Modifiable Factors Associated with Non-adherence to Antihypertensive or Antihyperlipidemic Drugs Are Dissimilar: a Multicenter Study Among Patients with Diabetes in Indonesia

We collected data from October 2018 to March 2019 from patients who met the following inclusion criteria: aged over 18 years, with a diagnosis of type 2 diabetes at least 1 year, were prescribed antihypertensive and/or antihyperlipidemic drugs for at least 3 months (prevalent users), and were literate. We excluded patients who had their medication picked up by someone else. The Health Research Ethics Committee of Universitas Padjadjaran approved the study protocol (no. 1137/UN6.KEP/EC/2018).

Adherence was assessed using the Medication Adherence Report Scale (MARS), which has shown to perform well on a number of psychometric indicators and internal-reliability.²⁸ The MARS has been translated and validated to Indonesian and showed to be valid and reliable.²⁹ The MARS contains one item that reflects unintentional non-adherence (“I forget to take my lipid-lowering medicines”) and four items that largely reflect different forms of intentional non-adherence (e.g., “I alter the dose of my lipid-lowering medicines”).²⁸ Patients indicate how often each item applied to them in the last 3 months on a 5-point Likert scale, where 5, “never”; 4, “rarely”; 3, “sometimes”; 2, “often” and 1, “always”.²⁸ Non-adherence is defined as a score of 1 to 3 on any of the items, and adherence as a score of 4 or 5 on all items allowing for rarely missing or changing a dose. We defined the subtypes of non-adherence a priori as follows:

Patients’ beliefs were assessed using Beliefs about Medicines Questionnaire (BMQ)-specific.²⁴ The Indonesian version of the BMQ-specific showed to be valid (correlation value of each question to the total score > 0.530) and reliable (Cronbach α coefficient of 0.835 and 0.811 for necessity and concern beliefs, respectively) (unpublished manuscript). The BMQ-specific contains five items about necessity beliefs (e.g., “My health at present depends on my lipid-lowering medicines”), five items about concern beliefs (e.g., “I sometimes worry about becoming too dependent on my lipid-lowering medicines”), and one item about side effects (e.g., “My lipid-lowering medicines gives me unpleasant side effects”). Patients indicate how often each item applied to them in the last 3 months on a 5-point Likert scale ranging from “strongly disagree” to “strongly agree” with an overall range from 5 (low necessity, low concern) to 25 (high necessity, high concern). We calculated the necessity–concern differential score by subtracting the scores of the concerns scale from the necessity scale (range − 20 to 20). A positive differential score indicates stronger beliefs in the necessity, while a negative score indicates stronger concern.²⁴ The item about experienced side effects was included because of its expected additional role in non-adherence.^{30, 31}

Sociodemographic factors included as non-modifiable factors were age at the completion of the questionnaire, gender, highest level of education completed (no formal education/elementary school, junior high school, senior high school, or university), and type of health insurance. Type of health insurance was classified as those whose insurance premium was paid by the government (BPJS-PBI), those whose insurance premium was paid by the patients themselves (BPJS-Non PBI), or those without health insurance. Clinical factors included as non-modifiable factors were obtained from medical records: time since diagnosis of diabetes, hypertension, and/or hyperlipidemia (years) and the most recent systolic blood pressure (SBP), diastolic blood pressure (DBP), and total cholesterol level in the past 3 months.

The pharmacist on duty at the CHCs screened the patients’ eligibility. Once a patient was deemed eligible, the pharmacist informed the researcher or research assistant to approach the patient and explain the study, and ask to sign informed consent. Consenting patients were asked to report the name of their antihypertensive or antihyperlipidemic drugs and subsequently filled in the MARS and BMQ-specific questionnaire. If patients used both antihypertensive and antihyperlipidemic drugs, the MARS and BMQ-specific questionnaires were administered for each therapeutic group. Patients were asked to complete the questionnaire independently. However, in some cases, elderly patients were allowed to complete the questionnaire verbally. Research assistants collected all other data from the medical records using a predefined data collection form. For those with incomplete or unavailable medical records, diagnostic data were obtained using patients printed record from the private laboratory they had visited.

A previous small-scale study showed that non-adherence rates among Indonesian patients with diabetes ranged from 50 to 65% using the MARS questionnaire.³² In studies elsewhere, similar and lower non-adherence rates have been found, also using the MARS questionnaire.^{26, 33} Therefore, a minimum sample size of 180 patients per therapeutic drug group was required based on the formula for prediction models with a binary outcome,³⁴ when including maximum of 9 possible independent variables in the multivariate analysis and assuming a proportion of non-adherence of 50%. With an expected distribution of 2:1 between patients receiving antihypertensive drugs and antihyperlipidemic drugs, 540 patients need to be recruited in the most conservative scenario of no overlap in the use of both therapeutic groups.

