Objective To categorise patient-reported outcome measures (PROMs) into their propensity to detect intentional and/or unintentional non-adherence to medication, and synthesise their psychometric properties.
Design Systematic review and regression analysis.
Eligibility Medication adherence levels studied at primary, secondary and tertiary care settings. Self-reported measures with scoring methods were included. Studies without proxy measures were excluded.
Data sources Using detailed searches with key concepts including questionnaires, reliability and validity, and restricted to English, MEDLINE, EMBASE, CINAHL, International Pharmaceutical Abstracts, and Cochrane Library were searched until 01 March 2022. Preferred Reporting Items for Systematic Reviews and Meta-Analyses 2020 (PRISMA-2020) checklist was used.
Data analysis Risk of bias was assessed via COnsensus-based Standards for the selection of health Measurement INstruments (COSMIN-2018) guidelines. Narrative synthesis aided by graphical figures and statistical analyses.
Outcome measures Process domains [behaviour (e.g., self-efficacy), barrier (e.g., impaired dexterity) or belief (e.g., perception)], and overall outcome domains of either intentional (I), unintentional (UI), or mixed non-adherence.
Results Paper summarises evidence from 59 studies of PROMs, validated among patients aged 18–88 years in America, the United Kingdom, Europe, Middle East, and Australasia. PROMs detected outcome domains: intentional non-adherence, n=44 (I=491 criterion items), mixed intentionality, n=13 (I=79/UI=50), and unintentional, n=2 (UI=5). Process domains detected include belief (383 criterion items), barrier (192) and behaviour (165). Criterion validity assessment used proxy measures (biomarkers, e-monitors), and scoring was ordinal, dichotomised, or used Visual Analogue Scale. Heterogeneity was revealed across psychometric properties (consistency, construct, reliability, discrimination ability). Intentionality correlated positively with negative beliefs (r(57)=0.88) and barriers (r(57)=0.59). For every belief or barrier criterion-item, PROMs’ aptitude to detect intentional non-adherence increased by β=0.79 and β=0.34 units, respectively (R2=0.94). Primary care versus specialised care predicted intentional non-adherence (OR 1.9; CI 1.01 to 2.66).
Conclusions Ten PROMs had adequate psychometric properties. Of the ten, eight PROMs were able to detect total, and two PROMs were able to detect partial intentionality to medication default. Fortification of patients’ knowledge and illness perception, as opposed to daily reminders alone, is most imperative at primary care levels.
- primary care
- public health
- clinical audit
Data availability statement
All data relevant to the study are included in the article or uploaded as supplemental information. Additional data can be accessed via the Dryad data repository at http://datadryad.org/ with the doi: 10.5061/dryad.8cz8w9gsq.
This is an open access article distributed in accordance with the Creative Commons Attribution Non Commercial (CC BY-NC 4.0) license, which permits others to distribute, remix, adapt, build upon this work non-commercially, and license their derivative works on different terms, provided the original work is properly cited, appropriate credit is given, any changes made indicated, and the use is non-commercial. See: http://creativecommons.org/licenses/by-nc/4.0/.
Statistics from Altmetric.com
If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.
Strengths and limitations of this study
While the processes leading to the final classification can be subjected to individual interpretation of the items’ construct and consequences to treatment default, the twofold approach to first, classify every PROM item, and second, to classify the overall PROMs into outcome domains depending on the dominance of item type, demonstrated good inter-rater reliability.
Using regression analysis of process and outcome domains, some insights were gained on the reasons for non-adherence. However, for future studies, further analysis with greater sample sizes are recommended.
The process domains of behaviour (habitual), barrier (tangible impediment), or belief (mindset) were not only difficult to assess by means of self-reported outcome measures, but were also revealed as the real challenges to fostering adherence.
While we had juxtaposed scores against proxy measures, as with any self-reported data, limitations existed with the accuracy of recall and fair reporting of events.
The act of taking medication is deemed fitting when it is done in accordance to instructions given by healthcare professionals. In general, the patient is instructed to take an accurate dose, at a predetermined time interval. Poor medication adherence poses a serious and expensive challenge to patients and healthcare systems. The Centers for Disease Control estimated that each year, the death of 125,000 persons in the USA alone was caused by treatment failure for chronic diseases and default in taking prescribed medications regularly and accurately. In addition, $300 billion USD (approx. £228 billion) was spent on hospitalisation, emergency admissions and clinic visits.1
Intentional and unintentional non-adherence are usually distinguished by patients’ medication-taking conduct, which is influenced by, firstly, processes such as pre-emptive thoughts or behavioural aspects, which eventually lead the patients to take their medication (e.g., to get better, habitual, or due to medication stock availability). Secondly, medication-taking conduct is distinguished by the outcome or action, i.e., patient ended up forgetting, skipping a dose, changing the amount, or delaying the time at which a medication should be taken, or did not take medication on purpose.2 As the nature of instructions for medication-taking is often complex and varied depending on an individual’s pre-existing medical conditions and concomitant medications, the criteria for adherence assessment are also subjective. Ideally, an adherence measure is developed to detect common predictors that may influence adherence for a diseased population.
