John J Sheehan, PhD

This study examined the relationship between medication adherence, cost sharing measured as out-of-pocket spending, and total annual spending in Medicare beneficiaries with type 2 diabetes (T2D) to evaluate whether pharmacy cost-sharing programs have the potential to decrease adherence. These programs may unintentionally increase the risk of medical complications and may result in higher spending overall.

This retrospective study used 2006 to 2009 Medicare claims data. The sample included patients 65 years or older with T2D (at least 1 claim with International Classification of Diseases, 9th Revision, Clinical Modification codes 250.x0 and 250.x2 and at least 1 antidiabetes drug claim). 

Medication adherence was measured as proportion of days covered over the first 12 months of observation. Spending and adherence outcomes were defined in deciles. 

The sample included 12,305 patient-year observations. Pharmacy spending for patients in the most adherent (10th) decile was 59% higher than that for patients in the least adherent (1st) decile ($4839 vs $3046). Yet, patients in the 10th decile had 49% lower total ($12,531 vs $24,468) and 64% lower medical spending ($7692 vs $21,421) than patients in the 1st decile. Greater out-of-pocket spending was correlated with lower adherence and higher total and medical spending. 

 This study describes a widespread variation in medication adherence, pharmacy cost sharing, and medical spending in a sample of Medicare beneficiaries with T2D. We found that lower adherence was correlated with higher cost sharing in the Medicare population, perhaps because of unobserved confounding factors. However, the existing literature on patients with employer-sponsored insurance suggests some of this correlation may be indicative of causal relationships.

		Increased cost sharing could decrease the likelihood that patients with type 2 diabetes will adhere to their medication and increase diabetes complications and total diabetes spending. 

		Using Medicare claims data (2006-2009), we examined the relationships between medication adherence, cost sharing measured as out-of-pocket spending, and total annual spending to evaluate whether pharmacy cost-sharing programs have the potential to decrease adherence. 

		We found a widespread variation in medication adherence, pharmacy cost sharing, and medical spending, and that lower adherence is correlated with higher cost sharing in the sample of Medicare beneficiaries with type 2 diabetes.

In 2014, nearly 30 million Americans (9.3%) were living with diabetes, including an estimated 11.2 million who were 65 years or older.

 In 2012, the total national spending on diabetes was $245 billion, including $176 billion in direct medical spending. Patients 65 years or older use more healthcare services than younger patients and incur the majority of costs associated with diabetes. Ultimately, the Medicare program—in which 28% of beneficiaries have diabetes&mdash;bears most of this burden.

 Interventions targeted at individuals using the most resources and who have the greatest expenditures have the potential for a disproportionate impact on spending.

Several studies evaluating the economic burden of diabetes have estimated that antidiabetes agents and supplies account for 12% of total healthcare spending.

 Payers have implemented various cost-sharing schemes to reduce pharmacy spending, including incentive-based formularies, coinsurance or co-payments, and benefit caps.

 Research consistently shows that higher co-payments or coinsurance reduce pharmaceutical utilization and adherence among patients with chronic diseases like type 2 diabetes (T2D), heart failure, and schizophrenia,

 and that reducing co-payments in employer-sponsored insurance plans can increase adherence.

Although cost sharing may reduce overall pharmacy spending, it could also have consequences for patient health and result in unintended increases in nonpharmacy healthcare spending, thereby offsetting the savings on pharmaceuticals. The limited existing research shows that cost sharing increases emergency department visits and hospitalizations, especially among chronically ill patients.

 The impact of cost sharing on therapeutic adherence is particularly relevant for patients with diabetes, given that improved adherence is associated with a greater likelihood of reaching glycated hemoglobin (A1C) targets—the primary measure of effective disease management

—and cost sharing has increased since the introduction of Medicare Part D.

 By some estimates, a 1% increase in antidiabetes prescription utilization would reduce medical spending for Medicare beneficiaries with T2D by 0.83%.

Despite the large overarching financial burden of T2D, few studies have examined the extent to which pharmacy and medical spending contribute to total diabetes-related spending.

 This retrospective claims study of a Medicare population examined the relationship between adherence, cost sharing as measured by out-of-pocket (OOP) spending, and total annual spending.

The study used claims data for a 20% random sample of age-eligible Medicare fee-for-service beneficiaries for the years 2006 to 2009. Data were acquired from the CMS Medicare Research Identifiable Files, which provide medical and pharmacy claims, including diagnoses, procedure codes, and measures of charges and spending. Overall spending in any category (total healthcare, medical, or pharmacy) equaled the sum of the patient OOP component and the payer-covered component.

A sample was created using the following criteria: 1) Medicare beneficiaries 65 years or older; and 2) with at least 1 claim with an 

International Classification of Diseases, 9th Revision, Clinical Modification

codes 250.x0 and 250.x2) and at least 1 claim for an antidiabetes medication in the period following the first observed claim with diabetes diagnosis. The index date for patients in the sample was the date of the first observed claim with a T2D diagnosis. Only patients 65 years or older at the time of the first observed claim with a diabetes diagnosis were included. The sample included patients with a T2D diagnosis or antidiabetes medication use in the year prior to the first observed claim in the database if the patient’s index date was less than 1 year after the date of Medicare enrollment. Patients with a follow-up period of less than 1 year after the index date or those who were pregnant or less than 6 months post partum were excluded.

Adherence was measured as the proportion of days covered (PDC), a well-established method that has been used for several decades to evaluate adherence in diabetes and other chronic conditions (eg, cardiovascular disease, dyslipidemia, hypertension).

 PDC was calculated as the total number days when a patient had medication on hand during the observation window (ie, date of first claim to exhaustion of last prescription, segmented into calendar years).

This retrospective study used descriptive analysis to evaluate the extent to which patients’ health status differed across deciles of spending. In particular, we calculated the prevalence of diabetic comorbidities, complications, and drug therapy type by decile of total annual spending. To investigate the relationship between medication adherence and spending, average pharmacy, medical, and total spending were calculated within each adherence decile. To examine the relationship between cost sharing—as measured by OOP spending (total, medical, and pharmacy)&mdash;and medication adherence, average OOP pharmacy, medical, and total spending were calculated within each decile of adherence.

