John Z. Ayanian, MD, MPP

 To compare use of diabetes medications between beneficiaries enrolled in Medicare Advantage (MA) and traditional Medicare (TM).

 Retrospective cohort analysis of Medicare enrollment and Part D event claims during 2015-2016.

 Data came from 1,027,884 TM and 838,420 MA beneficiaries who received at least 1 prescription for an oral or injectable diabetes medication. After matching MA and TM enrollees by demographic characteristics and geography, we analyzed use of medication overall, choices of first diabetes medication for those new to medication, and patterns of adding medications.

Overall and for patients on 1, 2, or 3 diabetes medications, use of metformin was higher in MA by about 3 percentage points, but use of newer medication classes was 5.1 percentage points higher in TM overall (21.3% vs 16.2%). Use of guideline-recommended first-line agents was higher in MA. For those who started metformin first, use of a sulfonylurea as a second medication was 7.8 percentage points higher in MA than TM (61.5% vs 53.7%), whereas use of medications from newer classes was 7.7 percentage points lower (22.0% vs 29.7%). Mean total spending was $149 higher in TM for those taking 1 medication and $298 higher for those taking 2 medications. Differences in spending among MA plans were of similar magnitude to the MA-TM differences.

 MA enrollees are more likely to be treated with metformin and sulfonylureas and less likely to receive costly newer medications than those in TM, but there also is substantial variation within MA. A limitation of the study is that we could not assess glucose control using glycated hemoglobin levels.

Medicare Advantage (MA) is associated with decreased utilization and higher quality relative to traditional Medicare (TM), but little is known about differences in medication management of chronic medical conditions such as diabetes. Using retrospective analyses of Medicare claims, we find that:

MA enrollees were more likely to use metformin and sulfonylureas and less likely to receive newer medications.

Mean plan and patient out-of-pocket spending for patients on 1, 2, or 3 medications was substantially higher for patients in TM.

There is substantial variation in the approaches to medication management across MA health plans.

Enrollment in Medicare Advantage (MA), Medicare’s managed care program, continues to expand, with more than one-third of Medicare beneficiaries currently enrolled in an MA plan.

 Relative to traditional Medicare (TM), care management by MA plans may enable them to treat patients at lower costs while attaining equal or superior quality through their control of enrollee benefits, network contracting, and coordination of care.

 This may be of particular relevance for patients with chronic medical conditions, for whom plans may employ a variety of care management approaches and financial incentives to promote the delivery of appropriate services and influence the quantity of services delivered.

 Prior research has demonstrated higher quality of ambulatory care for those enrolled in MA for process measures such as preventive care and monitoring of chronic diseases, as well as lower utilization of emergency department and inpatient services and of some elective services such as joint replacements.

 Differences in choices of medication might be another method by which MA plans achieve savings compared with TM. Most MA enrollment is in plans that include Part D coverage in an integrated product. TM enrollees generally enroll in stand-alone Part D prescription drug plans, although some have non-CMS coverage (retiree or other private plans) or none. However, no prior research that we are aware of has compared patterns of medication management in the 2 programs.

Diabetes is a prototypical example of a chronic and often progressive medical condition that may be affected by managed care approaches. Diabetes currently affects more than one-fourth of Medicare beneficiaries and is among the most common and costly chronic medical conditions affecting elderly Medicare enrollees.

 The principal medical treatment for diabetes is pharmacologic therapy aimed at controlling blood glucose levels and thereby preventing downstream cardiovascular, renal, and retinal complications that may arise from poorly controlled diabetes. The American Diabetes Association estimated that in 2017, almost $50 billion was spent on prescription diabetes medications and testing.

 Consensus recommendations uniformly support lifestyle modification plus metformin as the preferred initial treatment strategy for type 2 diabetes, but many classes of medication compete for use as second- or third-line therapy, including those new in the past decade.

 More recently, some emerging data suggest improved outcomes with several new agents in high-risk populations with cardiovascular or chronic kidney disease, but health plans and patients who use these newer medications, which have no generic alternatives, also face higher plan and out-of-pocket costs, respectively.

 For instance, according to GoodRx.com, mean monthly costs of glucagon-like peptide-1 (GLP-1) receptor agonists are approximately $800/month or higher compared with less than $100 for even the most costly dose of metformin. In addition, recommendations specifically for elderly patients include avoidance of long-acting sulfonylureas, which are associated with increased rates of hypoglycemia, and application of less stringent treatment goals for glucose control, particularly for those with limited life expectancy or multiple coexisting comorbid conditions.

 The large number of treatment options and potential intensity of treatment, as well as the wide range in prices of medications used to treat diabetes, present an opportunity for MA plans to improve the care of diabetes using approaches that are less expensive but similarly effective. Moreover, lower out-of-pocket costs to patients are associated with improved adherence to treatment recommendations, which ultimately can influence patient outcomes, as well.

Previous research indicates that diabetes is among the clinical areas in which MA plans achieve their highest rates of savings for medical services,

 but little research has examined how choices of medications for enrollees with diabetes differ for those enrolled in MA vs TM. Given the high cost of medical treatment of diabetes, we therefore used comprehensive data on beneficiaries being treated for diabetes in TM and MA to compare patterns of drug use in the 2 programs, as well as total pharmaceutical spending and adherence. We hypothesized that, relative to TM, MA plans would promote higher rates of use of medications that are recommended by consensus guidelines and that they also would promote use of less expensive generic medications when more expensive alternatives have not demonstrated clear benefits.