Existing measures have commonly explored forgetfulness as an obstacle to adherence and subsequently identified instances where forgetting may be more frequent, such as, when away from home, during weekends, or when travelling for work.3 Intuitively, many measures also explored psychosocial aspects, such as, attitudes and perceptions associated with belief (e.g., their medication is not working, they do not need to take ‘too much of’ the medication). Nevertheless, users have critiqued that although such patient-administered questionnaires addressed specific obstacles to medication adherence, they were disease-specific, and hence had limited use in the general clinical care setting.4 In addition, the vulnerable older adults, who most require the assessment, were often excluded from the surveys as they presented with visual or cognitive impairment, and could not communicate effectively.
Several self-reported adherence measures, such as the Morisky scale,5 were established to screen for the possibility of missed doses or timings, and have been widely incorporated in clinical practice because of their ease of use. However, measures that assess only the instances of non-adherence offer little insights into the myriad of possibilities contributing to such misconduct, such as, the lack of motivation or specific difficulties encountered that could impact one’s adherence.6 To the best of our knowledge, this is the first systematic review that characterises patient-reported outcome measures (PROMs) of medication non-adherence, and distinguishes between intentional and unintentional causes. For the purpose of this review, intentional non-adherence is defined as instances when a patient is in a conscious and aware state of mind, and subsequently decides against following the recommendations of healthcare professionals.7 The patient is mindful of his or her decision, and had consciously weighed the pros and cons of adhering to treatment. Unintentional or sporadic non-adherence, in contrast to intentional causes, is consequential to a passive process i.e., unlikely associated with one' s implicit judgement.8 Patients can unintentionally non-adhere to their medication, i.e., when forgetful, careless, or have limited physical dexterity.
The aim of the systematic review was to critically appraise and synthesise PROMs in order to categorise their data into intentional and unintentional non-adherence to medication for chronic illnesses. The scope of this review is threefold: first, to establish studies where the level of non-adherence to medication was scored using validated and published self-reported measures (where adherence was reported, the scoring was reversed); second, to categorise each of the measure’s items/criteria (adherence related) to one of the three process domains: behaviour, barrier or belief, and to one of the three outcome domains: intentional, unintentional, or mixed; third, to categorise every measure into intentional, unintentional or mixed, based on the dominance or leading domain of its criterion item.
The term, ‘medication-taking conduct’ shall be used henceforth, and is defined as whether or not a patient is capable of taking medication in a manner judged to be congruent with the governing norms.
This review was developed and reported considering the Preferred Reporting Items for Systematic Reviews and Meta-Analyses 2020 (PRISMA-2020) checklist9 (checklist in online supplemental appendices 1 and 2) and in accordance with the COnsensus-based Standards for the selection of health Measurement INstruments 2018 (COSMIN-2018) guideline for systematic reviews of PROMs10 (see section on assessment of methodological quality). The protocol for this systematic review is available at https://doi.org/10.5061/dryad.8cz8w9gsq.
The search strategy was aimed at retrieving articles published in the English language, on validated patient-reported outcomes, which were direct reports from patients about their medication-adherence status. Grey literature was excluded from the search criteria. The following databases were searched from January 1980 until 1 March 2022: MEDLINE (EBSCO), EMBASE (EBSCO), CINAHL (EBSCO), International Pharmaceutical Abstracts (EBSCO) and the Cochrane Library. We combined three groups of keywords: those relating to patient-reported outcomes (e.g., self-report$, patient-report$), those relating to conformity towards instructions pertaining to medication taking (e.g., complian$, adheren$, non-adheren$, poor adherence) and those relating to psychometric measures of adherence (e.g., validity, reliability, construct, instrument). The full strategy is available in online supplemental appendix 3. In the MEDLINE query strings, for example, we used free texts and keywords, and expanded the medical subject headings terms. The search terms included in the search strategy were agreed by two experts in the field of chronic illnesses (clinician) and COSMIN psychometric properties (biostatistician). The sensitivity search filter developed by Terwee et al11 was adapted and used on all the databases. All selected records were uploaded and managed within EndNote X20 (Clarivate Analytics, Philadelphia, PA, USA).
