Identifying determinants of adherence to adjuvant endocrine therapy following breast cancer: A systematic review of reviews

Abstract Background In oestrogen‐receptor positive breast cancer, daily oral adjuvant endocrine therapy (ET) for at least 5 years significantly reduces risks of recurrence and breast cancer‐specific mortality. However, many women are poorly adherent to ET. Development of effective adherence support requires comprehensive understanding of influences on adherence. We undertook an umbrella review to identify determinants of ET adherence. Methods We searched PubMed, Embase, CINAHL, PsycINFO, Cochrane and PROSPERO (inception to 08/2022) to identify systematic reviews on factors influencing ET adherence. Abstracted determinants were mapped to the World Health Organization's dimensions of adherence. Reviews were quality appraised and overlap assessed. Results Of 5732 citations screened, 17 reviews were eligible (9 quantitative primary studies; 4 qualitative primary studies; 4 qualitative or quantitative studies) including 215 primary papers. All five WHO dimensions influenced ET non‐adherence: The most consistently identified non‐adherence determinants were patient‐related factors (e.g. lower perceived ET necessity, more treatment concerns, perceptions of ET ‘cons’ vs. ‘pros’). Healthcare system/healthcare professional‐related factors (e.g. perceived lower quality health professional interaction/relationship) were also important and, to a somewhat lesser extent, socio‐economic factors (e.g. lower levels of social/economic/material support). Evidence was more mixed for medication‐related and condition‐related factors, but several may be relevant (e.g. experiencing side‐effects, cost). Potentially modifiable factors are more influential than non‐modifiable/fixed factors (e.g. patient characteristics). Conclusions The evidence‐base on ET adherence determinants is extensive. Future empirical studies should focus on less well‐researched areas and settings. The determinants themselves are numerous and complex in indicating that adherence support should be multifaceted, addressing multiple determinants.

Results: Of 5732 citations screened, 17 reviews were eligible (9 quantitative primary studies; 4 qualitative primary studies; 4 qualitative or quantitative studies) including 215 primary papers.All five WHO dimensions influenced ET non-adherence: The most consistently identified non-adherence determinants were patient-related factors (e.g.lower perceived ET necessity, more treatment concerns, perceptions of ET 'cons' vs. 'pros').Healthcare system/healthcare professional-related factors (e.g.perceived lower quality health professional interaction/relationship) were also important and, to a somewhat lesser extent, socio-economic factors (e.g.lower levels of social/economic/material support).
Evidence was more mixed for medication-related and condition-related factors, but several may be relevant (e.g.experiencing side-effects, cost).Potentially modifiable factors are more influential than non-modifiable/fixed factors (e.g.patient characteristics).