We analyzed Medicare claims and enrollment data provided by CMS for the period 2015-2016, the 2 most recent years available to us. Demographic and enrollment characteristics drawn from the Beneficiary Summary File were available for all enrollees. Data on prescription medication use from the Part D prescription drug event files were available for the subset (more than 67.8% in the study years) of the 20% random sample of TM beneficiaries enrolled in a Part D plan, and for a substantially higher percentage (94.6%) of those enrolled in MA plans, which typically are bundled with Part D coverage.

Our sample included all persons who were continuously enrolled in Medicare parts A and B throughout the year and were at least 65 years old on January 1. We excluded residents of long-stay nursing homes, whom we identified using a validated algorithm,

 because these beneficiaries rarely enroll in MA plans other than the dedicated special needs plans aimed at this population,

 and Medicare beneficiaries with end-stage renal disease. We also excluded MA enrollees in special needs plans that most commonly are restricted to low-income enrollees and have different rules related to enrollee cost sharing. We restricted some analyses focused on initiation or up-titration of prescription drugs to beneficiaries 66 years or older who had at least 1 year of continuous Part D enrollment prior to the respective study year.

Because medical claims data are not available for enrollees in MA plans, we restricted our analyses to those with diabetes who had received at least 1 prescription for a diabetes medication during the year of study.

 Thus, our study sample focuses on enrollees with diabetes being treated with medication and excludes those who received a diagnosis of diabetes but were being treated conservatively through lifestyle modification and diet.

Beneficiary and Health Plan Characteristics

The Medicare Master Beneficiary Summary Files provided data on age, sex, race/ethnicity, region of the country, and, when applicable, Medicaid coverage and disability as the original reason for Medicare eligibility.

Using CMS data, we categorized health plans (defined based on Medicare contracts) as large (> 25,000 enrollees) vs small, and we identified health plans new to MA since 2006.

Characterizing Medication Use and Guideline Concordance

We lacked data on clinical indicators, such as glycated hemoglobin levels, that typically drive decisions regarding use of medication; therefore, in addition to analyzing medication use overall, we focused on prespecified cohorts that we defined for this study to identify those with diabetes at a similar level of disease control for whom a clinician had decided that initiation or escalation of medical therapy was warranted. Medications were identified through drug or product names (see 

] for a full list and how they were classified).

For patients who received a new prescription for any diabetes medication in the year of interest and had received no diabetes medications in the prior year, we categorized the 

by class. These enrollees are likely those who previously had been treated with diet and lifestyle modification and who a physician now thought needed to start medication treatment. We considered metformin to be the only guideline-concordant initial choice of medication.

For those treated with a single drug in the prior year, we then characterized the 

 for those started on the 2 most commonly prescribed initial drug classes (metformin and sulfonylureas). Although guidelines regarding second medications allow for individualized use of medications from any of multiple classes, the guidelines also suggest that cost could be a consideration.

 Because of their less frequent use, we grouped into 1 category all relatively newly available medications without a generic alternative, such as sodium-glucose cotransporter-2 (SGLT-2) inhibitors, dipeptidyl peptidase-4 (DPP-4) inhibitors, and GLP-1 receptor agonists.

Finally, we examined whether those patients who were 

 received short-/rapid-acting, intermediate-acting, long-acting, or combination insulin, counting patients filling prescriptions for 2 types of insulin within 30 days as receiving combination therapy. When insulin therapy is necessary, guidelines consistently recommend beginning with basal or long-acting insulin. We did not differentiate between human and analogue insulin.

We based our classifications on largely consistent recommendations from the American College of Physicians, the American Diabetes Association, and the American Geriatrics Society for the treatment of elderly adults with diabetes during the time period of our study.

 Finally, we considered long-acting sulfonylureas to be guideline discordant at any stage of treatment because of increased risks of hypoglycemia in the elderly population.

These analyses included all patients taking any diabetes medication in a year. We calculated each beneficiary’s total, out-of-pocket, and Part D plan monthly diabetes drug spending over the year. Finally, for each medication taken, we calculated the medication possession ratio as a measure of adherence.

 This measure counts the number of days of supply that a patient had starting from the date of the first prescription fill to the date of the last prescription filled during the year plus the days supplied from that last prescription. Higher numbers indicate greater adherence.

We first compared the characteristics of enrollees in TM and MA who were identified as users of diabetes medications by available Part D prescription claims. As in previous studies, for all comparisons between the MA and TM samples, we first reweighted the TM sample to match the MA sample distribution by geography and demographic characteristics (age [in 5-year ranges], sex, race/ethnicity) and then aggregated these results to obtain national estimates that were weighted by the enrollment of MA plans.

 Matching on geography controlled for variation in practice patterns within Medicare across regions.

 By matching at the zip code level where possible (73% of the prevalent cohort), we also controlled for unmeasured socioeconomic characteristics associated with residence at this level of geography, as well as the proportion who were dually eligible for Medicaid. Of note, excluding those dually enrolled in Medicaid did not substantively affect our results, so dual-eligible enrollees were included in the study.

We tabulated characteristics of all enrollees with diabetes on medication as well as those initiating any drug treatment and those initiating treatment with insulin, testing differences between MA and TM proportions or means using χ

 tests as appropriate. We calculated the unadjusted and adjusted MA mean, the matched TM mean, the difference, and their respective standard errors, and we performed χ

 tests as appropriate to assess significance. Because of the size of our sample, most overall MA-TM differences that we report were statistically significant at the 

 < .001 level unless otherwise noted in the tables. Therefore, comparisons were not adjusted for multiple comparisons.

Lastly, as an illustrative example to examine variation in spending among MA plans, we modeled total spending for beneficiaries on 2 medications, using hierarchical linear regression models that included a random effect for each MA contract, as well as random effects for MA and TM at the state and hospital referral region levels. The random effects variance parameters quantify the variation across plans of savings relative to matched TM enrollees in its same service area after accounting for market- and state-level variation and sampling error.

