The Brief Medication Questionnaire and Morisky-Green Test to evaluate medication adherence

OBJECTIVE: To analyze the reliability and performance of the Portuguese version of questionnaires used to evaluate adherence to hypertensive treatment. METHODS: Hypertensive patients attending a primary healthcare unit in Porto Alegre, Southern Brazil, from January to September 2010, were randomly selected (n = 206). To evaluate adherence, Portuguese versions of the Morisky-Green test (MGT) and the Brief Medication Questionnaire (BMQ) were used. The analysis considered internal consistency, temporal stability and performance compared to three gold standards, which are: inadequate control of blood pressure (BP ≥ 140/90 mmHg); insuffi cient rate of medication acquisition at the institution’s pharmacy (<80%) and a combination of both factors. RESULTS: Of the patients studied, 97 only used medications dispensed by the Basic Health Unit. The tests showed good internal consistency by Cronbach’s α: BMQ 0.66 (95%CI 0.60 to 0.73) and the MGT 0.73 (95%CI 0.67 to 0.79). The BMQ Regimen Screen had a sensitivity of 77%, specifi city of 58%, and an area under the ROC curve of 0.70 (95%CI 0.55 to 0.86); for MGT sensitivity was 61%, specifi city 36% and area under the ROC curve 0.46 (95%CI 0.30 to 0.62). The correlation between the BMQ and the MGT was r=0.28, p> 0.001. Low adherence per the BMQ is associated with higher blood pressure levels when compared to adherent patients (148.4 [SD 20.1] vs 128.8 [SD 17.8]; p <0.001), but not for the MGT. CONCLUSIONS: The BMQ showed better performance than the MGT, with greater sensitivity and specifi city. Evaluation of adherence may help clinicians discriminate between inadequate use of medication and insuffi cient treatment regimen. DESCRIPTORS: Hypertension, therapy. Antihypertensive Agents, therapeutic use. Medication Adherence. Questionnaires, utilization. Sensitivity and Specifi city. Reproducibility of Results.


INTRODUCTION
Epidemiologic studies point to hypertension as a key risk factor for acute cardiovascular events. 24The control of blood pressure has decreased mortality from cardiovascular disease, and although the number of treated patients is increasing, more than 50% of hypertensive people have uncontrolled blood pressure. 25In observational studies, low adherence to hypertensive treatment has been considered a barrier to control of blood pressure. 6,8lthough there is no consensus, adherence to medications is understood as use of the prescribed medication at least 80% of the time, including hours, doses and length of treatment.Patients with a use less than 80% present four times the risk for acute cardiovascular risks. 13Various methods have been utilized to evaluate low adherence, such as self-report, manual and electronic counts of medication, retrieval of medication from pharmacies, laboratory tests for pharmaceuticals or metabolites of pharmaceuticals, and questionnaires. 10spite low sensitivity and accuracy, questionnaires are more often utilized due to low cost and ease of application in large populations.These instruments can be useful in differentiating between low adherence and non-response to antihypertensive medication, when used together with other methods such as electronic counts of pills (Medication Events Monitoring System -MEMS). 26The Morisky-Green Test (MGT), 16 the most utilized questionnaire in Brazil, performs poorly.This test, validated in the USA with hypertensives, utilizes control of blood pressure as the gold standard and presented a low sensitivity of 43.6% and reasonable specifi city of 81% in identifying non-adherent behavior.Evaluation of its performance in Portuguese was described in at least four studies with hypertensive people. 4,17,19,22We did not identify an evaluation of the Portuguese translation or its reliability in this language in the literature.The Brief Medication Questionnaire (BMQ) 23 was evaluated in English with 43 hypertensive patients using MEMS as the gold standard.The tool is divided in three screens that identify barriers to adherence of a drug regimen, beliefs and patient recall of medication treatment.The original study presented a sensitivity of 80% and specifi city of 100% in the regimen screen, although it has not yet been evaluated in Portuguese.Therefore, the BMQ tool appears potentially superior to the MGT, especially in screening for non-adherent behavior, but they were evaluated in different settings.Evaluation of both instruments in the same sample would improve understanding of their respective utility in the clinic and in research.
The objective of this study was to analyze the reliability and performance of these tools to evaluate adherence to antihypertensive treatment.