Dividing the sample into deciles created enough bins to characterize the spending distribution. Using deciles to map the distribution of total, pharmaceutical, and medical spending among Medicare beneficiaries with T2D also enabled the identification and description of the degree of skew in annual spending distribution.

Based on the inclusion and exclusion criteria, the sample included 12,305 patients with a mean follow-up period of 3.2 years and a mean age of 74.4 years. The sample was 64% female and 36% male.

Distribution of Total, Medical, and Pharmacy Spending

Considerably higher average total and medical spending were observed for patients in the upper deciles of total spending compared with patients in the lower deciles, but the between-decile difference in pharmacy spending for patients was not as substantial (Figure 1). The average annual total spend for patients in the lowest spending (1st) decile was $794; this increased by nearly 100-fold, to $78,641, for patients in the highest spending (10th) decile. The average annual medical spend for patients in the 1st decile was $417 and increased more than 170-fold, to $71,203, for patients in the 10th decile. However, the average annual pharmacy spend for patients in the 1st decile was $377 compared with $7439 for patients in the 10th decile, a less than 20-fold increase.

Patient Descriptive Statistics and Health Status by Spending

Patient age and sex were similar across all total annual deciles of spending; however, patients in the upper deciles were substantially more likely to have comorbid conditions, including congestive heart failure (74% in the 10th decile vs 17% in the 1st decile), cardiac arrhythmias (68% vs 24%), complicated hypertension (57% vs 17%), or moderate/severe renal impairment (18% vs 1%), among others. The use of certain antihyperglycemic drug therapies was also more prevalent in the upper deciles of total spending. Compared with patients in the 1st annual spending decile, patients with T2D in the 10th decile had a nearly 5-fold higher probability of using long- or rapid-acting insulin to treat their diabetes (31%-34% vs 3%-7%, respectively). Dipeptidyl peptidase-4 inhibitors, meglitinide, and thiazolidinedione use was also higher for patients in the upper deciles of total annual spending, whereas metformin (a biguanide) use was less common in the 10th versus 1st decile (31% vs 53%). Use of sulfonylureas and glucagon-like peptide-1 receptor agonists did not vary significantly across the deciles of total annual spending.

Mean PDC was divided into deciles and ranged from 0.20 for patients in the 1st decile to 1.00 for patients in the 10th decile. As shown in Figure 2, patients with greater medication adherence, as measured by PDC, had lower average annual medical and total spending and only moderately increased pharmacy spending. For patients in the 1st decile of PDC (lowest adherence) to the 10th decile, average annual total spending decreased 49% (from $24,468 to $12,531), and average annual medical spending fell 64% (from $21,421 to $7692). By comparison, pharmacy spending between the 1st decile of PDC and the 10th decile increased 59% (from $3046 to $4839).

An examination of the relationship between total OOP spending and adherence found analogous results. Average annual pharmacy OOP spending for patients in the 1st to 10th PDC decile increased 67% (from $441 to $737), but average annual total OOP spending decreased 36% (from $2116 to $1344), and average annual medical OOP spending decreased 64% (from $1675 to $607).

Based on the present study’s findings, Medicare patients with T2D have highly skewed total annual spending; patients in the upper deciles of total spending account for a disproportionate share of aggregate total spending. As previous studies had similar findings,

 this analysis confirms that medical spending accounts for the vast majority of the skew in total spending on T2D treatment, and that pharmacy spending is a relatively small proportion of total spending.

An analysis of comorbidities and medication use according to total spending showed that patients in the highest deciles were more likely to have comorbid conditions and to use certain antidiabetes drugs (in particular, long-acting, short-acting, and premix insulins). It is likely that the differences by decile in the prevalence of diabetic comorbidities, complications, and therapy utilization in patients play a large role in driving healthcare spending variation. This finding is supported by existing studies.

This study also examined the distribution of total, medical, and pharmacy spending in Medicare patients with T2D based on their level of medication adherence. More adherent patients had substantially lower total spending and medical spending, but moderately increased pharmacy spending, suggesting that, for Medicare patients with T2D, lower adherence is related to increased spending in general. Increased spending may be the result of poorer outcomes, increased comorbidities, and a need for additional treatment, as medication adherence has been significantly associated with improved health status, fewer complications, and less medical resource utilization in patients with T2D.

This study corroborates previous research and suggests that increasing adherence among Medicare beneficiaries with T2D could generate significant cost savings.

 The use of pharmacy cost sharing in Medicare has gone up since the introduction of Medicare Part D, which extended pharmacy benefit coverage and increased pharmaceutical utilization,

 but the availability of generic drugs has prevented significant increases in OOP spending. If this cost-sharing trend continues, adherence may fall and any savings on pharmacy spending could be offset by increased medical spending.

 Recent proposals to limit annual OOP spending for Medicare beneficiaries would reduce the financial burden on high-spend Medicare beneficiaries with T2D, potentially improving medication adherence and health outcomes.

This study has certain limitations. The analysis was descriptive, with data randomly drawn from an administrative claims database representing the years 2006 to 2009. Although the levels of pharmacy and medical spending may have changed since 2009, our qualitative results and the correlations between cost sharing, adherence, and medical spending should not be impaired by the age of the data, as many similar studies have found the same relationships in other patient samples and time periods.

 Additionally, because the study drew from the Medicare database, and because only patients 65 years or older at the time of the first observed T2D diagnosis claim were included, the mean age of the sample (74 years) was higher than the average age of T2D diagnosis (54 years) in the United States.

 Last, only limited medical history is available for patients in claims data research, thereby increasing the potential for confounding, and the date of patients’ initial T2D diagnosis could not be determined with certainty.

This study has several implications. First, the study results demonstrates the existence of a skew in spending on T2D treatment. Medicare patients in the highest deciles of total spending had dramatically higher total and medical spending, clearly indicating that most spending on diabetes-related treatment is from medical—and not pharmacy&mdash;spending among a subset of high-cost patients. In addition, the study expands on available studies of patients with employer-sponsored insurance to demonstrate that improved adherence is related to reduced resource use and spending in Medicare beneficiaries with T2D and suggest that interventions aimed at improving adherence among high-cost patients have the greatest potential to reduce health expenditures. Most importantly, the study results confirm the negative connection established in the existing literature between cost sharing and medication adherence, which varied widely among Medicare beneficiaries with T2D.

