In recent years, the use of oral anticancer treatment, particularly preventive treatments such as AI therapy, has been expanding and it is likely to further increase in the future . Patient adherence is essential for the success of disease management , but should not be taken for granted, as growing evidence on non-adherence indicates. Drawing conclusions on actual adherence rates in breast cancer patients is limited by considerable discordance in results due to methodological inconsistency across studies and by the various measurement methods employed.
Herein, we report on a comparison of four methodological approaches for adherence assessment in patients receiving AI treatment in order to investigate consistency of results determined by self-report, physician rating, refill records and the measurement of substance plasma concentrations. To the best of our knowledge, this is the first study directly comparing these four measurement approaches in this patient population.
The methods implemented in our study revealed adherence rates between 78% and 98%, reflecting well-known inconsistencies for estimations of adherence to endocrine agents. These numbers are in the higher bracket of the 20%-100% range reported in the literature [6, 7]. Chlebowski and colleagues (2006) quote adherence rates of 72%-77% across adjuvant clinical breast cancer trials and of 54%-80% in breast cancer prevention trials, regardless of method or type of endocrine agent. However, very recent studies implementing single methods of adherence assessment report highly contradictory results [24, 25]. This might be attributed to the heterogeneity of study designs. Therefore, because of the remarkable heterogeneity of reported adherence rates and of methods of determination used, a comparison of the different methods appears to be of particular relevamce.
Our study results suggest an at most modest concordance between adherence measures tested in a sample of breast cancer patients receiving endocrine treatment. Given correlations of 0.2-0.4, the comparability of results using these study methods is rather limited.
This observation is consistent with the literature on adherence in chronic disease. In a study by Dunbar-Jacobs and colleagues  the comparison of self-report (7-day recall), MEMS (Medication Event Monitoring System) and pill count to assess adherence to lipid-lowering medications revealed a lack of correlation between measures despite sample homogeneity (regarding observation period and treatment). Additionally, the association between self-report and MEMS in a sample of rheumatoid arthritis patients approached zero .
Evidence on the subject of comparability of methodological approaches for measurement of adherence to AIs is scarce. Ziller and colleagues [16, 27] reported a significant gap between self-reports and prescription refills for adherence to tamoxifen and anastrozole. They found non-adherence rates of 20%-31% on the basis of prescription refill records and perfect self-reported adherence (100%). In an earlier study by Waterhouse and colleagues  in 24 patients receiving tamoxifen, 98% were classified as adherent using self-reports and 92% using pill counts, but only 69% by means of MEMS. The self-report questionnaire and pill count taken together identified about 17% of patients with poor adherence, whereas MEMS found that 75% of the cohort was less than 80% adherent to the tamoxifen regimen.
Our findings suggest a strong dependence of the estimated adherence rates on the specific measurement method used. As assumed in previous studies [3, 16] methodological shortcomings assigned to each measurement approach may serve as an explanation for these discrepancies. The following limitations need to be considered when interpreting our findings:
Two major attributes of the determination of substance plasma concentrations are its high level of accuracy and objectiveness [28, 29]. Since concentrations of one single anastrozole dose can be detected several days after cessation (mean anastrozole elimination half-life 45.4 – 50 h ) the mere verification of the substance is not indicative of regular medication intake. On the other hand, inter-patient variations in AI elimination can cause highly variable plasma concentrations despite regular drug intake. Farmer et al.  suggested that this method simply shows whether the patient recently took a dose of the drug, but does not quantify the manner in which the patient took the drug or any fluctuation in medication intake. Concurrently, interindividual differences in drug pharmacokinetics challenge an appropriate definition of a threshold for plasma concentrations indicating (non-) adherence. Urquhart  claims biological variability in drug response to be the main obstacle to using biological markers for adherence.
Moreover, a lack of evidence on those plasma level relationship limits the interpretation of individual values. Beer and colleagues  suggested that those patients with values below or close to the 10th percentile of the study population are likely not adhering to the regime prescribed, regardless of time since intake. Nonetheless, they claim concentrations at the lower end of the distribution to be indicative, but not confirmative, of non-adherence and recommend repeated assessments as a more qualified strategy for assessing adherence. Therefore, classifying only those patients with plasma concentrations below the quantification limit as non-adherent we may have underestimated non-adherence behavior in this patient population. Moreover, this approach does not consider interindividual differences for metabolization. In order to be able to use the determination of AI plasma concentrations to evaluate adherence, clear thresholds for adequate plasma concentrations are needed. On the other hand by using such a conservative approach, we clearly reduced the site of overestimating adherence.
Results based on prescription refills strongly depend on the threshold for the medication-possession ratio defined as indicating adherence. The determination of an adequate medication-possession ratio is recommended inconsistently in the available literature, namely between 80% and 95%. The level of medication intake required to achieve a therapeutic efficacy of endocrine agents is still unclear and challenges the definition of this threshold. In contrast to Ziller and colleagues  and Waterhouse and colleagues , we defined a value of ≥ 90% as satisfactory following Dunbar-Jacob and Sereika , who claimed the majority of patients to adhere above the 90% level. This percentage appeared to be persistent across measures and over time and was thus regarded as adequate for this study. However, the key obstacle presented by this measure is the underlying premise that refill equals medication intake. This premise is invalid if patients refill their medication, particularly in the case of low out-of-pocket costs, but are reluctant regarding intake. Moreover, patients who never filled any prescription or developed metastatic disease are lost to this approach.
Lacking a validated self-report questionnaire specific for breast cancer patients receiving AIs we used the SMAQ, which considers the impact of iatrogenic harm. After adaptation for use in our patient sample, a German version of the questionnaire was deemed appropriate for patients receiving any type of endocrine treatment. Yet, we are aware of the shortcoming of lacking validation. In this context, the inconsistent use of a great variety of self-report measures further limits data comparability [31–34].
Proxy-rated adherence by the treating physician can be susceptible to bias in terms of overestimation. Treating clinicians seem to be at risk for inaccurately estimating adherence, either because they are unaware of their patient‘s poor intake behavior or simply underestimate it [35–37]. The fact that the second highest adherence rates in the present study were derived from the proxy ratings partly support this assumption.
Besides the limitations distinctive to each measurement method, there are some more general methodological aspects of this study which might contribute to the observed difference between methods. First, the selection of non-adherence cut-offs is somewhat arbitrary for all methods which impacts on prevalence rates. Nonetheless, the subject of defining satisfactory cut-off levels is a common problem in adherence research and has been heterogeneously discussed in the literature. Second, differing time-frames of adherence measurement methods could have impacted on the comparability of the approaches. However, we considered this subject by varying time-frames and could illustrate differing time-frames to have little effect on methodological comparability. Finally, acknowledging the shortcomings of a cross-sectional design for the investigation of an issue that might be influenced by time (of medication intake), our results elucidate the complexity of adherence measurement.