Personal Health Perceptions as Predictors to Medication Adherence in a Prospective Cohort Study Among Patients with Multiple Sclerosis

Background: Though adherence to disease-modifying therapies (DMTs) among persons with multiple sclerosis (PwMS) varies and is often below 80%, only few prospective studies on adherence examined predictors beyond demographic and clinical characteristics. Objectives: Identify antecedents to adherence and persistence in a prospective design among PwMS. Methods: PwMS (n=186) were prospectively assessed at three time points: baseline, 6 and 12 months later. Clinical, demographic information and patient-reported medication beliefs, illness perceptions, medication habits, perceived health and affect were assessed. Adherence and persistence were assessed by a combination of self-reports and retrospective review of medication claims. Findings: PwMS were 69.9% (Time 1) and 71% (Time 2) adherent to their DMTs and 64.5.9% were persistent. Beliefs about Medications were consistently predictive at both time points (baseline to Time 1 and Time 1 to Time 2) of medication adherence and persistence whereas other perceptions were predictive at some analyses; clinical and demographic characteristic were mostly not predictive of adherence nor persistence. The prospective association of beliefs about medication with adherence held also in multivariate analyses. Conclusions: Adherence and persistence are predicted by medication beliefs of PwMS. As medication beliefs are modiable, they should be assessed periodically and targeted as a focus of tailored interventions aimed to improve adherence and consequently health outcomes in PwMS.


Background
There is widespread recognition that adherence to medication is key to successful health care of persons with Multiple Sclerosis (PwMS) (1-3) yet reviews on DMT medication taking among PwMS estimate adherence as ranging between 41-88% (1) and persistence ranging from 16-27% (2). Adherence is especially challenging to PwMS taking DMTs: despite of the long term nature of the chronic condition that require medication taking for long periods, the need for DMTs may be less obvious during periods of disease inactivity which may turn PwMS complacent on adherence. Concomitantly, adverse side effects of DMTs reduce quality of life of PwMS (4) and are often associated with decreased adherence.
Medication nonadherence is widely recognized as a common and costly problem (5), as nonadherence reduces the patient's potential bene ts from treatment (6) and increases healthcare costs (7). Medication taking behavior has two main aspects: adherence and persistence (8); adherence refers to the extent of correspondence between medication taking behavior and the recommendations made by the provider with respect to the timing, dosage, and frequency whereas persistence refers to staying on (same) treatment (9,10).
The World Health Organization adherence model posits that adherence is determined by the interplay of ve sets of factors: social and economic (e.g., age, ethnicity, education), health care system (e.g., type of insurance), condition-related (e.g., duration, comorbidity), therapy-related (e.g., type of medication, complexity of regimen, side effects) and patient-related (5). Factors most often studied are social-economic and patient-related, as the rst is easier to measure and the latter is considered as potentially modi able in interventions, including perceptions on illness, medication beliefs, habits in medication and affective states (11,12). Most studies on adherence among PwMS examined the social-economic factor and only few studies investigated patientrelated factors (13)(14)(15) or therapy-related (2,(16)(17)(18).
Considerable variation is evident in the measurement of adherence, with no single gold standard (7,19). Hence, different measures (e.g., patients' reports, medication possession ration (MPR) and electronic monitoring devices (EMD)) are regarded as measuring different phenomenon, each with shared and unique variability related to clinical outcomes (19,20). Most studies, though, rely on a single method to evaluate adherence (21,22).
Only few studies in MS (15,23,24) has used multiple measures of adherence and longitudinally examined their association with varied potential antecedents. Thus, the present study aimed at assessing both adherence and persistence using multiple measures (patient reported outcomes and medication claims) and examined their association with diverse predictors among MS patients (see the multidimensional adherence model, Fig. 1).

