Factors associated with non-adherence to insulin and non-insulin medications in patients with poorly controlled diabetes

Abstract

Objectives:

To evaluate differences in factors associated with self-reported medication non-adherence to insulin and non-insulin medications in patients with uncontrolled type 2 diabetes.

Methods:

In this secondary analysis of a randomized trial in patients with obesity and uncontrolled type 2 diabetes, multivariable logistic regression was used to evaluate associations between several clinical factors (measured with survey questionnaires at study baseline) and self-reported non-adherence to insulin and non-insulin medications.

Results:

Among 263 patients, reported non-adherence was 62% (52% for insulin, 55% for non-insulin medications). Reported non-adherence to non-insulin medications was less likely in white versus non-white patients (odds ratio (OR) = 0.42; 95%CI: 0.22,0.80) and with each additional medication taken (OR = 0.75; 95%CI: 0.61,0.93). Non-adherence to non-insulin medications was more likely with each point increase in a measure of diabetes medication intensity (OR = 1.43; 95%CI: 1.01,2.03), the Problem Areas in Diabetes (PAID) score (OR = 1.06; 95%CI: 1.02,1.12), and in men versus women (OR = 3.03; 95%CI: 1.06,8.65). For insulin, reporting non-adherence was more likely (OR = 1.02; 95%CI: 1.00,1.04) with each point increase in the PAID.

Discussion:

Despite similar overall rates of reported non-adherence to insulin and non-insulin medications, factors associated with reported non-adherence to each medication type differed. These findings may help tailor approaches to supporting adherence in patients using different types of diabetes medications.

Introduction

An estimated 34.2 million people in the United States, or 10.5% of the population, have diabetes.¹ Diabetes is the 7th leading cause of death in the United States.² Medication non-adherence in type 2 diabetes is associated with worse glycemic control,³ and inadequate diabetes control is in turn associated with increased morbidity, mortality, and costs.¹ Despite advances in diabetes medications, overall glycemic control has not improved in recent years,⁴ in part because medication adherence remains suboptimal. Estimates of diabetes medication non-adherence vary widely from 6.9% to 61.5% in prior reports.⁵

Adhering to diabetes medications can be challenging, particularly when patients use complex treatment regimens that include insulin.⁶ The degree to which the prevalence of medication non-adherence varies across insulin-based and non-insulin treatment regimens is poorly understood.⁷ Equally unclear are contextual forces that influence patients’ ability to adhere to diabetes treatment, including factors related to therapy, comorbid conditions, healthcare providers and the health system itself.⁸ Part of the difficulty in understanding factors associated with medication non-adherence to insulin-based and non-insulin treatment regimens relates to the fact that objectively assessing insulin non-adherence is difficult.⁹ A deeper understanding of factors associated with insulin-based and non-insulin treatment regimens could facilitate development of patient-centered approaches to enhancing medication adherence, which in turn could reduce the burden of diabetes nationwide.

In order to address current gaps in our understanding of diabetes medication adherence, we used data from a randomized, controlled trial (RCT) in overweight patients with uncontrolled type 2 diabetes to explore rates of non-adherence to insulin and non-insulin based medications. Furthermore, we evaluate associations between a broad set of patient factors and self-reported non-adherence to insulin and non-insulin based medications. Our ultimate goal in conducting these post hoc analyses was to generate information to help tailor approaches to supporting adherence in patients using different types of diabetes medications.

Methods

Patient population

The Jump Start trial (ClinicalTrials.gov NCT01973972) randomized 263 veterans with type 2 diabetes who were also overweight or obese to one of two 48-week interventions: 1) group medical visits (GMVs) focusing on self-management counseling and medication management (control); or 2) GMVs focusing on low carbohydrate diet weight management counseling and medication management.¹⁰ Upon enrollment, patients were administered a baseline series of questionnaires; the present analyses focus on this baseline survey. All patients provided written informed consent at the in-person RCT enrollment visit. The study was approved by the Durham Veterans Affairs (VA) Health Care System Institutional Review Board (study ID# 01794).