Initially, the items for each measure were generated and content validity assessed. The research must measure what it claims to have measured, i.e., the construct of each measure must be valid for the data to be valid. Subsequently, the scale was constructed, the questions pretested, survey administered, number of items reduced and the number of factors the scale captured was studied. During the evaluation phase of the scale, the number of dimensions, reliability and validity were assessed.
We report on the types of validity measured in the quantitative studies. For example, whilst one study described the process of developing the measure’s scale, other studies assessed the reliability and validity in various settings. In such cases where several studies used the same PROM, attempts were made to select the study in which the PROM was first featured or developed and validated.
Additionally, the following criteria were considered:
Articles describing patient-reported and validated medication non-adherence measures development and/or validation.
Each item/criterion/subscale of a measure is guided by a statement or close-ended question. If the question is open-ended, the response is dichotomised.
At least two psychometric properties (reliability and/or validity) were assessed.
Included a proxy measure (employed a direct or indirect method to correlate adherence level with that evaluated by the measure).
Articles where the scoring method or criteria for distinguishing the level or type of non/adherence could not be sourced.
Studies published only as abstracts or protocols.
PROMs’ items consisting of open-ended questions which were without a method of rating or scoring the adherence measure. This is because without the specification for scoring (for example, an answer of ‘yes’ carried one mark, towards the sum score), an overall assessment of adherence, i.e., poor, moderate or good, could not be performed.
Translation from an original English version into a different language.
All titles and abstracts were independently screened by two reviewers, MLF and KMS. To identify relevant papers, the reference lists for articles that met the inclusion criteria were then reviewed. The reviewers were trained and a pilot with 30 papers was performed to guarantee an inter-reviewer agreement (until a kappa score of ≥0.75 was attained). Reasons for exclusions at the full-text screening process were recorded and discrepancies at any stage were resolved through discussions or a third reviewer consulted (KAMI) to reach a consensus.
A PRISMA flow chart summarises the process of selection (figure 1).
Data extraction and tabulation
Two reviewers (MLF and KMS) independently extracted the following data, where available, from the included articles:
General characteristics of the study setting (year of publication, age, country).
Characteristics of disease or condition: disease or condition studied, characteristics of the diseased respondents.
PROMs’ characteristics: methods of administration, availability of electronic administration, response scale, domains, scoring, number and type of items, proxy measures of adherence (indirect and direct methods).
Quality assessment of the psychometric properties of the PROMs (statuses on reliability and validity of measure/instrument).
Data from each study were tabulated using a standardised and pretested Word data collection form which was used for the purpose of achieving reliable inter-rater scores (online supplemental appendix 4). The corresponding author of the included articles was contacted via email for clarification or to request for additional information on any missing or unclear data. For the purpose of reporting, all terms related to adherence, such as compliance and concordance, were standardised and reported as adherence.
Data synthesis and analyses
Characteristics of participants: patients with chronic diseases and/or require drug therapy, including but not limited to recipients of renal transplants.
Characteristics of the proxy measures: used another direct (e.g., biomarkers) or indirect (e.g., pill count, electronic monitors) method in addition to administering the PROM.
Characteristics of the criterion items: statements, open or close-ended questions that can be binary coded, for example, medium and high adherence (adherent) versus low adherence (non-adherent).
Characteristics of the process: analyses of primary data and coded into ‘behaviour’, ‘barrier’ or ‘belief’.
Characteristics of the outcome: analyses of primary data and coded into ‘intentional (I)’, ‘unintentional (UI)’ or ‘mixed (I/UI)’ non-adherence.
Item classification (binning)
Items within outcomes and process domains were classified as follows:
(1) Outcome domains: intentional, unintentional or not reported, and (2) process domains: behaviour, barrier or belief.
Guided by the Medication Adherence Model12 and the Health Belief Model,13 items which were phrased to incorporate an objective reasoning or a predetermined consequence of a medication-taking decision were classified as intentional. An example of an item is, ‘Did you take less than the amount prescribed when you started to feel an unpleasant side effect?’