METHODS
The cross-sectional study used simple random selection of hypertensive people enrolled at least six months in the program to assist hypertensive and diabetic individuals (Hiperdia), in basic health units (unidade básica de saúde, UBS) of the city of Porto Alegre, Southern Brazil, from January to September of 2010.Of the The questionnaire included data on sociodemographics; clinical aspects such as comorbidities, physiology and name of the antihypertensives utilized; mode of obtaining medication; expenditures on antihypertensive medications and affi liation with a health service. 9Low adherence was evaluated utilizing two questionnaires: the Portuguese version of the MGT with four questions 16 and the newly translated Portuguese version of the BMQ with 11 questions (Annex).
The version of the MGT included the following questions: 1) Do you sometimes have diffi culty to remember to take your medication (Você às vezes tem problemas em se lembrar de tomar a sua medicação)? 2) Do you sometimes not pay attention to taking your medication (Você às vezes se descuida de tomar seu medicamento)?3) When you feel better, do you sometimes stop taking your medication?(Quando está se sentindo melhor, você às vezes para de tomar seu medicamento)?4) Sometimes, if you feel worse after taking medication, do you stop taking it (Às vezes, se você se sentir pior ao tomar a medicação, você para de tomá-la)?
The score obtained by the BMQ in each screen (regimen, beliefs and recall) were obtained by comparing patient responses to the prescription received.The medical prescription analyzed was documented in the charts of people affi liated with a UBS or a traditional prescription was provided by patients of other physicians or health services.
Other aspects encountered included: level of physical activity, measured by the short version of the International Physical Activity Questionnaire (IPAQ), 2 classifying individuals as sedentary and active; tobacco use defi ned as current use of any cigarettes; socioeconomic level established by criteria for Brazil 2008; a self-perceived health dichotomized in good/very good and fair/poor/very poor; 1 presence of common mental disorders according to the Portuguese version of the Self-Report Questionnaire (SRQ), 20 with a cutoff value of eight positive responses for women and seven responses for men.Comorbidities included obesity (body mass index, BMI ≥ 30 kg/m 2 ); cardiovascular disease (angina, infarction or intermittent claudication) evaluated by the Rose questionnaire 14 or history of cerebrovascular accident reported by the patient and/or recorded in medical records; chronic renal insuffi ciency (CRI) with endogenous creatine levels estimated by the Cockcroft-Gault formula at ≤ 30 ml/min/1.73m 2 ; 20 diabetes mellitus; glycemia ≥126 mg/dl, or use of hypoglycemia medications; dyslipidemia reported by the patient; use of cholesterol lowering drugs or record of total cholesterol ≥ 200 mg/dl, LDL ≥ 130 mg/dl; or triglycerides ≥ 150 mg/dl. 12ring interview, blood pressure was measured with an aneroid manometer (Missouri®, Brazil) using the average of two measurements.When the diameter of the upper arm was greater than 32 cm, a cuff for obese people was used.Height and weight were measured with a digital anthropometric scale (Welmy®, Brazil).Fasting serum levels measured included: creatine, glycemia, total cholesterol, HDL and triglycerides using the colorimetric enzyme method.
Questionnaire validation included: translation and back translation of the BMQ questionnaire and evaluation of internal consistency, temporal stability and performance in regards to the gold standards for the TMG and BMQ.
The BMQ was translated into Portuguese by two researchers and corrected by a third researcher that speaks native English.The questionnaire was then back translated into English by a translator.The version utilized is the result of comparing the original version to the back translation and correcting the differences identifi ed.This phase was performed by two physicians from Brazil with familiarity with English.Finally, the resulting version was administered to four individuals, not included in the study, for fi nal adjustments.A manual for applying the BMQ was developed to train interviewers.In the Portuguese translation of the BMQ (Annex), a simplifi cation was performed regarding the dosing of medications, since in the pilot study the patients could not recall the concentrations of medication per pill.In the original version of the questionnaire, the fi rst question asked the patient the name and dosage of the medications used.Failure in reporting any one of these items was considered a positive response and low adherence.In the translation, if the patient failed to report the name or class of medication, the response was considered positive for the fi rst item in the score of problems encountered by the BMQ for the regimen screen, irrespective of failure to recall dosage.It was assumed that content validity was performed by the authors of the original study.In the analysis of internal consistency, the correlation of each item with the sum of the items and inter item correlation, calculating a Cronbach α coeffi cient for each questionnaire.
For the analysis of temporal stability, patients with stable therapeutic schemes were retested at an interval of 14 to 30 days.Concordance between test and retest was evaluated by a gamma correlation coeffi cient.
The performance analysis for the BMQ and MGT used the descriptive statistics of sensitivity, specifi city and area under the ROC curve, considering three gold standards: 1-uncontrolled blood pressure (BP≥ 140/90 mmHg); 2-insuffi cient rate of medication acquisition from UBS pharmacy (acquisition <80% of medication in the period considered); and 3-combination of the fi rst two, which are uncontrolled blood pressure and insuffi cient acquisition of medication.Gold standards (2) and (3) were only considered in the sub-sample of patients whose therapeutic regimens included medications available from the UBS pharmacy during the entire study period (captopril, propranolol, furosemide and hydrochlorothiazide).
Gold standard 2 was calculated from the agreement between the three measures for acquisition of medicines as defi ned below: Continuous single-interval medication availability (CSA): number of days for which medications were provided divided by the interval of days between the last two acquisitions of medications.Dispensing is monthly.Month of interview was evaluated.
Medication possession ratio (MPR): number of days for which medications were acquired divided by the number of days between the fi rst and last acquisition during the six months before the interview.
Acquisition of Medication during the six previous months (AM6M): number of times the patient acquired medications in the pharmacy divided by six, considering the six months before the interview.
The Spearman coeffi cient was utilized to analyze the correlation between MGT and BMQ.The Kappa coeffi cient was used to analyze agreement between the two methods.Characteristics of hypertensive people were also described according to the level of adherence identifi ed by the MGT and BMQ.For the comparisons, chisquare tests, t tests and Mann-Whitney tests were used according to the distribution of variables.The significance level was 5%.To analyze potential confounding factors, the variables associated with low adherence at p< 0.1 in any of the questionnaires, were analyzed in a logistic regression model with the dependent variable as low adherence on the MGT or the BMQ.
The study was approved by the Research and Postgraduate Ethics Committee of the Hospital das Clínicas in Porto Alegre (appearance no.18883, on 14 July 2010).Participants signed voluntary informed consent forms.

