The DCSI is a useful tool for predicting direct healthcare costs and HRU, and for identifying at-risk patient groups to whom preventive therapies and interventions can be aimed. Further investigation into the association between pharmacy utilization and medical outcomes could elucidate whether relatively small increases in drug costs (through improved adherence or use of additional therapies) may result in total cost savings. Preventing or delaying the use of high-cost services, such as inpatient hospitalizations and ED visits, can play an important role in driving down costs in the diabetic population. Health plans may consider whether the construction of algorithms incorporating these demographic and clinical characteristics would provide added insights as to which patient groups should be offered more support in controlling their T2DM symptoms to prevent the onset or worsening of complications.

The authors wish to thank Mary Costantino of Comprehensive Health Insights and Tulay Cushman of Novo Nordisk for their assistance with medical writing and for their critical reviews of earlier drafts of this manuscript.

From Comprehensive Health Insights, Inc (LHF, CM, SLS, NCP), Louisville, KY; Cigna (YL), Columbia, SC; Magellan Health Services (DN), Glen Allen, VA; Humana Inc (YM, JBa), Louisville, KY; Novo Nordisk Inc (JBo), Princeton, NJ.

 This study was sponsored by Novo Nordisk Inc.

Drs Hazel-Fernandez, Moretz, Patel, and Slabaugh are employed by Comprehensive Health Insights, a wholly owned subsidiary of Humana; Comprehensive Health Insights was paid by Novo Nordisk for the development and submission of this manuscript. Mr Bouchard is a Novo Nordisk employee and stockholder. Drs Li and Nero were employed by CHI at the time the study data were collected and analyzed. Drs Nero and Meah and Ms Baltz report no relationship or financial interest with any entity that would pose a conflict of interest with the subject matter of this article. The study sponsor, Novo Nordisk, participated in the study design, interpretation of the results, and in the progress and critical revisions of the manuscript.

 Concept and design (LHF, DN, CM, SLS, YM, NCP, JBo); acquisition of data (YL, DN); analysis and interpretation of data (LHF, YL, DN, CM, SLS, YM, JBa, NCP, JBo); drafting of the manuscript (LHF, CM, JBa, NCP); critical revision of the manuscript for important intellectual content (LHF, YL, DN, CM, SLS, YM, JBa, NCP, JBo); statistical analysis (YL, DN, JBo); obtaining funding (JBo); administrative, technical, or logistic support (CM, NCP, JBo); and supervision (CM, NCP, JBo).

 Leslie Hazel-Fernandez, PhD, MPH, Comprehensive Health Insights, Inc, 515 W Market St, 7th Fl, Louisville, KY 40202. E-mail: 

1. Boyle JP, Thompson TJ, Gregg EW, Barker LE, Williamson DF. Projection of the year 2050 burden of diabetes in the US adult population: dynamic modeling of incidence, mortality, and prediabetes prevalence. 

2. American Diabetes Association. Economic costs of diabetes in the U.S. in 2012. 

Articles

This retrospective cohort study in a Medicare Advantage population posits that type 2 diabetes mellitus complications pose an excess burden on healthcare resource use and related costs.

Relationship of Diabetes Complications Severity to Healthcare Utilization and Costs Among Medicare Advantage Beneficiaries

 To assess the relationship between cost sharing and adherence to antidiabetic medications in patients with type 2 diabetes and to examine the relationship between medication adherence and outcomes, including complication rates, medical service utilization, and workplace productivity measures.

A retrospective, cross-sectional study analyzing the healthcare experience of patients with type 2 diabetes on oral antidiabetic medication (OAD) with or without insulin (n = 96,734) and patients on OAD only (n = 55,356) with employer-sponsored insurance in the 2003-2006 MarketScan Database.

Using a 2-stage residual inclusion model, the first stage estimated the effects of cost sharing on adherence to antidiabetic medications in an 18-month time frame (January 2003 through June 2004). Adherence was determined from the percentage of days covered. The second stage estimated the effects of adherence on complication rates (eg, retinopathy, neuropathy, peripheral vascular disease), medical service utilization rates, and measures of productivity (absence days and short-term disability days) in the subsequent 2 years (July 2004 through June 2006).

A $10 increase in the patient cost-sharing index resulted in a 5.4% reduction in adherence to antidiabetic medications for patients on OAD only and a 6.2% reduction in adherence for patients on OAD with or without insulin. Adherence was associated with lower rates of complications (eg, amputation/ulcers, retinopathy) and also was associated with fewer emergency department visits and short-term disability days.