We classified items as behaviour when the medication-taking conduct was attributed to one’s own responsibility. Where instructions on timings, doses and frequencies are adhered to, this behaviour was habitual and reflected self-efficacy, self-management skills, motivation, social support and discipline. An example of an item is, ‘I was careful not to miss a dose’. A tangible barrier comprised items addressing physical or practical aspects, for example, impaired manual dexterity, visual impairment, holidays, weekends, storage issues, prescription refills, cost and difficulty remembering, i.e., forgetting. Belief comprised items driving implicit judgements of the treatment or necessity beliefs—for example, perceptions, past negative experiences, knowledge on dosing, not taking when feeling better or taking only when feeling worse, trust and confidence in healthcare providers or the system. Online supplemental appendix 5 enlists the methods used to classify the PROMs’ items.
To ensure that the binning process was exhaustive, two independent reviewers (MLF and KMS) evaluated every PROM item for possible inclusion into any one of the three outcome and process domain categories depending on the nature of each criterion. The supplemental questions to the index question, for example, subquestions which explored reasons for ‘Did you forget to take your medication?’, such as ‘Did you forget to take when on holiday?—intentional (I)’ were identified as an additional item, which was then assigned to a domain.
In the final data synthesis, MLF and KMS independently assigned each PROM an overall category of either intentional (I), mixed of intentional and unintentional (I/UI) or unintentional (UI) depending on the coded items’ dominant or leading domain. Items of studies which aimed at measuring adherence (vs non-adherence) were reverse worded and scored. Any discrepancies were resolved via discussions or via a third reviewer (KAMI).
Assessment of methodological quality
MLF and KMS performed the quality assessment of the psychometric properties of the PROMs independently using the COSMIN guidelines.10 Each PROM’s development and its nine psychometric properties were assessed as ‘doubtful’, ‘inadequate’, ‘adequate’, ‘good’ or ‘very good’. In addition, we used Terwee et al’s11 psychometric quality criteria to assess and rate each criterion as ‘positive (+),’ ‘indeterminate (?)’ or ‘negative (−)’. In this review, we included PROMs which assessed and described a minimum of two psychometric properties.
Patient and public involvement
Patients or the public were not contacted nor involved in this study.
Kolmogorov-Smirnov test was used to test normality and subsequently the Spearman rank correlation was used to assess the correlation between outcome domain variables and process domains. Associations between measures’ increased intentionality (dependent variable) and their increasing behaviour or barrier or belief criterion items (independent variables) were studied in regression analyses. Type of patient care (dichotomous variable—primary care vs specialised care) was included in a logistic regression model, while the number of behaviour, barrier or belief criterion items (continuous variables) was included in linear regression models. Regression models were adjusted for age. Results from the logistic and linear regression analyses are presented as an OR and standardised beta coefficients (β) with 95% CIs, respectively. The level of statistical significance was set at p<0.05. The statistical analysis was performed with SPSS (version 24, SPSS Inc., Chicago, 147 IL, USA).
Search results and study characteristics
In total, 31,960 articles were retrieved from the five databases (figure 1). After removing duplicates, a further 28,600 articles were excluded based on the titles and abstracts (online supplemental appendix 6). Six hundred twenty relevant articles were identified from the reference lists. Hence, 3660 articles remained for full-text review. These articles were scrutinised against the paper’s inclusion and exclusion criteria. At this stage, 3601 articles were eliminated. In the span of four decades, i.e., from January 1980 through to March 2022, a total of 59 PROMs relevant to adherence were extracted—one from each of the remaining articles (n=59).
In instances where a number of studies utilised the same PROM, subsequent reporting for that PROM was based on the initial study in which the measure first appeared. As for methodological quality, the studies fulfilled the inclusion criteria of adequate description of at least two psychometric properties, for example, its content, construct, structural validity and/or reliability. Criterion validity was achieved by all included studies. This is because studies on PROMs that were juxtaposed against another standard or proxy measure were included, which meant that the studies included presented a correlate, i.e., a corresponding indirect or direct method employed in parallel with the PROMs to assess adherence. For instance, the pill count method, using an additional PROM, or data mining from medication possession ratio or from a medication events monitoring system, or direct measurements of biomarker plasma levels were undertaken.
The studies were conducted across regions of North and South America, Europe, Australasia, and in 16 countries, i.e., the USA (24), the United Kingdom (8), Canada (4), France (3), the Netherlands (3), Spain (3), Malaysia (2), Australia (2), China (1), Brazil (2), Hungary (1), Switzerland (2), Belgium (1), Germany (1), Romania (1) and Turkey (1). The respondents’ ages ranged from 18 to 88 years. The respondents included organ transplant recipients or patients receiving specified treatments for chronic conditions such as retroviral or cardiovascular diseases, kidney and endocrine disorders, ophthalmic conditions, gout, psychiatric illnesses or oncology. They were attending doctor appointments either at their general practice, or within secondary or tertiary care settings, or visiting primary healthcare facilities (community pharmacies, optometry services). Additionally, several patients were admitted from the emergency departments or hospitalised after surgery.