RESULTS
Of the 206 patients evaluated, 105 only used medications available in the UBS pharmacy (group 1) and 101 used other antihypertensives besides those available in the pharmacy (group 2) (Table 1).Eight patients were removed from the fi rst group, since they purchased the medications utilized, for a total of 97 patients in group 1.These patients were not included in group 2 due to the requirement of different prescription.Comparison between groups showed the similarity regarding sociodemographic characteristics, level of physical activity, tobacco use, self-perceived health and other comorbidities investigated.Group  As the number of positive responses to the questionnaires increased, the specifi city in screening for low adherence also increased in relation to the three gold standards (Table 2).The BMQ regimen screen with a cutoff value of >1 for the score of problems identifi ed by the BMQ presented better equilibrium between sensitivity and specifi city for the three gold standards.This cutoff value can be utilized in screening for low adherence.
To identify a more specifi c clinical standard associated with the low adherence identified by the questionnaires, the profi les of patients with low and high adherence were analyzed.We considered high adherence as negative responses to all questions and low adherence as two or more positive responses in the MGT and in the score of problems identifi ed by the BMQ regimen screen.There were sociodemographic and clinical aspects in regards to blood pressure levels and greater prevalence of CRI among patients identifi ed with low adherence by the regimen screen, but not among patients with low adherence identifi ed by the MGT (Table 3).This indicates a nexus of low adherence and clinical outcome that is only being measured by the BMQ.Hypertensive people with low adherence per the BMQ regimen screen presented a greater average number of medications than those with high adherence.A medication usage pattern was not identifi ed among patients with low adherence as identifi ed by MGT.These differences are compatible with the low correlation between BMQ and MGT (r=0.28;p<0.001).The two methods did not present agreement (r= -0.14, p=0.56).We identifi ed a lower percentage of patients with self-perceived good or very good health among patients with low adherence as identifi ed by both tools.There were no signifi cant differences between the questionnaires in regards to expenditures on medications, type of medication utilized, number of daily doses and other characteristics.
Logistic regression analysis was performed to assess confounding by the factors that were associated in univariate analysis to low adherence in one of the screening tests with p < 0.

