Healthcare providers’ adherence to breast cancer guidelines in Europe: a systematic literature review

We designed the search strategy and conducted the electronic searches, for both systematic reviews and primary studies, in MEDLINE (accessed through PubMed) and Embase (accessed through Ovid), from inception to May 2019. The strategy searched for terms related to adherence, clinical guidelines, and breast cancer. Search strategies are outlined in Additional file 2. Reference sections from retrieved systematic reviews were used to identify additional primary studies.

We sought for primary studies conducted in European countries. The addressed population was healthcare providers of breast cancer care (including primary care, oncology, radiotherapy, or others). Selected studies should measure the adherence of healthcare providers’ indications to breast cancer CGs, with any method (quantitative or qualitative) or source of data (self-reported, medical records’ assessment, interviews) or explore the barriers to adherence from healthcare providers’ perspective.

The extension of assessment could be only one recommendation or a complete guideline. We sought for studies reflecting usual care conditions. We excluded studies assessing patients’ adherence to treatment or indications, clinical trials, or studies to implement programmes to improve CGs adherence. One author (IR) screened titles and abstracts to select potentially relevant references to be evaluated on full text. Then, two authors (AVM, ENDG) independently assessed whether these studies met the eligibility criteria. Discrepancies were resolved through discussion. We managed references with Endnote version X7 software (Thomson Reuters, New York, USA).

We used tabular formats to extract main study characteristics (e.g. country, publication year, objective, year of study data, name of the guideline, guideline scope, adherence definition, number of patients, and patients’ characteristics). Primary outcomes were the proportion of patients receiving breast cancer adherent care according to CGs recommendations, by treatment modality, and the factors that impact on provider’s adherence.

We applied risk of bias assessment tools based on the study design: the AXIS tool for cross-sectional studies [16], the Quality Assessment Tool for Before-After (Pre-Post) Studies with No Control Group for non-controlled before-after studies [17], the Newcastle–Ottawa Quality Assessment Scale for prospective and retrospective cohort studies [18], and the Critical Appraisal Skills Programme (CASP) checklist for the qualitative study [19]. We followed each tools’ recommendations for the risk of bias assessment. The AXIS tool [16] assigns 0 to 10 scores and has three categories: low (1–4), moderate (5–7), and high (8–10). The Quality Assessment Tool for Before-After (Pre-Post) Studies With No Control Group [17] includes 12 criteria, and three final categories: good, fair, or poor. The Newcastle–Ottawa Quality Assessment Scale [18] uses scores from 0 (the highest risk of bias) to 9 (the lowest risk of bias). And, the Critical Appraisal Skills Programme (CASP) [19] includes ten domains, with three possible answers: “yes”, “can't tell”, and “no”.

Six authors (AVM, IR, ENDG, JP; MR, YS) were involved in the data extraction and risk of bias assessment. One author extracted data, and then two authors (YS, ENDG) contrasted the accuracy of data against full texts, cross-checked the risk of bias assessment, and completed missing information. Disagreements were solved by discussion or with the help of a third author (IR).

We considered inappropriate pooling quantitative findings due to the high variability in adherence definitions and healthcare processes assessed. Instead, we summarised results in a narrative synthesis, for which we defined and categorised breast cancer care as follows:

We grouped findings based on similarities of treatment modality. To summarise the proportion of adherence, we calculated descriptive statistics (the median, Interquartile range (IQR) (including the median), and range). Data analysis included estimating if possible the overall adherence when reported disaggregated (e.g. as subgroups); selecting the least restrictive definition when several definitions were reported; and selecting the most recent publication or the one with the longest period for studies with more than one publication with a similar objective. To explore barriers to adherence, we summarised the main findings from quantitative studies reporting factors that were significantly associated with non-adherence, and narratively integrated these findings with the qualitative findings. We classified them as internal or external (related to healthcare providers or not). We explored sources of heterogeneity by the source of data (i.e. self-reported or medical records based), treatment modality, and subtype of tumour. We used Microsoft Excel for data analysis. For data reporting, we present tabular summaries and graphical representations.

Our search yielded 10,558 references. After removing duplicates, 8137 references were screened based on titles and abstracts. Of these, 102 references were selected for full-text appraisal. We finally included 57 references representing 41 primary studies. Three studies were reported in more than one publication: BRENDA I [12, 13, 20‐28], BRENDA II [29‐31], and OncoDoc2 [32‐34]. (Fig. 1). Excluded studies alongside their exclusion rationale are available in Additional file 3.

Included studies were published between 1997 [35] and 2019 [36, 37]. These were conducted in eight European countries, with two countries leading in frequency: The Netherlands (n = 13, 31.7%), and Italy (n = 8, 19.5%). Most followed a retrospective cohort design (n = 19; 46.3%), the remaining included cross-sectional studies (n = 13; 31.7%); non-controlled before-after studies (n = 5; 12.2%); prospective cohorts (n = 4; 4.9%); a case study (n = 1; 2.4%), and a qualitative study (n = 1; 2.4%).