Many of the PROMs were predominantly intentional followed by mixed intentionality and unintentional. These 59 PROMs yielded a total of 740 items/criteria that were related to adherence. Of those items, 572 were intentional in nature, 90 were unintentional and 78 did not report. Tables 1–3 depict the process and outcome domains assigned to individual criterion items, as well as the overall categorisation of each measure.
Each measure or instrument was grouped into one of the three PROMs of non-adherence categories: table 1–intentional, I (n=44 of 59 instruments)5 6 14–55; or table 2–mixed, I/UI (n=13 of 59 instruments)55–67; or table 3–unintentional, UI (n=2 of 59 instruments).68 69 Of the total 572 intentional criterion items, I=491, I=79, and I=2 were split into the intentional, mixed intentional/unintentional, and in the unintentional categories, respectively. Of the total 90 unintentional criterion items, UI=35, UI=50, and UI=5 were split into the intentional, mixed intentional/unintentional, and in the unintentional categories, respectively. Finally, additional n=67, 9 and 2 items were not reported within each category.
The process domains which affected non-adherence were belief (383 items), barrier (192 items) and behaviour (165 items).
Correlation and regression analyses
There was a strong positive correlation between the outcome domain, intentional (I) and the process domains, belief (Spearman’s ρ=0.88, p<0.001) and barrier (Spearman’s ρ=0.59, p<0.001), i.e., the numbers of belief and barrier criterion items increased, along with the numbers of intentional criterion items.
The results of the regression analyses are displayed in table 4. Regression models were adjusted for respondents’ age. There was a positive and significant association between increased PROMs’ propensity to detect intentional non-adherence and primary care (vs specialised care) that patients received (OR=1.9; 95% CI=1.01 to 2.66), i.e., a measure was 1.9 times more likely to have detected an intentional non-adherence event among patients who attended primary care as compared to those attending secondary and tertiary care appointments, or who were hospitalised at these facilities. Similar findings were obtained for negative beliefs and barriers encountered. For every belief or barrier criterion item included, the PROMs’ aptitude for detecting intentional non-adherence increased by β=0.79 and β=0.34 units, respectively. Model R2=0.94, i.e., 94% of variance in the outcome variable intentional non-adherence can be explained by the predictor variables belief and barrier.
All studies included had a minimum of two psychometric properties assessed and described (online supplemental appendices 6–8). With reference to validity, hypotheses testing for content (including face) validity and/or construct validity were evaluated in a majority of the studies.6 18 20–22 24–27 31 39 42 43 46–50 52 57 58 60 64 66 With reference to reliability, the parameter internal consistency, was evaluated in 38 studies, respectively. Measurement error or cross-cultural validity/measurement invariance was not evaluated in any of the studies. In addition, many of the studies assessed PROMs’ development, but they were not necessarily of good methodological quality (for example, limited information on data validity as described in the eligibility section) and were hence excluded.
Defining and scoring adherence
There was heterogeneity in the term ‘adherence’ used in the studies. First, this was evidenced in the cut-off points used for determining level of non-adherence, for example, ‘habitual to occasional refusers of medications’,14 ‘high, medium or low adherence’,5 41 56 and ‘adherent’ or ‘non-adherent’.19 21 23–30 32–40 42–44 47 51 52 57–66 68–70
All PROMs that had a component of unintentional had binomial categories of ‘adherent’ versus ‘non-adherent’. While a handful of studies specified the score which determined ‘high’ score16 and hence ‘highly’ adherent, others reported that a higher score indicated better adherence.17 48 50 A majority of the measures (n=47 of 59 instruments) were worded as closed-ended questions and provided responses in the form of nominal or Likert scales; seven measures were open-ended and provided dichotomised Yes/No scale (which ultimately yielded a positive or negative overall score which determined level of adherence).5 14 15 18 21 34 44 Several measures incorporated an addition of Visual Analogue Scales (VAS).21 35 46
The measures for assessing adherence were patient or interviewer-administered and conducted via face-to-face interviews (real-time or virtual), telephone interviews, questionnaires, diaries or internet surveys.