Participants
Persons with RMS (relapsing multiple sclerosis) treated with DMTs at Carmel Medical Center's specialized MS clinic in Haifa, Israel: 186 at baseline (T0), 6 months later (Time 1) and 12 months (Time 2) since baseline. Recruitment is depicted at Fig. 2.
A prospective observational study design was used. Data were collected in a large single-center between February 2016 and August 2018. Inclusion criteria were: RMS diagnosis, being at baseline on DMT of Fingolimod, Dimethyl Fumarate, Interferon beta-1a and Glatiramer Acetate. Exclusion criteria at recruitment were: language literacy (n = 14), cognitive impairment (n = 3), and disinclination to participate (n = 2) and moving to another clinic (n = 8). The survey and neurological evaluation (see details below) were administered prospectively at baseline, 6 months (Time 1, median length of 6.9 months) and 12 months later (Time 2, median length of 6.8 months from Time 1). Medication possession data were retrieved retrospectively for the same periods.
The study was approved by an Internal Review Board of Carmel Medical Center (#0061-14-CMC) and registered (clinical trials registry #NCT02488343). All participants were provided written informed consent con rming that they were free to leave the study at any time.

Measures Adherence and Persistence
Medication withdrawal records were retrieved from the computerized dataset of 'Clalit Health Services'; these were available for 136 PwMS in the prospective study who are members of this Health Maintenance Organization (HMO) and not for 50 PwMS treated at the clinic yet are members of other HMOs. Based on medication withdrawal data, Medication Possession Ratio (MPR) was computed for each PwMS based on her/his medication type and the initial prescription: it was estimated as the total days with index medication supply within the re ll interval (six months between baseline and time 1 and six months between time 1 and time 2) divided by the number of days between the rst prescription data and the last prescription date. Using the commonly accepted threshold of MPR ≥ 80% (10), PwMS were considered adherent if they were above the threshold and non-adherent when they were below this threshold.
Patient-reported outcomes measures -Multiple Sclerosis Treatment Adherence Questionnaire (MS-TAQ; (25)) and Probabilistic Medication Adherence Scale (ProMAS; (26)). The items from MS-TAQ used in this analysis tapped whether the participant did not take a prescribed dose in the last four weeks and the reported number of these doses. In cases of reported non-adherence, the percentage was calculated per regiment. The ProMAS is an overall estimation 18-item questionnaire tapping adherence behaviors (e.g., "I have never changed my medicine use myself", "When I am away from home, I occasionally do not take my medicines") to which respondents indicate 'yes, true' (coded as 1) or 'no, not true' (coded as 0). Higher individual's adherence scores represent better adherence rates. Adherence categories are low (sum score 0-4), medium low (sum score 5-9), medium-high (sum score [10][11][12][13][14] and high (sum score [15][16][17][18]. Internal reliabilities of the ProMas were baseline = 0.83, Time 1 = 0.82 and Time 2 = 0.83. A score of adherence was constructed so that good adherence was de ned as either = > 80% medication claims per regiment (medication possession ratio (MPR)), or = > 80% self-reported medication use by MS-TAQ or being at the medium-high and high categories of ProMAS. Full details are described in a methodological report (24). Low adherence was de ned as the complement. Persistence was de ned as staying with the same medication from baseline till Time 2.

Predictors
Self Report Habit Index (SRHI; (27)) is a 12-item PRO assessing habit strength, speci cally repetition, automaticity of medication taking behavior and the sense of identity the medication behavior re ects (in either administration route). The items were measured on a ve-point bipolar scale, ranging from 'I completely agree' (4) to 'I completely disagree' (0). An overall score for habit strength was constructed (higher values denote less habit). Cronbach's internal reliabilities were α = 0.86, α = 0.88 and α = 0.86 for baseline and Time 1, respectively.
Belief about Medicine Questionnaire (BMQ; (28) is used to assess the cognitive represetations of medicines. The 18-item scale contains two ve-item subscales measuring Necessity and Concerns about medication and two four-item subscales measuring Harm and Overuse. Scores on this measure were constructed so that higher scores indicate stronger beliefs in the concepts represented by the scale. Internal reliabilities were α = 0.81 for both baseline and Time 1; internal reliabilities of the subscales ranged from α = 0.71 to α = 0.83.
Illness perceptions (Brief Illness Perception Questionnaire; (29)). The B-IPQ includes eight items graded on a linear 0-10 response scale assessing cognitive and emotional representations of illness. Each item refers to one dimension of illness perception (consequences, timeline, identity, personal control, treatment control and coherence, and the (two-item) dimension of emotional representation). The scale was scored so that higher scores represent more negative illness perceptions. Cronbach's internal reliabilities were α = 0.71, 0.76 at baseline and Time 1, respectively.
Emotional states (Hospital Anxiety and Depression Scale (HADS;(30) is a self-report 14-items depression and anxiety questionnaire widely used in medical settings and has been used in the past among PwMS (31,32).
Respondents rate the degree to which they have been experienced depression and anxiety over the last week. Reliabilities were α = 0.84 and α = 0.85 at baseline for depression and anxiety, respectively.
Background and clinical variables examined for this study included age, gender, marital status, educational attainment and subjective social economic status, ethnicity, comorbidity, MS duration, time on current DMT and type of DMT. Physical disability was assessed by a neurologist using a widely used scale of disease progression and neurological impairment (Kurtzke Expanded Disability Status Scale, EDSS; (33)).

