INTRODUCTION

Diabetes, most of which is type 2 (T2D), is a major cause of morbidity and mortality in the United States (US), with 34 million adults affected in 2020.1 Besides lifestyle modification, two-thirds of patients take medications to control their hemoglobin A1c (HbA1c).2 In the past two decades, many new antihyperglycemic drugs have been approved by the Food and Drug Administration (FDA)—the most recent American Diabetes Association (ADA) Standards of Medical Care in Diabetes includes 20 different compounds for T2D in 6 preferred noninsulin classes and 20 insulin products.3

Patients with diabetes need medication options that address their medical and financial needs. Some newer medications have several advantages, including less hypoglycemia and less weight gain. For example, some drugs within the glucagon-like peptide-1 receptor agonist (GLP-1 RA) or sodium glucose co-transporter 2 inhibitor (SGLT-2) class reduce cardiovascular events, slow the progression of kidney disease, and even prevent death.4,5,6,7,8,9,10,11,12,13,14 After metformin, these drugs are now recommended for patients with established cardiovascular disease (CVD), chronic kidney disease (CKD), or heart failure.3 However, newer products usually demand a higher price. All SGLT-2 inhibitors, GLP-1 RAs, DPP-4 inhibitors, and analog insulins are categorized as high cost in the ADA’s Standards of Care.3 SGLT-2 inhibitors cost 4 to 12 times as much as sulfonylureas and GLP-1 RAs cost 8 to 20 times as much.3 The lower prices of generic medications should promote greater use, while high prices of newer, patented medications might limit use, but actual use of diabetes drugs by cost is unknown.

Health disparities in diabetes risk, treatment, and outcomes have been increasingly appreciated.15 Racial/ethnic minority groups have higher diabetes incidence, prevalence, complications, and mortality than non-Hispanic Whites.16,17 Patients living below the federal poverty level or with less than a high school education have increased diabetes-related mortality,18 and having health insurance is associated with both receiving diabetes care and having better glycemic control.19 There were racial/ethnic and income differences in use or initiation of GLP-1 RAs, SGLT-2 inhibitors, or DPP-4 inhibitors in commercially insured patients20,21 and clinical trial participants.22 However, information on access to high-cost medications in the general adult population and its changes over time remained limited. Given the increasing number of new expensive medications with additional clinical benefits beyond glycemic control, a better understanding of who receives such medications can contribute to the design of policy interventions to reduce disparities.

To address this need, we used population-based data to (1) estimate trends in use of diabetes medications by cost and (2) examine use of high-cost medications by race/ethnicity, income, and insurance status among US adults with T2D. We divided our analyses between non-insulin drugs, where newer classes provide mortality and other benefits over older classes, and insulins, where newer analog products are only marginally safer and generally no more effective than older human insulins.

METHODS

Data Source, Analysis Sample, and Study Design

We conducted a cross-sectional analysis of continuous National Health and Nutrition Examination Survey (NHANES) data from 2003 to 2018. We included people aged ≥18 years who had ever been told they had diabetes, had an HbA1C ≥6.5%, or fasting plasma glucose ≥126 mg/dL. We excluded pregnant women and patients with probable type 1 diabetes (aged <20 years or diagnosed with diabetes before age 20, and receiving only insulin). Use of prescription medications was ascertained from self-reports with bottles inspected by the interviewers.

Cost Category

We divided diabetes medications into 2 mutually exclusive groups: low-cost and high-cost. Because there is no standard criterion to identify low- versus high-cost medications, we adapted a flexible definition that accounted for the change in drug patents among noninsulin therapies. Low-cost medications included any drugs that had at least one generic version approved by the FDA. Because of the delay between FDA’s approval date and actual market entry date, we added 6 months to the approval date to estimate the generic’s availability date. If the availability date preceded the midpoint of a survey cycle, the drug was considered low cost (Appendix Table 1). Information about generic approval date was obtained from the FDA website.23 Generics tend to be less expensive than brand-name medications and are preferred by insurance plans although price differences between generic and brand-name medications can vary.24 All low-cost noninsulin medications except miglitol had a median average wholesale price (AWP) of <$100/month,3 and all except miglitol and acabose were included on Walmart’s $4 drug list.25 For insulins, we classified human regular, NPH and premixed NPH/regular 70/30 insulins as low-cost. Although these insulins had a median AWP of >$100/month,3 they are available over-the-counter for $25/vial at Walmart stores.26 All other noninsulin medications and insulins were considered high-cost. Because we did not know how many NHANES participants used Walmart insulin, we conducted a sensitivity analysis in which all insulins were considered high-cost. For patients whose insulin was unspecified, we did not assign them to either cost group in the primary analysis but to the high-cost group in the sensitivity analysis. In addition, because most insulin products were classified as high-cost, people who used high-cost drugs were also more likely to use insulin. We conducted an additional sensitivity analysis excluding patients taking insulin.