Extent of non-adherence
Seven measures specified a time period for non-adherence, for example, ‘Did you miss a tablet–yesterday? last week? last month? or in the last 3 months?’ Fifteen adherence measures specified a frequency for skipping or missing doses, for example, ‘How many days over the past month did you take less than prescribed?’ The time frame specified ranged from 1 day to 12 months. Most of the items were interpreted from measures that state the number/amount of doses taken compared with those prescribed for that time period. Haag et al53 went on to stratify their analysis according to level of adherence for daily, two times per day and three times a day regimen.
Some of these adherence scales are disease-specific and thus explored common predictors that may influence adherence in those diseased populations. Examples of diseases or statuses explored were asthma,61 diabetes, AIDS,58 hypertension,57 68 chronic diseases, acute osteoporosis, eye-related conditions,16 cardiovascular, and organ transplant recipients.27 59
Quantification of adherence and non-adherence
Medication adherence has been defined as ‘the extent to which patients take medications as prescribed by their healthcare providers’,71 while its measurement is the process of systematically assigning numbers or words to tangible and intangible characteristics so that they can be defined, quantified and differentiated. In our review of 59 PROMs, these self-reported measures aimed at quantifying medication adherence based on the responses from a nominal or Likert-point scale and assigning categories to the total scores. A small number of adherence measures had taken a different approach to assigning the ‘adherent’ category. The Medication Adherence Questionnaire (MAQ), Morisky Medication Adherence Scale (MMAS), Brief Medication Questionnaire, Adherence Self-Report Questionnaire (ASRQ) and VAS categorised patients into levels of adherence. For example, the MAQ and MMAS categorised the population into high, medium and low levels of adherence whilst several others collectively grouped respondents with medium and low adherence scores into a broader ‘non-adherent’ class. For the Brief Medication Questionnaire, the study population was grouped into repeat, sporadic and no non-adherence. The ASRQ and VAS classified non-adherence into six and seven levels, respectively, based on the researchers’ expertise. Where reported, the most common cut-off point for non-adherence is the score below the corresponding point of another objective or self-reported measure. For instance, the cut-off was the point where it was revealed that the patients took 80% of their medication. The MMAS cut-off points, for instance, were selected based on the correlation with blood pressure control. Other adherence measures, such as the Drug Attitude Inventory, Adherence Attitude Inventory and Medication Adherence Self-Efficacy Scale-Revision first segregated the population into adherent and non-adherent based on responses to questions about whether medications were taken or not, and then compared the mean scores of the two adherence measures to determine the cut-off.
Since the publication of the inaugural study on adherence in 1968,72 many studies aimed at understanding factors related to poor adherence, yet they lacked crucial perspectives on the measurement tools used to evaluate non-adherence. The approach for combining adherence and its measurement is complex given that adherence impacts health-related clinical outcomes.73 The analyses of PROMs’ psychometric properties were hence essential, given the vast and multidimensional outcomes they evaluated. To ensure construct validity, several studies used exploratory factor analysis to ascertain that the smallest number of factors that best represented the items was retained.27 42 50 Reliability was evaluated using Cronbach’s alpha tests for internal consistency to assess how closely related a set of items were as a group and this was frequently reported and often yielded values of more than 0.7, indicating a relatively acceptable level of internal consistency and correlation between PROM items. In addition, the reliability of a measurement tool in consistently reproducing the same result across several measurements, denoted using a test/retest reliability score, were also not regularly reported.
It is crucial to evaluate PROM in the context of evaluating medication-taking conduct, i.e., its construct and structure need to be sensitive and adequately detect aspects of medication-taking behaviour. Medication-taking behaviours were often detected when using a direct, indirect or a combined method. The most precise approaches were postulated to have been direct methods, such as directly observed therapy, biological methods, and measurement of the level of medicine or metabolite, for example, blood or urine drug concentrations. Numerous indirect methods were equally proposed, such as the use of clinician reports, data from rates of prescription refills, pill counts, electronic medication monitors, patient diaries and patient self-report measures as well as healthcare provider-reported measures. When reporting on aspects of PROMs’ construct, many of the studies evaluated its criterion and/or discriminant validity. However, there is also a school of thought that described criterion validity as being inaccessible for evaluation as it requires a component of comparison with an established ‘gold’ standard. The authors synthesised the studies and deemed proxy measures as either direct, indirect or a combination of both, as standards against which the PROMs were validated.
