Ranking facilitators and barriers of medication adherence by patients with inflammatory arthritis: a maximum difference scaling exercise

Introduction Facilitators and barriers of adherence to disease-modifying anti-rheumatic drugs (DMARDs) have been identified by patients with inflammatory arthritis earlier. However, the relative importance from the patients’ perspective of these factors is unknown. Knowledge on this ranking might guide the development of interventions and may facilitate targeted communication on adherence. This study aims to examine 1) the relative importance patients attach to facilitators and barriers for DMARDs adherence, and 2) the relationship between patient characteristics and ranking of these factors. Methods One hundred twenty-eight outpatients with inflammatory arthritis; (60% female, mean age 62 years (SD = 12), median disease duration 15 years, IQR (7, 23) participated in a Maximum Difference scaling exercise and ranked 35 items based upon previously identified facilitators and barriers to medication adherence. Hierarchical Bayes estimation was used to compute mean Rescaled Probability Scores (RPS; 0–100) (i.e. relative importance score). Kendall’s coefficient of concordance was used to examine a possible association between patients’ characteristics (i.e. age, sex and educational level) and ranking of the items. Results The three most important items ranked by patients were: Reduction of symptoms formulated as “Arthritis medications help to reduce my symptoms” (RPS = 7.30, CI 7.17–7.44), maintaining independence formulated as “I can maintain my independence as much as possible” (RPS = 6.76, CI 6.54–6.97) and Shared decision making formulated as “I can decide –together with my physician- about my arthritis medications” (RPS = 6.48, CI 6.24–6.72). No associations between patient characteristics and ranking of factors were found. Conclusions Reducing symptoms, maintaining independency and shared decision making are patients’ most important factors for DMARDs adherence. This knowledge might guide the development of interventions and may facilitate communication between health professionals and their patients on medication adherence.


Key messages
-Reducing symptoms, maintaining independency and shared decision making are patients' most important factors for DMARDs adherence -Facilitators and barriers of DMARDs adherence do not seem to be related to sex, age or education Background Inflammatory arthritis (IA), including rheumatoid arthritis (RA), psoriatic arthritis (PsA) and ankylosing spondylitis (AS), is characterized by inflammation, impaired physical function, and ultimately progressive joint damage. New pharmacological treatments have led to improved outcomes and prevention of disability over the last decades. Obviously, disease modifying anti rheumatic drugs (DMARDs) only work if taken by a patient in accordance with the prescription. According to the World Health Organization (WHO): 'Across diseases, adherence is the most important modifiable factor that compromises treatment outcomes'. Indeed, adherence to DMARDs is still suboptimal. Non-adherence estimates between 30 and 90% [1][2][3], and consequently has been shown to have a negative impact on outcomes of the disease [4][5][6]. Adherence to treatment reflects the extent to which medication is taken as prescribed and the full benefits of DMARDs can be achieved if patients follow drug regimens [7][8][9][10][11]. Multiple factors have been identified that may affect medication adherence in patients with inflammatory arthritis (e.g. RA). Whereas medication beliefs, self-efficacy and the patient-clinician relationship are consistently related to medication adherence [12][13][14][15][16], other factors such as age, disease duration, and complexity of medication regimen show conflicting results [2,17,18]. Understanding which potentially modifiable factors are most important to patients and whether these factors are related to patient's characteristics is needed to develop targeted adherence interventions [2,[18][19][20].
In a previous qualitative study [20], we identified facilitators and barriers to DMARDs use in patients with inflammatory arthritis using an adapted version of the Theoretical Domains Framework (TDF) [21,22]. However, the relative importance of the identified facilitators and barriers of adherence for patients with inflammatory arthritis are unknown. The ranking by patients of these factors might help communication between physicians and their patients with regard to the initiation and follow up of a DMARD-treatment. Even more, a relation between the ranking of the facilitators and barriers and patient characteristics needs to be addressed; this may be useful for the development of targeted adherence interventions.
Therefore, the aim of this study is to study the ranking by arthritis patients themselves of facilitators and barriers for DMARDs adherence, using a preference scaling methodology. In addition we aim to examine whether this ranking is associated with demographic characteristics.

Design
This study is a cross-sectional study, ranking the importance of previously identified facilitators and barriers for adherence to DMARDs in patients suffering from inflammatory arthritis using a Maximum Difference Scaling (MaxDiff) exercise [20]. A MaxDiff exercise (or best-worst scaling) is a method that simplifies the ranking tasks for participants and provides insight into trade-offs in the decision making process (Sawtooth Software 1 (2013) The MaxDiff system/version 8 Technical Paper. Sequim: WA, Sawtooth Software. The MaxDiff method was proved to be efficacious in similar studies in rheumatology [23][24][25]. This study took place in 2018-2019.