Statistical Analysis

Descriptive statistics of study participants were presented. For each 2-year survey cycle, we calculated weighted average number of drugs and prevalence of use by cost category for the whole study population. We then aggregated data from the 8 survey cycles to estimate the weighted prevalence and average number of high-cost drugs and the proportion of any high-cost drugs (insulin and noninsulin) relative to total drugs at the individual level stratified by race/ethnicity, income, and insurance status. We followed the National Center for Health Statistics guidelines27 and assessed nonlinearity in trends using polynomial regression with the survey cycle as a continuous variable. If nonlinearity was detected, we used the National Center Institute’s Joinpoint Regression Program version 4.9.0.1 to identify joinpoints, which were then evaluated with piecewise regression. If nonlinearity was not detected, logistic and linear regression with mid-year of the survey cycle as an independent variable were used to assess the significance of trends. We also used multivariable logistic regression to assess the association between race/ethnicity, income, and insurance status with the use of any high-cost diabetes drug, adjusted for other patient characteristics and survey cycle. A 2-sided p-value of <0.05 was considered statistically significant. We analyzed data using Stata 16.1, accounting for the complex survey design of the NHANES in all analyses.

RESULTS

Diabetes Medication Use by Cost

The final sample included 7,394 patients with T2D, representing more than 27 million Americans. Fifty-two percent were male. Mean age was 59.7 (SD=13.7) years, mean BMI was 33.1 (SD=7.6) kg/m2, and mean HbA1C was 7.2% (SD=1.7) (Appendix Table 2). The weighted prevalence of patients using low-cost noninsulin drugs increased from 37% in 2003–2004 to 54% in 2013–2014 (p-value<0.001) and stable thereafter. The weighted prevalence of high-cost noninsulin drug use decreased from 2003–2004 to 2013–2014 (p-value<0.001) and then slowly increased although trend between 2013–2014 and 2017–2018 was not significant (p-value=0.247) (Fig. 1a). Use of metformin and DDP-4 inhibitors increased from 2003–2004 to 2017–2018 while use of sulfonylureas and thiazolidinediones decreased significantly (Appendix Figure 1). Use of other noninsulin drugs represented only a limited share. In contrast, use of low-cost insulin declined steadily, from 7% in 2003–2004 to 2% in 2017–2018 (p-value<0.001), while the weighted prevalence of patients using high-cost insulin quadrupled from 4 to 16% during the same period, although the rate of increase was higher between 2003–2004 and 2007–2008 than between 2007–2008 and 2017–2018. The pattern was similar when we limited the analysis to pharmacologically treated patients who received ≥1 drug or to patients who were on combination therapy (Appendix Figure 2).

Figure 1
figure 1

Use of diabetes drug in US adults with type 2 diabetes, 2003–2018. a Prevalence (%) of use by cost, and b proportion of high-cost antihyperglycemic drugs relative to total drugs used per person in patients on combination therapy.

Full size image

Overall, the weighted average number of diabetes medications used per person increased from 1.00 to 1.14 during 2003–2018 (p-value=0.007). The weighted mean number of low-cost noninsulin drugs increased while that of high-cost noninsulin drugs decreased from 2003–2004 to 2013–2014 (p-values<0.001); both trends remained stable between 2013–2014 and 2017–2018 (Appendix Figure 3). In contrast, the weighted mean number of low-cost insulin prescriptions decreased significantly throughout the whole study period, while prescriptions for high-cost insulin increased from 2003–2004 to 2013–2014 and were stable thereafter. In addition, among patients using combination therapy, the weighted proportion of drugs that were high-cost did not change during the study period (Fig. 1b). More than 40% of medications used in these patients were high cost.

Disparities in Medication Use

When stratified by race/ethnicity, the weighted prevalence and number of high-cost drugs was significantly higher for non-Hispanic Whites than other racial/ethnic groups (Tables 1 and 2). The use of low-cost drugs did not differ by race/ethnicity. People without insurance used fewer antihyperglycemic drugs than did their insured counterparts, including both low- and high-cost drugs. Use of both low- and high-cost diabetes drugs did not differ by income. In multivariable analysis adjusted for other patient characteristics, non-White patients had between 25 and 35% decrease in the odds of using any high-cost drug compared to non-Hispanic Whites (Table 3). Health insurance was associated with more than 2 times greater odds of using high-cost drugs compared to those with no insurance, but there was no difference by insurance type. Patients with higher HbA1c values or who were moderately obese (BMI ≥35 kg/m2) were also more likely to use a high-cost drug. Sex, income, and educational attainment were not associated with high-cost drug use in adjusted analysis.