 If pharmacy cost-sharing programs decrease medication adherence, initiatives that promote pharmacy cost sharing may unintentionally increase the risk of medical complications offsetting some (or in targeting cases, all) of the savings associated with lower pharmaceutical spending.&ensp;

 Precision Health Economics (JPM, JV-O, JS, DP), Los Angeles, CA; AstraZeneca (JJS, IK), Fort Washington, PA; University of California, Los Angeles (WY), Los Angeles, CA; University of Southern California (ALP), Los Angeles, CA.

 This study was supported by AstraZeneca.

 Dr MacEwan, Mr Sullivan, and Ms Peneva are employees of Precision Health Economics (PHE), which is a consultant for biotech/pharma companies and was compensated by AstraZeneca to perform the study. Dr Vanderpuye-Orgle was an employee of PHE during the time of the study and is currently employed by Amgen. Dr Yin is a consultant for PHE. Dr Sheehan has been employed by AstraZeneca, and Dr Kalsekar has been employed by and holds stock in AstraZeneca and Johnson & Johnson, which manufacture drugs for the treatment of diabetes. Dr Peters has been a consultant and grant recipient from numerous pharmaceutical companies including AstraZeneca, and receives editorial fees from Medscape.

 Concept and design (JPM, JJS, WY, JV-O, JS, IK, ALP); acquisition of data (JS, IK); analysis and interpretation of data (JPM, JJS, WY, JV-O, JS, DP, IK); drafting of the manuscript (JPM, WY, DP); critical revision of the manuscript for important intellectual content (JPM, JJS, WY, JS, ALP); statistical analysis (WY, JS); provision of patients or study materials (ALP); obtaining funding (JJS, IK); administrative, technical, or logistic support (DP, IK); and supervision (JJS, WY, JV-O, IK).

 Joanna P. MacEwan, PhD, Precision Health Economics, 11100 Santa Monica Blvd, Suite 500, Los Angeles, CA 90025. E-mail: joanna.macewan@pheconomics.com. 

1. National diabetes statistics report. CDC website. https://www.cdc.gov/diabetes/pubs/statsreport14/national-diabetes-report-web.pdf. Published October 24, 2014. Accessed January 21, 2015.

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

. 2013;36(4):1033-1046. doi: 10.2337/dc12-2625.

3. American Diabetes Association. Economic costs of diabetes in the US in 2007. 

. 2008;31(3):596-615. doi: 10.2337/dc08-9017.

4. Goldman DP, Joyce GF, Zheng Y. Prescription drug cost sharing: associations with medication and medical utilization and spending and health. 

5. Goldman DP, Joyce GF, Escarce JJ, et al. Pharmacy benefits and the use of drugs by the chronically ill. 

6. Sokol MC, McGuigan KA, Verbrugge RR, Epstein RS. Impact of medication adherence on hospitalization risk and healthcare cost. 

Articles

This study describes a widespread variation in medication adherence, pharmacy cost sharing, and medical spending. Increased cost sharing may decrease adherence and increase total diabetes spending.

The Relationship Between Adherence and Total Spending Among Medicare Beneficiaries With Type 2 Diabetes

In 2008, the FDA issued Guidance for Industry in Diabetes Mellitus: Developing Drugs and Therapeutic Biologics for Treatment and Prevention.

 The Guidance recommends that all new medications for the treatment of type 2 diabetes (T2D) be tested to ensure that they do not lead to an “unacceptable increase in cardiovascular risk” by including multiple cardiovascular (CV) endpoints in new trials or evaluating the frequency of CV events in ongoing or completed trials. The American Diabetes and American Heart Associations have also weighed in, issuing new guidelines on the prevention of CV disease (CVD) among patients with diabetes.

 Following the FDA recommendations, sponsors of all dipeptidyl-peptidase-4 inhibitors (DPP-4is) available in the United States designed at least 1 randomized controlled trial to evaluate the CV safety of these medications: saxagliptin (SAVOR),

 and linagliptin (CAROLINA and CARMELINA).

 Some of these CV outcomes trials (CVOTs) are complete (SAVOR, EXAMINE, and TECOS), while others are ongoing (CAROLINA and CARMELINA). 

The 5 CVOTs evaluated in this analysis differed considerably with respect to region, study size, primary outcomes, length of follow-up, and inclusion/exclusion criteria. Specific to enrollment criteria, the representativeness of a study, or the external validity, is the extent to which the results of the study apply to other and, or, larger populations.

 This analysis provides a quantitative assessment of the representativeness of the DPP-4i CVOTs populations, as a whole, and differences among the 5 trials. In this analysis, we used the National Health and Nutrition Examination Survey (NHANES) to determine the extent to which each CVOT is representative of the population of adults with T2D in the United States. We used 2 recent waves of NHANES data, 2009 to 2010 and 2011 to 2012, to estimate the proportion of Americans with T2D who would be eligible for enrollment in each of the 5 CVOTs.

This analysis used NHANES, a nationally representative survey designed to measure objective health data from the noninstitutionalized US population, combined with field surveys about health and health behavior. Within the “diabetes” component of the questionnaire, individuals are asked if they have been diagnosed with diabetes, whether they are on insulin, or whether they are “taking diabetic pills to lower blood sugar.” The NHANES prescribed medicine data is based on a combination of patient self-reports and examination of pill bottles. Patients also undergo a physical examination and complete a broad range of laboratory tests.

The target population of this analysis is adults (older than 18 years of age) with a diagnosis of T2D. We attempted to distinguish individuals with T2D from both pregnant women with gestational diabetes and individuals with type 1 diabetes (T1D).

Patients fulfilling all of the following criteria were included in the analysis: (i) Patients at least 18 years of age (ii) Patients with diagnosis of diabetes.

Patients fulfilling the following criteria were excluded from the analysis to limit the analysis to the target population of adults with T2D: (i) Patients younger than 18 years (ii) Patients who were pregnant (iii) Patients diagnosed with diabetes before age 30 years (iv) Patients currently on insulin and those who initiated insulin use within 1 year of diagnosis, indicating T1D.

 (v) Patients were excluded from the analysis if they had missing information on: a. Age b. Gender c. Other demographics d. Diabetes status e. Age of diagnosis and/or medication use f. Glycated hemoglobin (A1C) g. Body mass index (BMI) h. Smoking status i. Other CVOT inclusion/exclusion criteria variables.