All analyses were conducted using SAS software version 9.2 (SAS Institute). Our study protocol was approved by the Harvard Medical School Human Studies Committee. This work was funded by the National Institute on Aging, which had no role in the design of the study or in the collection, analysis, and interpretation of the data.

During 2015 and 2016, we studied 1,027,884 enrollees in TM and 838,420 enrollees in MA who were enrolled for the full year in Medicare and had available Part D pharmaceutical claims. Their respective characteristics are shown in 

. Among these enrollees, 62,265 from TM and 49,089 from MA initiated treatment with an initial diabetes drug (including all oral medications or insulin) during this time period, and, of these, 24,530 in TM and 16,635 in MA first started using insulin (including as an initial medication). In each treatment category, White enrollees were overrepresented in TM (near 75%) relative to MA (60%), and Hispanic enrollees were underrepresented in TM (7%-9%) relative to MA (21%-23%). More TM enrollees were eligible for Medicaid (25.7% vs 21.2% overall). After weighting, there were no observable demographic differences between the MA and TM populations.

Overall, use of metformin was lower in TM by about 3 percentage points (

). For example, among those on 1 medication, 59.8% of TM enrollees were on metformin compared with 62.7% of MA enrollees. Use of short-acting sulfonylureas was also lower in TM (19.7% vs 23.6%), but use of long-acting sulfonylureas, which should be avoided in elderly patients, was slightly higher in TM (17.1% vs 16.3%). Rates of insulin use were higher in TM (19.7% vs 17.4%). The largest TM-MA difference was for use of any of the new classes of medication (such as DPP-4 or SGLT-2 inhibitors), which was 5.1 percentage points higher in TM (21.3% vs 16.2%).

Consistent with guidelines for the initial treatment of older patients with diabetes, metformin was the predominant first medication, but it was used slightly less frequently in TM than in MA (71.7% vs 72.8%; absolute difference, 1.1 percentage points) (

). For the remaining approximately 30% of patients started on a medication, short-acting sulfonylureas were less often used in TM (6.9% vs 9.1%). These differences were offset predominantly by higher use in TM of a medication from a newer class (9.9% vs 6.8%). Thiazolidinediones were rarely the initial medication (less than 2% in both programs).

These differences varied somewhat by region and plan characteristics. Most notably, TM enrollees in the South were started on metformin 2.3 percentage points less frequently than MA enrollees, and TM enrollees in the West were started on a new medication 5 percentage points more frequently than those in MA. Those enrolled in large and more established plans were about 1 percentage point less likely to start a medication from another class.

Either a sulfonylurea (for those on metformin) or metformin (for those on a sulfonylurea) was the second-choice medicine for more than half of those given a second medicine (

). However, for those who started metformin first, use of a sulfonylurea as a second medication was 8 percentage points lower in TM (53.7% vs 61.5%), which was primarily offset by use of medication from a new class (29.7% vs 22.0%). Patterns were similar for those who started a sulfonylurea first (with metformin the most common choice of second drug), but the differences were about half the magnitude. Differences for smaller and newer plans were larger than for established and larger MA plans.

Among those new to insulin, the percentage of patients started on long-acting insulin was higher in TM than in MA (63.5% vs 58.2%) (

), with the largest differences in the Midwest.

Annual diabetes-related medication costs are summarized in 

, stratified by the number of medications being taken. In each stratum, the percentage of enrollees with total costs of more than $1000 was greater in TM than MA (differences of 4.0, 8.2, and 2.9 percentage points for those taking 1, 2, and 3 medications, respectively). Similarly, mean spending was $149 higher in TM for those taking 1 medication and $298 higher for those on 2 medications. These differences were largest in the West and Midwest, as well as in large and established health plans. Out-of-pocket spending showed similar patterns. Medication possession ratios generally were 90% or higher and generally did not substantively differ between TM and MA or by plan characteristics (

Analyses of estimated differences in mean spending for those taking 2 medications arrayed by MA plan (contract) showed that the variation in spending across MA plans after controlling for geographical factors was similar in magnitude (SD, $240) to the observed overall difference in spending between MA and TM (mean difference, $298, as noted above) (

Patterns of use of diabetes medications for beneficiaries enrolled in TM and MA differ, even after identifying enrollees at a stage of disease who either start medication therapy or require an additional medication. More importantly, these prescribing patterns have important implications for both the quality and costs of care. Metformin, which is recommended as first-line treatment in type 2 diabetes by all major guidelines and is available as a relatively inexpensive generic medication, was used more frequently in MA as both a first and second agent. In contrast, medication choices in TM generally were costlier, with greater use of expensive new classes of medications as both first- and second-line treatment leading to higher levels of both plan and out-of-pocket spending for TM beneficiaries. To achieve these savings, however, enrollees in MA plans more frequently receive sulfonylureas, which have the potential to cause hypoglycemia in older patients.

Our study found substantially greater use of new diabetes agents that do not yet have generic alternatives as both first- and second-line treatment options in TM, substantially contributing to higher spending for those in TM. The use of some of these newer classes of medications, however, could represent higher quality of care, particularly for higher-risk populations with established cardiovascular or chronic kidney disease. For instance, emerging evidence supports the use of SGLT-2 and GLP-1 inhibitors to treat specific high-risk populations with established cardiovascular disease or chronic kidney disease.

 We note, however, that studies have not found consistent advantages in these groups with respect to cardiovascular events or mortality and that these benefits have not been shown for average-risk patients.

