1

British Journal of Health Psychology (2018)

© 2018 The British Psychological Society
www.wileyonlinelibrary.com

Medication adherence for resistant hypertension:

Assessing theoretical predictors of adherence
using direct and indirect adherence measures
Hannah Durand1* , Peter Hayes2, Brendan Harhen3,
Ann Conneely4, David P. Finn4 , Monica Casey2 , Andrew W.
Murphy2,5 and Gerard J. Molloy1
1
Medication Adherence Across the Lifespan (MEDAL) Group, School of Psychology,
National University of Ireland, Galway, Ireland
2
Discipline of General Practice, School of Medicine, National University of Ireland,
Galway, Ireland
3
National Centre for Biomedical Engineering Science, National University of Ireland,
Galway, Ireland
4
Discipline of Pharmacology and Therapeutics, School of Medicine, National
University of Ireland, Galway, Ireland
5
Health Research Board Primary Care Clinical Trials Network, Galway, Ireland

Objectives. This study examined theoretical predictors of long-term medication

adherence (i.e., treatment-related beliefs, coherence of beliefs from experience with
medication, habit strength, and pill burden) for patients with apparent treatment-resistant
hypertension in primary care, using a composite adherence score derived from direct and
indirect measures (i.e., prescription refill, self-report, and bioanalytical assays of urine).
Design. Cross-sectional study.
Methods. Individual patient records were screened for prescription refill adherence.
Patients provided a urine sample for adherence screening and completed a battery of
psychometric scales, including two self-report adherence measures (N = 204). Conver-
gence of adherence measures was assessed, a composite adherence score was calculated,
and hierarchical multiple regression was used to examine the role of theoretical
predictors of adherence.
Results. Non-adherence estimates ranged from 20.3 to 41.1%, depending on the
assessment method used. Associations among adherence measures were weak to
moderate (q = .00–.53). Medication-taking habit strength was the strongest predictor of
adherence, explaining 19% incremental variance in adherence beyond treatment-related
beliefs. Beliefs and coherence did not predict adherence, even for patients with weaker
habits. Pill burden was not associated with habit strength or adherence for this sample.
Conclusions. Associations among unique adherence measures were weak overall,
providing further evidence that multiple measures are necessary to accurately assess
adherence. Habit strength is a key predictor of adherence for chronic conditions. Both

*Correspondence should be addressed to Hannah Durand, MSc, G050 AMBE, School of Psychology, National University of
Ireland, Galway H91 EV56, Ireland (email: [email protected]).

DOI:10.1111/bjhp.12332
2 Hannah Durand et al.

habit strength and pill burden represent important intervention targets for improving
long-term medication adherence. Longitudinal inception studies are needed to properly
test Common-Sense Model propositions and elucidate the role of beliefs, coherence, and
habits in predicting adherence at various stages of the chronic illness trajectory.

Statement of contribution
What is already known on this subject?
 Non-adherence to antihypertensives is a leading cause of apparent treatment-resistant hyperten-
sion (aTRH).
 Behaviour maintenance (vs. initiation) factors may be more predictive of long-term adherence.

What does this study add?

 Associations among direct and indirect measures of adherence are generally weak.
 Habit strength is the strongest predictor of long-term adherence for aTRH in primary care.
 Inception studies are needed to further validate Common-Sense Model propositions.

Medication non-adherence is a major contributor to negative health outcomes and elevated

health care costs. Poor adherence is of particular concern among hypertensive patients who
appear to be resistant to treatment. Resistant hypertension is characterized by uncontrolled
blood pressure despite concurrent treatment using three or more antihypertensive
medications of different classes, or concurrent use of four or more antihypertensive
medications regardless of control status (Calhoun et al., 2008; Mancia et al., 2013). Non-
adherence to antihypertensive medications precludes a diagnosis of true resistant hyperten-
sion; as such, until adherence is confirmed, patients are instead classed as having apparent
treatment-resistant hypertension (aTRH). Although patients who are non-adherent cannot be
classed as truly resistant to treatment and are instead classified as ‘pseudoresistant’ (Calhoun &
Grassi, 2017), these patients remain at elevated risk of adverse cardiovascular events as long as
their non-adherence is not effectively managed.
One potentially useful theoretical framework for describing and predicting non-
adherence among patients with aTRH is the Common-Sense Model of Self-Regulation
(CSM; Leventhal, Phillips, & Burns, 2016). The CSM posits that individuals create mental
representations of their illness based on the sources of information available to them in
order to make sense of and manage the health problem (Hagger, Koch, Chatzisarantis, &
Orbell, 2017). The CSM proposes that components of the individual’s illness represen-
tation (identity, causes, consequences, timeline, and controllability) affect how they will
behave in response to a health threat, for example by adhering to a prescribed treatment.
Although the CSM has been widely used to predict adherence behaviour, many CSM-
inspired studies focus on behaviour initiation factors (e.g., illness and treatment-related
beliefs) over other theoretically important processes involved in behaviour maintenance.
Recent empirical work (Phillips, Cohen, Burns, Abrams, & Renninger, 2016; Phillips,
Leventhal, & Leventhal, 2013) examined the role of two such processes in predicting long-
term adherence: ‘coherence’ of beliefs (i.e., a certainty in one’s beliefs regarding the
illness and treatment that results directly from personal experience that a treatment does
what it is expected to do), and medication-taking habit strength. This work draws on dual-
process theories (Hagger, 2016), which purport that individuals’ behaviour is controlled
by both deliberative/reflective processes (rational and conscious decision-making
influences on action), and implicit/impulsive processes (well-learned, spontaneous,
and non-conscious influences). Phillips et al. (2013) posit that long-term medication
adherence occurs if: (1) adherence behaviour is successfully initiated due to formation of
 Theoretical predictors of adherence for aTRH 3