| INTRODUCTION
Breast cancer is the most commonly-diagnosed cancer worldwide, with recent data suggesting there are over 2 million breast cancer cases diagnosed annually-equating to 11.7% of all cancer cases. 1 Although survival outcomes for breast cancer continue to improve, each year there are still around 685,000 deaths globally. 2While there are different subtypes of breast cancer, the most common type (~75%) is oestrogen receptor (ER) positive-meaning the growth of the cancer is dependent upon the presence of oestrogen. 3Women with ER positive breast cancer are usually recommended to take oral adjuvant endocrine therapy (AET) for a number of years following surgery and/or radiotherapy.][6] Indeed, extending the treatment duration of AET from 5 to 10 years, has been shown to further improve outcomes and reduce disease recurrence. 7,8Examples of AET include the aromatase inhibitors (e.g.letrozole, anastrozole or exemestane), which are generally used for postmenopausal women, and tamoxifen which is generally used for pre-menopausal women.
Despite the well-recognised benefits of AET, there is evidence to suggest that people do not take it every day, while some people stop taking the AET altogether.Indeed, current estimations suggest that between 30 and 60% of women take less AET than recommended, while between 30 and 70% of women prematurely stop their AET at the end of 5 years treatment. 9There is strong evidence that not-adhering to AET is associated with an increased risk of breast cancer reoccurrence and mortality. 10,11Adherence is a complex behaviour influenced by many different factors.In response, the World Health Organization developed a model outlining five dimensions of medication adherence, which highlighted this issue is not exclusively a patient-driven problem. 12Numerous studies have attempted to examine the broader factors why women do not adhere to AET.Indeed, several systematic reviews have been published on this topic that have synthesised and summarised this complex information.While systematic reviews are widely recognised as a robust way to appraise and synthesise available evidence, existing reviews on AET adherence are highly variable and have often focused exclusively on qualitative literature (e.g.Clancy et al. 13 ) or quantitative literature (e.g.Cahir et al. 14 ).Furthermore, some systematic reviews have focused on specific aspects of treatment adherence (e.g.only exploring adherence in the context of side effects from AET 15 ), making it challenging to understand the broader factors underpinning AET adherence.A broad and comprehensive review identifying and summarising the different AET adherence determinants is needed to understand how best to support women to continue to take their AET, as well as identifying evidence gaps and informing future research endeavours.In this paper, we undertake a 'review of reviews' to examine different determinants of adherence to AET following breast cancer.A review of reviews is an established and effective way to bring together and summarise a broad evidence base, and this approach has been used to report a number of complex topics relating to adherence (e.g. 16,17).
Outcomes: Determinants of adherence to AET reported at the systematic review level (e.g.personal, clinical, economic, psychosocial).Adherence could be measured objectively (e.g.prescription encashment/refill data) or self-reported through completion of questionnaires or lived experience.
Study design: Systematic reviews of (i) qualitative studies, (ii) quantitative studies or (iii) qualitative or quantitative studies.To meet the definition of a systematic review, four of the following five criteria had to be present: • Inclusion/exclusion criteria were reported; • The search strategy was reported; • The included studies were synthesised; • The quality of the included studies was assessed; and • Sufficient details about the individual studies were presented.

| Exclusion criteria
Systematic reviews investigating prevalence of (non-) adherence to AET or exploring the effectiveness of interventions to promote adherence to AET were excluded.Systematic reviews published as conference abstracts only were also excluded, as were those published in languages other than English.

| Search strategy
A search strategy was developed with support from an information specialist and executed in the following databases: PubMed, Embase, Cumulative Index to Nursing and Allied Health Literature (CINAHL), PsycINFO, Cochrane (Breast Cancer Group) and PROSPERO.The searches were carried out from database inception 3 August 2022.The search strategy is included in Supplementary File 2. Reference lists of included systematic reviews were further hand searched for relevant articles.

| Study selection, and data extraction
Titles and abstracts were screened by a single reviewer (CW).
The full text of all reviews identified for inclusion, together with any for which a decision on inclusion was not possible, were obtained for a more detailed examination.The full-text articles were then assessed by one reviewer (CW or AT) and checked by another (ZM or JSB).Discrepancies at the full-text screening stage were resolved through discussion.If agreement was not possible, a third reviewer was consulted (LS) who had consensus.If needed, authors of eligible reviews were contacted to confirm details relevant to eligibility; if no response was received within 2 weeks, a decision on inclusion was made based on the published information.Guided by the inclusion criteria, reasons for exclusion were recorded at the full text stage.
For each review, all data extraction was completed by one author (LMcG, LS, CC, SJS) and checked by a second (AT, EW, MW).Any disagreements were discussed with the wider group to reach consensus.Data extracted included bibliographic details and characteristics of the systematic review, methodology employed, primary studies included in each systematic review, synthesised determinants of adherence, conclusions.No data were abstracted from any of the original papers included in the eligible systematic reviews.

| Synthesis
Narrative synthesis was undertaken at the systematic review level (i.e. the synthesis focussed on the reported findings, as stated in the eligible reviews; the original primary studies included in the reviews were not examined or extracted).Factors reported to be associated with adherence (reviews including quantitative studies) or themes/constructs reported to influence adherence (reviews including qualitative studies) were mapped to the World Health Organization's (WHO) five dimensions of adherence, 12 and an updated conceptual model of these five dimensions. 18ach potential determinant was extracted and tabulated under the domains(s) it related to (namely: conditionrelated factors; medication-related factors; healthcare system/healthcare professional-related factors; patientrelated factors; socio-economic factors) and categorised as to whether the relevant systematic review stated that: it was seldom or never reported to influence adherence; the evidence for association was mixed; or it was associated with non-adherence.To aid interpretation, and as our focus was on influences on non-adherence, if a determinant was reported to be associated with (better/higher) adherence, we recorded the inverse as being associated with non-adherence (e.g. if early-stage disease was reported to be associated with better adherence, we recorded that later stage disease was associated with non-adherence).We did not seek to distinguish between different types of nonadherence (e.g.sub-optimal implementation/early discontinuation) because the reviews did not report this.