 Moreover, older patients with more limited life expectancy may not receive these potential benefits. Consequently, guidelines continue to recommend that the choice of a second medication should be individualized based on patient preferences and, potentially, cost.

 Although our study could not assess patient preferences, MA treatment regimens in most plans demonstrated lower mean costs for both plans and patients, even among patients using the same number of medications, and these differences could be substantial. Whether these differences will translate into improved outcomes is not clear and will require future research.

Surprisingly, we found more frequent use of basal insulin in TM than in MA when first starting treatment with insulin. Multi-injection insulin regimens may be difficult for elderly patients to manage and administer, with little incremental value over once-daily administration of a single long-acting agent, while also carrying increased risk for hypoglycemia.

 Although this finding might reflect that MA physicians may turn to insulin later in the course of treatment, the reasons for this difference are not clear.

Our study results suggest that managed care approaches to diabetes can bring about care that is both more guideline concordant (at least in the case of first-line medications) and less expensive, although this was not uniformly the case for all MA health plans. How might some health plans achieve such results? Although our study did not explore mechanisms, there are several potential explanations. Most MA enrollees choose integrated MA–Part D plans, which may facilitate a more integrated approach to the management of all aspects of diabetes, including the use of medications, than is practiced by the stand-alone Part D plans that are used by TM beneficiaries.

Some health plans employ various approaches such as increased use of guidelines, audits, and feedback to encourage a more uniform approach to treatment within the plan, including the use of less costly medications as initial selections.

 Our results showed greater TM-MA differences for larger and more established health plans, which might be more likely to have the capacity for managing such programs.

 Second, health plans can employ financial incentives to physicians and physician organizations that encourage use of preferred lower-cost agents, thus delegating responsibility for management of prescribing patterns to the providers. For instance, some MA health plans have entered into capitation-type arrangements with physician organizations that also put these organizations at risk for drug spending. Plans using management mechanisms like these might choose to discourage adoption of some newer and more expensive medications until their marginal value is clearly proven in the literature. Of note, many of the studies noted above—those supporting the use of some newer agents in higher-risk populations—came out after the period of our study. However, it will be important to monitor choices of medications in such plans to ensure that patients are not denied more expensive medications when they may be of benefit. Finally, plans also have direct control over their formularies, and integrated plans may make different choices related to inclusion, tiering, or cost sharing that can also influence use of medications.

Our study has several potential limitations. First, our analyses were limited to claims data, so we could not assess the adequacy of diabetes control using measures such as glycated hemoglobin levels. Second, we lacked data on functional status and frailty, each of which may affect decision-making and treatment goals. Third, because we lacked information on the date of first diagnosis, we could not control for the duration of diabetes when treatment began, which might have differed for MA and TM. For this reason, we focus much of our analyses on initial treatment choices for patients who had previously not been on a medication, as well as patterns of adding medications for those who had been on a single agent, which effectively controls for disease status. Fourth, we were unable to assess formulary status of medications, which also can influence medication choice. Fifth, our data extend back almost 5 years, so these findings might differ using more current data. In particular, use of newer agents with no generic alternatives might have accelerated over this time period. Finally, because we lacked information on comorbid conditions for MA enrollees (for whom we lacked claims), we could not examine the extent to which comorbid conditions such as cardiovascular disease or heart failure influenced medication choices, nor could we examine longer-term complications of diabetes such as retinopathy, nephropathy, or cardiovascular disease.

We found substantial differences in patterns of medication use between beneficiaries in TM and enrollees in MA health plans, although with substantial variation among MA plans. Costly and complex regimens are used more frequently in TM than MA across all treatment levels, reflecting more frequent use within MA of medications that can cause hypoglycemia (eg, sulfonylureas) and less frequent use of newer agents, which might be of particular benefit for higher-risk patients. These findings suggest that managed care approaches adopted by at least some plans influence diabetes treatment practices. As our health system strives to find opportunities to lower health spending and improve the value of care, chronic diseases such as diabetes present a substantial opportunity for cost savings that increase the value of care, but they also raise concerns about stinting on care that might ultimately benefit some subgroups of patients.


 Department of Health Care Policy, Harvard Medical School (BEL, AMZ, JS), Boston, MA; Division of General Medicine and Primary Care, Beth Israel Deaconess Medical Center and Harvard Medical School (BEL), Boston, MA; Institute for Healthcare Policy and Innovation; Division of General Medicine, Medical School; Department of Health Management and Policy, School of Public Health; Gerald R. Ford School of Public Policy, University of Michigan (JZA), Ann Arbor, MI.

 Supported by a grant from the National Institute on Aging (P01 AG032952).

 The authors report no relationship or financial interest with any entity that would pose a conflict of interest with the subject matter of this article.

 Concept and design (BEL, JZA); acquisition of data (BEL); analysis and interpretation of data (BEL, AMZ, JS, JZA); drafting of the manuscript (BEL); critical revision of the manuscript for important intellectual content (AMZ, JS, JZA); statistical analysis (BEL, AMZ, JS); obtaining funding (BEL); administrative, technical, or logistic support (BEL).

 Bruce E. Landon, MD, MBA, Department of Health Care Policy, Harvard Medical School, 180 Longwood Ave, Boston, MA 02215. Email: landon@hcp.med.harvard.edu.

1. 2019 Medicare Advantage and Part D prescription drug program landscape. CMS. September 28, 2018. Accessed February 9, 2021. https://www.cms.gov/newsroom/fact-sheets/2019-medicare-advantage-and-part-d-prescription-drug-program-landscape

2. Landon BE, Zaslavsky AM, Saunders R, Pawlson LG, Newhouse JP, Ayanian JZ. A comparison of relative resource use and quality in Medicare Advantage health plans versus traditional Medicare. 