Statistical Analysis
Descriptive analyses for background and clinical characteristics were conducted and reported for all participants. For categorical variables, counts and percentages are provided whereas means and standard deviations (SDs) are presented for continuous variables. Adherence was constructed such that non-adherence was de ned as either detected/reported by one of the PRO or MPR (24); it is presented across Time 1 and Time 2 and also by DMT administration route. Persistence is reported as staying with the same medication between baseline and Time 2 and reasons for discontinuation are described.
Then, adherent and non-adherent PwMS were compared in their background and clinical characteristics as well as their perceptions. Categorical variables were analyzed using a chi-square test, and continuous variables were analyzed using the t-test or Mann-Whitney U test (depending on the normality of distribution, tested using Kolmogorov-Smirnov test). Statistical signi cance was set for p < 0.05. The relative contribution of variables found to be signi cantly different among the two groups were further evaluated using binary logistic regression analysis while adjusting also for age and gender. A similar analysis -comparing those who persisted with their medication to those who did not persist in their background and clinical characteristics as well as their perceptions -was conducted.

Patient characteristics
The study cohort consisted of 186 PwMS meeting the inclusion criteria and having follow-up data. Their demographic and clinical characteristics at baseline are depicted in Table 1 Table 2). It ranges 66.3-73.8%. Persistence is at lower levels (64.5%) and reasons for discontinuation were: clinical and/or MRI deterioration (n = 15), pregnancy planning (n = 14), laboratory abnormal results (n = 7) and patient-reported non-tolerability (n = 30).
Adherent persons at Time 1 and Time 2 were compared to non-adherent patients on background (baseline), clinical characteristics and perceptual characteristics (baseline characteristics to Time 1 adherence and Time 1 characteristics to Time 2 adherence). There was one statistically signi cant difference between adherent and non-adherent PwMS on the background and clinical characteristics, and it was not consistent across time.
Speci cally, adherence at Time 1 was more frequent among PwMS who had higher social economic status (than lower social economic status), though not at Time 2.
Consistent statistically signi cant differences between the two groups were uncovered in perceptual characteristics (see Table 3). Speci cally, adherent PwMS, compared to non-adherents, believed their medication to be less overtreatment and less harmful, both at Time 1 and Time 2. Self-rated health, illness perception (general score and components), habits and affective states (depression and anxiety) at baseline and time 1 did not differ signi cantly between the adherence groups at Time 1 and Time 2, respectively.
Beliefs about medication were then tested as a predictor of adherence in a multivariate analysis, controlling for traditional background variables of age and gender. Persistence, just as adherence, was not predicted by background and clinical characteristics. Persistence, measured at Time 2, was predicted by perceptions at baseline, speci cally concerns about medication and anxiety (see Table 5). A multivariate analysis which also controlled for age, gender and included anxiety and concerns about medication resulted in gender being the only predictive variable (OR = 2.34, 95% CI 1.07-5.14, p = 0.034). As concerns about medication and anxiety were highly correlated (r = 0.52, p < 0.001), the regression was also run controlling only for age and gender, and both concerns about medication, age (younger) and male gender were signi cantly predictive of persistence (see Table 4).