| Assessment of methodologic quality and study overlap
The quality of each systematic review was assessed by one author (LS, CC, SJS, LMcG) and checked by a second (EW, RN) using the JBI checklist for systematic reviews. 19Any disagreements were discussed with other AT who had consensus.Reviews were not excluded from the synthesis on the basis of low-quality assessment.The overlap of primary studies was reported in a citation matrix and the corrected coverage area (CCA) 20 was calculated for all reviews combined, and by study design(s) included in the reviews (i.e.quantitative only, qualitative only, qualitative or quantitative).The CCA is calculated using the following equation: CCA = N − r ∕ (rc − r), where N is the total number of primary publications from the included systematic reviews (including double counting); where r is the number of primary publications; and, c is the number of included systematic reviews. 20According the CCA, a value of 0-5 represents 'slight' overlap; 6-10 'moderate' overlap; 11-15 'high' overlap; and greater than 15, 'very high' overlap.

| RESULTS
After de-duplication, searches provided 5732 records for screening; 92 full text papers were assessed.Screening reference lists of eligible reviews provided a further 179 citations of which 18 were reviewed in full text.The study selection process and reasons for exclusions are shown in Figure 1.Overall, 17 systematic reviews, reporting unique 215 primary studies, were eligible and included in the narrative synthesis. 9,13-15,21-33
All reviews aimed to synthesise determinants of adherence to AET, although several focussed on particular groups of determinants, such as adverse effects, 15,22 psychosocial factors 30 or those which are potentially modifiable. 14,29All reviews limited eligibility to primary studies published in English.One review limited consideration to studies of women with postmenopausal breast cancer 21 ; 11 reviews focused only on women with breast cancer and excluded men (or populations who were predominately male) 9,14,15,[21][22][23][24]26,29,31,32 ; and some excluded studies of Reports not retrieved (n = 0) DCIS or people with Stage 4 cancer 14,15,22 to maintain a focus on AET.All reviews undertook a narrative synthesis with three of the four qualitative reviews using a thematic synthesis approach 13,15,31 and the fourth undertaking meta-synthesis based on grounded theory. 33Two reviews conducted a meta-analysis but could only do this for one 24 or three 14 potential modifiable adherence determinants.Nine reviews made some attempt to distinguish determinants of different types of non-adherence (e.g.missing doses of AET or stopping AET earlier than planned), but the terms used were not consistent across the reviews. 9,13,14,22,24,26,27,30,33Of these nine reviews, one review of qualitative studies organised their findings in relation to different forms of non-adherence, with different themes focusing on initiation, adherence and persistence, 33 while a review of quantitative studies conducted meta-analyses according to non-adherence or non-persistence. 13

| Review quality
Overall, the quality of the reviews was mixed (Supplementary File 4).Four 14,15,31,32 had ten 'yes' ratings out of a possible eleven; a further three 22,23,26 had nine and three more 13,29,33 had eight.One review 30 had five 'yes' ratings, while one review 25 had three 'yes' ratings.The reviews were most often rated 'unclear' in response to questions on whether: the likelihood of publication bias assessed; critical appraisal conducted by two or more reviewers independently; recommendations for policy and/ or practice supported by the reported data.