treatment-favourable beliefs; (2) performance of the treatment provides evidence that the
treatment is working, thereby confirming the individual’s beliefs, resulting in a ‘coherent’
CSM and increased likelihood of treatment persistence; and (3) the medication-taking
routine is repeated enough to develop automatic contextual cues that maintain the
behaviour. Evidence for a variety of behaviours, including adherence, would suggest that
the effect of patients’ beliefs and intentions on adherence is moderated by habit strength,
such that when habits are strong, beliefs are weakly (if at all) predictive of behaviour
(Bolman, Arwert, & V€ ollink, 2011; Sheeran, 2002). Furthermore, beliefs, coherence, and
habit should have differential influences on intentional versus unintentional non-
adherence: habit strength should theoretically be more strongly related to unintentional
non-adherence, because habit strength reflects automatic processes that eliminate the
need to remember to take medication, whereas treatment-related beliefs and coherence
should be more reflective of intentional non-adherence, because both represent
deliberative, intentional processes. In testing this proposed model of long-term
adherence, Phillips et al. (2013, 2016) found that habit strength was the strongest
predictor of medication adherence, and that treatment-related beliefs and experiences did
not predict adherence even for patients with weaker medication-taking habits. Patients’
experiences with medication were shown to predict intentional non-adherence, whereas
habit strength predicted unintentional non-adherence.
Phillips’s work emphasizes habit strength as a key factor in maintaining adherence
behaviour for long-term treatments. Another important factor that has been consistently
associated with poor adherence is daily pill burden and regimen complexity (Burnier,
2006; Mathes, Jaschinski, & Pieper, 2014). Pill burden refers to the number of pills that a
patient takes on a regular basis. As this number increases, associated efforts (e.g., with
regard to storing, organising, consuming, and understanding various medications) also
increase, thus making the daily routine of medication-taking more complex (Gerbino &
Shoheiber, 2007; Ingersoll & Cohen, 2008). By definition, patients with aTRH are
prescribed multiple antihypertensive medications, which may have implications for habit
development. Taking multiple medications is accompanied by increased cognitive load
associated with monitoring and selecting the correct pills, thereby increasing the
behavioural complexity of the task. Development of automaticity for complex
behaviours, such as taking multiple daily medications, may take longer or reach a lower
level of automaticity than for simple behaviours (Lally, Van Jaarsveld, Potts, & Wardle,
2010; Verplanken, 2006). Therefore, we hypothesize an association between greater
antihypertensive pill burden and weaker habit strength.

Study overview
This article aimed to replicate and extend existing work examining theoretical predictors
of long-term medication adherence for patients with aTRH in primary care. Specifically,
we draw on previous research by Phillips et al. (2016, 2013), which used the CSM
extended to include coherence and habit as the guiding theoretical framework to
understand medication adherence in patients with hypertension. We aim to extend this
work by examining the effect of antihypertensive pill burden on habit and adherence in a
specific sample of hypertension patients (i.e., those who appear to be resistant to
treatment). Furthermore, this work will incorporate expert recommendations (DiMatteo,
2004; Vrijens, Antoniou, Burnier, de la Sierra, & Volpe, 2017) to assess adherence using
multiple direct and indirect measurement methods (i.e., self-report, pharmacy refill data,
and urine assay). Previous research has suggested that, while measurement type
4 Hannah Durand et al.

significantly moderates adherence estimates (Durand et al., 2017), combining measures

of adherence improves measurement sensitivity and increases diagnostic accuracy (Liu
et al., 2001; Sch€afer-Keller, Steiger, Bock, Denhaerynck, & De Geest, 2008). Therefore,
this study will utilize a composite adherence measure that incorporates both direct and
indirect measures of adherence to provide a comprehensive adherence assessment. The
hypotheses to be tested are as follows:

Hypothesis 1 (H1): Habit strengthandCSM coherence will eachaccount for incremental variancein
patient adherence to that accounted for by patients’ treatment-related beliefs,
with habit strength accounting for the greatest amount of unique variance.
Hypothesis 2 (H2): Patients’ treatment-related beliefs and medication-taking habit strength will
interact such that beliefs will predict adherence only for patients with weaker
habits.
Hypothesis 3 (H3): Habit strength will be more strongly related to unintentional non-adherence
than to intentional non-adherence; conversely, CSM coherence and treat-
ment-related beliefs will be more strongly associated with intentional non-
adherence than with unintentional non-adherence.
Hypothesis 4 (H4): Habit strength will be weaker for those with higher antihypertensive pill
burden.

Method
Design
The study design was cross-sectional with retrospective measurement of certain
outcomes (i.e., adherence as measured by prescription refill).

Procedure
Ethical approval was granted by the Clinical Research Ethics Committee at Merlin Park
University Hospital, Galway (Ref: C.A. 1386). Forty general practices in the University-
affiliated research network WestREN (Kavanagh, O’Brien, Glynn, Vellinga, & Murphy, 2010)
were invited to participate in a two-phase study of resistant hypertension in primary care. In
Phase 1, patient records were screened in Socratesâ patient management software, first using
a standard Anatomical Therapeutic Chemical drug search to identify patients on any
antihypertensive medication as defined by the British National Formulary 69th Edition (Joint
Formulary Committee, 2015). Records of patients identified as being on one or more
antihypertensive medications were then reviewed by two researchers (PH and MC) to
determine whether patients were hypertensive or not, and what antihypertensive medica-
tions they were currently prescribed. A detailed account of the patient record screening
process is available in Hayes et al. (2018). Patients meeting the criteria for aTRH were sent a
letter introducing the study and inviting them to participate in Phase 2. This involved patients
attending a clinical appointment at their GP clinic, during which they completed a
psychometric questionnaire, were fitted with a 24-hr ambulatory blood pressure monitor and
provided a urine sample for biochemical analysis.

Participants
Participants were primary care patients identified as having aTRH in a review of patient
records in 16 participating GP clinics in the west of Ireland. In Phase 1, 646 individual
 Theoretical predictors of adherence for aTRH 5

patient records were identified as meeting diagnostic criteria for aTRH. All patients had
been prescribed ≥3 antihypertensive medications for at least 3 months. Of the patients
screened during Phase 1, 95 (14%) were not invited to participate at the advice of their GP
due to significant morbidity and/or infirmity (e.g., nursing home residents, patients
undergoing active chemotherapy or radiotherapy for cancer, patients with severe
psychiatric illness, and patients who were housebound due to, e.g., neurological illness,
dementia, cardiac or renal failure). From a single practice, which was outside of the
referral area of a resistant hypertension clinic (Hayes et al., 2018), 77 patients were also
not invited to participate in Phase 2. A further 21 patients were not invited to take part
owing to a variety of reasons (i.e., moving practice, no longer taking antihypertensive
medications, or death). Of the remaining 453 patients, 239 (52.75%) agreed to participate
in Phase 2 of the study. Responders and non-responders did not differ significantly on any
demographic (i.e., age, sex, and GMS status) or clinical variables (i.e., blood pressure,
number of antihypertensive medications, duration of antihypertensive treatment, classes
of antihypertensive agents, kidney function, heart failure, and diabetes mellitus).
Prescription refill records were available for 517 patients (i.e., 80% of patients identified
in Phase 1). Self-report data were provided by 237 patients. A spot urine sample for
bioanalytical testing was provided by 235 patients. Adherence data using all assessment
methods were available for 204 patients.

Measures
Adherence measures
Self-reported medication adherence. Two scales were used to assess patients’ self-
reports of adherence: the Morisky Medication Adherence Scale (MMAS; Morisky, Ang,
Krousel-Wood, & Ward, 2008) and the Medication Adherence Report Scale (MARS; Horne,
2004). The MMAS (Morisky et al., 2008) is an eight-item measure with seven yes/no items
and with answer options ranging from ‘always’ to ‘never’ on a five-point scale. The MMAS
has been extensively validated (Krousel-Wood et al., 2009; Morisky & DiMatteo, 2011).
Conventional scoring was used to create a composite variable of the eight MMAS items, with
higher scores indicating better adherence. Internal consistency for the MMAS was .62. The
MMAS was used with permission from its developer. The MARS (Horne, 2004) is a five-item
measure with answer options ranging from ‘always’ to ‘never’ on a five-point scale, where
higher scores indicated better adherence. Internal consistency for the MARS was .77.