3. Newhouse JP, McWilliams JM, Price M, Huang J, Fireman B, Hsu J. Do Medicare Advantage plans select enrollees in higher margin clinical categories? 

2013;32(6):1278-1288. doi:10.1016/j.jhealeco.2013.09.003

4. Newhouse JP, Price M, McWilliams JM, Hsu J, McGuire TG. How much favorable selection is left in Medicare Advantage? 

. 2015;1(1):1-26. doi:10.1162/AJHE_a_00001

5. Ayanian JZ, Landon BE, Zaslavsky AM, Saunders RC, Pawlson LG, Newhouse JP. Medicare beneficiaries more likely to receive appropriate ambulatory services in HMOs than in traditional Medicare. 

. 2013;32(7):1228-1235. doi:10.1377/hlthaff.2012.0773

6. Landon BE, Zaslavsky AM, Saunders RC, Pawlson LG, Newhouse JP, Ayanian JZ. Analysis of Medicare Advantage HMOs compared with traditional Medicare shows lower use of many services during 2003-09. 

. 2012;31(12):2609-2617. doi:10.1377/hlthaff.2012.0179

7. National diabetes statistics report 2020: estimates of diabetes and its burden in the United States. CDC. 2020. Accessed February 9, 2021. https://www.cdc.gov/diabetes/pdfs/data/statistics/national-diabetes-statistics-report.pdf

8. The cost of diabetes. American Diabetes Association. Accessed June 5, 2020. https://www.diabetes.org/resources/statistics/cost-diabetes

9. White WB, Cannon CP, Heller SR, et al; EXAMINE Investigators. Alogliptin after acute coronary syndrome in patients with type 2 diabetes. 

. 2013;369(14):1327-1335. doi:10.1056/NEJMoa1305889

10. Green JB, Bethel MA, Armstrong PW, et al; TECOS Study Group. Effect of sitagliptin on cardiovascular outcomes in type 2 diabetes. 

. 2015;373(3):232-242. doi:10.1056/NEJMoa1501352

11. American Diabetes Association. 10. Older adults. 

. 2015;38(suppl 1):S67-S69. doi:10.2337/dc15-S013

12. American Diabetes Association. 7. Approaches to glycemic treatment. 

. 2015;38(suppl 1):S41-S48. doi:10.2337/dc15-S010

13. Reusch JEB, Manson JE. Management of type 2 diabetes in 2017: getting to goal. 

. 2017;317(10):1015-1016. doi:10.1001/jama.2017.0241

14. Qaseem A, Wilt TJ, Kansagara D, Horwitch C, Barry MJ, Forciea MA; Clinical Guidelines Committee of the American College of Physicians. Hemoglobin A1c targets for glycemic control with pharmacologic therapy for nonpregnant adults with type 2 diabetes mellitus: a guidance statement update from the American College of Physicians. 

. 2018;168(8):569-576. doi:10.7326/M17-0939

15. Kalyani RR, Golden SH, Cefalu WT. Diabetes and aging: unique considerations and goals of care. 

. 2017;40(4):440-443. doi:10.2337/dci17-0005

16. Humphrey LL, Kansagara D, Qaseem A; High Value Care Committee of the American College of Physicians. World Health Organization guidelines on medicines for diabetes treatment intensification: commentary from the American College of Physicians High Value Care Committee. 

. 2018;169(6):398-400. doi:10.7326/M18-1148

17. Mann JFE, Ørsted DD, Brown-Frandsen K, et al; LEADER Steering Committee and Investigators. Liraglutide and renal outcomes in type 2 diabetes. 

. 2017;377(9):839-848. doi:10.1056/NEJMoa1616011

18. Wiviott SD, Raz I, Bonaca MP, et al; DECLARE-TIMI 58 Investigators. Dapagliflozin and cardiovascular outcomes in type 2 diabetes. 

. 2019;380(4):347-357. doi:10.1056/NEJMoa1812389

19. Neal B, Perkovic V, Mahaffey KW, et al; CANVAS Program Collaborative Group. Canagliflozin and cardiovascular and renal events in type 2 diabetes. 

. 2017;377(7):644-657. doi:10.1056/NEJMoa1611925

20. 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. 

. 2013;369(14):1317-1326. doi:10.1056/NEJMoa1307684

21. Marso SP, Daniels GH, Brown-Frandsen K, et al; LEADER Steering Committee; LEADER Trial Investigators. Liraglutide and cardiovascular outcomes in type 2 diabetes. 

. 2016;375(4):311-322. doi:10.1056/NEJMoa1603827

22. Wanner C, Inzucchi SE, Lachin JM, et al; EMPA-REG OUTCOME Investigators. Empagliflozin and progression of kidney disease in type 2 diabetes. 

. 2016;375(4):323-334. doi:10.1056/NEJMoa1515920

23. Zinman B, Wanner C, Lachin JM, et al; EMPA-REG OUTCOME Investigators. Empagliflozin, cardiovascular outcomes, and mortality in type 2 diabetes. 

. 2015;373(22):2117-2128. doi:10.1056/NEJMoa1504720

24. Yun H, Kilgore ML, Curtis JR, et al. Identifying types of nursing facility stays using Medicare claims data: an algorithm and validation. 

. 2010;10:100-110. doi:10.1007/s10742-010-0060-4

25. Medicare Advantage. Kaiser Family Foundation. June 2019. Accessed February 9, 2021. http://files.kff.org/attachment/Fact-Sheet-Medicare-Advantage

26. Chronic Conditions Data Warehouse. Accessed February 9, 2021. https://www.ccwdata.org/web/guest/home

27. Zaslavsky AM, Ayanian JZ, Zaborski LB. The validity of race and ethnicity in enrollment data for Medicare beneficiaries. 