Medical plans, employers, and policy makers should consider implementing interventions targeted to improve antidiabetic medication adherence, which may translate to better outcomes.

	This retrospective, cross-sectional study investigated the relationships between cost sharing and adherence to antidiabetic medication, and among adherence, diabetes-related complications, and utilization measures.

		Higher levels of patient cost sharing for antidiabetes medications were associated with lower levels of adherence (percentage of days covered) to antidiabetic medications. For example, among patients on oral antidiabetic medications only, those whose cost-sharing amount was $10 higher had adherence rates that were 5.4% lower.

		Antidiabetic medication adherence was associated with lower rates of many diabetes-related complications (eg, amputation/ulcer and retinopathy), and with fewer inpatient visits, emergency department visits, and short-term disability days.

		Medical plans, employers, and policy makers should consider ways to promote adherence to medications, including reduced patient cost sharing.

	Approximately 23.6 million individuals in the United States have diabetes, a serious metabolic disorder that affects all systems of the body, particularly the neurologic and cardiovascular systems.

 Proper medication management among patients with diabetes is a key component of preventing diabetes-related complications.

	Inadequate medication adherence among patients with chronic illness has been associated with increased healthcare utilization, costs, and risk for adverse health outcomes.

 Encinosa and colleagues studied adherence among privately insured patients with diabetes on oral antidiabetic medication (OAD), finding that improved adherence was associated with decreased hospitalization and emergency department (ED) costs.

 Similarly, Sokol and colleagues found that a high level of medication adherence among privately insured patients with diabetes was associated with decreased condition-specific medical spending.

 Regarding health outcomes, Ho and colleagues found that nonadherence to OAD among privately insured patients with diabetes was associated with an increase in glycosylated hemoglobin (A1C) levels, all-cause hospitalizations, and all-cause mortality.

 Such evidence underscores the value of medication adherence to patient health.

	These findings also suggest that important differences may exist between patients who are adherent and those who are nonadherent, and these potential selection effects may affect the relationship between adherence and outcomes. Thus, rigorous methods must be adopted to account for possible endogeneity in the relationship between adherence and outcomes.

	Barriers to medication adherence such as high patient cost sharing for prescription drugs inhibit patient adherence.

 Few studies have linked the effects of cost sharing to adherence, and subsequently, to outcomes. Hunt and colleagues measured the relationships between cost sharing and adherence and between adherence and glycemic control among patients with diabetes, finding a positive relationship between cost sharing and A1C level.

 However, the ultimate effects of cost sharing on diabetes-related complications, through adherence, are not yet well known.

	This study contributes to the literature in several ways. First, we investigated the relationships between cost sharing and adherence to antidiabetic medication. Second, we addressed the relationship among adherence, diabetes-related complications, and utilization measures using rigorous methods. We assumeda more comprehensive view of the impact of medication adherence, also incorporating measures of productivity for employees with diabetes. We used 2-stage residual inclusion models to estimate these  relationships and produce consistent estimates.

	This study is based upon the Thomson Reuters MarketScan Database, 2002-2006, which represents the healthcare experience of more than 21 million enrollees with  employer-sponsored benefits annually. Inpatient medical, outpatient medical, and outpatient pharmacy claims, absenteeism data (dates of absence from work due to illness), and claims for short-term disability benefits were linked to enrollment information to create the analytic data set.

	A retrospective, cross-sectional study was conducted among patients aged 18 years or older with diabetes who used OAD (sulfonylureas, meglitinides, biguanides,  thiazolidinediones, or alpha-glucosidase inhibitors) and were continuously enrolled from January 1, 2002, through June 30, 2006. Patients were included if they filled at least 2  prescriptions for an antidiabetic agent from January 1 through June 30, 2003, and had a diagnosis of diabetes (

International Classification of Diseases, Ninth Revision, Clinical Modification 

] code 250.xx) as indicated by 1 ED visit, 1 inpatient admission, or 2 outpatient visits at least 30 days apart. Patients also were required to have the first  observed fill for an OAD occurring in 2002 or the first quarter of 2003. Patients with a diagnosis of type 1 diabetes or gestational diabetes (

 code 648.8x) were excluded. The final samples consisted of 96,734 patients with type 2 diabetes on OAD with or without insulin in the full sample and 55,356 patients who filled an OAD prescription only (no insulin). We created 2 additional subsets of patients who were employees with available absenteeism data (n = 1829; n = 1040 OAD only) and short-term disability data (n =   3027; n = 1753 OAD only).

	Adherence was measured as the percentage of days covered (PDC), expressed as the percentage of days with an antidiabetic medication on hand during the 18-month time period (January 2003 through June 2004).

