Understanding Patients’ Adherence-Related Beliefs about Medicines Prescribed for Long-Term Conditions: A Meta-Analytic Review of the Necessity-Concerns Framework

1Prescribing medicines is fundamental to the medical management of most long-term conditions. However, approximately half of this medication is not taken as directed, representing a failure to translate potentially effective treatment into optimal outcomes for patients and society,. Where prescriptions are appropriate, this level of nonadherence has potentially serious consequences, both for individual patients, in terms of lost opportunities for health gain with increased morbidity and mortality, and for the health care system, in terms of wasted resources, increased use of services and hospital admissions.

2In the absence of a single definitive intervention to address nonadherence, the NICE Medicines Adherence Guidelines amalgamate insights from trials of interventions and explanatory studies of nonadherence. They apply a perceptions and practicalities approach recognising that nonadherence may be both unintentional and intentional. Unintentional nonadherence occurs when the patient wants to adhere but is unable to because they lack capacity or resources. For example, they may not have understood the instructions, cannot afford copayment costs, or find it difficult to schedule, administer or remember the treatment. Intentional nonadherence occurs when the patient decides not to follow the recommendations. It is best understood in terms of the perceptual factors (e.g. beliefs and preferences) influencing motivation to start and continue with treatment.

3Prescribing consultations do not occur in a vacuum. Patients (and prescribers) bring pre-existing beliefs about the illness and treatment, which influence the patient’s evaluation of the prescription, their adherence and even beneficial or adverse outcomes. Interventions to optimise adherence tend to be more effective if they are tailored to the needs of the individual taking account of the perceptions of the treatment as well as practical abilities and resources that enable or impede their adherence. Although the perceptual and practical dimensions of adherence are influenced by the social, cultural, economic and healthcare system contexts, taking account of the patient’s beliefs about the prescribed medication is fundamental to shared-decision making and supporting adherence,.

4Research conducted with patients with a variety of long-term conditions suggests that the key beliefs influencing patients’ common-sense evaluations of prescribed medicines can be grouped under two categories: perceptions of personal need for treatment (Necessity beliefs) and Concerns about a range of potential adverse consequences,,. This ‘Necessity-Concerns Framework (NCF)’ potentially offers a convenient model for clinicians to elicit and address key beliefs underpinning patients’ attitudes and decisions about treatment.

5Over the past decade, a number of studies have been conducted, using a validated questionnaire, the Beliefs about Medicines Questionnaire to quantify Necessity beliefs and Concerns in order to explore the relationship between these beliefs and adherence. This research spans a range of long-term medical conditions, across different settings and within various cultural groups. Many of the individual studies have demonstrated the utility of the NCF in explaining nonadherence to medication (e.g. –). It is therefore timely that a meta-analysis is performed to consolidate the results from these studies and to examine the explanatory value of the NCF in predicting adherence to medication prescribed for long-term medical conditions. In line with the underlying theory, we hypothesized that adherence in long-term conditions would be associated with stronger perceptions of Necessity for treatment and fewer Concerns about adverse consequences.

6This review was conducted in line with the MOOSE guidelines for meta-analysis of observational trials.

7A computerised literature search was conducted by the investigators on April 22nd, 2013 using EMBASE, Medline, PsycInfo, CDSR/DARE/CCT and CINAHL. The search strategy included the following terms:

8BMQ or belief$

9and

10treatment$ or medicine$ or medication$

11and

12adheren$ or complian$

13The search was limited to studies published from the year 1999 onwards (the year in which the BMQ was published). Duplicates were removed.

14Identified studies were included in the meta-analysis if they met the following criteria:

15participants were suffering from a long-term condition

16participants were taking medication

17participants were adults

18the article was published in a peer-reviewed journal

19the Necessity and/or Concerns subscales of the BMQ were used

20a measure of adherence was employed

21There were no restrictions based on language, or on cultural or geographical factors.

22Titles and abstracts were screened for relevance, and the full text of relevant articles was obtained. Data from each article was extracted as described below.