Intentional and unintentional non-adherence. To represent intentional versus

unintentional non-adherence, composite scores were calculated from the MMAS and
MARS by averaging the Z-scores of intentional (skipping, altering, or adding doses) and
unintentional (forgetting) items (Phillips et al., 2013). Internal consistency for intentional
and unintentional non-adherence composites was .77 and .72, respectively. For both
composites, items were ordered so that higher scores indicate greater non-adherence.

Prescription refill records. Prescription refill records were available for patients on the
General Medical Services (GMS) scheme (a means-tested scheme providing free health
care services and medication cover for eligible individuals in Ireland, e.g., those aged
≥70 years or on a reduced income) via Socratesâ practice management software.
6 Hannah Durand et al.

A majority of patients were in receipt of quarterly scripts (i.e., prescriptions for

medications to cover a 3-month period) as opposed to monthly scripts, thus precluding
the use of the typical <80% cut-off for non-adherence (Karve et al., 2009). Therefore,
using this measure, patients were considered adherent if they had collected ≥75% of their
printed scripts from the practice over the last 12 months (i.e., three quarterly scripts/nine
monthly scripts).

Biochemical assay of urine. Adherence was assessed by bioanalytical screening for 25 of

the most commonly prescribed antihypertensive medications or corresponding metabolites
(see Table 1) in spot urine using high-performance liquid chromatography coupled to mass
spectrometry (HPLC-MS/MS), using a methodology similar to Jung et al. (2013) and
Tomaszewski et al. (2014). Additional information regarding sample preparation and
analysis is provided in Appendix S1. Total non-adherence was defined as complete absence
of any prescribed antihypertensive medications (or their metabolites, where appropriate) in
a spot urine sample on screening. Patients whose urine analysis confirmed the presence of
fewer medications than prescribed were classified as partially non-adherent. Five medica-
tions (lercanidipine, nebivolol, spironolactone, felodipine, and candesartan) could not be
detected using HPLC-MS/MS and were therefore excluded from analysis. In line with Jung
et al. (2013), adherence ratios were calculated for each patient – based on the medications
that could be detected – by dividing the total number of antihypertensive medications
detected in spot urine into the total number of detectable antihypertensive medications
prescribed, whereby total non-adherence was equal to 0 and perfect adherence was equal to
1. Sensitivity analyses (i.e., excluding all participants for whom fewer than three
antihypertensives could be detected using HPLC-MS/MS) revealed that this did not
significantly bias the results; therefore, all participants were included in the analyses.

Adherence composite. As each adherence measure has established construct validity

but is also subject to varying kinds of measurement biases (Osterberg & Blaschke, 2005), a
unit-weighted composite adherence score was calculated by standardising and summing
scores from each individual adherence measure (Bobko, Roth, & Buster, 2007). The use of
multiple methods to assess medication-taking behaviour is recommended in this literature
due to the inherent limitations of each individual measure (Bond, 2016).

Pill burden. Antihypertensive pill burden was operationalized as the number of daily
antihypertensive medications prescribed, which was obtained from patient records.

Treatment-related beliefs. Health beliefs relevant to the patients’ hypertension med-

ication were measured using the Illness Perceptions Questionnaire–Revised (IPQ–R)
treatment control items (Moss-Morris et al., 2002) and the Beliefs about Medicines
Questionnaire (BMQ; Horne, Weinman, & Hankins, 1999). Both scales have response
options ranging from ‘strongly disagree’ (= 1) to ‘strongly agree’ (= 5). Scores on the IPQ-
R subscale represent how well patients think their medication can control their
hypertension, with higher scores indicating better control. Scores on the BMQ Specific
Necessities subscale represent how necessary patients believe their hypertension
medication is for their health, with higher scores indicating a greater necessity. Scores
 Theoretical predictors of adherence for aTRH 7

Table 1. Antihypertensive medications and/or their metabolites examined in urine

Antihypertensive medication
No. (metabolite) Ion mode

1. Amiloride Positive
2. Amlodipine Positive
3. Atenolol Positive
4. Bendroflumethiazide Negative
5. Bisoprolol Negative
6. Bumetanide Positive
7. Candesartan Positive
8. Diltiazem Positive
9. Enalapril Positive
10. Doxazosin Positive
11. Felodipine Positive
12. Furosemide Negative
13. Hydrochlorothiazide Negative
14. Indapamide Positive
15. Lercanidipine Positive
16. Lisinopril Positive
17. Losartan Positive
18. Nebivolol Positive
19. Olmesartan Positive
20. Perindopril Positive
21. Ramipril (ramiprilat) Positive
22. Spironolactone (canrenone) Positive
23. Telmisartan Positive
24. Valsartan Positive
25. Verapamil Positive

Note. Depending on their chemical structure (i.e., positively or negatively charged ions), medications/
metabolites were detected by positive or negative ion mode scanning, respectively.

on the BMQ Specific Concerns subscale represent how concerned patients are about their
hypertension medication, with higher scores indicating greater concern. A composite
variable (average) of treatment-specific health beliefs was used, consisting of the IPQ scale
items, the BMQ Specific Necessity scale items, and reverse-scored BMQ Specific Concern
scale items. This measurement approach replicates the approach taken by Phillips et al.
(2013), wherein items were selected for their theoretical alignment with the core
constructs and processes specified in the CSM.1 Internal consistency for the treatment-
related beliefs composite was .67.

Illness coherence. Two survey items assessed illness coherence according to the CSM:
‘Have you noticed the positive benefits of the hypertension medicine? Yes/No’, and ‘Have
you experienced any solid (convincing) evidence that the hypertension medication does

1
This approach to assessing treatment-related beliefs was first utilized by Phillips et al. (2013) for its theoretical alignment with
the CSM as it was intended by its developer, Professor Howard Leventhal. Although this approach has previous predictive validity,
we are not aware of any study that has specifically set out to validate this approach to operationalizing treatment-related beliefs.
Until further psychometric evaluation of this approach is conducted, results regarding the role of treatment-related beliefs should
be interpreted with caution.
8 Hannah Durand et al.

what it is supposed to do? No evidence/Some evidence/Solid evidence’ (Phillips et al.,

2013). These items were standardized and averaged to make a composite score for CSM
coherence, with higher scores indicating more coherent CSM beliefs. Internal consistency
of the coherence composite was .61.