Preparatory work
For this study, facilitators and barriers for DMARD use, as identified in a previous qualitative study using focus groups and a questionnaire in patients with inflammatory rheumatic diseases was used [20]. In that study, 82 facilitators and barriers were identified, collated in the 11 domains from the adjusted Theoretical Domain Framework (TDF) [21,22], and subsequently grouped into the higher order components Capability, Opportunity, and Motivation of the Behaviour Change Wheel (BCW) [22]. Factors were selected from these three COM components by three researchers (MV, JV, BvdB), yielding a feasible, neutrally phrased, and representative set of 35 items related to the originally identified facilitators and barriers. The 35 items were used as input for the Maximum Difference Scaling exercise. (see Table 1).

Pilot testing
The maximum difference scaling exercise was pilot tested among a small group of patients and care providers (patients with inflammatory arthritis, n = 4; psychologists, n = 2; and pharmacists, n = 2). Based on the pilot results, small changes in the wording of the items and the instructions were made to avoid misinterpretation.

Sample size
A standard method for sample size calculations for discrete-choice experiments is lacking [26]. Neither is there a guidelines for sample size calculations when using a maximum difference scaling method [27]. Previous MaxDiff studies reported that a minimum of 100  [25]. For our study, we therefore chose to recruit a convenience sample of at least 100 consecutive patients with IA from one clinic.

Patients and measures
Patients were recruited from the rheumatology clinic (Arthritis Centre Twente, Medisch Spectrum Twente, Enschede in the Netherlands). Inclusion criteria were: 18 years or older; a diagnosis of inflammatory arthritis; using one or more DMARDs; the ability to communicate in Dutch; and informed consent. Patients with a planned visit to their rheumatologist at the hospital, were informed about this study by their rheumatologist or rheumatology nurse. If they showed interest and fulfilled the inclusion criteria, they received a letter explaining the study in more detail. The letter comprised an internet link that led them to the web-based Maximum Difference Scaling exercise, which they could complete at home. This study protocol (K17-08) was presented to the Medical Ethical Committee of the Medisch Spectrum Twente Hospital in Enschede, the Netherlands. Formal ethical approval for this study is not required under Dutch Law. Still, in accordance with the Personal Data Protection Act, informed consent was sought before the start of the Maximum Difference Scaling exercise. Taking into account an expected response rate of 20-30% [28], invitation letters were sent by mail to 450 outpatients with inflammatory arthritis using one or more DMARDs.

Online survey
The Sawtooth software's Lighthouse Studio MaxDiff 500 SSI Web version 8 was used to develop an online survey comprising a MaxDiff exercise (Sawtooth Software Inc. 2013) and a questionnaire to collect patient characteristics. In the MaxDiff exercise participants are shown several subsets of the possible items pertaining to medication adherence and are asked to indicate the most and least important item in each subset. In this study, each subset contained four items related to adherence. Part 1 of the online survey assessed the following patient characteristics: sex, age, diagnosis, level of education, disease duration, and DMARDs use (which type of DMARD).
Part 2 of the online survey comprised the choice task using the maximum difference scaling method: 27 subsets were presented. Each subset contained four items related to DMARD use (Fig. 1). This approach simplifies the ranking tasks for participants, enables discrimination between ratings of different items involved in complex decisions and is not influenced by scale related biases [29]. Sawtooth software creates an optimal design of subsets based on 20,000 iterations of the exercise to ensure variation in the combination of items. An open link was created to be disseminated to the patients.
Each of the subsets had a different order of various items, to ensure that every participant would rank the item three times (27 × 4/35 = 3) and to avoid that higher importance was given to the first mentioned item. Participants were asked to choose the most important and the least important item, by answering the question: "In order to take my arthritis medication as prescribed by

Statistical analysis
Descriptive analyses were used to describe patient characteristics. Where appropriate, mean and standard deviation (normally distributed continuous variables), median and interquartile range (25th-75th percentile; not normally distributed continuous variables), or percentages (binary variables) were calculated. Level of education was categorized according to the International Standard Classification of Education (ISCED). To enhance international comparability, three broad groups of educational categories were constructed: no or preprimary education, primary education and lower secondary education (ISCED 0-2); upper secondary education (ISCED 3-4); and tertiary education (ISCED 5-6). Age was grouped into tertiles (0-58/59-68/69-84 years old). The rank order and the relative importance of each item were generated by the Sawtooth software's Lighthouse Studio, using hierarchical Bayes modelling. To facilitate interpretation, the scores were rescaled to a Rescaled Probability Score (RPS) on a scale from 0 to 100: the higher the score, the more important the item. (Sawtooth Software 1 (2013) The MaxDiff System/version 8 Technical Paper. Sequim: WA, Sawtooth Software, 2013). The rescaled probability scores added up to a value of 100, which reflects the relative importance of the item. Patients that gave inconsistent answers on the maximum difference scaling exercise (Root Likelihood below the recommended cut-off of 0.336) were excluded from the analysis. To explore a possible association between the ranking of the RPS scores and patients' characteristics (i.e., sex, age, educational level), Kendall's coefficient of concordance (Kendall's W) was computed [30]. Kendall's W, a non-parametric statistic, is used to determine the degree of agreement between groups when working with ranked data at an ordinal level of measurement. It is a normalization of the statistic of the Friedman test, and can be used for assessing agreement among raters. Kendall's W ranges from 0 (no agreement) to 1 (complete agreement). A p-value < 0.05 allows to reject the null hypothesis that there is no agreement in ranking between groups. SPSS version 24.0 was used.