Discussion
Adherence to DMTs in this sample of PwMS, assessed by the combination of measures, medication claims and patients' reports (24), fell within the range reported in previous studies on medication adherence (2,21). The innovative nding of the present study is that adherence and persistence were consistently associated prospectively with patient-related factors, speci cally perceptions of medication -beliefs on the harm medication cause, their over-use and general concern. Other patient-reported perceptions (i.e., anxiety) was prospectively associated with adherence or persistence at one of the measurement points. Adherence and persistence were largely not predicted by demographic nor MS clinical characteristics. Interestingly, the habit of medication taking increased from Time 1 to Time 2 among most PwMS, even those less adherent.
Most (64.5%) PwMS in our sample persisted in the medication they were taking at the follow-up period. Two ndings on persistence are noteworthy. First, almost half of all non-persisters (n = 66) stopped taking the medication following reported complaints on non-tolerability (n = 30). Though the decision to discontinue the medication is shared by the physician and the patient, the move is driven by patients' perceptions, highlighting the importance of patients' perceptions. Secondly, though both adherence and persistence were predicted by beliefs about medication, the speci c beliefs differed: harm and overtreatment were prospectively associated with adherence whereas concerns were prospectively associated with persistence.
The current study addressed four of the ve factors in the WHO multidimensional model of medication adherence(34): social-economic, therapy-related (e.g., type of medication and administration route, naivety), patient-related (e.g., affective states, illness and medication perceptions) and condition-related (e.g., condition duration, EDSS). It did not address health care system characteristics (e.g., monetary issues that could be related to affording a medication), as the study was monocentric, conducted in a socially-nanced healthcare system.
The lack of association between background characteristics and adherence is different from some previous studies (17,18,35) yet similar to other (36). Indeed, a review on adherence in autoimmune conditions also concluded that economic, demographic, and clinical characteristics were only moderately linked to adherence or persistence (21). The ndings on the prospective association of beliefs about medication with adherence and persistence in medication taking is congruent with previous work in other conditions (37)(38)(39). Still, beliefs about medication were scarcely studied among PwMS; the only study that examined adherence to DMTs was cross-sectional and found no association between medication beliefs and adherence (13). Hence, the current study is the rst to demonstrate such an association with adherence and persistence. Recent work among PwMS that delved into reasons for non-adherence to DMTs (40) reported avoidance, side effects, cost and mild course of illness; the work reported that the avoidant group could not be characterized. The present work succeeded in characterizing this group and suggests that people's beliefs about medication at onset could predict their adherence and persistence.
The study is hampered by several limitations. First, the study's N is relatively small and hence the results of the present study should be veri ed in future studies with bigger cohort of patients. However, this study used medication claims as only one indicator and relied also on patients' reports of adherence. Secondly, the study was carried out at only one medical center. This may bias the ndings, as the patient -practitioner communication (40) and the organizational climate of the speci c specialty clinic may not be representative.
Lastly, the study reported on a one-year follow-up; adherence may still change in a longer follow-up.
The strengths of the study are manifold. First, it relied on multiple measures of adherence: two PROs and on medication claims. Second, it measured both adherence and persistence. Third, it included an array of predictors, focusing on patients' perceptions previously examined only scarcely in adherence to medication among PwMS. Lastly, the study used a prospective design that allowed to conclude on prediction.

Conclusions
To conclude, PwMS' perceptions of their medication consistently predict adherence and persistence in medication taking. These ndings are similar to conclusions in other medical conditions (41)(42)(43). Other perceptions, such as on one's health or on one's illness also predict either adherence or persistence at some time points. Importantly, perceptions are malleable and can be targeted for potential interventions aimed at increased adherence. Clinicians should therefore discuss with patients their beliefs on their prescribed medication. Lastly, beliefs about medication should be considered as part of a routine PRO battery so as to continually monitor patients' perceptions about their medication and be able to intervene, if needed. A detailed assessment of beliefs about medication can guide a speci c intervention strategy. These are part of implementation of patient empowerment, participatory medicine and patient-centered approaches (44,45) Figure 1 Conceptual adherence model.

Figure 2
Enrollment of Participants