In summary, the review identified 59 unique PROMs of medication adherence. The selected 59 PROMs demonstrated ease of administration primarily in community dwellers seeking services at primary healthcare facilities. Patients who visited secondary or tertiary care outpatient clinics, blue collar workers undergoing health screening and patients hospitalised following emergency department visits were end-users of the PROMs. These measures were tested primarily on adherence to pharmacotherapy for psychiatric disorders, and for retroviral and cardiovascular diseases. This was an indication that the PROMs had good usability and credibility, and were meaningful to be used for their targeted populations. Eight of these outcome measures had adequate, evidence-based psychometric properties, and had the aptitude to identify those who intentionally failed to adhere to good medication-taking conduct. They were the Compliance Questionnaire Rheumatology,22 Immunosuppressant Therapy Barrier Scale,27 Medication Adherence Rating Scale,20 Treatment Satisfaction Questionnaire for Medication,25 Self-Efficacy for Appropriate Medication Use Scale,32 Medication Adherence Assessment Tool,50 Patient-Medication Adherence Instrument and Healthcare Professional-Medication Adherence Instrument52 and Barriers to Oral short-Term antibiotic Adherence53 questionnaires. Two measures with adequate, evidence-based psychometric properties had the aptitude to identify those who intentionally and unintentionally failed to adhere to good medication-taking conduct. They were the Adherence Starts with Knowledge-2066 and Hill-Bone Scale57 questionnaires.
Intentionality (intentional and/or unintentional)
While the advantages of those psychometric properties were recognised, authors seldom weighed in on the inherent ability of self-reports to assess intentionality. A patient’s belief, past experiences and cognition level may equally come into play, influencing medication-taking conduct. Decisions by patients may be well-informed or ill-informed driven by concerns over the benefits and risks of their medication.74 Using theories of Planned Behaviour and the Common Sense Self-regulatory Model of Illness, several attempts were made to explain variation in adherence to medication.17 A social cognition model that aims to explain one’s intentional non-adherent conduct was developed in 1999 by Horne17 and was founded on the Health Belief Model75 and the Illness Perceptions Model.76 According to Horne, adherence to medication is based on patients’ deeply-rooted belief concerning perceived severity of illnesses, i.e., perceptions and their understandings, and hence, knowledge of the nature of the illness, and treatability.17 In the context of adherence, belief is usually associated with intentional non-adherence because the belief can strongly lead to an active decision either to follow or not to follow instructions by healthcare professionals.
Self-efficacy, on the contrary, is key for habitual non-adherence and is a term that encompasses self-management skills, discipline and the state of being self-sufficient.77 78 In our review, we observed more of such terms in the later versions of the health beliefs model to increase the explanatory power of the theory, placing emphasis on patient accountability .79 80 In recent times, during the COVID-19 pandemic, the health promotion model is also referenced to as it emphasises self-efficacy and purports that judgement about one’s ability to successfully perform challenging tasks is based on four types of criteria: (a) behavioural engagement and evaluating performance, (b) observation of how others perform, (c) verbal persuasion by others about one’s ability, and (d) anxiety, fear, calm, tranquillity, or other physiological states associated with self-judgement of competencies.81 82
When used effectively, measurement tools such as self-reports essentially screen for these predictors (be it one’s belief system—perceptions, experiences, knowledge and trust towards healthcare professionals or one’ behaviours—self-efficacy and habits). These perspectives are important for better understanding of the underlying reasons contributing to non-adherence. As this review demonstrates, distinguishing intentional from unintentional non-adherence is pertinent. Only then can appropriate and targeted interventions be provided to the patient.73 83
Although we successfully classified each item of every measure or instrument based on intentionality for non-adherence, future research using a similar approach may reveal varied findings. This is because the process of classification was largely dependent to a certain extent, on one’s interpretation of the criterion items’ construct and thereby the consequences of treatment default. Our findings also cannot be used to ascertain the phase of medication taking, whether ‘initial, i.e., starting a recommended medication regimen’, or ‘implementation, i.e., executing the prescribed dosage schedule’ or ‘persistence, i.e., length of time on regimen before discontinuation’.84 In addition, while there is a body of evidence to suggest that past negative experiences with the healthcare systems were attributed to the intention to default treatment, the key role of healthcare professionals in advocating positive habits for medication adherence remains vital. In other words, past negative experiences could become positive and the new experience should equally influence medication-taking conduct.
Recommendations for best practice
Many of the studies identified PROMs of adherence towards antiretroviral and post-organ transplant therapies. End-users of these measures can easily be subjected to stigmatisation, and as such, undergoing a series of validity tests including cross-cultural validity, can prove beneficial for the development of individual measure items. This is also true for measures where the appeal, social desirability and sensitivity are of concern, and yet the assessment needs to be conducted without compromising the method in which the latent trait of the measure is being measured across cultures and populations. For instance, as proposed by Stirratt et al,84 respondents with cognitive impairment may require other approaches such as daily texts, interactive voice response surveys, or computer-assisted measures.