Patients included in this study fit the following criteria: 18–75 years old, diagnosis of type 2 diabetes (based on ICD-9 codes 250.x0 or 250.x2 or ICD-10 E11.xxx), hemoglobin A_1c ≥ 8.0% (≥7.5% if age <55 years), BMI ≥27 kg/m2, expressed an interest in losing weight, agreed to attend regular study visits per protocol, had access to a telephone and reliable transportation, and received primary care at a VA Medical Center. Patients were excluded if they had a condition making hemoglobin A_1c measurement unreliable, type 1 diabetes, chronic kidney disease, unstable coronary heart disease, uncontrolled hypertension (blood pressure ≥160/100 mmHg), uncontrolled hyperlipidemia (triglycerides ≥600 mg/dL or LDL-C ≥190 mg/dL), dementia, psychiatric illness, substance abuse, were enrolled in another weight loss program/study, or were pregnant or breastfeeding.¹⁰

Patient factors evaluated

The World Health Organization (WHO) has identified five sets of factors that may contribute to medication non-adherence and informed this analysis: social and economic factors, therapy-related factors, condition-related factors, patient-related factors, and health care team/system-related factors.⁸ Factors relating to the health-care team/health system were not captured within the Jump Start Study and were not addressed in this analysis.

Socioeconomic factors included age, sex, and race. Therapy-related factors included hypoglycemia,¹¹ total number of medications, and diabetes medication effect score (MES).¹² For hypoglycemia, we created a dichotomous variable indicating if participants reported having one or more hypoglycemic events at baseline. The total number of medications captured the patients’ diabetes and cardiovascular medications. The MES measures diabetes regimen intensity by considering the number, total daily dosage and potency (in terms of hemoglobin A_1c lowering ability) of medications used within that regimen. To calculate the MES, for each diabetes medication, a patient’s dose was calculated as a percent of the maximum daily dose for that medication. Insulin was considered to have a maximum daily dose of 1 unit/kg. Each medication was then multiplied by an adjustment factor based on its hemoglobin A_1c lowering potential. The total MES is the sum of the adjusted percentages for each medication.

Condition-related factors included hemoglobin A1c and EQ-5D.¹³ Hemoglobin A_1c was measured for each patient by the Durham VAMC Central Laboratory at the time of the baseline study assessment. The 5 L version of the EQ-5D- was utilized in this study. There are 5 questions in the EQ-5D, with responses on a 5-point Likert scale. Each question represents a different health state: mobility, self-care, usual activities, pain/discomfort and anxiety/depression. A single index value is created by mapping the responses to these 5 questions.

Patient-related factors that we evaluated were the Problem Areas in Diabetes (PAID) scale¹⁴ and the Patient Health Questionnaire-2 (PHQ-2).¹⁵ The PAID scale is a 20-item measure of self-reported diabetes-related distress related to chronic disease management. Each item is scored from 0 (not a problem) to 4 (serious problem) and the sum of all item scores is multiplied by 1.25 for the total PAID score. The PAID scores range between 0–100, and patients scoring ≥40 may be particularly at risk for distress relating to their diabetes. The PHQ-2 is a 2-item screening tool for depression and inquires about the frequency of depressed mood and anhedonia over the previous 2 weeks. Each item is scored from 0 (not at all) to 3 (nearly every day) and totaled; a score of ≥3 is the optimal cut point for screening for major depression.

Self-reported medication non-adherence outcome

Non-adherence was measured using the first version of a self-reported measure developed by Voils¹⁶ of 3 items assessing the extent of non-adherence to diabetes medications over the previous 7 days (non-insulin medications and insulin were assessed separately). This measure has been validated for use in chronic medical conditions, such as hypertension,¹⁶ dyslipidemia,¹⁷ and hepatitis C virus.¹⁸ In these conditions, the extent of non-adherence items produced reliable scores (alphas ranging from 0.84 to 0.91). The wording of the 3 items is generalizable to both insulin-based and non-insulin-based diabetes medication regimens.

The 3 extent of non-adherence statements were the following: “I missed my medicine”, “I skipped a dose of my medicine”, and “I did not take a dose of my medicine”. Responses were recorded based on a 5-point scale ranging from “none of the time” (1) to “every time” (5). Scores were recorded as the average of the 3 responses. One missing value was allowed, and in this case the score was recorded as the average of the 2 non-missing responses. Patients were considered non-adherent based on a score of >1 and if non-adherent to either insulin or non-insulin medications if taking both.

Statistical analysis

We fit separate logistic regression models to examine the association between patient factors and non-adherence for participants using insulin-based and non-insulin-based regimens. We included age, sex, race, hypoglycemia, total number of medications, MES, A1c, the EQ-5D score, the PAID score, and the PHQ-2 score as independent variables. Non-adherence, as measured by the Voils self-report measure, was the dependent variable. One model was fit to the sample of patients that were prescribed non-insulin medications (these patients could also be prescribed insulin but the analysis only used the non-adherence measure for prescribed non-insulin medications) for their diabetes (n = 244). The second model was fit to the sample of patients prescribed insulin (these patients could also be prescribed non-insulin medications but the analysis only used the non-adherence measure for prescribed insulin) for their diabetes (n = 165). Due to sample size constraints and to avoid model over fitting, model specification was identified a priori. In all models, linearity of the logit assumptions for continuous covariates were assessed, and higher-order terms were included for the PAID, PHQ and EQ-5D scores to meet these assumptions. All analyses were conducted in SAS version 9.4 (SAS Institute, Cary, NC).