Results
A total of 128 patients took part in the maximum difference exercise (response rate 28%). Three patients had a RLH below 0.336 and were therefore excluded from the analyses. Table 2 shows the characteristics of the participants. Most participating patients were women (60%) and most participating patients had rheumatoid arthritis (89%). The median disease duration was 15 years, IQR (7,23). These patient characteristics were not significantly different from the former study in which the items were identified [20].
In Table 3, the 35 items with their relative probability score are displayed and grouped according the COM-B model. The five highest ranked items were: "Arthritis medications help to reduce my symptoms" (RPS = 7.30, CI 7.17-7.44), "I can maintain my independence as much as possible" (RPS = 6.76, CI 6.54-6.97) and "I can

Agreement in ranking between patient subgroups
Kendall's W analyses showed that all subgroups ranked the items at a high level of agreement (Kendall's W ranged between .97 to .99 for sex (p = .000), age (p = .000) and level of education (p = .001)) indicating that no significant difference based on sex, age or level of education, in ranking the relative importance of items important to adherence were found. Kendall's W ranges from 0 (no agreement) to 1 (complete agreement). A p-value < .05 allows to reject the null hypothesis that there is no agreement in ranking between groups.

Discussion
This study ranked the earlier identified facilitators and barriers of adherence to arthritis medication by patients. The five highest ranked items are "Arthritis medications help to reduce my symptoms", "I can maintain my independence as much as possible", "I can decide -together with my physician-about my arthritis medications", "My general health (apart from IA)", and "I expect the arthritis medications will help to reduce my symptoms". All these factors are related to reducing symptoms, maintaining independency and shared decision making. The  Our results are in line with a recently published qualitative focus group study using nominal group technique [31]. Although this latter study had an inductive approach and was consequently not based on a theoretical model, several findings were comparable. Items like the importance of patient-physician interaction and efficacy of medication were ranked high. Our study also indicated the absence of a relation between ranking of factors pertaining to adherence and demographics: sex, age and education.
Our study also confirmed that patients prefer to discuss and decide the choice of medications in close collaboration with their physician [32][33][34]. Patients are also more likely to adhere to a treatment that matches their preferences [35]. Tailoring treatment to patients' medication preferences shows to be a promising strategy to improve adherence [28]. Therefore, providing the patient with unambiguous, high quality information [36] creates optimal circumstances for adherence to medication. Although practical issues such as the availability of aids to facilitate adherence were in this study ranked relatively low, such aids (e.g., application of electronic messages or reminders) can have a significant effect on adherence [37].

Strengths and limitations
This study is the first patient oriented and theory based study ranking facilitators and barriers for adherence that can help for the development of targeted interventions to improve adherence. These findings are helpful information for healthcare professionals to guide the conversation on adherence with their patients. The mentioned factors influencing adherent behaviour can be used as a checklist for both the health professional and the patient to discuss potential adherence problems when needed. The factors ranked as most important were related to patients' motivation, indicating that motivational support, such as motivational interviewing, could be useful to enhance adherence [38]. Furthermore, the findings of this study can be used for the development of future adherence interventions. There are some limitations in this study. The first limitation is that a selection bias in the recruitment process could have occurred. One may assume that more motivated patients, and more adherent patients were inclined to participate in this study. However, our study sample was representative for patients with inflammatory arthritis using DMARDs in the Netherlands. The second limitation is that all items related to facilitators and barriers needed to be rephrased. Obviously, rephrasing was necessary to reduce bias by using neutral wordings taking away any positive or negative valence of each facilitators and barriers [39,40]. Most patients included in this mono centre study had rheumatoid arthritis, used conventional synthetic DMARDs (csDMARDs) or biological DMAR Ds (bDMARDS), and had a long disease duration, limiting the generalization of our findings. Furthermore, patients with different levels of disease activity, with different routes of administration or with comorbidities may also rank these items differently.
To examine whether the outcome of the ranking process differs in subgroup of patients (e.g., other rheumatic conditions, other DMARDs such as targeted synthetic DMARDs (tsDMARDs), patients with short disease duration, patients with different levels of disease activity) or in other countries (e.g., with different healthcare systems and insurances), future research is warranted.

Conclusions
Our study showed the ranking of adherence-related factors by patients with inflammatory arthritis. Reducing symptoms, maintaining independency and shared decision making are considered most important for DMAR Ds adherence by patients. This finding can be of value to