. 2012;47(3, pt 2):1300-1321. doi:10.1111/j.1475-6773.2012.01411.x

28. Bonito AJ, Bann C, Eicheldinger C, Carpenter L. Creation of new race-ethnicity codes and socioeconomic status (SES) indicators for Medicare beneficiaries: final report. Agency for Healthcare Research and Quality. January 2008. Accessed November 20, 2014. http://archive.ahrq.gov/research/findings/final-reports/medicareindicators/medicareindicators.pdf

29. American Diabetes Association. Approaches to glycemic treatment. 

. 2016;39(suppl 1):S52-S59. doi:10.2337/dc16-S010

30. Standards of medical care in diabetes—2016: summary of revisions. 

. 2016;39(suppl 1):S4-S5. doi:10.2337/dc16-S003

31. Kirkman MS, Briscoe VJ, Clark N, et al; Consensus Development Conference on Diabetes and Older Adults. Diabetes in older adults: a consensus report. 

. 2012;60(12):2342-2356. doi:10.1111/jgs.12035

Articles

Medicare Advantage enrollees are more likely to be treated with metformin and sulfonylureas and less likely to receive costly newer medications than those in traditional Medicare.

Use of Diabetes Medications in Traditional Medicare and Medicare Advantage

 Michigan expanded Medicaid under the Affordable Care Act (ACA) through a federal waiver that permitted state-mandated features, including an emphasis on primary care. We investigated the factors associated with Michigan primary care providers (PCPs)’ decision to accept new Medicaid patients under Medicaid expansion.

Statewide survey of PCPs informed by semistructured interviews.

 After Michigan expanded Medicaid on April 1, 2014, we surveyed 2104 PCPs (including physician and nonphysician providers, such as nurse practitioners and physician assistants) with 12 or more assigned Medicaid expansion enrollees (response rate, 56%). To guide survey development and interpretation, we interviewed a separate group of 19 PCPs with Medicaid expansion enrollees from diverse urban and rural regions. Survey questions assessed PCPs’ current acceptance of new Medicaid patients.

 Of the 2104 surveyed PCPs, 78% reported that they were currently accepting additional Medicaid patients; 58% reported having at least some influence on the decision.

Factors considered very/moderately important to the Medicaid acceptance decision included practice capacity to accept any new patients (69%), availability of specialists for Medicaid patients (56%), reimbursement amount (56%), psychosocial needs of Medicaid patients (50%), and illness burden of Medicaid patients (46%). PCPs accepting new Medicaid patients tended to be female, minorities, nonphysician providers, specialized in internal medicine, paid by salary, or working in practices with Medicaid-predominant payer mixes.

 In the era after Medicaid expansion, PCPs placed importance on practice capacity, specialist availability, and patients’ medical and psychosocial needs when deciding whether to accept new Medicaid patients. To maintain primary care access for low-income patients with Medicaid, future efforts should focus on enhancing the diversity of the PCP workforce, encouraging healthcare professional training in underserved settings, and promoting practice-level innovations in scheduling and integration of specialist care.

In the era after Medicaid expansion, the factors considered important to Michigan primary care providers (PCPs) when deciding whether to accept additional Medicaid patients included the capacity to accept any new patients (69%), availability of specialists for Medicaid patients (56%), reimbursement amount (56%), psychosocial needs of Medicaid patients (50%), and illness burden of Medicaid patients (46%).

PCPs continuing to accept new Medicaid patients tended to be female, minorities, nonphysician providers, specialized in internal medicine, paid by salary, or working in practices with Medicaid-predominant payer mixes.

Policy makers should consider these factors in addition to reimbursement policies to ensure adequate PCP capacity in states with expanded Medicaid coverage.

As state and federal governments debate the future of the Affordable Care Act (ACA) and its Medicaid expansion, it seems likely that expanded Medicaid coverage will remain, but it will likely be modified by greater state flexibility in coverage and implementation.

 The impact of expanded Medicaid in any form depends on several factors, including, crucially, the acceptance of Medicaid by healthcare providers and systems.

Payment has long been emphasized as a driver of physician participation in Medicaid. Prior studies have found that reimbursement level is important to healthcare providers’ decisions to accept Medicaid.

 However, it is also important to carefully consider both the financial and nonfinancial factors that may influence providers’ participation in Medicaid and other programs.

Since the 1980s, research has examined several nonfinancial factors associated with physician Medicaid acceptance, including characteristics of “high-share” Medicaid providers, such as younger age, female gender, and nonwhite race.

 However, few studies have comprehensively examined which provider characteristics and practice settings may be associated with provider willingness to accept new Medicaid patients after the 1990s, when this was last examined in depth following the rapid growth of managed care. Given that influences on healthcare providers’ willingness to see Medicaid patients may have changed in the era after ACA Medicaid expansion, this question is worth revisiting.

Because primary care providers (PCPs) deliver frontline access to care for patients, understanding the factors associated with an adequate supply of PCPs who accept Medicaid patients is critically important. Access to care for Medicaid patients depends on several dimensions, including (1) providers’ willingness to accept any Medicaid patients, (2) the geographic distribution of available providers in relation to patients’ location, and (3) providers’ capacity and willingness to accept a greater number of Medicaid patients as demand rises (eg, after insurance coverage expansion). In the setting of Michigan’s Healthy Michigan Plan (HMP)—which expanded Medicaid under the ACA through a federal section 1115 waiver that permitted state-mandated features, including an emphasis on primary care

—prior studies have examined the first 2 dimensions of provider Medicaid acceptance. With regard to whether primary care practices accepted any Medicaid patients, these studies demonstrated increased availability of PCP practices offering appointments to HMP patients after Medicaid expansion

 despite rapid enrollment in the program,

 which was consistent with trends observed in 10 other states.