Habit strength. The strength of patients’ antihypertensive medication-taking habits was

assessed using the Self-Report Behavioural Automaticity Index (SR-BAI; Gardner,
Abraham, Lally, & de Bruijn, 2012). The SR-BAI consists of four items drawn from the
Self-Report Habit Index (Verplanken & Orbell, 2003) that were most confidently and
consistently judged to capture behavioural automaticity. The SR-BAI has been demon-
strated to be a reliable, valid, and parsimonious self-report measure of habit (Gardner
et al., 2012). Items are rated on a seven-point response scale ranging from ‘strongly
disagree’ (=1) to ‘strongly agree’ (=7), with higher scores indicating stronger medication-
taking habits. Internal consistency for the SR-BAI was .87.

Statistical analyses. Analyses were conducted in SPSS 23. Missing self-report data were
imputed using expectation maximization. Analyses were conducted both with and
without the imputed data included to ensure that the imputation did not bias the data.
Associations among predictor and criterion variables were examined using Spearman’s
rho correlations. Hierarchical linear regression was conducted for the composite measure
of adherence and each continuous outcome (i.e., adherence measured by self-report and
urine assay). Logistic regression was conducted for dichotomous outcome variables (i.e.,
adherence measured by prescription refill). Analyses for individual adherence measures
are presented in Tables S1–S4. As per Phillips et al. (2013), the treatment-related beliefs
variable was entered in the first step of the regression, the CSM coherence variable was
entered in the second step, and the habit strength variable was entered into the last step.
For moderation analysis, the independent variable and moderator of interest were mean-
centred and entered together in the first step, and the interaction of the two variables (the
product of the mean-centred variables) was entered in the second step of the regression.

Results
Sample characteristics
A total of 204 participants with data available for all adherence measures were included in
this analysis. Sample characteristics are described in Table 2.

Prevalence of non-adherence to medication

Non-adherence estimates ranged from 20.3 to 41.1%, depending on the method of
measurement (see Table 2). The lowest and highest non-adherence estimates observed
were for prescription refill and self-report (MMAS), respectively. Using a cut-off of 23 on
the MARS2 (determined using Receiver Operating Characteristic analysis; Kumar &

2
The cut-off for the MARS (and the MMAS) to denote non-adherence was used for descriptive purposes only. The use of MARS
scores to produce a dichotomous adherent/non-adherent variable is not typical, nor is it recommended. Continuous MARS scores
should continue to be utilized in future predictive studies of adherence.
 Theoretical predictors of adherence for aTRH 9

Table 2. Sample characteristics

Possible range/observed
Variables M (SD)/% range

Female sex 42.2%

Age 69.86 (10.69) 18+/32–96
Married or in a relationship 60.3%
Completed secondary education 81.5%
No. of antihypertensive medications 3.68 (0.70) 3+/3–8
Adherence
MARS 24.25 (1.75) 5–25/5–25
MMAS 6.34 (1.06) 0–8/0–8
Prescription refill (% adherent) 79.7%
Urine assay
Total non-adherence 2.1%
Partial non-adherence 23.8%
Adherence ratio 0.84 (0.23) 0–1/0–1

Indrayan, 2011) and 6 on the MMAS (Morisky et al., 2008) to denote non-adherence, the
respective self-reported non-adherence estimates were 36.7% on the MARS and 41.1% on
the MMAS. According to the urine assay measure, 26% of patients had evidence of non-
adherence to at least one of their prescribed antihypertensive medications.

Associations among adherence measures

Adherence summary statistics are shown in Table 2. Both self-report and urine assay
measures of adherence were significantly skewed, thereby violating the assumption of
normality. As such, Spearman rank correlation coefficients (q) were used. As expected,
the MMAS and the MARS were significantly correlated (q = .52, p < .001). Both self-report
measures were significantly associated with prescription records; however, the corre-
lation coefficients suggested only weak to moderate associations. There was no significant
association between adherence measured by urine assay and any other measure of
adherence. The composite adherence score was significantly correlated with each
component measure (p < .001). Correlation coefficients are displayed in Table 3.

Composite adherence score

All adherence values were standardized, and analysis of internal consistency was carried
out to establish that individual adherence items were sufficiently related to form a
composite score. All items were positively correlated and Cronbach’s a remained stable
when each item was iteratively removed, suggesting that all adherence measures could be
reliably included in a composite measure. The scores for each component measure were
converted to Z-scores, and the sum of the standardized scores was calculated to produce a
unit-weighted composite score (i.e., all component measures were equally weighted). A
unit-weighted composite was deemed the most appropriate, given the measures are
theoretically related and each has its own valid strengths and limitations. Internal
consistency for the composite adherence measure was .76. Correlation coefficients
between the composite measure and each individual measure of adherence are displayed
in Table 3.
10 Hannah Durand et al.

Table 3. Correlations among predictor and criterion variables (N = 204)

1 2 3 4 5 6 7 8 9

1. Prescription refill –
2. MARS .30‡ –
3. MMAS© .17* .53‡ –
4. Urine assay .00 .01 .04 –
5. Composite .44‡ .74‡ .82‡ .25‡ –
6. Beliefs .02 .19† .19† .05 .22† –
7. Coherence .02 .04 .13* .04 .11 .32‡ –
8. Habit strength .08 .36‡ .35‡ .02 .36‡ .22† .06 –
9. Pill burden .01 .00 .08 .07 .08 .05 .05 .02 –

Note. The use of the MMAS© is protected by US copyright laws. Permission for use is required. A licence
agreement is available from: Donald E. Morisky, MMAS Research) LLC 14725 NE 20th St. Bellevue WA
98007.
*p < .05; †p < .01; ‡p < .001.

Associations among predictor variables

No strong relationships were observed between independent variables (r < .4). Variance
inflation factor values (<5) and tolerance values (>.1) for all independent variables were
adequate, thereby showing that there was no issue with multicollinearity in the data.

Hypothesis testing
The subsequent analyses (H1–H3) utilize a composite measure of adherence. Regression
tables for each individual measure of adherence are available in Tables S1–S4. Results for
H1 and H2 are displayed in Table 4. Results for H3 are displayed in Table 5. Results for H4
are displayed in Table 3.

H1: The first hypothesis was partially supported. Favourable treatment-related beliefs
significantly predicted adherence in the first step. Counter to the hypothesis, CSM coherence
did not have any significant predictive value beyond treatment-related beliefs. As hypoth-
esized, the incremental variance explained by the measure of habit strength in the third step
of the regression was significant. Habit strength explained 19% of the variance in patient
adherence in addition to that already explained by beliefs and coherence.

Table 4. Hierarchical regression analysis of theoretical predictors of adherence (H1, H2)

b Adj. R2 DR2 F DF

H1
Treatment-related beliefs .14 .04 .05 7.59† 7.59†
CSM coherence .01 .04 .00 3.88* 0.20
Habit strength .44‡ .22 .19 15.51‡ 36.96‡
H2
Beliefs 9 habit strength .07 .23 .01 16.48‡ 0.95

Note. *p < .05, †p < .01, ‡p < .001.