| Overlap between reviews
In total, the 17 systematic reviews included 215 unique primary papers in their syntheses.Almost half (n = 96; 46%) of primary studies were cited in only one review, 41 (19%) were cited in two, 28 were cited in 3 (13%) and the remaining 50 (23%) were cited in 4 or more reviews.Considering all included systematic reviews, the CCA was 0.09, representing slight overlap.For the reviews that included only quantitative primary papers, the CCA was 0.13; for the reviews that included both qualitative and quantitative, it was 0.12; and for the reviews that included only qualitative primary papers, it was 0.48; all of these are considered slight overlap (Supplementary File 4).
In terms of breast-cancer related characteristics, including markers of disease severity (e.g.larger tumour size, lymph node involvement) and history of previous cancer-directed treatment, the evidence regarding adherence was inconsistent: some reviews reported an association with non-adherence, 21,28,32 some reported mixed findings, 26,32 while others reported neutral findings. 26egarding patient-specific factors, reviews concluded that the presence of co-morbidities was either associated with non-adherence 21,25,28,32 or the findings were mixed. 26,27imilarly, there was inconsistent evidence on the associations with mental health conditions, receipt of psychological help, and use of psychotropic medication and non-adh erence. 22-24,27,29,30,32

| Medication-related factors
Fourteen reviews (seven quantitative, three qualitative and four mixed) synthesised evidence on AET-related factors (Table 2).In terms of medication regimen, five reviews concluded there was mixed evidence of an association between the patient experiencing polypharmacy and non-adherence. 9,14,23,26,28Switching type of AET was linked with non-adherence by three reviews 9,27,32 but a fourth review judged the evidence as mixed. 26Two reviews reached different conclusions about non-adherence and using tamoxifen versus aromatase inhibitors: Yussof et al. reported that using tamoxifen was associated with non-adherence (compared with aromatase inhibitors), 32 while Moon et al. reported the findings were mixed. 26ight reviews (including two of qualitative evidence) reported that experiencing treatment-or disease-related side-effects influenced non-adherence 14,15,23,25,27,28,32,33 ; three further reviews concluded evidence was mixed. 9,22,29inks between non-adherence and lack of preparedness for side-effects and poor side-effect management of coping strategies were each reported in two reviews. 13,15,30,33In terms of medication properties, four reviews observed that higher AET cost was related to non-adherence 23,27,28,33 ; while another review concluded the evidence on this was mixed. 26 3.6 | Healthcare system/healthcare professional-related factors Fourteen reviews (six quantitative, four qualitative, four quantitative or qualitative) reported on health professional or healthcare system factors 9,[13][14][15]23,[25][26][27][28][29][30][31][32][33] (Table 3). Nine of 0 reviews concluded that patient perception of lower quality interactions, relationships or communication with health professionals was related to non-adherence 13,15,23,[25][26][27][28]30,31 All eight reviews which commented on support from health professionals observed that lack of support was associated with non-adherence.13,15,23,26,[30][31][32][33] Non-adherence was also judged to be associated with: insufficient information provided from health professionals, or information that patients found hard to understand 13,15,23,26,30,31,33 ; not valuing or trusting the health professionals' opinions 26,29,33 ; and lack of involvement of a breast cancer specialist in follow-up care.14,15,23,26,32,33 Fewer reviews commented on health system-related factors: two concluded that irregular or inconsistent follow-up care was associated with nonadherence 23,32 and two reported more hospitalisations were linked to non-adherence.26,32

| Patient-related factors
Reviews included a wide range of patient-related factors; these were reported in 14 reviews (five quantitative, four qualitative, five quantitative or qualitative) 9,13,15,21,23, [25][26][27][28][29][30][31][32][33] (Table 4).6][27] Lack of knowledge or understanding about breast cancer or AET was linked with nonadherence in four reviews, 23,28,31,33 and belief that missing doses will not impact efficacy in three reviews. 23,31,33Not fearing cancer recurrence, or perceiving self as at low risk of recurrence was judged to be associated to nonadherence in five reviews 13,15,25,27,30,33 ; two further reviews considered evidence on this mixed. 26,29[29][30][31] Non-adherence was linked to forgetfulness in four reviews, 21,23,30,33 poor medication-taking routines in two reviews, 23,27 and practical problems with medication-taking in one review. 26s regards priorities, eight of nine reviews reported that non-adherence was associated with negative T A B L E 1 Condition-related factors reported at the systematic review level presented according to their association with AET adherence.