 Regarding provider distribution, these increases in appointment availability appeared to be geographically concentrated in regions with lower pre-expansion Medicaid acceptance.

The current study examines the third dimension of providers’ capacity and willingness to accept a greater number of Medicaid patients: whether PCPs who already have some Medicaid patients are willing to accept more in response to patient demand. In a recent study, Neprash and colleagues found that most physicians maintained or slightly increased their Medicaid participation rates after ACA Medicaid expansion, with greater increases observed in expansion states.

 Of note, approximately 20% of PCPs accounted for the care of 60% of Medicaid patients in that study. In other words, most increases in PCP Medicaid participation occurred among providers already caring for Medicaid patients. The authors of that study concluded that policy efforts to increase access to primary care should target existing Medicaid providers. Thus, focusing on PCPs already caring for Medicaid patients, we investigated what factors were associated with their continued acceptance of new Medicaid patients.

As part of a formal evaluation under contract with the Michigan Department of Health and Human Services (MDHHS) and required by CMS for a section 1115 waiver, we conducted a statewide survey of PCPs regarding their experiences with HMP enrollees, new practice approaches and innovations adopted or planned in response to HMP, and future plans regarding acceptance and care of HMP patients.

 As an evaluation of a public program, the University of Michigan and MDHHS institutional review boards deemed the study exempt.

The PCP sample was drawn from the MDHHS Data Warehouse, which stores data generated from encounters on all Medicaid and HMP enrollees and their providers, including provider demographics, specialty (assessed by linking National Provider Identifier [NPI] numbers from the Data Warehouse to each respondent’s National Plan and Provider Enumeration System profile), practice setting, and health plan participation. From the Warehouse, 7360 NPI numbers were identified as the assigned PCP for at least 1 HMP managed care enrollee (adult Medicaid expansion group) in April 2015, 1 year after the HMP launch. Nurse practitioners (NPs) and physician assistants (PAs) who were listed as the assigned PCP were also included in the sample. We considered PCPs with at least 12 assigned members (an average of 1 HMP enrollee per month) eligible for the survey in order to assess the perspectives of PCPs who had sufficient experience with HMP patients to provide informed input on the primary care—oriented features of the program for another related study.

 Thus, 2813 PCPs with fewer than 12 assigned members were excluded. Of the remaining 4547 PCPs, exclusions included 25 with an NPI entity code that did not reflect an individual provider (20 organizational NPIs, 4 deactivated, and 1 invalid), 161 with only pediatric specialty, 4 University of Michigan physicians involved in the HMP evaluation, and 35 with out-of-state addresses greater than 30 miles from the Michigan border. After exclusions, 4322 PCPs (3686 physicians and 636 NPs or PAs) remained in the survey sample.

Although we focused on a PCP sample that already had assigned Medicaid patients, we considered the representativeness of the sample relative to the typical PCP. As the average US primary care panel size is approximately 1200 to 1900 patients,

 the median number of assigned HMP members per PCP in our sample was 53; with HMP representing approximately one-third of all Michigan Medicaid enrollees, the estimated Medicaid proportion of the panel was approximately 6% to 15% of the total panel size. This is consistent with the national average Medicaid proportion of PCP panels, which has been reported to range from 9% to 20%.

The survey included standard measures of PCP demographic, professional, and practice characteristics, as well as items from prior surveys on decision making related to Medicaid patients.

 New items related to PCPs’ decisions to accept Medicaid patients in the context of the state Medicaid expansion were developed based on qualitative interviews. Items were subsequently cognitively pretested with 2 PCPs (1 physician from a safety net clinic and 1 PA from a private practice) to ensure understanding prior to survey administration. The final survey was also pretested with 1 PCP to ensure appropriate timing and flow.

 To guide survey development and interpretation, we conducted 19 semistructured interviews with a separate group of PCPs caring for Medicaid/HMP patients between December 2014 and April 2015. These interviews were conducted in 5 geographic regions across Michigan, purposefully selected to include racial/ethnic diversity and a mix of urban, suburban, and rural communities: city of Detroit, western Michigan, central lower Michigan, northeastern Michigan, and Michigan’s Upper Peninsula. Interviewees included both physicians and nonphysician providers (ie, NPs/PAs) who worked at small private practices, federally qualified health centers (FQHCs), free/low-cost clinics, hospital-based practices, or rural practices (

Interview topics included (1) provider awareness of patients’ insurance and experiences caring for HMP patients; (2) PCP involvement in decision making about whether to accept Medicaid/HMP patients; and (3) factors that may affect PCPs’ Medicaid acceptance decisions in the future, including knowledge of reimbursement changes such as the Medicaid primary care rate bump to the level of Medicare rates in 2013-2014.

Interviews were audio recorded and transcribed verbatim. Each transcript was reviewed iteratively by at least 2 members of the research team, with in-depth coding and thematic analysis

 performed using Dedoose software (SocioCultural Research Consultants, LLC; Manhattan Beach, California).