 Theoretical predictors of adherence for aTRH 11

Table 5. Comparison of prediction of unintentional versus intentional non-adherence (H3)

Unintentional non-adherence Intentional non-adherence

b t (203) p b t (203) p

Beliefs .05 0.79 .43 .07 1.01 .32

Coherence .04 0.56 .58 .03 0.35 .73
Habits .45 7.04 <.001 .22 3.08 <.01

H2: Contrary to the hypothesis, the interaction between patients’ treatment-related beliefs
and habit strength was not significant for the composite adherence measure.
H3: The third hypothesis was partially supported. As hypothesized, habit strength was more
strongly associated with patients’ unintentional non-adherence, b = .45, t (203) = 7.04,
p < .001, than intentional non-adherence, b = .22, t (203) = 3.08, p < .01. However,
counter to the hypothesis, neither treatment-related beliefs nor CSM coherence was
associated with intentional or unintentional non-adherence.
H4: The fourth hypothesis was not supported. As seen in Table 3, there was no association
between pill burden and habit strength (q = .05) and no association between pill burden and
adherence (q = .08) in this sample.

Discussion
The current study examined theoretical predictors of medication adherence (i.e.,
favourable beliefs, experiential feedback, and habit strength) for patients with aTRH. As
previously demonstrated by Phillips et al. (2016, 2013), habit strength was the strongest
predictor of a composite measure of adherence consisting of self-report, prescription
refill, and urine assay data (see Tables S1–S4 for results using individual measures). Also in
line with Phillips et al. (2013), treatment-related beliefs and habit strength did not
interact, suggesting that medication-taking habits are more predictive of adherence than
treatment-related beliefs even when the habit is weak. Contrary to previous findings,
treatment-related beliefs and CSM coherence were not more predictive of intentional non-
adherence than habit strength; in fact, habit strength was the only significant predictor of
either intentional or unintentional non-adherence, although the relationship was stronger
for unintentional non-adherence. Finally, there was no association between habit strength
and the number of antihypertensive medications prescribed, nor between pill burden and
adherence. Cumulatively, the current findings provide additional support for the
extended CSM proposed by Phillips et al. (2013) for predicting medication adherence
in the long-term. However, as discussed in detail in the limitations section below, more
research is needed to further validate this model.
The current study also examined concordance between prescription refill records,
self-report, and bioanalytical assessment of spot urine using HPLC-MS/MS for the
assessment of antihypertensive medication adherence in primary care. It is repeatedly
argued that there is no gold standard measure of adherence and that, in order to overcome
each measure’s limitations with respect to validity and reliability, a combination of
measures should be used (DiMatteo, 2004; Durand et al., 2017; Gupta, Arshad, & Poulter,
2010; Lam & Fresco, 2015; Vrijens et al., 2017). Results indicated significant associations
between indirect measures of adherence for a sample of aTRH patients prescribed
multiple daily medications. There was no significant relationship between bioanalytical
measures and any other measure of adherence. Despite this, analysis of internal
12 Hannah Durand et al.

consistency revealed that all adherence measures were sufficiently related to produce a
meaningful and useful composite adherence score. Composite measures of adherence
have the potential to overcome limitations of each individual adherence measure while
also combining their strengths, potentially resulting in the most comprehensive
adherence assessment available. Composite adherence measures have been utilized in
previous research and shown to be highly sensitive and demonstrate good specificity (Liu
et al., 2001; Sch€afer-Keller et al., 2008). Our analyses show that multiple adherence
measures may be statistically compiled to provide a comprehensive estimate of adherence
behaviour that can be utilized in similar predictive studies. Future research should
consider such an approach to further evaluate the utility of composite adherence
measures, ideally utilizing longitudinal designs and incorporating additional validated
measures such as electronic monitoring.
Urine assays were not associated with any other measure of adherence. Barring direct
observation, bioanalytical testing of bodily fluids is the only adherence assessment that
can verify ingestion of medication. However, these measures provide only a single
snapshot in time and may not be reflective of typical adherence behaviour. It is possible
that the act of ingesting medication is too distinct from the adherence behaviours assessed
by indirect measures (e.g., collecting a prescription) to produce a statistical association.
Interestingly, although urine assays were not associated with either self-report measure of
adherence, they were significantly associated with items reflecting intentional non-
adherence across both the MARS and the MMAS (q ≥ .15, p < .05). Relatively few patients
reported intentional non-adherence, which, although consistent with previous adher-
ence research (Moon, Moss-Morris, Hunter, & Hughes, 2017), may explain why a
significant association did not emerge for total self-report adherence scores. Similarly, only
27 patients reported not taking their medication the day before the clinical appointment
(i.e., the period for which the urine analysis is most relevant), which may suggest a
potential reporting bias. Although further research with greater variability in terms of
intentional versus unintentional adherence is needed (see limitations below), this may
have important clinical applications for patients who are deliberately non-adherent but
choose not to report this to their GP. Objective physical measures may allow clinicians to
identify adherence issues for patients who are reluctant to discuss, for example, their
concerns about treatment or related side-effects. This approach may be particularly cost-
effective for aTRH patients, for whom the cost of an assay may be less than a monthly
supply of antihypertensives (Gupta et al., 2017). Future research should evaluate the
feasibility of integrating biochemical assessment of antihypertensive adherence into
routine primary care.
Habit strength was the strongest predictor of adherence, that is overall adherence,
intentional non-adherence, and unintentional non-adherence. This is consistent with
previous findings and provides additional support for an extended CSM that includes habit
strength as a core component (Phillips et al., 2013, 2016). Contrary to the fourth
hypothesis, however, pill burden was not associated with habit strength or adherence.
This conflicts with the existing literature, which suggests that habit formation is
theoretically more difficult for those taking complex medication regimens. The absence of
association between pill burden and habit strength in this study may be due to the
restricted range of pill burden in the current sample. All patients were prescribed at least
three antihypertensive agents, in addition to their other medications. Future studies may
benefit from samples that include the full range of pill burden, particularly the lower end
of this spectrum, to adequately test this hypothesis. An increasing number of patients are
being treated with multimedication regimens, due in part to the increase in
 Theoretical predictors of adherence for aTRH 13