DISCUSSION
Reliability as evaluated by the analysis of internal consistency is ideal when the Cronbach α coeffi cient is greater than 0.7, but acceptable when above 0.6. 7oth questionnaires were correlated to the sum of their items, meaning that, within each questionnaire, the items measure the same concept.The MGT presented greater internal consistency than the BMQ, although lower temporal stability.
The analysis of BMQ showed that the regimen screen performed better than the other screens and the MGT for the identifi cation of low adherence among people with uncontrolled hypertension.This fi nding is similar to the original study performed in the USA, which used a more reliable gold standard (MEMS) than the insuffi cient acquisition of medications and uncontrolled blood pressure, although it was obtained in a smaller sample of patients (43 vs 206).We did not encounter studies evaluating BMQ in relation to control of blood pressure.In the present study, the BMQ regimen screen presented lower performance than in the original study -sensitivity of 80% vs 77% and specifi city of 100% vs 58.3% for the combined gold standard.This may be due to differences in the sample, culture, the gold standards and in the system that records acquisition of medications.Furthermore, in addition to the adaptations during Portuguese translations of the BMQ, a simplifi cation was introduced in regards to dosage.Nonetheless, we cannot affi rm that the validation using this adaptation infl uenced the different performance in comparison to the original BMQ. 23e MGT presented variable performance in previous studies, with a sensitivity of 43% 16 to 73.5% 19 and specifi city of 81% to 45.3%.The association identifi ed for BMQ between low adherence and the effects of treatment (blood pressure and presence of CRI) was absent for the MGT in the present study and in other studies. 17,22is fi nding reinforces the impression that the BMQ can discriminate between people with uncontrolled hypertension who do and do not take medications.This characteristic can make it useful to differentiate between low adherence and inadequate prescription, which is a frequent clinical dilemma.
The lack of correlation of the MGT with clinical outcomes motivated Morisky to broaden the questionnaire, adding  15 approximating our fi nding with the BMQ regimen screen (81.2%; p<0.001).In addition, the format of questions on the recall of medication usage, in the BMQ regimen screen, may facilitate its use in clinical practice.
The study presents some limitations.The lack of an appropriate gold standard and the multiple determinants of adherence complicate the analysis of results.There is no consensus on a gold standard method to evaluate adherence. 6Direct methods are used (serum measures of medications or metabolites) which are onerous and diffi cult to execute in the case of multiple medications, as is the case with hypertension.Indirect methods include measurement of medication dispensing, pill counts, questionnaires and clinical response to the medications, although indirect methods are more subject to measurement bias.Studies show low to moderate correlation between the methods, which can be attributed to the fact they measure different dimensions of the same construct, to different cutoff values for non-adherence, to limitations of the methods or to diffi culty in controlling the subjective factors related to the patient. 8Therefore, the selection of a way to evaluate adherence should consider the available resources in the health services and the strategies utilized and the strategies selected should follow the basic psychometric norms of reliability and validity. 10Finally, since no strategy is considered excellent, a multi-method approach has been uses in studies and in clinical practice. 6,10,11 could not evaluate the acquisition of medications for all patients, since half of patients utilized medication not available in our pharmacy.These fi ndings relative to the availability of antihypertensives were similar to the fi ndings of Bertoldi et al in regards to medication utilized for acute and chronic illnesses, which showed that only 51% of patients receive prescribed medication from the public system. 3Due to this limitation we evaluated less patients using gold standards 2 and 3, which may have infl uenced the results.The use of gold standard 2 was possible due to pharmacy information systems, which presents limitations in comparison to MEMS but have a satisfactory correlation with pill counts as shown by Steiner 21 (r=0.68;p=0.001).