Table 1 Prevalence of Low- and High-Cost Diabetes Drug Use by Race/Ethnicity, Insurance, and Income in US Adults with Type 2 Diabetes, 2003–2018
Full size table
Table 2 Use of Any and High-Cost Diabetes Drugs by Race/Ethnicity, Insurance, and Income in US Adults with Type 2 Diabetes, 2003–2018
Full size table
Table 3 Association Between Race/Ethnicity, Income, and Insurance Status with the Use of Any High-Cost Diabetes Drugs Among US Adults with Type 2 Diabetes, 2003–2018
Full size table

Sensitivity Analysis

When we included all insulins in the high-cost medication group, findings were largely the same as in the main analysis. The proportion of any high-cost drugs relative to total drugs used per person did not change and disparities remained (Appendix Figure 4). Non-Hispanic Whites used more high-cost drugs compared to non-Hispanic Blacks, and uninsured people used fewer low- and high-cost drugs than their insured counterparts (Appendix Table 3). Use of diabetes drugs did not differ by income level regardless of their cost. In adjusted analysis, race/ethnicity and insurance were statistically significantly associated with increased odds of using high-cost drugs (Appendix Table 4).

When patients on insulin were excluded, use of high-cost drugs was significantly different by race/ethnicity, insurance, and income (Appendix Tables 5 and 6). In multivariable model, patients with a family income-to-poverty ratio of ≥4 had an increased odds ratio than those with a ratio of <1. In addition, insurance but not race/ethnicity was significantly associated with use of high-cost drugs (Appendix Table 7).

DISCUSSION

Using data from a nationally representative population-based survey, the 2003–2018 NHANES, we estimated trends in antihyperglycemic medication use by cost and examined disparities in receiving high-cost drugs. Overall, prescriptions for diabetes medications increased throughout the period, but patterns differed by cost. Prescriptions for low-cost noninsulin drugs increased steadily, while those for high-cost noninsulin drugs decreased through 2013–2014 and then increased as newer drugs came to market and cardiovascular outcomes trials demonstrated their benefits.4,9 The pattern for insulin was the opposite. Use of low-cost insulins fell to near zero and use of high-cost insulins quadrupled. High-cost drugs were particularly prevalent as add-on therapy. By 2017–2018, almost half of the drugs used in combination therapy were high cost. Consequently, patients with poor glycemic control and those with moderate obesity were more likely to receive high-cost medications. Other cost-related disparities were more troubling. Hispanic and other non-White patients were much less likely to receive high-cost drugs than non-Hispanic Whites, even after adjusting for insurance and income. This same disparity was not seen for low-cost drugs. Uninsured patients received fewer drugs, both high- and low-cost, compared to those with insurance.

Following the ADA’s recommendation, use of low-cost noninsulin drugs increased as physicians embraced metformin as first-line therapy. The decrease in use of high-cost noninsulin drugs from 2003–2004 to 2013–2014 coincided with sulfonylureas and pioglitazone going off patent during this period. After 2013–2014, high-cost noninsulin prescriptions slowly increased again as newer and more expensive drugs were approved. Interestingly, loss of patent did not stimulate use of sulfonylureas and thiazolidinediones.2,28 Instead, their use decreased gradually even before they went off patent, presumably due to concerns about potential side effects and lack of marketing. With the ADA’s latest recommendations to use GLP-1 RAs and SGLT-2 inhibitors for patients with CVD, heart failure, and CKD,3 the overall use of high-cost noninsulin drugs will likely continue to increase for the foreseeable future, as they will not become generic for at least 10–15 years. With the prevalence of diabetes increasing, the proliferation of high-cost noninsulin drugs will likely present a challenge for the health system as pharmaceutical expenditures impose a burden on all payers. Because our findings included periods before data on the additional health benefits of GLP-1 RAs and SGLT-2 inhibitors became available, the trends may be exacerbated in the near future.