In addition, the testing for criterion and construct validity was often conducted using proxy measures as opposed to direct measures. Although our review demonstrated the link between intentional non-adherence and one’s belief system, and in theory, responses to sociocognitive domains can be used as proxy measures, we recommend further statistical analyses, for example sensitivity and positive predictive value calculations, to assess the concordance of direct and proxy measures of individual measure items. Despite having the advantage of producing quick results, a PROM needs to be reliable, its results reproducible, and have an ability to capture a single underlying aspect of behaviour, barrier, or belief, and map that construct onto a valid measurement scale.
Our findings demonstrated that existing measures had sufficient evidence-base and psychometric properties, and therefore, should first be used, as opposed to developing a new measure, unless the study subjects do not correspond with the tested population.85 Consensus of a panel of experts is necessary for determining criteria for scoring and cut-off for adherence versus non-adherence. They need to consider frequency and dose (and the allowable deviation in the amount prescribed whenever the patient decides to lower or increase dose), as well as determine habitual versus occasional refusers. Additionally, developing response items that differentiate adherence in ordinal terms (e.g., anchored Likert rating scale), nominal or VAS (e.g., estimated level of doses taken) may help decrease ceiling effects. To curb recall bias, phrases like ‘during the last weekend’ and ‘in the past 7 days (to include one weekend)’ were preferred. A 30-day recall period, as opposed to brief intervals, can also be considered in view of its advantageous impact on ceiling effect reduction.
Finally, future research could underscore the adjudication of measurement errors and their sources of variability, as there is a paucity of information in these areas. Elucidating responsiveness of PROMs and their ability to detect changes in clinical conditions of patients are also potential areas for psychometric characteristic evaluation.
The data spanning four decades revealed that PROMs were developed to assess medication non-adherence, which had increasingly become more intentional. Of the 59 studies, ten PROMs had adequate, evidence-based psychometric properties. Of the ten, eight and two PROMs were able to detect total and partial intentionality to medication default, respectively. Measures with matching intentionality properties will better detect non-adherence, thereby relieving clinicians from having to contemplate on ineffective medications or therapeutic failure due to drug resistance. Finally, recommendations for patients to simply use daily reminders to consume medication should be brought to a halt as it seems counterintuitive to tackle the premeditated intentional non-adherence. Instead, because our data demonstrated a link between intentional non-adherence and one’s belief, empowering patients with the appropriate knowledge, helping them to better manage their illnesses, and strengthening their trust in medications and healthcare providers could be more reliable impetuses to drive positive attitudes and perceptions. However, these need to commence at the primary care level.
Data availability statement
All data relevant to the study are included in the article or uploaded as supplemental information. Additional data can be accessed via the Dryad data repository at http://datadryad.org/ with the doi: 10.5061/dryad.8cz8w9gsq.
Patient consent for publication
We thank colleagues at the Faculty of Pharmacy, UiTM for their continued support. Our gratitude is extended to Dr Ungku Ahmad Ameen Ungku Mohd Zam for his expert advice.
This web only file has been produced by the BMJ Publishing Group from an electronic file supplied by the author(s) and has not been edited for content.
MLF and KMS are joint first authors.
Contributors MLF acts as the guarantor. MLF and KMS conducted the search, juxtaposed the articles against the review’s inclusion and exclusion criteria, and performed the item binning, coding and data analysis. MLF, KMS and KAMI analysed and interpreted the binned data. MLF, ASA and FN were responsible for database access and funding acquisition. MLF, KMS and FR drafted the manuscript. All authors read and approved the final manuscript.
Funding This work was supported by the Special Research Grant (600-RMC/GPK 5/3 (107/2020)), Universiti Teknologi MARA, Malaysia.
Competing interests None declared.
Patient and public involvement Patients and/or the public were not involved in the design, or conduct, or reporting, or dissemination plans of this research.
Provenance and peer review Not commissioned; externally peer reviewed.
Supplemental material This content has been supplied by the author(s). It has not been vetted by BMJ Publishing Group Limited (BMJ) and may not have been peer-reviewed. Any opinions or recommendations discussed are solely those of the author(s) and are not endorsed by BMJ. BMJ disclaims all liability and responsibility arising from any reliance placed on the content. Where the content includes any translated material, BMJ does not warrant the accuracy and reliability of the translations (including but not limited to local regulations, clinical guidelines, terminology, drug names and drug dosages), and is not responsible for any error and/or omissions arising from translation and adaptation or otherwise.