23Fifteen studies reported multiple associations of beliefs related to different adherence measurements (details reported in Table 1). Where the choice was between adherence measures, the most objective measure was selected for the meta-analysis. Therefore, electronic monitoring of adherence and prescription redemption data were chosen over self-report. Where data was presented for both ‘on demand’ and prophylactic medications, data for the prophylactic medication data were chosen,, for consistency with medications prescribed for other long-term conditions. In studies where cross-sectional and longitudinal data were both available, longitudinal data was used within the analysis, –. Where one group provided cross-sectional data at multiple timepoints, the timepoint with the fewest missing data points was selected. If the choice was between two self report measures of adherence, we used the more commonly used measure. Thus the Morisky Medication Adherence Scale (MMAS) was chosen over the Brief Medication Questionnaire and the ACTG adherence measure was used over the Walsh VAS scale. Where patients within a sample were taking multiple medications and individual associations were provided for each medication,, the mean association was used within the meta-analysis but individual effect sizes are reported in Table 1 to facilitate comparison. Where data on two samples are reported within the same study, we included both associations within the analysis.

24The following information was extracted from papers onto coding forms: author names, date of publication, the country in which the research was conducted (dichotomized into UK or non-UK), sample size, illness group, sex (% male), mean age, study design (cross-sectional, longitudinal or prospective), the number of Necessity and Concerns items included (since items may be added specific to the medication prescribed), the adherence measure used, information (means and standard deviations, odds ratios and 95% confidence intervals or correlation coefficients) to calculate the effect size between adherence and Necessity beliefs and Concerns, and the p-value. Where the full required statistics were not reported, authors were contacted for further information.

25A simple methodology assessment tool was devised for this study. Methodology was assessed by two of three independent expert raters (SC, RP and VC) using the following parameters:

26study location (UK or non-UK)

27study design (cross-sectional or longitudinal/prospective)

28measure of adherence (self-report or objective measure [electronic monitors, prescription redemption, blood test results]).

29sample size (<82 = 0 or ≥82 = 1). This was based on the sample needed to detect a medium effect size for a correlation (r = 0.3) with an alpha level of 0.05 and 80% power.

30Ratings were completed independently and then combined. There were no disagreements regarding ratings.

31The primary outcome measure was adherence to medication. For each study, the effect size was expressed as an odds ratio with 95% confidence intervals. Where studies reported the standard mean difference or correlation coefficient, the effect size was converted into an odds ratio, using the Comprehensive Meta-Analysis program. We used a random effects model to accommodate heterogeneity between studies which was anticipated due to differences with respect to sample characteristics, study design and the adherence measure used.

32The presence of significant heterogeneity across studies was examined using the chi-squared statistic (Q). The magnitude of this heterogeneity across studies was estimated using the I 2 statistic, which assesses the percentage of variance among studies which is not due to chance.

33Sensitivity analyses were conducted to ascertain whether the effect sizes seen were robust when individual studies, or studies grouped based on the methodological factors described above were excluded.

34Orwin’s fail-safe N, was calculated to estimate the number of unpublished studies necessary to reverse any conclusion that a significant effect exists (based on the conservative assumption that unpublished studies would have effect sizes of equal magnitude but opposite direction to the overall effect size in this meta-analysis). Egger’s t-test and funnel plots were also used to test for publication bias, in line with recent recommendations.

35Ninety-four percent (3554) of the 3775 studies retrieved were rejected after checking the titles and abstracts against the selection criteria above (Figure 1). 223 relevant articles were identified. A search of the reference lists of these articles revealed one further relevant study.

36Of the 223 studies identified, a further 129 were excluded (Figure 1). Thirty of these were unpublished studies and conference proceedings. These were investigated further and authors were contacted where necessary to clarify whether unpublished work had led to publications –. Sixteen studies, – had since been published, fifteen of which already formed part of the included list and one additional eligible study was available online early. Six papers reported data on samples which overlapped with included studies –, and four were protocols for ongoing studies –.