Results

The majority of participants were non-White (57%, primarily African-American) and male (89%) with a mean age of 60.7 years (Table 1). The mean hemoglobin A_1c was 9.1%. For the 244 (95%) patients prescribed non-insulin medications, 55% reported non-adherence. For the 165 (64%) patients prescribed insulin, 52% reported non-adherence.

Table 1.

Jump start patient baseline characteristics by insulin use status.

Variable	Overall N = 263	Insulin users N= 171	Non-insulin users N = 92~
Social and economic factors:
Mean age (SD)	60.7 (8.2)	60.8 (8.1)	60.5 (8.4)
Sex
Male, n (%)	235 (89.4)	150 (87.7)	85 (92.4)
Female, n (%)	28 (10.6)	21 (12.3)	7 (7.6)
Race
White, n (%)	112 (42.6)	71 (41.5)	41 (44.6)
Non-White, n (%)	151 (57.4)	100 (58.5)	51 (55.4)
Therapy-related factors:
Any hypoglycemia, n (%)*	66 (25.1)	47 (28.1)	19 (21.1)
Mean total number of medications (SD)	5.7 (1.9)	6.2 (1.8)	4.8 (1.6)
Mean medication effect score (MES) (SD)	2.3 (1.1)	2.6 (1.0)	1.7 (0.9)
Condition-related factors:
Mean hemoglobin A1c (SD)	9.1 (1.3)	9.3 (1.3)	8.8 (1.1)
Mean EQ-5D (SD) – higher values indicate better quality of life^	0.7 (0.2)	0.7 (0.2)	0.8 (0.2)
Patient-related factors:
Mean Problem Areas in Diabetes (PAID) scale (SD) – higher score means greater emotional distress about diabetes#	30.5 (21.6)	30.2 (21.5)	30.9 (22.0)
Mean Patient Health Questionnaire-2 (PHQ-2) (SD) – higher score means greater depression symptoms+.	1.6 (1.7)	1.6 (1.7)	1.5 (1.7)

^~1 patient was not on any diabetes medication at baseline.

^*6 patients are missing data on hypoglycemic events (4 insulin users and 2 non-insulin users).

^{^}3 patients are missing data for the EQ-5D (3 non-insulin users).

^#4 patients are missing data for the PAID score (1 insulin user and 3 non-insulin users).

⁺11 patients are missing data for the PHQ score (5 insulin users and 6 non-insulin users).

Associations with self-reported medication non-adherence

In the model for those prescribed non-insulin medications (n = 244), the odds of reporting non-adherence were lower (OR = 0.42; 95% CI: 0.22, 0.80) in whites versus non-whites and lower (OR = 0.75; 95% CI: 0.61, 0.93) with each additional medication taken. The odds of reporting non-adherence were higher (OR = 1.06; 95% CI: 1.02, 1.12) with each one-point increase in the PAID score and higher (OR = 1.43; 95% CI: 1.01, 2.03) with every 1 point increase in the MES score. Additionally, the odds of reporting nonadherence were higher (OR = 3.03; 95% CI: 1.06, 8.65) in males versus females.

In the logistic regression model for those prescribed insulin (n = 165), the odds of reporting non-adherence were modestly higher (OR = 1.02; 95% CI: 1.00,1.04) with every one-point increase in the PAID score (Table 2). No other significant associations were found.

Table 2.

Logistic regression model results for the association between baseline patient factors and self-reported medication non-adherence for combined non-adherence to insulin or antihyperglycemic pills, non-adherence to insulin only, and non-adherence to antihyperglycemic pills only.