 The PDC was first calculated for each therapeutic class (ie, sulfonylureas, meglitinides, biguanides, thiazolidinediones, alpha-glucosidase  inhibitors, insulin) based on prescription drug claims, using the fill date and the intended days supply of medication from each claim. Information from prescriptions filled in the fourth quarter of 2002 provided prescription utilization information for early 2003. If a patient refilled the same medication before the end of the days supply from the previous prescription fill, then the days supply for the new prescription was appended to the end of the previous fill. If a patient switched medication within a therapeutic class, the remainder of the previous prescription was discarded and coverage commenced with the supply of the new medication. Days of hospitalization were counted as adherent days (to at least 1 antidiabetes medication), as some of the factors that influence patient medication adherence may change during days of hospitalization. For insulin prescriptions, where the days supply may not indicate the full extent of coverage, we multiplied the days supply of each fill by 1.5.

 We produced results using various algorithms for insulin adherence and adherence to OAD alone for those patients who used insulin with no material change in findings.

	After calculating the number of days that medication was on hand in each medication class, we compiled the information from all therapeutic classes into a single measure and measured adherence based on whether a patient had any antidiabetes medication on hand on each day. Consistent with previous studies, patients were classified as adherent to antidiabetic medications if the PDC equaled or exceeded 80%,

 a threshold at which clinical benefits are most likely to occur.

	Complications were recorded using dummy variables that indicated the complication occurred on a claim from July 2004 through June 2006. Codes for each complication are noted in 

. Utilization events for July 2004 through June 2006 included physician office visits, ED visits, and inpatient admissions, coded as the number of events in the 24-month  period. Patients with productivity data available (with days of absence and short-term disability reported) were included in the productivity analysis, even if zero days were reported.

	Patient cost sharing for antidiabetic medications was measured, using a cost-sharing index created for each employer/plan combination. The index was based on the average  cost-sharing amount (ie, copayment, coinsurance) per prescription (standardized to a 30-day supply) for brand and generic drugs in each antidiabetic medication class. The index  aggregated the brand and generic copayments using weights, developed from the overall proportion of utilization of brand and generic drugs within each medication class.

 If we were to examine cost sharing for the medications filled by each patient, this method might introduce selection bias, as more adherent consumers might be selecting medications (eg, generics) on the basis of cost-sharing amounts. Thus, aggregating cost sharing into levels for each employer/plan combination reduced the effects of selection bias related to actual, individual-level cost sharing and related choices. The coinsurance flag indicated which plans used coinsurance (a percentage of the total payment) for   prescription drug cost sharing versus a flat fee copayment. Any mail-order use in the past year, which can be associated with higher levels of adherence,

 also was indicated. As outpatient physician visit cost sharing has been associated with reductions in the use of prescription drugs,

 the per-visit amount was included in the models.

	Other explanatory variables associated with medication adherence were included in the models. Patient-level sociodemographic characteristics included sex, age, census region,  urban residence, and median income in the patient’s area of residence (by ZIP code) from the US Census files. An indicator for employee status (vs spouse/dependent) was  included, along with plan type (eg, health maintenance organization, preferred provider organization). An indicator for a visit to a related specialist (ie, cardiologist, endocrinologist) in the past year accounted for disease severity and possible differences in practice patterns.

	To account for differences in health status, the following explanatory variables were incorporated into the models as well. Because newly diagnosed patients may have different utilization patterns than patients with existing disease,

 a flag indicated whether a patient was newly diagnosed with type 2 diabetes in 2003. Scores on the Charlson Comorbidity Index, a numeric scale reflecting the risk of death or serious disability in the next year based on the presence of a diagnosis for 1 of 19 conditions (eg, diabetes, heart disease,  cancer) in the 12-month preindex period, were included.

 The Charlson Comorbidity Index is associated with cost and other utilization measures such as length of stay and readmissions.

 The total patient out-of-pocket burden for chronic conditions (medical and drug) in the preindex year also may reflect lower health status. For prescription drug claims, chronic out-of-pocket spending was classified per claim by using the National Drug Code and Redbook for drug claims (claims with a drug type of “maintenance”) and for  medical claims by using the chronic conditions listed in the Charlson Comorbidity Index (as identified by a nosologist).

	We analyzed the relationships among patient cost sharing for antidiabetic medication, adherence, and diabetes-related events or outcomes by first estimating the relationship  between patient cost sharing for antidiabetic medication and adherence to antidiabetic medications (controlling for covariates) during an 18-month time frame (January 2003 through June 2004). We then estimated the relationship between adherence in the initial time frame and outcomes, controlling for covariates in the subsequent 2 years, July 2004 through June 2006.