Condition-related factors a
Neutral or seldom Mixed

Associated with non-adherence
Disease control ---

Disease characteristics
Positive lymph nodes b The factors presented in italics were taken directly from the conceptual model of adherence, 18 which was used to guide the synthesis of the review of reviews.b The determinants presented under each factor are taken directly from the included systematic reviews and, as such, have a cancer specific context.
decisional balance (i.e.cons for taking AET outweighing pros). 13,15,23,25,26,30,31,33Three concluded that a desire to 'move on' from breast cancer, 'get back to normal' and/or discounting future benefits influenced nonadherence 23,25,31 ; two stated that fertility preservation influenced non-adherence 23,26 ; and one judged that the treatment interfering with lifestyle was important. 28n terms of personal and family/caregiver characteristics, younger age was linked to non-adherence in four of six reviews, 21,27,28,32 as was older age in two of four reviews. 28,32Being unmarried was reported to influence non-adherence in three reviews. 27,28,32

| Socio-economic factors
Evidence on socio-economic factors influencing adherence was reported in 11 reviews (four quantitative, three qualitative, four quantitative or qualitative) 13,15,23,[25][26][27][28][29][30]32,33 (Table 5). The mst extensively evaluated socio-environmental factor was lack of social, emotional or material support, which seven of eight reviews concluded there was associations with non-adherence.23,[25][26][27]30,32,33 Single reviews reported links between non-adherence and: having a family member with breast cancer 23 ; lack of partner support 13 ; T A B L E 2 Medication-related factors reported at the systematic review level presented according to their association with AET adherence.

Medication regimen
Polypharmacy The factors presented in italics were taken directly from the conceptual model of adherence, 18 which was used to guide the synthesis of the review of reviews.b The determinants presented under each factor are taken directly from the included systematic reviews and, as such, have a cancer specific context.
T A B L E 3 Healthcare system/healthcare professional-related factors reported at the systematic review level presented according to their association with AET adherence.

HCP characteristics
Perceived lower quality of HCP interaction/communication/ relationships b The factors presented in italics were taken directly from the conceptual model of adherence, 18 which was used to guide the synthesis of the review of reviews.b The determinants presented under each factor are taken directly from the included systematic reviews and, as such, have a cancer specific context.

T A B L E 4
Patient-related factors reported at the systematic review level presented according to their association with AET adherence.and not feeling an obligation to family members. 15One review commented on lifestyle factors, observing that smoking was related to non-adherence. 28In terms of economic factors, three reviews stated that there were relationships between lack of health insurance and nonadherence, 23,27,33 while single reviews reported links between non-adherence and higher education, 27 being in paid employment, 27 working in a medical-related The factors presented in italics were taken directly from the conceptual model of adherence, 18 which was used to guide the synthesis of the review of reviews.b The determinants presented under each factor are taken directly from the included systematic reviews and, as such, have a cancer specific context.profession, 23 having a burdensome work schedule 28 and lower financial status. 32

| DISCUSSION
The high incidence of ER + ve breast cancer coupled with the high prevalence of non-adherence (both suboptimal implementation and early discontinuation) means adherence to AET is an important public health problem.This umbrella review identifying determinants of adherence to AET included 17 systematic reviews and showed evidence for factors within all five of the WHO dimensions of medication adherence.The most consistently identified determinants were patient-related factors (e.g.cognitive and psychological, such as having a lower perceived necessity for AET).Healthcare system/healthcare professional-related factors (e.g.healthcare professional characteristics, such as having a perceived lower quality interaction or relationship with a healthcare professional) were also important and, to a somewhat lesser extent, socio-economic factors (e.g.social/environmental such as a patient having lower levels of social support).Evidence was more mixed as regards medication-related and condition-related factors.The primary conclusion from our synthesis is that nonadherence to AET in women with breast cancer is complex and affected by multiple determinants.This echoes conclusions from umbrella reviews on adherence to medications for cardiovascular conditions and diabetes, and for medications more generally. 17,34,35Modifiable factors appear to be much more important than non-modifiable factors.The former include some that are 'inherent' to the individual (such as how they weigh the pros and cons of taking the medication, or their self-efficacy to take it), and external factors including both the individual's 'personal' context (such as social, emotional or material support) and the wider healthcare/health system context (such as T A B L E 5 Socioeconomic factors reported at the systematic review level presented according to their association with AET adherence.