From the pooled NHANES data for 2009-2010 and 2011-2012, we had a sample with non-missing data on all the variables required to determine eligibility in each of the 5 CVOTs for 3439 individuals aged 18 years and older. Overall, 654 of these patients reported having a diagnosis of diabetes. We identified 17 of these patients as having T1D, leaving 637 patients with T2D.

Using the NHANES sample we identified adults with diagnosed T2D. Next, based on the enrollment criteria for each CVOT, we identified individuals within the adult T2D sample who were eligible for each of the CVOTs based on published descriptions of the study designs.

The CVOT inclusion and exclusion criteria were based on patient’s A1C, age, cardiovascular event history, smoking status, renal function, blood pressure, blood cholesterol, body weight, liver function, and prescription medication use. From these counts, ie, the number of patients with T2D and the number of patients with T2D meeting the enrollment criteria for each CVOT, we estimated the share of all adults with T2D each CVOT represents.

Individual trials included enrollment criteria regarding the timing of prior CV events, which was not directly available in NHANES. The timing of these events was imputed based on patient responses to questions about their health history and laboratory files. To identify patients in NHANES who had a myocardial infarction (MI) or a hospitalization for unstable angina in the 15 to 90 days prior to completing their survey, we used information from the physical functioning questionnaire in the 2007-2008 NHANES survey—a recent wave of NHANES to ask patients how many days they have experienced specific health problems, including heart and stroke problems, and imputed the number of days a recent MI or stroke survivor with T2D had a heart or stroke problem as a function of age, gender, race, and other comorbidities using multivariate regression analysis. We then used these estimated parameters to predict the duration of heart and stroke problems among patients with T2D who reported an MI or stroke in the previous year in the 2009-2010 and 2011-2012 NHANES samples, which were used in the primary analysis.

Patients in the NHANES sample who met the criteria for at least 1 of the 5 CVOTs were older (P <.0001), more likely to be in Medicare (P <.0001), and had more numerous comorbidities (eg, hypertension (P = .0010), high cholesterol (P = .0225), heart failure (P = .0345), and stroke (P = .0097), compared with the overall NHANES population with T2D (

Overall, 28% of the T2D population met the enrollment criteria for at least 1 of the 5 CVOTs (FIGURE 1). Among DPP-4i CVOTs, the proportion of adult Americans with T2D who would have met enrollment criteria varied substantially. Approximately 21.4%, 9.0%, 5.3%, 4.3%, and 0.2% of patients with T2D met the criteria for SAVOR, CARMELINA, CAROLINA, TECOS, and EXAMINE, respectively (

). Due to large differences in the enrollment criteria, particularly with the most representative study, SAVOR, most patients (60%) who met enrollment criteria for a DPP-4i CVOT met the enrollment criteria for just 1 trial and no patients met the criteria for all 5 trials. Nearly 40% of patients with T2D eligible for any DPP-4i CVOT were eligible for SAVOR alone, whereas 12% were eligible for CARMELINA only, and 17% were eligible for SAVOR and CARMELINA, the 2 CVOTs with the greatest overlap in eligibility (

FDA requires CVOTs in the interest of patient safety, but provides little guidance on the design of such studies. Different study designs ask and answer different questions, and the enrollment criteria component of the design greatly influences the external validity of a study. On the whole, the representativeness of a study, or the external validity, is the extent to which the results of the study apply to other and/or larger populations.

 In the case of T2D CVOTs, a trial design will have greater external validity if the trial participants are representative of the T2D population, as a whole, something stakeholders may want to consider in the decision-making process. The less generalizable each CVOT design, the more dissimilar the participants of each study may be, and the less comparable their results. This analysis provides a quantitative assessment of these differences.

All 5 CVOTs in this analysis were large randomized trials with excellent internal validity, but each CVOT faced tradeoffs between trial efficiency and external validity. The enrollment criteria among the 5 trials used different age, CVD risk, glycemic control (as measured by A1C), medication history, and renal impairment requirements. Collectively, the enrollment criteria of DPP-4i CVOTs also differ substantially from those used in the Glycemia Reduction Approaches in Diabetes comparative effectiveness study (GRADE), which aims to compare the efficacy and CV safety of 4 different classes of antidiabetic medications over a 4- to 7-year follow-up period, but excluded patients with recent CV events and those with diabetes for 5 or more years.

 Some studies focused on the adult population, while others focused on middle-aged and older populations. The allowable A1C ranges also varied substantially among the 5 CVOTs. CARMELINA, TECOS, and SAVOR excluded patients with renal impairment or chronic kidney disease (CKD), but each study used a different threshold and/or different biomarkers to define renal impairment. The CV risk factors also varied considerably among the 5 CVOTs; SAVOR and CAROLINA enrolled patients with established CVD, as well as broad, but differing definitions of CV risk factors related to age, sex, hypertension, or current smoking. The remaining trials exclusively enrolled patients with established CVD and the EXAMINE study restricted enrollment solely to patients 15- to 90-days post ACS.

These differences likely contributed to our finding that, of those patients who met the criteria for any of the 5 CVOTs, most patients met the enrollment criteria for only 1 and none met the criteria for all 5 CVOTs. Patients who met the enrollment criteria for at least 1 of the CVOTs tended to be older, more likely to have experienced a CV event, and had more numerous comorbid conditions. In light of these findings, caution in extrapolating from the results of a CVOT—which may be representative of the health characteristics of patients referred to clinical trials by physicians&mdash;to the general population with T2D and comparisons between CVOTs is warranted.

One strength of this analysis is that NHANES is nationally representative and the individuals with T2D in NHANES are generally representative of the adult population with T2D in the United States. In addition, NHANES contains nearly all of the information needed to apply the CVOT enrollment criteria, making NHANES better suited to this analysis than other survey or claims databases, which have far less granular information on patient health characteristics and behaviors.