Two cross-sectional studies were based on a national audit [38, 39]. Still, they reported findings for two different subpopulations. Similarly, eleven retrospective studies used data from the National Cancer Registry of the Netherlands but analysed different regions or periods [40‐50].

Most studies (n = 36, 87.8%) assessed adherence based on medical records (from hospital databases, specific registers, or national cancer registries), and their sample sizes ranged from 131 [51] to 104,201 records [44]. The remaining five studies (12.2%) were based on self-reported data, through surveys or interviews to healthcare providers [52‐56], and their samples sizes ranged from 10 [56] to 202 participants [53]. Adherence to CGs recommendations was the primary outcome in 38 studies (92.6%), while three focused on factors that influence adherence [32, 47, 56]. Sixteen studies (39.0%) measured adherence to CGs for more than one process of care (Table 1, Additional file 4).

Thirty-one studies (75.6%) reported how researchers measured the adherence to CGs. Most of these studies (22/31, 71%) assessed adherence contrasting healthcare providers’ indications for breast cancer patients against a selection of CGs recommendations. Six studies (6/31, 19.4%) addressed adherence as a process indicator integrated in the quality assurance programme of their institution [57‐62]. In three studies (3/31, 9.7%), a “tumour board”, or a multidisciplinary team of physicians, was involved in the clinical pathway of treatment decision [29, 31, 32, 63]. The definitions of adherence were variable across studies. They included one or more of the following components: non-adherence definition, classification for non-adherent treatment, threshold criterion, reference to the currency of guideline, or to the relevance of patients’ profile (Additional file 5).

We assessed the risk of bias for 40 studies; the case study [32‐34] was not considered as suitable for this assessment. Most were at low (25/40, 62.5%) or moderate risk of bias (14/40, 35.0%). Only one study (1/40, 6%) was considered to be at high risk of bias, mainly due to concerns regarding selection bias in data analyses [64] (Table 1, Additional file 6).

Main findings reported by treatment modality, and clinical guidelines are available in Table 2 and Fig. 2.

Sixteen studies described and analysed potential barriers for adherence to CGs, and factors associated with non-adherent indications. These barriers were categorised as internal and external factors [6]; the latter included patient-related factors and structural/organisational barriers (Table 3, Fig. 3).

Internal factors, represented by healthcare provider-related factors for non-adherence to CGs, include their perceptions, preferences, knowledge, and attitudes regarding CGs recommendations. A qualitative study [56] explored the barriers for the implementation of CGs for preventive treatment for women at increased risk of breast cancer. According to providers’ perceptions, the reasons for non-adherence were the perceived lack of benefit of the interventions, being poorly informed and finding difficulties interpreting recommendations [56]. Another retrospective study reported that non-adherent indications in surgery reflected providers’ preferences for using specific techniques [45].

In other cases, these non-adherent indications would be intentional and conscious healthcare providers' decisions, as described in a case study [34] conducted within an [“optimal”] setting, where a CGs-based clinical decision support system (OncoDoc2) was routinely applied. They found that all non-compliant decisions concerned mostly a group of patients and decisions (i.e. elderly patients in pre-surgery decisions, patients with micro-invasive tumour in re-excision decisions, and patients with positive hormone receptors and HER2+ in adjuvant decisions). These discordances were found mainly in areas where scientific evidence is lacking, and the non-adherence behaviour, in this case, was actually intentional and conscious [34] (Table 3, Fig. 3).

External factors include the patient-related factors and structural factors. The patients’ age was the patient-related factor that appeared consistently associated with non-adherent treatment [20, 21, 30, 31, 33, 34, 40, 45, 48, 50, 63, 64, 68, 72]. In comparison with younger patients, older women were less likely to receive CGs concordant surgery, CT and RT, but were more likely to receive guideline-concordant ET. The intrinsic tumour characteristics were also associated with non-adherent behaviours, such as the gene expression profile [47] having triple-negative breast cancer [12, 68] or being diagnosed with higher tumour stages [20, 63]. Non-adherence was also associated with comorbidities [20, 27, 29, 32, 40, 64] [e.g. in patients with cardiovascular diseases, it was more frequent to withhold trastuzumab [40], or in patients with cognitive impairment it was more frequent the delay starting chemotherapy [29]]. Other patient-related factors included the quality of life (QoL) and low socioeconomic status (SES) [31]. Poor QoL was associated with non-adherence to ET [31]. If the QoL was good, older age was not related to deviation [30]. Low SES patients were more likely to be undertreated with CT than patients with higher SES; however, an association with ET was not found [29, 44]. Previous treatment also influenced non-adherence, e.g. RT was associated with an increased follow-up consultation [42], the absence of prior axillary surgery, with non-adherence to adjuvant decisions [33], and the type of CT received [37] with non-indicating therapies to prevent side effects.

The structural factors represent the environmental or organisational characteristics of the healthcare system were associated with non-adherent behaviour. One study reported that non-compliant decisions were mainly “choices of multidisciplinary staff meetings” [34]. Other factors associated with adherence to CGs recommendations were the academic activity in the organisation [38], geographical location [63], and the volume of departments [72] (Table 3, Fig. 3).