37Thirteen studies were excluded because they did not include a measure of medication adherence –. Two of these included separate assessment modes for intentional and unintentional adherence but no overall adherence assessment,. Fifty-five studies did not use the BMQ Specific scales –. Four studies were excluded because the relationship between treatment beliefs and adherence behaviour was not reported, –. Two articles were conducted in acute rather than long-term condition samples (influenza and antibiotic use ) and one article was excluded because parental beliefs about medicine were measured. Thirteen studies study met the inclusion criteria but the article did not contain the required statistical information. We contacted the authors but were unable to obtain the relevant data, –. Thus, once screened against the inclusion criteria, 94 articles were retained for inclusion in the meta-analysis. Table 1 provides a summary of each of the studies included in the meta-analysis.

38Three of the included studies,, reported associations between adherence and Necessity beliefs, but not Concerns. The authors of these articles were contacted, but the data for Concerns was unavailable. Two studies, reported two largely non-overlapping samples for both Necessity beliefs and Concerns. Thus, data for 91 studies and 93 comparisons for Concerns, and data for 94 studies and 96 comparisons for Necessity beliefs, were included in the meta-analysis.

39The mean age of participants in the 94 included studies ranged from 24.0 to 74.2, with an overall mean age of 55.8 (it was not possible to calculate the mean age in 13 studies). The percentage of males ranged from 0–100% (breast cancer and haemophilia samples respectively), with an overall percentage of males of 49.7% male (excluding 3 studies where it was not possible to calculate the number of males). Sample sizes ranged from 16 to 1871.

40The total sample, N = 25,072, encompassed patients with asthma, renal disease, organ transplantation, dialysis chronic pain, kidney transplantation, cancer, cardiovascular disorders, Marfan’s syndrome, depression, haemophilia, diabetes, HIV, rheumatoid arthritis, osteoporosis, thalassemia, inflammatory bowel disease, bipolar disorder, schizophrenia, epilepsy, migraine, back problems, glaucoma and mixed chronic illness.

41Thirty-three studies (35.1%) used the MARS to measure adherence, 20 used the Morisky Medication Adherence Scale (21.2%), 3 used pharmacy refill (3.2%), 3 used electronic monitoring (3.2%) and two or fewer studies used the remaining measures.

42There was a significant relationship between Necessity beliefs and adherence, OR = 1.742, 95% CI [1.569, 1.934], p<0.0001. There was significant heterogeneity between the 96 comparisons from 94 studies, Q(95) = 422.662, p<0.001, which was substantial in magnitude, I 2 = 77.52%.

43Figure 2 presents the individual effect-size estimates and shows that the relationship between Necessity beliefs and adherence was significant (p<0.05) for 49 (51.0%) of the included studies. Sensitivity analyses revealed that the overall result was not affected when any single finding was omitted.

44There was a significant relationship between Concerns and adherence and fewer Concerns about adverse effects, OR = 0.502, 95% CI: [0.450, 0.560], p<0.0001. There was significant heterogeneity among the 93 comparisons from 91 studies, Q(92) = 481.84, p<0.001, suggesting that factors other than chance accounted for a moderate-substantial amount of variance, I 2 = 80.91%.

45Figure 3 presents the individual effect-size estimates and shows that the relationship between concerns and adherence was significant (p<0.05) for 53 (57.0%) of the included studies. Sensitivity analyses revealed that the overall result did not change when any single finding was omitted.

46See Tables 2 and 3 for OR stratified by different long-term conditions and adherence measures. Two few studies reported data on the majority of conditions and measures to allow statistical tests for heterogeneity.

47See Table 4 for sensitivity analyses.

48Most studies were conducted outside of the UK (n = 62; 66.0%). Stronger effects were apparent for both Necessity and Concerns for studies conducted in the UK relative to studies conducted outside of the UK, however the relationship between Necessity and Concerns was significant for both locations. Substantial and significant heterogeneity was present in all analyses.

49The majority of studies (n = 77, 81.9%) were cross-sectional, with few studies using longitudinal or prospective designs (n = 17; 18.1%). Effect sizes were similar for longitudinal/prospective and cross-sectional designs for both Necessity and Concerns. Substanital and signficant heterogeneity was present in all analyses.

50Eighty-three studies (88.3%) employed measured adherence using self-report, while 11 (11.7%) used other methods. The association between adherence and Concerns was smaller, but still significant, when objective measures were used, and the heterogeneity around this estimate was small. The association between Necessity beliefs and adherence did not differ if objective or subjective adherence measures were used. Heterogeneity around the subjective measures estimates and the objective Necessity estimate was substantial.