While NHANES, overall, may be representative of the US population and patients with T2D, within the subsample eligible for a particular CVOT we may have relatively few observations to estimate sample means, and across-group comparisons may not be representative of the whole T2D population. In addition, diagnosis of T2D is self-reported in NHANES, and there is a small chance that diagnosis is misreported. Since respondents are surveyed in their homes, NHANES may not be entirely representative of an inpatient population like that targeted in EXAMINE, either. Another limitation of the analysis is that the timing of the ACS event is imputed in our sample based on data from the 2007-2008 NHANES. We cannot completely separate T1D and T2D, as patients are not asked to distinguish between the 2 types in NHANES. Also, NHANES data do not contain information on bariatric surgery, and thus, we cannot exclude patients with T2D who have undergone bariatric surgery as done in the CARMELINA trial. However, bariatric survey is relatively uncommon, even among individuals with T2D (only 0.4% of T2D patients in the 2009-2011 Medical Expenditure Panel Survey reported having an “other gastric operation”). Also, the date of the T2D diagnosis cannot be determined with certainty; we can only identify the year of T2D diagnosis. Lastly, TECOS excluded patients with 2 or more hypoglycemic events requiring third-party assistance in the preceding 12 months.

However, NHANES does not contain any information that allowed us to identify these patients. This criterion would only exclude more patients, so to that extent, our estimate of the proportion of patients with T2D that would meet the enrollment criteria for TECOS is an upper bound.

The results of the DPP-4i CVOTs speak to the mid-term (1.5 to 3 years) CV safety of DPP-4is among a subset of adult patients with T2D at increased CV risk. They do not establish the potential long-term macrovascular effects of DPP-4is—positive or negative&mdash;among the majority of patients with T2D. Manufacturers should consider and balance the pros (increased representativeness) and cons (expense and duration) of enrolling lower-risk patients. Similarly, clinicians should consider these enrollment criteria when interpreting the applicability of the results of these trials to their patients and consider differences in the enrollment criteria when comparing the results of individual DPP-4i CVOTs. 

JPM is and employee, JVO is a past employee, and AM and ALP are consultants at Precision Health Economics, which was compensated by AstraZeneca to perform the study. ALP has consulted for Amgen, AstraZeneca, Abbott Diabetes Care, Boehringer Ingelheim, Becton Dickinson, Janssen, Merck, Novo Nordisk, Lilly, Biodel, Lexicon, Sanofi and Takeda. She has received research funding from Medtronic and Janssen. Precision Health Economics provides consulting and other research services to pharmaceutical, device, governmental, and non-governmental organizations. JS and IK are past employees of AstraZeneca.

 Joanna P MacEwan, PhD, is a research economist at Precision Health Economics, LLC (PHE). John J Sheehan, PhD, RPh, was employed with AstraZeneca at the time of the research. Anne Peters, MD, is a professor of Medicine at the Keck School of Medicine at the University of Southern California. Jacqueline Vanderpuye-Orgle, PhD, was employed with PHE at the time of the research. Iftekhar Kalsekar, PhD, was employed with AstraZeneca at the time of the research; Anup Malani, PhD, JD, is professor of Law at the University of Chicago School of Law. Funding: AstraZeneca supported this study.

Joanna P. MacEwan, PhD, 11100 Santa Monica Blvd, Suite 500, Los Angeles, CA 90025. Joanna.MacEwan@PHEconomics.com. 

  1. US Department of Health and Human Services Food and Drug Administration Center for Drug Evaluation and Research (CDER). Guidance for industry diabetes mellitus —evaluating cardiovascular risk in new antidiabetic therapies to treat type 2 diabetes. Rockville, MD. 2008. FDA website.http://www.fda.gov/downloads/Drugs/GuidanceComplianceRegulatoryInformation/Guidances/ucm071627.pdf. Published December 19, 2008. Accessed August 16, 2016.

  2. Fox CS, Golden SH, Anderson C, et al; American Diabetes Association. Update on prevention of cardiovascular disease in adults with type 2 diabetes mellitus in light of recent evidence: a scientific statement from the American Heart Association and the American Diabetes Association [published online August 5, 2015]. Diabetes Care. 2015;38(9):1777-1803. doi: http://dx.doi.org/10.2337/dci15-0012.

  3. Scirica BM, Bhatt DL, Braunwald E, et al; SAVOR-TIMI 53 Steering Committee and Investigators. Saxagliptin and cardiovascular outcomes in patients with type 2 diabetes mellitus. N Engl J Med.2013;369(14):1317-1326. doi: 10.1056/NEJMoa1307684.

  4. Scirica BM, Bhatt DL, Braunwald E, et al. The design and rationale of the saxagliptin assessment of vascular outcomes recorded in patients with diabetes mellitus—thrombolysis in myocardial infarction (SAVOR-TIMI) 53 Study. Am Heart J. 2011;162(5):818-825.e6. doi: 10.1016/j.ahj.2011.08.006.

  5. White WB, Cannon CP, Heller SR, et al. Alogliptin after acute coronary syndrome in patients with type 2 diabetes. N Engl J Med. 2013;369(14):1327-1335. doi: 10.1056/NEJMoa1305889.

  6. White WB, Bakris GL, Bergenstal RM, et al. EXamination of cArdiovascular outcoMes with alogliptIN versus standard of carE in patients with type 2 diabetes mellitus and acute coronary syndrome (EXAMINE): a cardiovascular safety study of the dipeptidyl peptidase 4 inhibitor alogliptin in patients with type 2 diabetes with acute coronary syndrome. Am Heart J. 2011;162(4):620-626.e1. doi:10.1016/j.ahj.2011.08.004.

  7. Raz I, Hanefeld M, Xu L, Caria C, Williams-Herman D, Khatami H. Efficacy and safety of the dipeptidyl peptidase-4 inhibitor sitagliptin as monotherapy in patients with type 2 diabetes mellitus. Diabetologia. 2006;49(11):2564-2571. doi :10.1007/s00125-006-0416-z.

  8. Freeman MK. Efficacy and safety of linagliptin (tradjenta) in adults with type-2 diabetes mellitus. PT. 2011;36(12):807-842. 

  9. Rosenstock J, Marx N, Kahn SE, et al. Cardiovascular outcome trials in type 2 diabetes and the sulphonylurea controversy: rationale for the active-comparator CAROLINA trial. Diab Vas Dis Res. 2013;10(4):289-301. doi: 10.1177/1479164112475102.