Variable	Insulina (n = 165c)			Antihyperglycemic pillsb (n = 244)
	OR	95% CI	p value	OR	95% CI	p value
Social and economic factors:
Age	1.52	0.92–2.49	0.100	0.96	0.92–1.00	0.076
Male Sex	1.01	0.36–2.85	0.984	3.03	1.06– 8.65	0.038
White Race	0.62	0.29–1.31	0.207	0.42	0.22–0.80	0.008
Therapy-related factors:
Hypoglycemia	1.26	0.58–2.69	0.556	0.95	0.46–2.00	0.902
Number of medications	0.87	0.70–1.08	0.211	0.75	0.61–0.93	0.008
Medication effect score (MES)	1.30	0.90–1.89	0.165	1.43	1.01–2.03	0.043
Condition-related factors:
Hemoglobin A1c	0.97	0.75–1.25	0.817	1.00	0.79–1.28	0.970
EQ-5D	1.00	0.97–1.02	0.750	0.56	0.31–1.00	0.051
Patient-related factors:
PAID Score	1.02	1.00–1.04	0.040	1.06	1.02–1.12	0.009
PHQ-2	N/Ad			1.85	0.90–3.78	0.092

^aC statistic for Insulin Non-adherence model = 0.705.

^bC statisitic for Antihyperglycemic pills model = 0.781.

^c6 patients are missing data on the adherence measure including the one patient not on diabetes medication at baseline.

^dPHQ-2 was omitted from the insulin non-adherence model to avoid overfitting the model.

Discussion

Medication non-adherence remains a major problem for management of chronic diseases, including diabetes. Because the relative prevalence of non-adherence to insulin and non-insulin diabetes medications remains unclear, and because factors associated with non-adherence to insulin and non-insulin medications may differ, we conducted post hoc analyses of RCT data to explore differences between clinical factors associated with non-adherence to insulin-based and non-insulin-based regimens. We found similar rates of non-adherence to insulin versus non-insulin medications (52% versus 55%, respectively), but different factors associated with non-adherence to insulin versus non-insulin medications. The odds of non-adherence to non-insulin medications were greater for patients who were male, non-white, taking fewer medications, had higher diabetes medication scores (from MES), and higher diabetes distress (from PAID) scores. The odds of non-adherence to insulin were greater for higher diabetes distress (from PAID) scores.

Implications of findings

While exploratory, our findings point to possible differences in factors associated with adherence to non-insulin medications and insulin, and identify important areas for future research. As such, the present study represents an important addition to the existing literature in this domain. A prior systematic review¹⁹ assessing factors associated with medication non-adherence to insulin in type 2 diabetes found only one study²⁰ that directly compared factors associated with non-adherence to insulin versus non-insulin medications. This study analyzed a large administrative database to examine the association between racial/ethnic and regional differences and the medication possession ratio (MPR) obtained from pharmacy records. A particular strength of our current study is its examination of prospectively-collected survey data, which are not available in electronic health records or administrative databases, and are difficult to collect retrospectively.

Although little is known about differences in factors associated with non-adherence to insulin and non-insulin medications, a deeper understanding of these factors could enhance clinical care. A prior study identified patient resistance as a major barrier to insulin initiation among primary care providers, such that providers often choose non-insulin regimens when insulin non-adherence is a concern.²¹ Another study identified 31% of women prescribed insulin would intentionally omit doses due to fear of weight gain, hypoglycemia, and diabetes-specific distress.²² With newer agents such as sodium glucose co-transporter 2 (SGLT2) inhibitors and glucagon-like peptide-1 receptor agonists (GLP-1 RA) available that are associated with weight loss and lower risk of hypoglycemia, non-insulin regimens are more feasible than ever when insulin non-adherence is suspected. The findings of the current study could not only help target intervention approaches for non-adherence in patients using different types of diabetes medications, but could also help guide medication selection in clinical practice.

Findings in the context of current literature

Our findings suggest that the odds of reporting non-adherence to non-insulin medications are 58% lower in white versus non-white patients (who were predominantly African-American in this study). These findings are highly concerning; prior studies have shown a higher risk of death from diabetes among African-American patients, and it is plausible that medication non-adherence contributes to this disparity.²³ A similar trend in racial differences was seen for non-adherence to insulin, although the results did not reach statistical significance. Factors like health literacy, differences in beliefs and concerns about the safety of medications, out-of-pocket costs, and patient-provider relationships,²⁴ may potentially mediated the observed association between race and non-adherence, so represent possible targets for interventions designed to reduce racial disparities in medication adherence in clinical practice.

Multiple studies have shown greater non-adherence with a higher number of diabetes medications, more complex regimens, and multiple daily doses.^25–27 Interestingly, our current findings showed that greater reported non-adherence to non-insulin medications was associated with higher diabetes medication intensity (as measured by the MES), but also with fewer combined cardiovascular disease and diabetes medications. A similar trend was seen for insulin, though this did not reach statistical significance. One possible explanation for this seemingly paradoxical finding may be that this study excluded patients with poorly-controlled hypertension and blood cholesterol; selecting for patients with better-controlled cardiovascular risk factors may have enriched our population with individuals successfully taking a higher number of cardiovascular medications, who also evinced better adherence. It is also possible that patients with more comorbid health problems and medications more readily realize the complexity of their medical condition, which may improve adherence. Individuals who are taking multiple cardiovascular medications may be more likely to take their diabetes medications as well and their regimens may not have been intensified due to non-adherence.