REFERENCES
An inherent bias in the cross-sectional study design is the bias of reverse causality, since we evaluated exposure and outcomes at the same period and lost temporality 18 .This bias may have occurred in the present study when considering patients who needed to purchase antihypertensives and who more often presented with CRI and elevated blood pressure levels.The signifi cance of this fi nding may be related to greater gravity of the disease, inadequate treatment or low treatment adherence.Likewise, we found a smaller percentage of patients with good or very good self-perceived health among those with low adherence identifi ed by the two questionnaires.It was not possible to establish if this association, described by DiMatteo in a meta-analysis, 7 was caused by low adherence or resulted from perception of poor health.
Another limitation is selection bias due to the large number of losses, which may have occurred due to: incorrect information received or compiled, patients enrolled in a unit to receive medications but affi liated with other services and a population resistant to the regionalization of health services.The patients encountered showed a high percentage of affi liation with the health service, which may differ from the patients not encountered and limits the fi ndings to patients that access the service.This study also did not evaluate adherence to non-pharmaceutical treatment, which may be a confounding factor.Nonetheless, the variable for level of physical activity and the obesity percentage can be indirect measures to evaluate adherence to nonpharmaceutical treatment, which were not associated to adherence in our study.
In the BMQ regimen screen, we found 48.1% of patients with low adherence.The clinical profi le of these patients (higher blood pressure, greater prevalence of CRI and worse self-perceived health) is insuffi cient to identify low adherence; therefore a more objective method to evaluate adherence is needed, potentially using the BMQ regimen screen which was strongly associated with control of blood pressure.From the point of view of program planning in public health, such evaluation can indicate which patients should receive educational reinforcement, pharmaceutical support and multidisciplinary care and which require adjustment of therapeutic regimens.

Figure .
Figure.Performance of the tests according to the gold standard of uncontrolled arterial pressure (1) and the combined gold standard (3, uncontrolled arterial pressure + insuffi cient uptake of antihypertensive medication).Porto Alegre, Southern Brazil, 2010.

Table 1 .
Characteristics of all hypertensive individuals evaluated, patients that only use medications provided at basic health units (group 1) and patients that use other medications in addition to those available in the health unit (group 2).Porto Alegre, Southern Brazil, 2010.
* Tests utilized: chi-square for dichotomous variables; t Test for continuous variables with a parametric distribution and Mann-Whitney for continuous variables with a non-parametric distribution; AP controlled: controlled arterial pressure < 140/90 mmHg , Classes D and E: economic classifi cation criteria for Brazil 2008; R$: reais; BMQ: Brief Medication Questionnaire.

Table 2 .
Performance of the questionnaires according to the number of positive responses in the Morisky-Green Test and the Brief Medication Questionnaire in comparison to the three gold standards utilized.Porto Alegre, Southern Brazil, 2010.
Se=sensitivity; Sp=specifi city; gold standard 1: uncontrolled arterial pressure >140/90mmHg; gold standard 2: insuffi cient use of medications from the basic health unit pharmacy <80%; gold standard 3: combination of gold standards 1 and 2

Table 3 .
Characteristics of hypertensive patients according to level of adherence in the Morisky-Green Test and Brief Medication Questionnaire.Porto Alegre, Southern Brazil, 2010.