multimorbidity internationally (Barnett et al., 2012; Fortin, Bravo, Hudon, Vanasse, &
Lapointe, 2005; Glynn et al., 2011); therefore, the role of pill burden with regard to both
habit formation and adherence, as well as potential strategies to reduce pill burden
warrants further investigation. Polypills (i.e., pills that combine multiple active pharma-
ceutical ingredients) may present a solution to some of the challenges of increased pill
burden. Evidence suggests that polypills improve adherence, are generally well-tolerated,
and reduce risk factor levels for cardiovascular disease (Gupta et al., 2010; Munger, 2010;
Sherrill, Halpern, Khan, Zhang, & Panjabi, 2011; Webster et al., 2016); however, more
research is needed to evaluate the clinical effectiveness and feasibility of implementing
this type of intervention on a large scale (Coca et al., 2017; Huffman, Xavier, & Perel,
2017; Webster, Castellano, & Onuma, 2017). Reminder packaging has also been used to
improve adherence where pill burden is an issue (Dupclay, Eaddy, Jackson, Raju, & Shim,
2012), which may be a more feasible solution to the problem of pill burden. Collaboration
with clinical and community pharmacists may facilitate implementation of these
strategies to reduce pill burden and thereby improve adherence. Further research is
needed among patients with a variety of chronic conditions and treatment regimens to
examine the impact of pill burden and the effectiveness of efforts to reduce pill burden for
improving adherence and treatment outcomes.
This study is not without limitation. First, despite recruiting a sample almost three
times larger than that of Phillips et al. (2013), our sample was also highly adherent with
relatively strong habits. Greater variability on these constructs would allow a better test of
these theoretical hypotheses, particularly with regard to CSM coherence, which was not
significantly predictive of adherence in any of these analyses. This relatively restricted
sample is theoretically more likely to be a similar stage in the dynamic process outlined in
the CSM (Leventhal, Weinman, Leventhal, & Phillips, 2008; Phillips et al., 2013) and
therefore may provide limited insight into processes that occur at earlier stages in the
model, for example gaining feedback on the efficacy of treatment. Although the current
study has important theoretical ramifications, the cross-sectional design limits the
certainty with which relationships between CSM constructs and adherence can be
determined. Therefore, longitudinal research among incident samples utilizing longitu-
dinal measurement of CSM constructs is now needed to properly test CSM propositions.
Following newly diagnosed patients from treatment initiation to a behaviour maintenance
stage is essential to properly assess the importance of treatment-related beliefs and other
reflective processes, in combination with automatic factors, for predicting long-term
adherence behaviour. Second, the current study is subject to certain measurement
challenges. Certain adherence measures, specifically self-report and urine assay, may have
been subject to the toothbrush effect, whereby patients may ‘load up on’ or adhere to
medication regimens a few days before the clinical appointment (Pullar, 1991).
Prescription records were taken over a 12-month period; however, given the nature of
the prescribing records available (i.e., quarterly prescription refills for most patients, thus
prohibiting the use of ≥80% as a standard cut-off), prescription refill adherence was
operationalized as a dichotomous adherent/non-adherent variable. It is possible that, had
more detailed prescription refill information been available, stronger correlations
between adherence measures might have been observed. Furthermore, although we
screened for 25 of the most commonly prescribed antihypertensive medications using
HPLC-MS/MS, there were five medications that could not be reliably detected. As such,
there were a number of participants (n = 61) for whom not all prescribed antihyperten-
sive medications were assessed. Had all prescribed antihypertensives been assessed, non-
adherence estimates may have been higher for the urine assay measure. Furthermore,
14 Hannah Durand et al.

although treatment-related beliefs were operationalized consistently with Phillips et al.

(2013), this approach to assessing beliefs as conceptualized by the CSM requires further
psychometric validation. Finally, it is possible that had adherence and/or habit strength
been assessed for individual antihypertensive medications, as opposed to antihyperten-
sive treatment in general (e.g., patients may be habitual in taking their beta blocker but
intentionally skip their diuretic on occasion), a different pattern of results might have
emerged. This is an important measurement consideration that future research examining
the role of pill burden for adherence should address.
Limitations notwithstanding, this study makes several important contributions to the
literature. First, it provided an opportunity to compare adherence scores as measured by
drug levels in urine, self-report, and prescription refill records within a large sample of
aTRH patients in primary care. Second, this study used rigorous and systematic statistical
methods informed by psychometric theory (Bobko et al., 2007) to create a composite
adherence score for a large sample of patients that can be replicated for future cohorts.
Third, our replication of previous empirical findings using an enhanced methodology of
adherence assessment by including a physical measure of patient adherence in the form of
urine assays provides additional evidence for the key role of habit strength in predicting
long-term adherence behaviour. Together these findings highlight important targets for
further research and intervention (i.e., habit development and pill burden) for improving
adherence to antihypertensive medications. Future research should validate these
findings using inception samples with longitudinal measurement of CSM constructs and
multiple methods of adherence measurement to properly test CSM propositions, and
ultimately use the extended CSM as a framework for the development of behaviour
change interventions to enhance adherence.

Conflict of interest
All authors declare no conflict of interest.

Acknowledgements
The authors are indebted to all the general practice staff and patients who participated. This
research is supported by the Health Research Board Patient-Oriented Research Award (Ref:
HRA-POR-2014-615). The use of the MMAS© is protected by US copyright laws. Permission for
use is required. A licence agreement is available from: Donald E. Morisky, MMAS Research LLC
14725 NE 20th St. Bellevue WA 98007.

References
Barnett, K., Mercer, S. W., Norbury, M., Watt, G., Wyke, S., & Guthrie, B. (2012). Epidemiology of
multimorbidity and implications for health care, research, and medical education: A cross-
sectional study. The Lancet, 380, 37–43. https://doi.org/10.1016/S0140-6736(12)60240-2
Bobko, P., Roth, P. L., & Buster, M. A. (2007). The usefulness of unit weights in creating composite
scores: A literature review, application to content validity, and meta-analysis. Organizational
Research Methods, 10, 689–709. https://doi.org/10.1177/1094428106294734
Bolman, C., Arwert, T. G., & V€
ollink, T. (2011). Adherence to prophylactic asthma medication: Habit
strength and cognitions. Heart & Lung: The Journal of Acute and Critical Care, 40(1), 63–75.
https://doi.org/10.1016/j.hrtlng.2010.02.003
 Theoretical predictors of adherence for aTRH 15