Socioeconomic factors a
Neutral The factors presented in italics were taken directly from the conceptual model of adherence, 18 which was used to guide the synthesis of the review of reviews. b The determinants presented under each factor are taken directly from the included systematic reviews and, as such, have a cancer specific context.
perceived quality of communication and relationships with healthcare professionals or lack of involvement of breast cancer specialists in the individual's follow-up).The implications of this are two-fold: first, that predicting whether any specific patient prescribed AET will face problems with adherence based on their sociodemographic characteristics will be challenging and, second, that adherence support needs to be multifaceted.It is clear that there is a significant body of literature on influences on adherence to endocrine therapy.The 17 included systematic reviews included 215 unique primary papers.In addition to these systematic reviews, there are several other reviews which did not meet our rigorous definition for a systematic review (see, e.g.Moore, 36 Hadji 37 ).This volume of primary and secondary data raises the question of whether more primary research is needed on this topic.However, it is worth noting that much of the evidence-base accrues from the USA where the complexity of the healthcare system may drive some of the findings of the systematic reviews and hence of our synthesis.An example is the higher cost of the medication which several reviews concluded was associated with non-adherence 23,27,28,33 -this may be less of an issue in a universal healthcare system.Further research on determinants is most warranted in healthcare settings or certain clinical contexts that have been less extensively investigated; for example, low and middle income countries or younger people using AET who wish to become pregnant.Of note, the POSITIVE (Pregnancy Outcome and Safety of Interrupting Therapy for Women with Endocrine Responsive Breast Cancer) clinical trial seeks to address the question if AET can be paused, while a person aims to get pregnant. 38ur synthesis also showed that some categories of potential adherence determinants have not yet been fully investigated.For example, tamoxifen is metabolised by various cytochrome P450, UDP-glucuronosyltransferase and sulfotransferase enzymes; all enzymes in the pathway are encoded by polymorphic genes. 39Enzymes in these and other pathways also metabolise aromatase inhibitors. 40Only two systematic reviews mentioned genetic factors in adherence, concluding that higher CYP2D6 activity was associated with non-adherence. 9,32his reflects the paucity of primary research on this topic.Better understanding of the role of these factors in influencing adherence-either directly or indirectly via other determinants (e.g.treatment adverse-effects)-would be worthwhile.Lifestyle factors were mentioned in one review, which concluded that smoking was associated with non-adherence. 28Smoking is associated with worse breast cancer survival 41 and the finding regarding adherence suggests an explanation for that association.Various other lifestyle factors (including body fat and physical inactivity) 42 have been associated with breast cancer survival and understanding whether those are also related to (non-)adherence would be of value.
Although individual reviews were generally appraised as being of reasonable quality, it was striking how little overlap there was in the primary studies that they included.This is demonstrated in the overall CCA of 0.09 and the fact that almost half of the primary studies were included in only one systematic review.While some systematic reviews had a more specific or narrow focus (e.g.treatment adverse-effects, or psychosocial factors) the lack of overlap was surprising.It may be that primary papers on this topic are difficult to identify systematically; a variety of terms for 'non-adherence' are used in studies and reviews and the studies themselves may be published in journals with a diverse range of disciplines.To advance understanding in this area, and to aid future syntheses of findings, it vital that future primary studies, and systematic reviews, adopt standardised terminology (e.g. the European consensus on taxonomy and terminology of adherence). 43t was striking that only two of the reviews which included quantitative studies undertook meta-analysis. 14,24 small number of reviews were reasonably explicit about how they had assessed the primary evidence (e.g.Moon et al. 26 ), but most simply provided a narrative synthesis of studies in which it was unclear how they had judged whether there was 'sufficient' evidence (e.g.measured in terms of number of studies, consistency of findings) to conclude any particular factor was associated with (non) adherence.This should be borne in mind when interpreting our synthesis.A further issue to consider is that it is highly likely that the 'same' determinant was assessed or interpreted in different ways in the primary studies included in the eligible reviews; we did not assess this as our synthesis was at the systematic review level.Thus, we concur with the comments from Leslie et al. 34 on adherence to cardiovascular medicines-that (AET) adherence research is highly heterogeneous and efforts to standardise it are needed to improve comparability and facilitate future efforts to more precisely synthesise findings.