  10. TrialTrove. Citeline website. http://www.citeline.com/products/trialtrove/. Accessed October 15, 2014.

  11. Boehringer Ingelheim, Eli Lilly and Co. CARMELINA: A multicenter, international, randomized, parallel group, double-blind, placebo-controlled, cardiovascular safety and renal microvascular outcome study with linagliptin, 5 mg Once daily in patients with type 2 diabetes mellitus at high vascular risk. ClinicalTrials.gov website. https://clinicaltrials.gov/ct2/show/NCT01897532. Accessed October 15, 2014.

  12. Boehringer Ingelheim, Eli Lilly and Co. A multicentre, international, randomised, parallel group, double blind study to evaluate cardiovascular safety of linagliptin Versus glimepiride in patients with type 2 diabetes mellitus at high cardiovascular risk. ClinicalTrials.gov website. https://clinicaltrials. gov/ct2/show/study/NCT01243424. Accessed October 15, 2014.

  13. Marx N, Rosenstock J, Kahn SE, et al. Design and baseline characteristics of the CARdiovascular Outcome Trial of LINAgliptin Versus Glimepiride in Type 2 Diabetes (CAROLINA®). Diab Vas Dis Res. 2015;12(3):164-174. doi: 10.1177/1479164115570301.

  14. Kane MT. A sampling model for validity. Appl Psychol Meas. 1982;6(2):125-160. 

  15. Menke A, Orchard TJ, Imperatore G, Bullard KM, Mayer-Davis E, Cowie CC. The prevalence of type 1 diabetes in the United States. Epidemiology. 2013;24(5):773-774. doi: 10.1097/ EDE.1090b1013e31829ef31801a.

  16. Green JB, Bethel MA, Paul SK, et al. Rationale, design, and organization of a randomized, controlled trial evaluating cardiovascular outcomes with sitagliptin (TECOS) in patients with type 2 diabetes and established cardiovascular disease. Am Heart J. 2013;166(6):983-989.e7. doi: 10.1016/j. ahj.2013.09.003.

  17. Nathan DM, Buse JB, Kahn SE, et al; GRADE Study Research Group. Rationale and design of the Glycemia reduction approaches in diabetes: a comparative effectiveness study (GRADE). Diabetes Care. 2013;36(8):2254-2261. doi: 10.2337/dc13-0356.

  18. Green JB, Bethel MA, Armstrong PW, et al. Effect of Sitagliptin on cardiovascular outcomes in type 2 diabetes. N Engl J Med. 2015;373(3):232-242. doi: 10.1056/NEJMoa1501352.

Given the growing prevalence of type 2 diabetes (T2D) and its contribution to cardiovascular disease, cardiovascular outcomes trials should aim to be more representative of the average patient with T2D.  

Representativeness of Dipeptidyl-Peptidase-4 Inhibitor Cardiovascular Outcomes Trials

 To measure the impact of state Medicaid formulary policies on costs for patients with schizophrenia and bipolar disorder.

 Retrospective analysis of medical and pharmacy claims for patients diagnosed with schizophrenia or bipolar disorder in 24 state Medicaid programs.

We combined information on formulary restrictions in Medicaid with the medical and pharmacy claims of 117,908 patients with schizophrenia and 170,596 patients with bipolar disorder in Medicaid who were single-eligible, and newly prescribed a second-generation antipsychotic from 2001 to 2008. We tested the impact of formulary restrictions on the medical costs and utilization of patients in the 12 months after the index prescription. To capture social costs in addition to medical expenditures in Medicaid, we estimated the incremental costs of incarcerating patients with schizophrenia and bipolar disorder associated with formulary restrictions.

Patients with schizophrenia subject to formulary restrictions were more likely to be hospitalized (odds ratio 1.13, 

 <.001), had 23% higher inpatient costs (

 <.001). Similar effects were observed for patients with bipolar disorder. Our estimates suggest restrictive formulary policies in Medicaid increased the number of prisoners by 9920 and incarceration costs by $362 million nationwide in 2008.

 Applying formulary restrictions to atypical antipsychotics is associated with higher total medical expenditures for patients with schizophrenia and bipolar disorder in Medicaid. Combined with the other social costs such as an increase in incarceration rates, these formulary restrictions could increase state costs by $1 billion annually, enough to offset any savings in pharmacy costs.

Applying formulary restrictions to atypical antipsychotics is associated with higher total medical expenditures for patients with schizophrenia and bipolar disorder in Medicaid.

		Adherence to medication declined due to formulary restrictions.

		Because a longer half-life means an active drug remains in the circulation longer after the last dose, oral antipsychotics with longer half-lives might be less impacted by partial nonadherence than those with shorter half-lives.

		Because the prescriber can individualize a patient’s treatment, our results suggest autonomous prescribers constitute an asset to payers, since these prescribers achieve lower hospitalization rates than prescribers who operate within a restricted payer environment.

In an effort to contain rising expenditures on prescription drugs, state Medicaid programs around the United States have increasingly adopted formulary restrictions designed to restrict access to specified high-cost medications.

Second-generation antipsychotics (SGAs), or “atypical” antipsychotics, have been among the most frequently targeted class of drugs.

 Since the first atypical antipsychotics were introduced in the 1990s, they have increasingly replaced first-generation or “typical” antipsychotics and in 2005 accounted for over 15% of all Medicaid spending on pharmaceuticals.

 Their high cost and rapid growth within Medicaid made them natural targets for cost containment. More than one-third of state Medicaid programs and Medicare Part D (MPD) prescription drug plans now require prior authorization or some other restriction for at least 1 atypical antipsychotic.

 Formulary restrictions on atypical antipsychotics are controversial because they affect patients with severe chronic mental illness, including schizophrenia and bipolar disorder. While these conditions are relatively uncommon in the general population—schizophrenia affects about 1.1% of people in the United States

 while bipolar disorder affects about 2.6%

—they are more common in the Medicaid population. Estimates suggest that the prevalence of diagnosed schizophrenia in the Medicaid population is approximately 1.7%, and it is estimated that over 30% of individuals with schizophrenia are Medicaid beneficiaries.

 Formulary restrictions influence clinical decisions about the type of atypical antipsychotic that a patient receives, but it is known that the efficacy and tolerability of these agents varies substantially from one patient to another.

 This heterogeneity across different treatments means that providers and patients sometimes need to try different treatment regimens to attain desired clinical outcomes, and restricting access to some medications can lead to possible noncompliance with therapy or treatment non-response.