Bond, C. M. (2016). Adherence: The holy grail? BMJ Quality & Safety, 25, 736–738. https://doi.org/
10.1136/bmjqs-2015-004886
Burnier, M. (2006). Medication adherence and persistence as the cornerstone of effective
antihypertensive therapy. American Journal of Hypertension, 19, 1190–1196. https://doi.org/
10.1016/j.amjhyper.2006.04.006
Calhoun, D. A., & Grassi, G. (2017). True versus pseudoresistant hypertension. Journal of
Hypertension, 35, 2367–2368. https://doi.org/10.1097/HJH.0000000000001568
Calhoun, D. A., Jones, D., Textor, S., Goff, D. C., Murphy, T. P., Toto, R. D., . . . Carey, R. M. (2008).
Resistant hypertension: Diagnosis, evaluation, and treatment. Circulation, 117, e510–e526.
https://doi.org/10.1161/CIRCULATIONAHA.108.189141
Coca, A., Agabiti-Rosei, E., Cifkova, R., Manolis, A. J., Red on, J., & Mancia, G. (2017). The polypill in
cardiovascular prevention: Evidence, limitations and perspective–position paper of the
European Society of Hypertension. Journal of Hypertension, 35, 1546–1553. https://doi.org/
10.1097/HJH.0000000000001390
DiMatteo, M. R. (2004). Variations in patients’ adherence to medical recommendations: A
quantitative review of 50 years of research. Medical Care, 42, 200–209.
Dupclay, L., Eaddy, M., Jackson, J., Raju, A., & Shim, A. (2012). Real-world impact of reminder
packaging on antihypertensive treatment adherence and persistence. Patient Preference and
Adherence, 6, 499–507. https://doi.org/10.2147/PPA.S31417
Durand, H., Hayes, P., Morrissey, E. C., Newell, J., Casey, M., Murphy, A. W., & Molloy, G. J. (2017).
Medication adherence among patients with apparent treatment-resistant hypertension:
Systematic review and meta-analysis. Journal of Hypertension, 35, 2346–2357. https://doi.
org/10.1097/HJH.0000000000001502
Fortin, M., Bravo, G., Hudon, C., Vanasse, A., & Lapointe, L. (2005). Prevalence of multimorbidity
among adults seen in family practice. The Annals of Family Medicine, 3, 223–228. https://doi.
org/10.1370/afm.272
Gardner, B., Abraham, C., Lally, P., & de Bruijn, G.-J. (2012). Towards parsimony in habit
measurement: Testing the convergent and predictive validity of an automaticity subscale of the
Self-Report Habit Index. International Journal of Behavioral Nutrition and Physical Activity,
9(1), 102. https://doi.org/10.1186/1479-5868-9-102
Gerbino, P. P., & Shoheiber, O. (2007). Adherence patterns among patients treated with fixed-dose
combination versus separate antihypertensive agents. American Journal of Health-System
Pharmacy, 64, 1279–1283. https://doi.org/10.2146/ajhp060434
Glynn, L. G., Valderas, J. M., Healy, P., Burke, E., Newell, J., Gillespie, P., & Murphy, A. W. (2011). The
prevalence of multimorbidity in primary care and its effect on health care utilization and cost.
Family Practice, 28, 516–523. https://doi.org/10.1093/fampra/cmr013
Gupta, A. K., Arshad, S., & Poulter, N. R. (2010). Compliance, safety, and effectiveness of fixed-dose
combinations of antihypertensive agents. Hypertension, 55, 399–407. https://doi.org/10.1161/
HYPERTENSIONAHA.109.139816
Gupta, P., Patel, P., Strauch, B., Lai, F. Y., Akbarov, A., Gulsin, G. S., . . . Tomaszewski, M. (2017).
Biochemical screening for nonadherence is associated with blood pressure reduction and
improvement in adherence. Hypertension, 70, 1042–1048. https://doi.org/10.1161/
HYPERTENSIONAHA.117.09631
Hagger, M. S. (2016). Non-conscious processes and dual-process theories in health psychology.
Health Psychology Review, 10, 375–380. https://doi.org/10.1080/17437199.2016.1244647
Hagger, M. S., Koch, S., Chatzisarantis, N. L. D., & Orbell, S. (2017). The common sense model of self-
regulation: Meta-analysis and test of a process model. Psychological Bulletin, 143, 1117–1154.
https://doi.org/10.1037/bul0000118
Hayes, P., Casey, M., Glynn, L. G., Molloy, G. J., Durand, H., O’Brien, E., . . . Murphy, A. W. (2018).
Prevalence of treatment-resistant hypertension after considering pseudo-resistance and
morbidity: A cross-sectional study in Irish primary care. British Journal of General Practice,
68, e394–e400. https://doi.org/10.3399/bjgp18X696221
Horne, R. (2004). The medication adherence report scale. Brighton, UK: University of Brighton.
16 Hannah Durand et al.

Horne, R., Weinman, J., & Hankins, M. (1999). The beliefs about medicines questionnaire: The
development and evaluation of a new method for assessing the cognitive representation of
medication. Psychology & Health, 14(1), 1–24. https://doi.org/10.1080/08870449908407311
Huffman, M. D., Xavier, D., & Perel, P. (2017). Uses of polypills for cardiovascular disease and evidence
to date. The Lancet, 389, 1055–1065. https://doi.org/10.1016/S0140-6736(17)30553-6
Ingersoll, K. S., & Cohen, J. (2008). The impact of medication regimen factors on adherence to
chronic treatment: A review of literature. Journal of Behavioral Medicine, 31, 213–224.
https://doi.org/10.1007/s10865-007-9147-y
Joint Formulary Committee (2015). British national formulary 69. London, UK: BMJ Publishing
and the Royal Pharmaceutical Society.
Jung, O., Gechter, J. L., Wunder, C., Paulke, A., Bartel, C., Geiger, H., & Toennes, S. W. (2013).
Resistant hypertension? Assessment of adherence by toxicological urine analysis. Journal of
Hypertension, 31, 766–774. https://doi.org/10.1097/HJH.0b013e32835e2286
Karve, S., Cleves, M. A., Helm, M., Hudson, T. J., West, D. S., & Martin, B. C. (2009). Good and poor
adherence: Optimal cut-point for adherence measures using administrative claims data. Current
Medical Research and Opinion, 25, 2303–2310. https://doi.org/10.1185/03007990903126833
Kavanagh, K. E., O’Brien, N., Glynn, L. G., Vellinga, A., & Murphy, A. W. (2010). WestREN: A
description of an Irish academic general practice research network. BMC Family Practice,
11(1), 74. https://doi.org/10.1186/1471-2296-11-74
Krousel-Wood, M., Islam, T., Webber, L. S., Re, R., Morisky, D. E., & Muntner, P. (2009). New
medication adherence scale versus pharmacy fill rates in hypertensive seniors. The American
Journal of Managed Care, 15(1), 59–66.
Kumar, R., & Indrayan, A. (2011). Receiver operating characteristic (ROC) curve for medical
researchers. Indian Pediatrics, 48, 277–287. https://doi.org/10.1007/s13312-011-0055-4
Lally, P., Van Jaarsveld, C. H. M., Potts, H. W. W., & Wardle, J. (2010). How are habits formed:
Modelling habit formation in the real world. European Journal of Social Psychology, 40, 998–
1009. https://doi.org/10.1002/ejsp.674
Lam, W. Y., & Fresco, P. (2015). Medication adherence measures: An overview. Biomed Research
International, 2015, 1–12. https://doi.org/10.1155/2015/217047
Leventhal, H., Phillips, L. A., & Burns, E. (2016). The Common-Sense Model of Self-Regulation (CSM):
A dynamic framework for understanding illness self-management. Journal of Behavioral
Medicine, 39, 935–946. https://doi.org/10.1007/s10865-016-9782-2
Leventhal, H., Weinman, J., Leventhal, E. A., & Phillips, L. A. (2008). Health psychology: The search
for pathways between behavior and health. Annual Review of Psychology, 59, 477–505.
https://doi.org/10.1146/annurev.psych.59.103006.093643
Liu, H., Golin, C. E., Miller, L. G., Hays, R. D., Beck, C. K., Sanandaji, S., . . . Wenger, N. S. (2001). A
comparison study of multiple measures of adherence to HIV protease inhibitors. Annals of
Internal Medicine, 134, 968–977. https://doi.org/10.7326/0003-4819-134-10-200105150-
00011
Mancia, G., Fagard, R., Narkiewicz, K., Redon, J., Zanchetti, A., B€ ohm, M., . . . Zannad, F. (2013). 2013
ESH/ESC guidelines for the management of arterial hypertension: The Task Force for the
Management of Arterial Hypertension of the European Society of Hypertension (ESH) and of the
European Society of Cardiology (ESC). Blood Pressure, 22, 193–278. https://doi.org/10.3109/
08037051.2013.812549
Mathes, T., Jaschinski, T., & Pieper, D. (2014). Adherence influencing factors – a systematic review
of systematic reviews. Archives of Public Health, 72, 37. https://doi.org/10.1186/2049-3258-
72-37
Moon, Z., Moss-Morris, R., Hunter, M. S., & Hughes, L. D. (2017). More than just side-effects: The role
of clinical and psychosocial factors in non-adherence to tamoxifen. British Journal of Health
Psychology, 22, 998–1018. https://doi.org/10.1111/bjhp.12274
Morisky, D. E., Ang, A., Krousel-Wood, M., & Ward, H. J. (2008). Predictive validity of a medication
adherence measure in an outpatient setting. The Journal of Clinical Hypertension, 10, 348–
354. https://doi.org/10.1111/j.1751-7176.2008.07572.x
 Theoretical predictors of adherence for aTRH 17