If so, the high cost of new antihyperglycemic medications could worsen disparities in diabetes treatment. Hispanics and non-Hispanic Blacks have a higher incidence of diabetes and are less likely to achieve control of HbA1c, blood pressure, and cholesterol than non-Hispanic Whites.16,29,30,31 Other socioeconomic disadvantages, such as low income, low educational attainment, and insecure occupation, are also associated with diabetes and its complications.18,29 In addition, uninsured patients have fewer office visits, receive fewer medications, and have worse glycemic control than insured patients.32,33 Regarding access to newer, higher cost antihyperglycemic drugs, among commercially insured patients, non-White and lower income patients were less likely to take GLP-1 RAs and SGLT-2 inhibitors than Whites and high-income people.20,21 Among Look AHEAD participants, racial/ethnic minorities had a lower rate of initiating GLP-1 RAs, SGLT-2 inhibitors, or DPP-4 inhibitors than Whites.22 Initiation of GLP-1 RAs and SGLT-2 inhibitors was also greater in older patients with higher income or commercial insurance versus Medicare Advantage.34 Our study confirmed some of these disparities in access in a general adult population with T2D. We found that more disadvantaged groups (non-Whites and the uninsured) did not have comparable access to these drugs that provide health benefits beyond glycemic control. Reasons for this were unclear. In multivariable analysis, neither income nor education was associated with receipt of high-cost drugs, and there was no difference by type of insurance among insured patients. Non-Hispanic Whites might be more aware of newer drugs and may have requested them, or they may have attended practices that routinely prescribe higher-cost drugs. We did find that patients with health insurance, regardless of whether it was public or private insurance, received more diabetes drugs, both high-cost and low-cost, than did uninsured patients. While reassuring to see that government insurance was not a barrier to receipt of newer medications, the low rate of all treatment for uninsured patients was concerning.

While our findings regarding non-insulin drugs seem intuitive—use of less expensive drugs increased consistently and drugs that are more expensive increased when they were demonstrated to produce significant health benefits—our findings regarding insulin are harder to explain. Human insulin, which is 90% less expensive than analogs35 yet equally effective in glucose control for T2D,36 saw a precipitous decline in use throughout the study period. Use of insulin analogs, promoted by their manufacturers as safer and more convenient, quadrupled in frequency. This trend looks set to continue in future years, as younger generations of physicians lack training on how to prescribe human insulin. This is of concern especially when the price of insulin has been increasing over the past two decades.37,38 The average list price of insulin increased 15–17% annually from 2012 to 2016 and even more so in recent years, straining Medicare spending and patient out-of-pocket expenditures,39 and attracting the attention of Congress.40 In 2013, per-patient spending on insulin exceeded by 50% the expenditures on all other diabetes drugs combined.38 High prices prompt underuse, resulting in poor glycemic control or even death for low-income patients who cannot afford their insulin.41 In response, some health systems, most notably Kaiser Permanente, have turned to less expensive human insulins for T2D. If the HbA1c goal is not attainable with oral medications, patients are offered bedtime NPH insulin and insulin lispro as add-on therapy.42 Insulin detemir or glargine is covered only in cases of recurrent nocturnal hypoglycemia or severe hypoglycemia with NPH insulin.43,44,45 Starting in January 2021, the Centers for Medicare and Medicaid Services encouraged sponsors of part D program to participate in a Senior Savings Model which will design beneficiaries plan choices with a maximum co-pay of $35 per month for multiple insulin types.46 In addition, the retailer Walmart created its own ReliOn brand of low-cost NPH and regular insulin a few years ago and recently added insulin aspart in June 2021.47 In July 2021, the FDA approved the first interchangeable biosimilar insulin glargine.48 In states with mandatory substitution laws, this could reduce insulin costs by 25%, but it may take years to determine the impact on average insulin costs.

Strengths of our study include use of a nationally representative sample and having HbA1c data and detailed drug information. One important limitation was that NHANES included medication data only through 2018. Recent studies that demonstrated the benefits of newer, noninsulin medications on kidney disease, CVD, and mortality likely have spurred additional uptake of those drugs. In addition, NHANES does not distinguish between brand name and generic drugs, leading to potential misclassification bias. We assumed that patients took generic drugs, if available, as most insurance plans require generic substitution, but this may not always have been correct. Finally, patients might not be aware of the less expensive human insulin option at Walmart and might pay more for human insulin. We would have inadvertently included these individuals in the low-cost group. However, our sensitivity analysis that considered all insulin as high-cost was generally consistent with the main analysis.

In conclusion, over the past 15 years, there was a shift in drug use from high- to low-cost noninsulin medications, but trends will likely reverse as use of newer drug classes with cardiovascular, renal, and mortality benefits increases. Newer, high-cost insulin analogs completely displaced human insulins. Receipt of newer drugs differs by race/ethnicity, and this cannot be explained by income, education, or insurance type. These disparities may require interventions to ensure that disadvantaged populations have appropriate access to newer treatment options.