The survey was initially mailed to the PCP sample (N = 4322) in June 2015 and included a personalized cover letter describing the project, a fact sheet about HMP, a paper copy of the survey, a $20 bill, and a postage-paid return envelope. The cover letter also gave information on the option to complete the survey online. Two additional mailings with the same contents were sent to nonrespondents in August and September 2015. Data from mailed and online surveys returned by November 1, 2015, were included in the analysis.

 for the full survey instrument. The dependent variable was PCPs’ current acceptance of new Medicaid or HMP patients (“Are you currently accepting new patients with…[Medicaid; Healthy Michigan Plan; private insurance; Medicare; no insurance (ie, self-pay)]?”). In addition to standard PCP and practice characteristics, survey items measured independent variables of PCP attitudes regarding the importance of various patient- and practice-level factors in their practice’s decision to accept new Medicaid/HMP patients (eg, reimbursement amount, practice capacity to accept any new patients, specialist availability, and illness needs of Medicaid/HMP patients) and their experiences caring for and expressed commitment to caring for underserved populations.

Descriptive statistics were used to report PCP characteristics, current acceptance of new patients, and responses to other individual survey items. Bivariate and multivariable logistic regression analyses were used to assess the association of independent variables with the dependent variable of continued PCP acceptance of new Medicaid patients (yes/no).

PCP personal, professional, and practice characteristics with statistically significant associations at 

 <.01 were included as covariates in multivariable analyses, except for cases with expected collinearity (eg, nonphysician provider variable included but not specialty, payer mix variable included but not payment arrangement). For inclusion of items assessing attitudes toward underserved patients in the regression analyses, we created an index across all underserved attitudes items and calculated a score based on agreement level for multiple items (a score of 1 was assigned to a response of strongly disagree, 2 for disagree, 3 for neither agree nor disagree, 4 for agree, and 5 for strongly agree). Scores for individual items were summed to produce a scaled score for which higher numbers represented stronger agreement with commitment to caring for the underserved. A 2-sided 

 <.05 was considered statistically significant. All analyses were performed using Stata version 13 or 14.2 (Stata Corp; College Station, Texas).

Characteristics of PCP Survey Respondents

After excluding PCPs with undeliverable addresses (n = 501) and who were ineligible (n = 27; eg, retired, moved out of state) from the 4322 surveys sent, the final response rate was 56% (54% for physicians, 65% for NPs/PAs; n = 2104). See 

 for a flowchart of survey response rates. There were no significant differences between respondents and nonrespondents with regard to gender, age, number of affiliated Medicaid managed care plans, and practice setting in an FQHC; PCPs with internal medicine specialty were less likely to respond (see 

 for comparison of PCP survey respondents and nonrespondents).

Nearly half (45%) of respondents were female and 79% were white (

). Nonphysician providers (NPs/PAs) represented 17% of respondents. Family medicine (53%) and internal medicine (27%) were the most common specialties, and 82% of respondents were board certified. Most respondents (74%) had been in practice for at least 10 years. Fifteen percent practiced in an FQHC, and 35% had a payer mix that was predominantly Medicaid.

Although all survey respondents had at least 12 assigned patients with Medicaid/HMP coverage based on the survey sampling, 78% reported continuing to currently accept new Medicaid or HMP patients compared with 87% accepting new patients with private insurance, 84% with Medicare, and 76% with no insurance (

Attitudes Influencing PCPs’ Medicaid Acceptance Decision

Most PCP survey respondents reported having at least some influence on the decision to accept or not accept Medicaid or HMP patients; 23% reported “the decision is entirely mine,” 14% reported “I have a lot of influence,” 21% reported “I have some influence,” and 43% reported “I have no influence.”

In interviews, the 19 PCPs described influences on the Medicaid acceptance decision at various levels (see themes in 

). At the provider level, the illness burden and psychosocial needs of prospective Medicaid patients influenced PCPs’ decisions about whether to accept them, particularly for patients with complex chronic pain or mental health needs. At the practice level, the decision to accept new patients depended on the practice’s capacity to provide sufficient care to established patients, including timeliness of appointments and ability to provide high-quality care. At the health system level, PCPs’ decision making depended on the resources and administrative structure of their health system, including whether specialists were available to see Medicaid patients. Lastly, with regard to the policy environment, although most PCPs thought that reimbursement was important, many lacked knowledge of the 2013-2014 ACA primary care rate bump or other payment details yet continued to accept new Medicaid patients.

Based on these themes identified in the qualitative interviews, we asked PCP survey respondents to rate the importance of several patient- and practice-level factors to their practice’s decision to accept new Medicaid or HMP patients (“Please indicate the importance of each of the following for your practice’s decision to accept new Medicaid or Healthy Michigan Plan patients”) (

). Factors considered very/moderately important to the Medicaid acceptance decision included the capacity to accept new patients with any type of insurance (69%), availability of specialists who see Medicaid patients (56%), reimbursement amount (56%), psychosocial needs of Medicaid patients (50%), and illness burden of Medicaid patients (46%).

We also asked PCP survey respondents about their prior experience and attitudes toward caring for poor or underserved patients. More than half (57%) of respondents reported providing care in the past 3 years in a setting that serves poor and underserved patients with no anticipation of being paid. Nearly three-fourths (73%) felt a responsibility to care for patients regardless of their ability to pay, and nearly three-fourths (72%) agreed or strongly agreed that all practitioners should care for some Medicaid patients (see 

 for full descriptive statistics of PCP attitudes about caring for underserved patients).

Factors Associated With PCPs’ Medicaid Acceptance

In the era after Medicaid expansion, primary care providers placed importance on practice capacity, specialist availability, and reimbursement when deciding whether to accept new Medicaid patients.

John Z. Ayanian, MD, MPP

Articles

Use of Diabetes Medications in Traditional Medicare and Medicare Advantage

Factors Influencing Primary Care Providers’ Decisions to Accept New Medicaid Patients Under Michigan’s Medicaid Expansion

A Comparison of Relative Resource Use and Quality in Medicare Advantage Health Plans Versus Traditional Medicare