| Implications
Increasing the effectiveness of adherence interventions may have a far greater impact on the health of the population than any improvement in specific medical treatments. 44Despite clinical trial data highlightingtwo decades ago-that non-adherence of AET was a significant problem, 45,46 how best to support women to adhere remains unclear.Guidance for development of complex interventions notes the importance of having a comprehensive understanding of determinants of the behaviour. 47We undertook this umbrella review to bring together the evidence on determinants to inform development of a complex intervention to support adherence to AET (see 48 for more information).
Early interventions in this area were largely based on 'education' and were ineffective. 49While four of the eligible reviews found that lack of understanding or knowledge about breast cancer or AET was associated with non-adherence, 23,28,31,33 it is generally accepted that improving knowledge is insufficient on its own to influence behaviour. 50Rather, as suggested by our synthesis, any intervention will need to be multifaceted seeking to address a wide variety of determinants.Our synthesis suggests that an intervention will need to pay particular attention to patient-related cognitive and psychological factors, including women's perceptions of their need to take the therapy (including their perceptions of their recurrence risk) and the potential benefits of doing so, as well as any negative beliefs or concerns.These factors should be acknowledged and addressed by any healthcare professional with responsibility for initiating and prescribing AET.Building self-efficacy while also addressing negative decisional balance (i.e.how women weigh the pros and cons of the therapy), and finding ways to positively frame adherence for women whose primary desire is to 'move on' from breast cancer will also be important.Other elements might include specific strategies to address forgetfulness and non-intentional adherence.Given the importance of the healthcare context and some social/environmental factors, an intervention may also need to include elements that would lead to improved patient-healthcare professional communication or relationships and more (or better quality) social and/or emotional support.

| Strengths & limitations
We followed systematic review methods, searching multiple databases and applying clear criteria for what we considered a systematic review.Nonetheless, there are some limitations.We did not independently double screen all citations, but at least two authors assessed each paper at the full text stage.We did not search databases of PhD theses or conduct forward citation searching from the eligible reviews.Moreover, we excluded reviews not published in English (this restriction was also applied in all of the eligible systematic reviews).It is therefore possible that we may have missed eligible reviews.We also acknowledge that the quality of included reviews was mixed.
We chose to use the term '(non-)adherence' in our synthesis to capture all the different ways patients may fail to take AET as prescribed and because most of the eligible systematic reviews lacked a clear definition of adherence.As others have noted, research on adherence to AETand medications in general-is bedevilled by inconsistent definitions and terminology. 17We were therefore unable to evaluate whether the determinants of different types of non-adherence (i.e.non-initiation, sub-optimal implementation and early discontinuation) differ.Some data indicate that women who have suboptimal implementation of AET are at increased risk of future early discontinuation 51 suggesting that the primary focus of adherence interventions should be to prevent suboptimal implementation.However, from the published data included in our synthesis is it impossible to clearly conclude which of the array of determinants identified here should be targeted to prevent this as opposed to early discontinuation.
While the WHO's five dimensions of adherence offer a useful approach to summarise correlates of non-adherence to date, the utility of this framework as an approach to guide intervention development has limitations 52 largely because the dimensions do not provide any understanding of the mechanisms through which each factor acts on adherence.Future work could therefore consider the review findings alongside the context of theoretical models, such as the Perceptions and Practicalities Approach (PaPA), 53 which help explain how these factors act on adherence through either a patients' motivation or ability.

| Conclusion
The evidence-base exploring and investigating AET determinants-at the condition-related, medication-related, healthcare system/healthcare professional-related, patient-related, socioeconomic-related levels-is extensive.Future research should focus on gaps and inconsistencies in this evidence base (e.g. economic and lifestyle factors), rather than repeat what is already known and well established.In terms of implications, adherence to AET is multifactorial and adherence support should acknowledge and address this complexity.

1
Study selection and exclusion according to the PRISMA statement.