Simplification of diabetes medication regimens may be a worthwhile target for interventions to improve adherence. Studies have shown improved adherence in patients requiring escalation of their diabetes regimen when switching to a fixed-dose combination therapy compared to combination therapy with multiple agents.²⁶ Similarly, studies have shown improved adherence and similar clinical outcomes when initiating insulin in insulin-naïve patients by starting once daily basal insulin vs multi-dose regimens.^27,28 When intensifying a patient’s diabetes regimen, prioritizing regimen simplicity may lead to overall improved adherence, decreased costs, and potentially similar or even improved clinical outcomes. Further, a number of small randomized controlled trials have shown that deprescribing interventions, as was performed in the dietary intervention study arm of the Jump Start Study, improve adherence.^29–33 Therefore, for patients who are identified as at risk or having difficulty with adherence to their diabetes medications, clinicians should likely simplify regimens when clinically appropriate and feasible.

Similarly, patients with suboptimal quality of life (as measured with the EQ-5D in this study) may have difficulty adhering to complex medication regimens. Although not significant, the results of this analysis suggested a trend toward lower reported non-adherence with higher EQ-5D scores (indicating a better quality of life and better scores in the 5 domains of mobility, self-care, usual activities, pain/discomfort, and anxiety/depression).

Managing the complex lifestyle and medication regimens associated with diabetes often leads to diabetes-related distress. Half of patients with type 2 diabetes develop diabetes distress at some point during the course of the illness.³⁴ Diabetes distress reflects negative reactions to the diagnosis of diabetes, potential or actual complications, self-management burden, difficult relationships with health-care providers, and problematic social support.³⁵ The present analysis suggests that the likelihood of reporting non-adherence to both insulin and non-insulin medications was greater for higher diabetes distress scores, indicating that interventions targeting reduction in diabetes distress could improve adherence. A meta-analysis of 30 randomized controlled trials assessing the effects of different psychological interventions for diabetes-related distress in type 2 diabetes found a benefit on self-efficacy and hemoglobin A_1c.³⁴ Medication adherence was not an included outcome, but it is plausible that the observed improvements in hemoglobin A1c could be in part mediated by improved adherence, given the benefit seen in self-efficacy. In this analysis, the PAID score was associated with both non-adherence to insulin and non-insulin medications, implying that interventions aimed at improving diabetes distress could be beneficial, regardless of diabetes medication regimen.

In contrast, male sex was strongly associated with reporting non-adherence to non-insulin medications but not non-adherence to insulin. Given the small number of women in this study, the significance of this finding is unclear. Possible differences in the association between sex and adherence to non-insulin versus insulin medication should be a topic for future research using more balanced populations.

The remaining factors found to be significantly associated with reported non-adherence to non-insulin medications had trends in the same direction for insulin; the lack of statistical significance in these cases may represent type II error, given the smaller number of patients taking insulin versus non-insulin medications.

Limitations

Beyond those discussed above, the present study has additional limitations. This analysis was based on a population of US Veterans and therefore may not be generalizable to other populations. Hemoglobin A_1c was not associated with reported medication non-adherence in this study, potentially because the two measures covered different time frames; hemoglobin A_1c is a 3-month marker of glycemic control and the adherence measure used in this study covered the previous 7 days. The outcome measure of self-reported medication non-adherence may itself be subject to recall or social desirability bias. However, rates of non-adherence reported in this study are similar to previously reported studies of patients with poorly controlled type 2 diabetes with non-adherence measured by self-report.³⁷ Additionally, the self-reported measure of medication non-adherence used for this analysis has not specifically been validated among patients using injectable medications. Lastly, there may be other factors associated with diabetes medication non-adherence that could not be evaluated in the present analysis.

Conclusions

Among a population of overweight patients with type 2 diabetes and suboptimal glycemic control, we found similar rates of reported non-adherence to both insulin-based and non-insulin-based diabetes medication regimens. Despite similar rates in non-adherence, we found several factors that were associated with non-adherence to non-insulin medications (male sex, non-white race, taking fewer medications, more intense diabetes regimens, and higher diabetes distress) that were variably associated with non-adherence to insulin. These findings may help target approaches for supporting adherence in patients using different types of diabetes medication regimens.