51Eighteen (19.1%) of the studies were classed as having small samples (less than 82). The size of the associations between Necessity and Concerns and adherence were similar for smaller and larger studies. Heterogeneity estimates indicated that variability around the larger samples estimates was substantial. However, the smaller sample estimates were less heterogeneous, with I2 values in the small range for Concerns and the moderate range for Necessity beliefs.

52The fail-safe N (Nfs) was 96, indicating that there would need to be ≥96 unpublished findings of an equal magnitude but opposite direction, to reverse our conclusion that a significant effect exists. Inspection of the funnel plot suggested asymmetry (see Figure 4), however Duval and Tweedie’s trim and fill method did not suggest that studies should be added/removed. Egger’s t-test was significant, t(94) = 1.60, p<0.001, suggesting the presence of asymmetry.

53The fail-safe N (Nfs) was 94, indicating that there would need to be ≥94 unpublished findings of an equal magnitude but opposite direction, to reverse our conclusion that a significant effect exists. Funnel plot inspection suggested the presence of asymmetry (see Figure 5), which was confirmed by a significant Egger’s t-test, t(91) = 1.80, p<0.001. Further, Duval and Tweedie’s trim and fill method suggested 13 studies should be added/removed to make the funnel plot symmetrical. The location of the imputed studies indicated that the asymmetry may arise from a lack of reporting of studies which find a negative relationship between concerns and adherence. However, the similarity between the adjusted OR 0.567 95% CI [0.507, 0.634], which includes the imputed trimmed and filled studies, and the observed OR 0.504 95% CI [0.450, 0.564], suggests that any bias does not have a large impact on the findings.

54This meta-analytic review indicates that the Necessity-Concerns Framework (NCF) is a potentially useful model for understanding patients’ evaluations of prescribed medicines. The magnitude of the aggregate effect sizes indicates that, for each standard deviation increase in Necessity beliefs, the odds of adherence increases by a factor of 1.7. Conversely, for each standard deviation increase in Concerns, the odds of adherence decreases by a factor of 2.0.

55The sensitivity and publication bias analyses conducted confirm our hypothesis that Necessity beliefs and Concerns are associated with adherence/nonadherence to medicines, across a wide range of conditions, medications, and study locations. No research synthesis can transcend the limitations of the primary studies. However, sensitivity analyses confirmed that this association is robust across methodological features; remaining when small, underpowered studies were removed, when only longitudinal/prospective designs were included, and when self-report and non self-report adherence assessments were included separately. The majority of the studies relied solely on self-reported adherence. Self-report measures have high face validity and high specificity for nonadherence, however they may be subject to self-presentation and recall bias. Thus some people may be reporting higher adherence rates than they actually attain. This bias does not diminish our confidence in the finding that beliefs were related to adherence, as there is no evidence that such a bias would be associated with medication beliefs. Indeed some patients with high Concerns and low Necessity beliefs may be expected to incorrectly report high adherence in order to present themselves positively. This pattern would attenuate the relationship found between adherence and medication beliefs, making it less likely that we would find an association between beliefs and adherence. Moreover, given that this relationship remained when non-self report measures were used, we are confident that the observed relationships between beliefs and adherence are not an artifact arising from the limitations of self-report. Only published studies were included, creating a possible bias, since studies submitted for publication may be more likely to have positive results and larger effect sizes. Since for both Necessity beliefs and Concerns, the fail safe N indicated that the number of additional negative findings required to accept our null hypothesis was similar to the number of studies included in this meta-analysis, and there was little suggestion of publication bias through funnel plot analysis, our findings appear to reflect a true relationship between beliefs and adherence.

56Stratifying by long-term condition and adherence measurement revealed a need for further studies using objective measures, and highlighted some conditions, for example epilepsy and functional pain syndromes where further research is needed. We do not know whether the Necessity-Concerns Framework will be of equal utility across medications administered by different routes e.g. depot injections, or if practical barriers to care may be of relatively greater importance in some groups using medications administered through different routes.