Given the vulnerable nature of these patients, there is concern that such disruption could have adverse consequences. For example, one study of medication nonadherence for individuals with schizophrenia found a 50% increase in the risk of a mental health hospitalization in the first 10 days following a missed prescription refill.

 Other studies have found that adherence to atypical antipsychotics fell

 There is some evidence of a small decrease in pharmacy expenditures associated with formulary restrictions for patients with bipolar disorder, but this was also associated with an increase in treatment discontinuation.

While informative, these studies are limited in that each focuses on only small groups of states (usually 1). More work on a larger number of states with a more diverse set of policy changes is needed to understand the impact of formulary restrictions on health outcomes and medical costs for patients with severe mental illness.

 Moreover, while Medicaid expenditures are an important metric of social costs, there are others. Disruptions in antipsychotic therapy could lead to more acute psychotic episodes by affected patients. This can lead to public unrest or even violence, which can be reflected as an uptick in incarceration rates among the mentally ill, driving up the burden of mental illness on society.

This study examines the relationship between formulary restrictions and healthcare utilization and expenditures for patients with schizophrenia and bipolar disorder from 24 state Medicaid programs. We combined information on formulary restrictions for atypical antipsychotics in these states with data on all of the medical and pharmacy claims in Medicaid for patients with schizophrenia and bipolar disorder from 2001 to 2008. We combined our estimates of formulary restriction effects on healthcare costs with evidence of their effect on incarceration costs. This provides a more complete accounting of the potential unintended consequences of trying to contain costs by restricting access to the full range of therapies for mental illness.

We used claim-level data on inpatient, outpatient, longterm care, and pharmacy claims from the Medicaid Analytic eXtract (MAX) files. Our data included all patients diagnosed with schizophrenia or bipolar disorder from 24 states from 2001 to 2008. (The states are: Alabama, Arkansas, California, Florida, Idaho, Illinois, Indiana, Kentucky, Louisiana, Massachusetts, Maryland, Minnesota, Mississippi, North Carolina, New Jersey, New Mexico, New York, Ohio, Pennsylvania, South Carolina, Texas, Virginia, Wisconsin, and West Virginia. We did not receive 2008 data for Pennsylvania or Wisconsin because they were unavailable at the time of data extraction.) We selected these because other states often deliver services to Medicaid beneficiaries through the use of managed care, and there are issues related to the availability and reliability of claims data for enrollees in Medicaid-funded managed care plans.

The study sample includes all working-age adult enrollees (aged 18-64 years) diagnosed with schizophrenia or bipolar disorder who were first prescribed an atypical antipsychotic (olanzapine, risperidone, quetiapine, aripiprazole, or ziprasidone) during the study period (N = 375,934). We restricted the study sample to patients newly prescribed an atypical antipsychotic, because these patients were more likely to be newly treated for their condition, and thus were most likely to be affected by formulary restrictions. An index date was identified using the start date of the first filled prescription for the first atypical prescribed. We tracked all medical and pharmacy claims 6 months before and 12 months after the index date. Patients with schizophrenia were identified based on medical claims with primary or secondary 

International Classification of Diseases, Ninth Edition

-CM 295.x during the study period (2001-2008).

Patients with bipolar disorder were identified based on 

-CM codes of 296.0x, 296.1x, and 296.4x-296.8x. We excluded patients with both schizophrenia and bipolar diagnoses (N = 106,314, 28%), because these patients likely had more uncertainty over their disease status, which could have affected treatment patterns (including use of atypical antipsychotics). We also excluded enrollees who were dually eligible for Medicare because of changes in prescription drug coverage that occurred with the introduction of MPD in 2006, which could have affected atypical antipsychotic use. We also excluded patients with prior first-generation antipsychotic use or a prescription for clozapine prior to the index date, because it is unclear how these patients would be affected by the restrictions (eg, clozapine is only approved for treatment-resistant schizophrenia, so there are already significant restrictions on its use). Finally, we excluded any patients who died in the 12-month follow-up period. Our final study samples included 117,908 patients with schizophrenia and 170,596 patients with bipolar disorder.

Data on formulary restrictions were obtained from a 2009 survey of state Medicaid programs that asked about their formulary policies affecting atypical antipsychotics and other drugs treating mental illness. The survey, used in prior work,

 covered 30 states (including the 24 states in our study sample) from 1999 to 2008 and asked whether specified restrictions applied to a list of drugs identified by US brand name, including the atypical antipsychotics examined in our study. Programs were asked to identify whether a given type of restriction applied to each drug in the list and over which years it applied. Survey responses were supplemented using the Medicaid pharmacy program websites for relevant documents (eg, preferred drug lists) and with direct contact with Medicaid program personnel.

The survey collected information about the most commonly used restrictions to contain Medicaid pharmacy costs. Prior authorization requires providers to obtain prior permission to prescribe a specified medication, or the Medicaid program will not guarantee reimbursement. Quantity limits are direct restrictions to how many units of a drug may be dispensed in a fixed time period, usually a month or day (this includes daily quantity limits—sometimes referred to as dose optimizations&mdash;that operate in the same way as monthly limits). Step therapy requires that medication be prescribed only after a provider tries other selected medications, usually cheaper alternatives. Under these policies, providers must document that the patient has had unsatisfactory responses, and needs the nonpreferred medication. Other, less common policies were also surveyed, including age edits, which restrict use by patients in certain age groups.

Following prior work, we implemented our measure of these policies using binary indicators when any of the policies affected patients using a drug in a state in a year.

 summarizes the share of states that had a restriction by drug and year in our sample. The figure shows that the use of formulary restrictions grew rapidly over the study sample. The most commonly restricted drug in the sample was olanzapine, particularly early on, followed closely by aripiprazole. By 2008, both olanzapine and aripiprazole were targeted in more than half of the states in our sample.

As a sensitivity analysis, we examined how the effects differed based on the type(s) of atypical antipsychotics targeted. Specifically, in our empirical analysis, we created binary categories to allow us to test for the effects of:

		Any of the policy types against any of the 5 molecules

		Any of the policy types targeting olanzapine only

		Any of the policy types targeting olanzapine and any of the other 4 molecules, or

		Any of the policy types targeting any of the 4 molecules but not olanzapine.

