The instrument consists of seven dichotomous items and a five-point Likert scale. The questions were formulated to avoid a ‘yes-saying’ bias, so one point is assigned to each ‘no’ answer, except in item 5, which is reversed. The eighth question has five possible answers (scores from 0 to 1, in 0.25 point steps). At the end, patients are classified according to the score obtained as low adherers (score <6), medium adherers (score 6 to <8) and high adherers (score 8).6 The original authors also proposed a dichotomous cut-off of low (score <6) vs. high/medium adherence (score ≥6) to facilitate statistical analysis.

A process of translation and back-translation was performed according to international guidelines.15,16 Three bilingual translators who were aware of the goals of the study produced three independent translations of the original questionnaire into European Portuguese. The three versions were compared to generate a consensus version. The back-translation, from Portuguese to English, was conducted by another bilingual translator who was not involved in developing the initial version and who was not aware of the objectives of the study. This new English back-translated version was compared to the original version, and a few discrepancies were corrected. Finally, a cross-culturally adapted version was obtained through a consensus meeting attended by two experts in pharmacology and one Portuguese language expert who evaluated its semantic, idiomatic and cultural aspects and conceptual equivalence.

To ensure the instrument's face validity, a pilot test was performed on a Portuguese population (n=20), checking patient understanding and possible doubts and difficulties in the use of the questionnaire. After small adjustments, the final European Portuguese version of the questionnaire was obtained. The patients who participated in the pilot were not included in the following phases of study.

The MAT8 is a validated Portuguese questionnaire developed to assess patients’ adherence to medication. It consists of seven items scaled according to a six-point Likert scale ranging from ‘always’ (1) to ‘never’ (6). The level of adherence is obtained by adding together the points from each item and then dividing by the total number of items. Higher scores are indicative of greater levels of adherence. Patients are classified as adherent or non-adherent according to the median value of the population assessed.

The HKT17 is a 21-item questionnaire developed to assess patients’ knowledge about hypertension, its causes and treatment, and ways to prevent and control high blood pressure. It is divided into two parts: 12 true/false questions and nine multiple-choice questions. Levels of knowledge are calculated by assigning one point to each correct answer, obtaining a total score that ranges from 0 to 21. The higher the score, the greater the patient's knowledge about hypertension.

Data were collected between March 2014 and September 2015. Patients were invited to participate if they were over 18 years of age and were taking at least one antihypertensive drug. After informed consent was obtained, a trained pharmacist interviewed the patients in a private office in which data on personal and family history were self-reported, and the MMAS-8, MAT and HKT instruments were administered. The existence of hypertension, stroke, diabetes and dyslipidemia were considered proxies of cardiovascular disease.

To examine the factor structure of the European Portuguese version of the MMAS-8, a confirmatory factor analysis (CFA) was performed using AMOS version 20.0 (IBM SPSS AMOS, Meadville, PA, USA). Models were estimated using the maximum likelihood method. Model fit was assessed using the chi-square goodness-of-fit statistic, the comparative fit index (CFI), the standardized root mean square residual (SRMR), and the root mean square error of approximation (RMSEA). Model fit was considered to be very good when the CFI was above 0.95, the SRMR was below 0.10 and the RMSEA was below 0.5.18 Whenever a model did not fit the data well, alternative models were also tested. The development of competing models was based on theoretical considerations and analysis of the data, and two models were tested: one that included all eight items of the MMAS-8 loading on one global factor, overall adherence to antihypertensive medication, and another that included two correlated subscales, one on unintentional non-adherence behaviors with items 1, 2, 4, 5 and 8, and the other on intentional non-adherence behaviors with items 3, 6 and 7.

Internal consistency was examined via Cronbach's alpha, to assess whether various items that set out to measure the same general construct produce similar scores. To assess the correlation between MMAS-8 constructs and other similar adherence measure constructs, convergent validity was assessed by comparing the MMAS-8 and the MAT. Since patients’ knowledge about hypertension has been associated with better adherence levels, concurrent validity was assessed by comparing the MMAS-8 and the HKT, to assess whether the MMAS-8 correlates well with a hypertension knowledge measure.

A total of 472 patients were enrolled in the study, mean age 68.2±10.6 years, 243 (51.5%) female. The mean time since hypertension diagnosis was 10.9±8.5 years, with a maximum disease duration of 50 years. Among the most prevalent comorbidities, 255 (54%) had dyslipidemia, 133 (28%) diabetes, 127 (27%) heart disease, and 40 (8%) had had a stroke. The mean MMAS-8 score obtained was 6.74±1.39. Table 1 presents the characteristics of the study population using dichotomous cut-offs for adherence. Using the three-level cut-offs, 132 (28%), 181 (38.3%) and 159 (33.7%) of patients were in the low, medium and high adherer groups, respectively. Table 2 describes the answers obtained for the MMAS-8. The mean MAT score was 5.74±0.31, and mean HKT score was 15.35±2.79.

The first CFA model (model 1) included all eight items of the MMAS-8 loading on one global factor, ‘overall adherence to antihypertensive medication’. The model revealed a poor fit to the data, with chi-square 132.13, p<0.001; CFI 0.81; RMSEA 0.11 (90% confidence interval [CI] 0.09-0.13); and SRMR 0.07. Analysis of modification indices and assessment of the phrasing of each item suggested that the introduction of error correlations between items 2 and 5 and between items 6 and 7 would improve model fit, and these were therefore included in the model. The fit indices of this final unidimensional model (chi-square 48.465, p<0.001) suggest a very good fit, with CFI 0.95, RMSEA 0.06 (90% CI 0.04-0.08), and SRMR 0.04.

A second CFA model (model 2) was tested, examining the fit to the data of a theoretical model of two correlated subscales: subscale 1 with items on unintentional non-adherence behaviors and subscale 2 with items on intentional non-adherence behaviors. The same error covariances as in model 1 were introduced in this model. The final two-dimensional model (chi-square [17] 45.90, p<0.001) fit indices suggest a very good fit, with CFI 0.95, RMSEA 0.06 (90% CI 0.04-0.08), and SRMR 0.04.

The differences in chi-square values between the two prior final models revealed that the model fit for the two models was not statistically different (Δ chi-square 1.6, p=0.206).

The Cronbach's alpha for all items of a global scale was 0.60; removing any item did not affect alpha significantly. The total correlation coefficient between items for the eight items ranged from 0.11 to 0.64.

Considering the two-dimensional model (model 2), Cronbach's alpha was 0.65 for unintentional non-adherence behaviors and 0.31 for intentional non-adherence behaviors. The total correlation coefficient between items for unintentional non-adherence behaviors ranged from 0.26 to 0.62 and from 0.10 to 0.24 for intentional non-adherence behaviors (Table 3).

The MMAS-8 was highly and significantly correlated with the MAT (0.67, p<0.001), confirming that both instruments are assessing correlated constructs and have good concurrent validity.

The MMAS-8 was significantly correlated with the HKT (0.14, p=0.014), demonstrating a high convergent validity. Using Morisky et al.’s dichotomous classifications of patients as adherent with MMAS-8 score ≥6 and non-adherent with MMAS-8 score <6,6 there were significant differences in the HKT score between the two groups (t[295] 2.123, p=0.035). Adherent participants had greater knowledge about hypertension than non-adherent patients (HKT score 15.56±2.77 and 14.79±2.78, respectively).

The selection of a sample of hypertensive patients may be a limitation of this study, preventing generalization to other conditions.