Morisky, D. E., & DiMatteo, M. R. (2011). Improving the measurement of self-reported medication
nonadherence: Response to authors. Journal of Clinical Epidemiology, 64, 255–263. https://
doi.org/10.1016/j.jclinepi.2010.09.002
Moss-Morris, R., Weinman, J., Petrie, K., Horne, R., Cameron, L., & Buick, D. (2002). The revised
illness perception questionnaire (IPQ-R). Psychology & Health, 17(1), 1–16. https://doi.org/10.
1080/08870440290001494
Munger, M. A. (2010). Polypharmacy and combination therapy in the management of hypertension
in elderly patients with co-morbid diabetes mellitus. Drugs and Aging, 27, 871–883. https://doi.
org/10.2165/11538650-000000000-00000
Osterberg, L., & Blaschke, T. (2005). Adherence to medication. New England Journal of Medicine,
353, 487–497. https://doi.org/10.1056/NEJMra050100
Phillips, L. A., Cohen, J., Burns, E., Abrams, J., & Renninger, S. (2016). Self-management of chronic
illness: The role of ‘habit’ versus reflective factors in exercise and medication adherence.
Journal of Behavioral Medicine, 39, 1076–1091. https://doi.org/10.1007/s10865-016-9732-z
Phillips, L. A., Leventhal, H., & Leventhal, E. A. (2013). Assessing theoretical predictors of long-term
medication adherence: Patients’ treatment-related beliefs, experiential feedback and habit
development. Psychology & Health, 28, 1135–1151. https://doi.org/10.1080/08870446.2013.
793798
Pullar, T. (1991). Compliance with drug therapy. British Journal of Clinical Pharmacology, 32,
535–539. https://doi.org/10.1111/j.1365-2125.1991.tb03948.x
Sch€afer-Keller, P., Steiger, J., Bock, A., Denhaerynck, K., & De Geest, S. (2008). Diagnostic accuracy
of measurement methods to assess non-adherence to immunosuppressive drugs in kidney
transplant recipients. American Journal of Transplantation, 8, 616–626. https://doi.org/10.
1111/j.1600-6143.2007.02127.x
Sheeran, P. (2002). Intention—behavior relations: A conceptual and empirical review. European
Review of Social Psychology, 12(1), 1–36. https://doi.org/10.1080/14792772143000003
Sherrill, B., Halpern, M., Khan, S., Zhang, J., & Panjabi, S. (2011). Single-pill vs free-equivalent
combination therapies for hypertension: A meta-analysis of health care costs and adherence. The
Journal of Clinical Hypertension, 13, 898–909. https://doi.org/10.1111/j.1751-7176.2011.
00550.x
Tomaszewski, M., White, C., Patel, P., Masca, N., Damani, R., Hepworth, J., . . . Williams, B. (2014).
High rates of non-adherence to antihypertensive treatment revealed by high-performance liquid
chromatography-tandem mass spectrometry (HP LC-MS/MS) urine analysis. Heart, 100, 855–
861. https://doi.org/10.1136/heartjnl-2013-305063
Verplanken, B. (2006). Beyond frequency: Habit as a mental construct. British Journal of Social
Psychology, 45, 639–656. https://doi.org/10.1348/014466605X49122
Verplanken, B., & Orbell, S. (2003). Reflections on past behavior: A self-report index of habit
strength. Journal of Applied Social Psychology, 33, 1313–1330. https://doi.org/10.1111/
j.1559-1816.2003.tb01951.x
Vrijens, B., Antoniou, S., Burnier, M., de la Sierra, A., & Volpe, M. (2017). Current situation of
medication adherence in hypertension. Frontiers in Pharmacology, 8, 100. https://doi.org/10.
3389/fphar.2017.00100
Webster, R., Castellano, J. M., & Onuma, O. K. (2017). Putting polypills into practice: Challenges and
lessons learned. The Lancet, 389, 1066–1074. https://doi.org/10.1016/S0140-6736(17)30558-5
Webster, R., Patel, A., Selak, V., Billot, L., Bots, M. L., Brown, A., . . . on behalf of the SPACE
Collaboration (2016). Effectiveness of fixed dose combination medication (‘polypills’)
compared with usual care in patients with cardiovascular disease or at high risk: A
prospective, individual patient data meta-analysis of 3140 patients in six countries.
International Journal of Cardiology, 205, 147–156. https://doi.org/10.1016/j.ijcard.2015.
12.015

Received 15 December 2017; revised version received 1 June 2018

18 Hannah Durand et al.

Supporting Information
The following supporting information may be found in the online edition of the article:
Appendix S1. Urine sample preparation and analysis.
Table S1. Hierarchical Regression Analysis of Theoretical Predictors of Adherence
Measured by Self-Report using the MARS (H1, H2).
Table S2. Hierarchical Regression Analysis of Theoretical Predictors of Adherence
Measured by Self-Report using the MMAS (H1, H2).
Table S3. Logistic Regression Analysis of Theoretical Predictors of Adherence
Measured by Prescription Refill Records (H1, H2).
Table S4. Hierarchical Regression Analysis of Theoretical Predictors of Adherence
Measured by Urine Assay (H1, H2).