We used these specific categories rather than breaking down the analysis to separately consider policies targeting all 5 drugs because there were an insufficient number of drugpolicy combinations.

We measured healthcare utilization and spending over a range of potential outcomes. This includes: whether we observe any hospitalization episode for the patient in the follow-up period, the total Medicaid dollars spent on inpatient spending, total medical spending (defined as the sum of inpatient and outpatient spending), total prescription drug spending (including spending on atypical antipsychotics), and total Medicaid spending from all sources. Dollar values were normalized to 2008 using the medical care component of the Consumer Price Index. Finally, we tested the impact of formulary restrictions on adherence to the index atypical antipsychotic, defined as the proportion of days covered (PDC) over the 365 day period after the index date.

To identify the effects of formulary restrictions on outcome, we employed a difference-in-difference study design using multivariate regression techniques.18 Specifically, we examined the difference in outcomes for patients prescribed an atypical antipsychotic in states with and without formulary restrictions before and after those restrictions were adopted. This method allowed us to estimate the impact of state policies while controlling for other state and time heterogeneity that could confound the results.

If individual patient characteristics systematically differed in states with and without restrictive formulary policies, this could confound our analysis. In addition to state and year fixed effects, which would capture fixed state differences or general trends, we also used patient demographics and claimsbased measures of patient health at baseline to control for potentially confounding variation. The MAX files provided basic demographics including age, gender, and race. The zip code of residence for each enrollee was used to measure median household income in their area to provide information on socioeconomic status of patients and their families. Zip code was also used to categorize the primary residence of the enrollee into a rural/urban location.

To control for patient health at baseline, we used medical claims from the 6-month period prior to the index date to construct a claims-based index of individual health using the Charlson-Deyo method.

 We also included an indicator for whether the patient experienced an all-cause hospitalization prior to baseline. Because atypical antipsychotics are associated with elevated metabolic risk, we included controls for the presence of metabolic syndrome (ICD-9-CM code of 277.7), dyslipidemia (272.xx), diabetes mellitus (250.xx), or overweight and obesity (278.xx) prior to baseline. We also controlled for other, nonmetabolic, comorbid chronic health conditions, including: stroke, vascular disease, chronic obstructive pulmonary disease, asthma, malignant cancer, gastric acid disorder, anemia, arthritis, allergic rhinitis, and epilepsy. Finally, we used the claims data to control for the severity of mental illness, based on whether or not the individual saw a psychiatrist in the preindex period. For patients with bipolar disorder, we also control for the use of mood stabilizers in the pre-index period.

We use logistic regression to estimate the effect of formulary policies on hospitalization rates and ordinary least squares to estimate the effect on expenditures and adherence to therapy. We logged the expenditures to correct for the skewness in medical expenditures,

and added $1 to expenditures to avoid dropping zero values. To describe the magnitude of the expenditure differences associated with formulary restrictions, we predicted values of different expenditure categories with and without formulary restrictions based on our regression models holding other covariates at mean values. We used mixed effects models—specifically, models that included both fixed and random effects at the state level&mdash;to reflect the fact that our key policy variables were at the state level, which can cause correlation in the error terms (or “clustering”) within groups.

 A smearing estimate was used to eliminate retransformation bias when calculating predicted values.

Measuring Societal Costs of Formulary Restrictions 

To study incarceration costs, we relied on past work that used logistic regression to estimate the association between formulary restrictions in Medicaid and the share of the prison population with mental illness in a national survey.

 This work found that formulary restrictions increase the share of the prison population afflicted with mental illness by up to 17%. Approximately 12% of the prison population is affected by schizophrenia or bipolar disorder, implying a 2 percentage point increase in the overall prison population. We combined these estimates with data from the Bureau of Justice Statistics Expenditure and Employment Extracts on prison costs by state and year. With these data, we estimate the incremental costs that arose as a result of states that had restrictive formulary policies.

summarizes outcomes and covariates for the schizophrenia and bipolar patient samples. Average medical expenditures in the follow-up period for both sets of patients were high for both schizophrenia ($18,896) and bipolar ($18,770) patients. Pharmacy costs were also high, at $5597 (schizophrenia) and $5036 (bipolar). Other characteristics were similar, although schizophrenia patients were more likely to be non-white and more likely to be male (

 reports estimated effects from the regression models. Patients with schizophrenia subject to formulary restrictions were more likely to experience a hospitalization (odds ratio 1.13,

 <.001), and had 16% higher total costs (

<.001). For patients with bipolar disorder, those subject to formulary restrictions were also more likely to experience a hospitalization (odds ratio 1.07, 

= .016), had 20% higher inpatient costs (

 <.001), and had 10% higher total costs (

 <.001). Formulary restrictions were not associated with statistically significantly lower pharmacy expenditures for either patient sample. Patients with schizonphrenia subject to formulary restrictions had worse adherence (—0.02 lower PDC, 

 <.001). For patients with bipolar disorder, formulary restrictions had no significant effect on adherence.

In the subanalysis by drug categories, restrictions that target olanzapine alone had no consistently positive or significant effect on medical costs in either patient sample. Policies targeting other molecules, with or without olanzapine, were associated with significantly higher hospitalization rates and medical expenditures and lower adherence in both patient samples. Predicted inpatient expenditures for patients with schizophrenia increased from $2983 to $3740 with a formulary restriction in place (

Predicted medical expenditures increased from $10,852 to $13,299 and predicted total expenditures increased from $16,171 to $18,897. The savings in medication costs were small: predicted pharmacy spending declined from $5806 to $5598. The effects on patients with bipolar disorder were similar but smaller overall (

). Predicted inpatient expenditures for patients with bipolar disorder increased from $3918 to $4770 and total expenditures increased from $16,958 to $18,771. The predicted expenditures on prescription drugs rose slightly with formulary restrictions in place.

Retrospective analysis of medical and pharmacy claims for patients diagnosed with schizophrenia or bipolar disorder in 24 state Medicaid programs.

John J Sheehan, PhD

Articles

The Relationship Between Adherence and Total Spending Among Medicare Beneficiaries With Type 2 Diabetes

Representativeness of Dipeptidyl-Peptidase-4 Inhibitor Cardiovascular Outcomes Trials

Formulary Restrictions on Atypical Antipsychotics: Impact on Costs for Patients With Schizophrenia and Bipolar Disorder in Medicaid