We synthesised the literature exploring the healthcare providers’ adherence to breast cancer CGs published in the last 20 years across eight European countries. Most followed a retrospective cohort design and were considered to have a low or moderate risk of bias. Adherence for overall breast cancer care process (from diagnosis to follow-up) ranged from 54 to 69%, for the overall treatment process (including surgery, CT, ET, and RT) the median adherence was 57.5% (IQR 38.8–67.3%), while for systemic therapy it was 76% (IQR 68–77%). The median adherence values for the processes assessed individually were higher, ranging from 74% (IQR 10–80%) for the follow-up to 90% (IQR 87–92.5%) for ET. Factors that potentially impact on healthcare providers’ adherence were internal: their perceptions, preferences, lack of knowledge, or intentional decisions; and external: the patient-related and structural factors. The most consistent factor for non-adherence was the age of patients.

Previous systematic reviews addressing adherence of healthcare providers to CGs recommendations but out of the scope of breast cancer care [76‐79], found similar results to our review, these studies reported a wide range of adherence values, depending on the specific criteria evaluated or the treatment modality. From 8.2 to 65.3%, for a set of CGs [76]; from 5 to 95% for different recommendations in the management of acute coronary syndromes [77]; and from 0.3 to 100% in antibiotic prophylaxis [78]. One study found 50% of adherence for cancer-associated thrombosis [79], lower than the median estimated for the overall treatment process in breast cancer care (58%).

Previous systematic reviews have also explored the main barriers to healthcare providers’ adherence in other healthcare areas [77, 78, 81, 82]. The factors they reported are consistent with our findings: the internal factors related to health professionals' limited skills or competences to use a therapy, and their poor knowledge, a low level of awareness, familiarity or confidence with the commonly used therapy (prior to CGs-recommended therapy), or a perceived lack of benefit of the "new" therapy [80‐82]. They also pointed out as internal factors those related to intentional non-adherence [77]. Those highlighted as external factors include patient-related factors, such as comorbidities and patient-level barriers [78, 80], as well as structural factors, such as organisational characteristics like being a teaching hospital [78].

Other additional factors identified in previous studies included poor organisational- or institutional-level support, inadequate peer support among health professionals, the complex nature of some therapies or guidelines [80], the expectation that compliance is mandatory [4], or the patients' preferences or demands [4, 81].

Our study has several strengths. The broad eligibility criteria we applied helped to capture the complete picture of the healthcare providers’ adherence to breast cancer CGs in Europe. We included studies with any design reporting the measure of adherence even for single CGs recommendations. The detailed data extraction process helped us to provide a synthesis of adherence proportions by treatment modality, although definitions or selected recommendations were variable. We could identify different reporting elements across definitions and methods to measure adherence. For factors associated with healthcare providers’ adherence, we adapted a conventional classification, which highlights factors to be considered by breast cancer CGs developers and other relevant stakeholders.

Our study also has some limitations. Included studies were conducted in eight European countries, with more than half of them coming from two countries (The Netherlands and Italy). Therefore, we should be cautious when interpreting results and cannot extrapolate our findings to other European countries. Since we included studies of patient cohorts from the 1990s, when the implementation of guidelines was starting, we might be underestimating CGs adherence values. We identified only one qualitative study exploring factors for non-adherence to breast cancer CGs, possibly explained by the type of databases we selected. Another important limitation is that guideline non-adherence data did not differentiate deliberate guideline deviations from unjustified practice variation. This was not possible to explore as most of the included studies did not specify the reasons for non-adherence. Other factors, like changes between versions or discrepancies between several national and international guidelines, could have potentially influenced guideline deviations. However, this analysis was not feasible.

This systematic review summarises for first-time healthcare providers’ adherence to breast cancer guidelines in Europe. Even though observed median proportions seem to be acceptable for most specific treatments, there is still room for improvement for healthcare providers’ adherence, especially for supportive measures during therapy as well as during follow-up. Advances in breast cancer screening and treatment have reduced breast cancer mortality across the age spectrum in the past decade [73, 74]. However, we identified that the most consistent external factor associated with non-adherence to CGs was the older age of patients. Breast CGs might not adequately address this subpopulation, or they may represent a population where the evidence to develop breast cancer guidelines is scarce.

We found high variability in the methods used to measure, define, and report adherence. Future studies should provide the rationale to define adherence with enough transparency and should always consider the strength of a recommendation in their selection. In usual care, healthcare providers may not be aware of the standards required for CGs and may be tempted to select low-quality recommendations [74]. Hence, to facilitate the use of evidence supporting healthcare recommendations, guideline developers should use rigorously developed presentation formats (e.g. decisions aids, clinical decision trees). In agreement with other authors [75, 76], we believe it is necessary to provide a selection of relevant, reliable, and reproducible definitions for unwarranted clinical variation in healthcare [76], in this case for breast cancer care. Use of robust methods to measure adherence, avoiding the selection or performance bias, with appropriate blinding of assessment, will help to evaluate this process in a more reliable and reproducible way. The development of more qualitative research to capture breast cancer healthcare providers' perspective should be fostered. Furthermore, health care providers should register the reasons for non-adherence to facilitate that real-world data inform guideline updating.