57Eighteen studies assessed whether Concerns and Necessity beliefs could predict adherence using longitudinal/prospective designs. The relationship was not reduced in these studies, supporting the proposal that medication beliefs can influence later adherence as part of the self-regulation of illness. We did not restrict our inclusion criteria to studies published in English. However, our search only identified one study published in any other language, despite the fact that the BMQ was translated into the native language for the study. Cultural values can impact on the way in which individuals interact with the healthcare system. However, variations in treatment necessity and concerns and association between these beliefs and adherence were noted across different countries, languages and cultures. We found that studies outside the UK, where the BMQ and it’s disease-specific modifications have been predominantly developed, found reduced associations between necessity and concerns beliefs and adherence. Further work is needed to investigate potential cultural variations in medication beliefs.

58The development of more effective methods for addressing nonadherence is a priority for research and practice,. Our findings suggest, that novel interventions to support informed choice and optimal adherence to appropriately prescribed medicines are likely to be more effective if they take account of patients’ beleifs about the treatment and how they judge their personal need for the prescription relative to concerns about ponteial adfverse consequences of taking it. Necessity beliefs and Concerns may trigger intentional nonadherence, for example, if patients decide not to take their medication due to concerns regarding potential or actual adverse consequences, and unintentional nonadherence, (e.g. if patients who believe a medicine is not important for their health forget to take it). Beliefs can have counter-balancing effects on adherence, such as when patients continue to take a medication they believe is essential for their health despite concerns regarding adverse effects 15. The challenge now is to develop effective interventions to address patients’ doubts about the necessity for treatment and concerns about adverse consequences in order to enhance adherence. The challenge goes beyond ‘getting patients to take more medicines’. Our findings show that many patients harbour significant, unresolved doubts and concerns about prescribed treatment suggesting a fault-line between patients’ and prescribers’ cultural perceptions of the treatment. Viewed from the perspective of biomedicine, nonadherence may seem irrational. However, from the patients’ perspective, nonadherence may be a ‘common-sense’ response to their implicit appraisal of the treatment. For some patients nonadherence might represent an informed choice. In this case the outcome of ‘adherence support’ would be to avoid prescribing an unwanted treatment, to the relief of patient and payer. However, for others, evaluations of treatment necessity and concerns may be based on misconceptions about the illness and treatment.

59More detailed studies of patient representations illness and treatment show that, even when treatment evaluations are based on misconceptions they appear to draw on a ‘common-sense’ logic,,. For example, the need for daily medication may seem less salient when symptoms are absent or cyclical –. Concerns about prescribed medication are not just related to side effects but are common, even when the medication is well tolerated. They are often related to beliefs about the negative effects of medication and include worries about long-term effects, dependence, cost of medication and dislike of having to rely on medicines,. Concerns are related to more general beliefs about pharmaceuticals as a class of treatment which are often perceived as intrinsically harmful and over-prescribed by doctors,. The package information leaflets, dispensed with many prescription medicines may exacerbate concerns as they list all possible side effects, leaving patients with outstanding questions and making it difficult to understand the likely risk and place them in context with potential benefits.

60Nonadherence is often a hidden problem. Patients may be reluctant to express doubts or concerns about prescribed medication and to report nonadherence; sometimes because they fear that this will be perceived by the prescriber as a lack of faith in them. The first step to facilitating adherence is therefore to take a ‘no-blame approach’ and encourages an honest and open discussion to identify nonadherence and the reasons for nonadherence. Adherence support should be tailored to the needs of the individual addressing perceptions (e.g. necessity beliefs and concerns) as well as practicalities (e.g. capacity and resources). This can be approached in a three stage process: 1) communicating a common-sense rationale for personal need that takes account of the patient’s perceptions of the illness and symptoms expectations and experiences 2) eliciting and addressing specific concerns and 3) making the treatment as convenient and as easy to use a possible. Interventions attempting to improve adherence by applying these approaches have had encouraging results,. Nonadherence remains a fault-line in clinical practice. Consideration of patients’ perceptions of treatment necessity and concerns in prescribing and treatment review is essential to support informed choice and optimal adherence to appropriately prescribed treatment.