Analyses were conducted per therapeutic group. When a patient used both antihypertensive and antihyperlipidemic drugs, they were included for both therapeutic groups. Clinical factors related to hypertension or hyperlipidemia, such as duration of hypertension and/or hyperlipidemia and the most recent SBP, DBP, or total cholesterol level, were included only for the related therapeutic group. Descriptive statistics were used to summarize the patient characteristics. Pearson χ² tests, Mann–Whitney tests, or Kruskal–Wallis tests were used to assess univariate associations of patient characteristics with outcomes. Since there were few missing data regarding the MARS and BMQ, we conducted complete-case analyses. However, information about the number of medications and comorbidities could not be obtained for all patients due to incompleteness of medical records. The potential factors found to be associated with the outcomes at a significance level of p < 0.25 in univariate analyses (Tables 2 and 3) were included in the initial multivariate models. Two regression models were built for both therapeutic groups. Due to collinearity, the necessity–concern differential score was analyzed in a separate model not including the necessity and concern beliefs. Models with a higher R-squared value were then selected. In the first model, binary logistic regression with being adherent or non-adherent as the outcome was conducted to obtain odds ratio (OR) with a 95% confidence interval (CI) with manual backward elimination. In the second model, multinomial logistic regression with being adherent, unintentional non-adherent, intentional non-adherent, and in part intentional non-adherent as the outcomes was conducted to obtain OR and 95% CI. All statistical analyses were carried out using SPSS software (version 25.0; IBM, Armonk, NY, USA).

A total of 571 diabetes patients who were prescribed antihypertensive drugs (492 patients) and/or antihyperlipidemic drugs (245 patients) participated in this study (response rate of 97.1%) from Bandung City (6 CHCs; 133 patients), Makassar City (3 CHCs; 67 patients), Samarinda City (5 CHCs; 162 patients), and Yogyakarta City (18 CHCs; 209 patients). The mean values of MARS scores for those who were prescribed antihypertensive and antihyperlipidemic drugs were 22.2 and 22.1, respectively (Table 1). Less than half of the patients were male and most of patients were aged between 60 and 69 years and graduated from senior high school (Table 1). Patients included in the analyses with antihypertensive drugs had shorter diabetes duration than those in antihyperlipidemic drug analysis. More than half of the patients who were prescribed antihyperlipidemic drugs also received antihypertensive drugs, while one-third of those who were prescribed antihypertensive drugs received antihyperlipidemic drugs (Table 1). The median scores of necessity beliefs and concern beliefs were 15.0 (range 12.0–18.0) and 16.0 (range 12.0–18.0) to antihypertensive drugs, and 14.0 (range 12.0–17.0) and 16.0 (range 13.0–19.0) to antihyperlipidemic drugs, respectively.

Around half of patients were non-adherent to antihypertensive and to antihyperlipidemic drugs (45.5% and 52.7%, respectively) (Table 2). Patients were further classified as unintentional (14.4%, mean score 22.3), intentional (13.2%, mean score 20.4), and in part intentional (17.9%, mean score 18.3) non-adherent to antihypertensive drugs, and as unintentional (18.1%, mean score 22.6), intentional (6.6%, mean score 21.4), and in part intentional (28.0%, mean score 18.6) non-adherent to antihyperlipidemic drugs (Table 3).

From the univariate analyses, gender, age, last education level, specific co-medication, necessity beliefs, concern beliefs, side effects, and necessity–concern differential were selected as potential factors associated with non-adherence (Table 2). In the multivariate model, older age (60–69 years) (OR, 5.65; 95% CI, 2.68–11.92) was associated with non-adherence to antihypertensive drugs, while higher necessity beliefs (OR, 0.92; 95% CI, 0.88–0.95) was associated with less non-adherence (Table 4). The goodness-of-fit p value of this model was .351 with an R-squared value of 15.2%. The model including the necessity–concern differential had a lower R-squared value of 12.7% (Table S1 in Supplementary data). Similar patterns were seen for the subtypes of non-adherence. Patients with higher necessity beliefs were less likely to be unintentional (OR, 0.91; 95% CI, 0.86–0.97), intentional (OR, 0.93; 95% CI, 0.87–0.98), and in part intentional non-adherent (OR, 0.92; 95% CI, 0.87–0.97) (Table 4). Patients aged 60–69 years showed the highest odds of being unintentional, intentional, and in part intentional non-adherent (Table 4).

From the univariate analyses, gender, age, type of insurance, last education level, duration of hyperlipidemia, total cholesterol level, specific co-medication, necessity beliefs, concern beliefs, and necessity–concern differential were selected as potential factors associated with non-adherence (Table 2). In the multivariate model, significant factors associated with non-adherence to antihyperlipidemic drugs were higher concern beliefs (OR, 1.10; 95% CI, 1.03–1.18) and female gender (OR, 1.84; 95% CI, 1.03–3.27), while higher necessity beliefs (OR, 0.89; 95% CI, 0.83–0.96) was associated with less non-adherence (Table 5). The goodness-of-fit p value of this model was .716 with an R-squared value of 9.1%. The model including the necessity–concern differential had a lower R-squared value of 6.6% (Table S2 in Supplementary data). Regarding different subtypes of non-adherence, 1-year increase on the duration of diabetes was associated with an increase in the likelihood of intentional non-adherence (OR, 1.16; 95% CI, 1.04–1.30). Furthermore, patients with higher necessity beliefs were less likely to be unintentional non-adherent (OR, 0.89; 95% CI, 0.80–0.98), and patients with higher concern beliefs were more likely to be both unintentional (OR, 1.11; 95% CI, 1.01–1.22) and intentional (OR, 1.19; 95% CI, 1.03–1.37) non-adherent (Table 5).