	Several hypotheses based on existing evidence that higher patient cost sharing reduces medication adherence

 informed our approach. We anticipated that adherence would be associated with lower rates of ED visits and hospital admissions,

 and would reduce the risk of diabetes-related events or outcomes, particularly for sequelae likely to develop in the near term.

 For employees with productivity data, we hypothesized that adherence might have translated into higher levels of productivity.

	A 2-stage model was used to estimate the relationship between antidiabetic medication (OAD with or without insulin) adherence and health outcomes.

. A logistic regression model was estimated for adherence (PDC >80%) in the 18-month time period (January 2003 through June 2004). Explanatory variables included sociodemographic variables, employee status, health plan type, physician specialist visit, health status, and cost sharing.

Second Stage: Adherence and Health Outcomes.

 In the second stage, generalized linear models estimated events between July 2004 through June 2006. All models included the first-stage explanatory variables, except for the cost-sharing variables. The cost-sharing variables (cost-sharing index, coinsurance flag, office visit cost sharing, and previous  mail-order use) served as instruments in the first-stage equation and thus were not included in the second stage. Using a 2-stage residual inclusion model, actual patient   adherence and the residual from the first-stage model also were included in the second-stage model. The residual was included to produce consistent estimates in the nonlinear second stage and to adjust for any unobservable confounding in the second stage.

 The coefficient of the residual term in the second stage also allowed an indication of   endogeneity. If the coefficient of the residual is statistically significant, the inclusion of the residual term aids in reducing selection effects, making the estimates consistent. If the coefficient of the residual term is not statistically significant, consistent estimates are produced, possibly with a loss of efficiency.

	Each complication model was estimated using a logit link and a binomial family for the binary outcomes. Using 2 versions of the complication models to focus on the development of complications in the measurement period, complication models were estimated (1) controlling for previous complications by including indicator variables for the presence of each complication in 2002 and (2) for the subset of patients with no complications in 2002. The utilization and productivity models were estimated using a log link and a negative binomial model to reflect outcomes expressed as counts.

	We also estimated the reduced form of each of the outcome equations to determine the strength and direction of the association between copayments and each outcome.

	The characteristics of the full and OAD-only samples are presented in 

. The full sample was 46% female and had an average age of 52.2 years. The majority of the full sample lived in the South, and three-fourths lived in urban areas. The median income in the patients’ area of residence averaged about $44,000. More than two-thirds of patients were employees (67.5%), and the majority of patients were enrollees in a preferred provider organization. The mean cost-sharing index amount was $11.45. Seventeen percent of individuals were subject to coinsurance, more than one-quarter (28.5%) had used a mail-order pharmacy in the past year, and the average physician office visit copayment was  about $20. Similar demographics were observed among OAD-only users.

	Almost three-quarters of patients were adherent (PDC >80%) from January 2003 through June 2004. Complication rates for OAD-only users ranged from 1.4% for an acute myocardial infarction (AMI) to 7.3% for retinopathy. Complication rates were higher (2.0% for an AMI and 13.5% for retinopathy) for all OAD users. Patients visited the ED on average less than once and had inpatient admissions less frequently than ED visits in the 2-year time frame. Employees reported slightly more than 30 days absent. OAD-only users had  13.3 short-term disability days, and the full sample had 17.4 short-term disability days.

 presents results from the first-stage logistic regression estimates for adherence. Notably, as the level of prescription drug cost sharing (measured as the cost-sharing index) increased, adherence decreased. For all OAD users, the odds ratio (OR) was 0.974 (95% confidence interval [CI] = 0.970, 0.978). For OAD-only users, the OR was 0.978 (95% CI = 0.973, 0.984). Higher cost sharing for physician visits also was associated with lower levels of adherence, although the effects were not as large. For all OAD users, the OR was 0.996 (95% CI = 0.994, 0.997). For OAD-only users, the OR was 0.995 (95% CI = 0.993, 0.996). Patients with coinsurance had higher levels of adherence, as did patients who had previous mail-order use. All instruments were individually and jointly statistically significant in the first-stage (adherence) models (

Higher cost-sharing levels reduced adherence to antidiabetic medications in patients with type 2 diabetes.

Jonathan R. Bouchard, MS, RPh

Articles

Relationship of Diabetes Complications Severity to Healthcare Utilization and Costs Among Medicare Advantage Beneficiaries

Cost Sharing, Adherence, and Health Outcomes in Patients With Diabetes