Breast Cancer Research and Treatment

, Volume 138, Issue 1, pp 325–328

A meta-regression analysis of the available data on adherence to adjuvant hormonal therapy in breast cancer: summarizing the data for clinicians

Authors

    • Unité de Soutien Méthodologique, CHU La Réunion—CHFG
    • INSERM, U912 (SESSTIM)
    • Université Aix Marseille, IRD, UMR-S912
  • Cyril Ferdynus
    • Unité de Soutien Méthodologique, CHU La Réunion—CHFG
  • Roch Giorgi
    • INSERM, U912 (SESSTIM)
    • Université Aix Marseille, IRD, UMR-S912
    • Service de Santé Publique et d’Information Médicale, Hôpital de la Timone, Assistance Publique—Hôpitaux de Marseille
Letter to the Editor

DOI: 10.1007/s10549-013-2422-4

Cite this article as:
Huiart, L., Ferdynus, C. & Giorgi, R. Breast Cancer Res Treat (2013) 138: 325. doi:10.1007/s10549-013-2422-4

To the Editor,

A recent, systematic, qualitative, review by Murphy et al. [1] has provided a complete overview of the available data on persistence and adherence to hormonal therapy for breast cancer, as well as an extensive description of their determinants. This review demonstrates that treatment intakes are largely suboptimal. It also illustrates the heterogeneity of available studies, with treatment discontinuation rates ranging from 31 to 73 % over the treatment period. This wide range of values may limit the usefulness of raw data for clinicians. In order to provide measures clinicians might use effectively, we conducted a meta-regression analysis [2] that summarizes results on adherence and persistence to hormonal therapy, based on the data rigorously selected in the review by Murphy et al. This meta-regression analysis was conducted moreover to assess the different sources of variability in measurements of persistence.

We reviewed the 29 studies selected in the comprehensive review by Murphy et al. [1], and discounted three of these. We excluded two studies [3, 4] in which results were pooled for tamoxifen and aromatase inhibitor (AI) treatment, and one [5] in which follow-up was limited to the first 4 months of treatment. We reviewed all 26 studies to record possible sources of heterogeneity in treatment adherence. These sources include: study type (cross-sectional or cohort study), data source (medical records, self-reported data, population-based database, specific health-coverage database), age of patients, measure of outcome (use of self-reported data or measures of refill gaps), and type of analysis (taking into account longitudinal data or not).

Persistence to hormonal therapy and level of adherence over a 5-year period were studied separately, using a mixed model for longitudinal meta-analytic data [6]. We used a log(−log) transformation of the measure of outcome to normalize its distribution [7]. Each study was weighed by the inverse variance of the transformed outcome [2, 6].

All statistical analyses were performed using the following software programs: SAS 9.2 (SAS Institute Inc.), including the SAS procedure PROC MIXED, and R (R Development Core Team http://www.R-project.org), including the package metafor [8].

Among the 26 selected studies, 9 (34.6 %) reported data on both adherence and persistence [917], 9 (34.6 %) reported data only on adherence [1826], and 8 (30.8 %) on persistence alone [2734]. The meta-regression model estimated adherence to range from 79.6 % (95 % CI: 68.2–87.3) at 1 year to 68.3 % (95 % CI: 52.4–79.9) at 5 years. These numbers varied from 79.2 % (95 % CI: 67.5–87.0) to 64.6 % (95 % CI: 47.8–77.2) for tamoxifen treatment, and from 80.1 % (95 % CI: 68.8–87.7) to 71.8 % (95 % CI: 56.2–82.6) for AI therapy. Overall, non-persistence to hormonal therapy ranges from 13.6 % (95 % CI: 11.1–16.6) at 1 year to 40.9 % (95 % CI: 34.5–47.9) at 5 years. Among the possible sources of heterogeneity tested, data source was the only source of variation in persistence. Table 1 shows results for non-persistence data, adjusted for data source, and illustrates that persistence is significantly poorer for tamoxifen than for AI therapy in the long term. In the final model, data source explained 27.2 % of the variations observed between studies in the measures of persistence to tamoxifen and 68.1 % in the measures of persistence to AI.
Table 1

Non-persistence to tamoxifen and AI therapy estimated from a meta-regression model comprising data from 17 studies, stratified for time and treatment and adjusted for data source

Treatment discontinuation

Tamoxifen

AI

% (95 % CI)

% (95 % CI)

At 1 year

13.6 (11.4–16.2)

11.7 (9.4–14.4)

At 2 year

22.1 (18.8–25.9)

16.6 (13.4–20.3)

At 3 year

32.1 (27.5–37.2)

22.3 (18.2–27.1)

At 4 year

37.7 (32.5–43.4)

26.7 (21.9–32.2)

At 5 year

47.1 (41.1–53.5)

31.3 (25.9–37.5)

Our meta-regression analysis illustrates the importance of considering data source in the assessment of persistence and adherence, as it constitutes the major source of variation between studies. We provide summary estimates of persistence and adherence, based on the available studies found in the review by Murphy et al. [1] and taking into account source of variation. These 5-year estimates, along with the comprehensive work of Murphy et al. on which they are based, can offer clinicians information that will help them monitor patients’ treatment.

Reply to: A meta-regression analysis of the available data on adherence to adjuvant hormonal therapy in breast cancer: summarizing the data for clinicians

Caitlin C. Murphy, L. Kay Bartholomew, Melissa Y. Carpentier, Shirley M. Bluethmann and Sally W. Vernon

Center for Health Promotion and Prevention Research, The University of Texas School of Public Health, 7000 Fannin, Ste. 2556B, Houston, TX 77030, USA

*Corresponding author: Caitlin C. Murphy

E-mail: caitlin.c.murphy@uth.tmc.edu

To the Editor,

We commend Huiart and colleagues for using the data from our systematic review to investigate the different sources of variability in studies that measure adherence and persistence to adjuvant hormonal therapy. The wide variation in reported prevalence from observational studies underscores potential differences in data collection and quality and highlights the need for better summary effect size estimates. In a meta-regression of 26 of the 29 studies included in our review [1], Huiart et al. provided effect size estimates of adherence to (79.6 % at year 1 to 68.3 % at year 5) and discontinuation of (13.6 % at year 1 to 40.9 % at year 5) adjuvant hormonal therapy among breast cancer survivors in clinical practice. Combined with the findings of our review, Huiart’s meta-regression provides additional evidence to support the conclusion that adherence and persistence to adjuvant hormonal therapy is suboptimal. This is especially concerning in light of recent findings from the ATLAS Trial [2] demonstrating that continuing tamoxifen for 10 years, rather than stopping at 5 years, provides an additional overall and disease-free survival benefit. High non-adherence rates may reduce the benefit of treatment observed in clinical trials when findings from trials are translated into clinical practice.

Many studies have identified sociodemographic and treatment-related factors that may affect adherence, but, as we emphasize in our review, we still lack information on modifiable determinants of treatment adherence. Thus, a critical research need is to identify modifiable determinants that influence adherence to develop interventions to improve it. Limited research suggests that lower perceived benefit of treatment, perception of less than optimal role in the treatment-decision making process, and low social and/or material support are associated with non-adherence to tamoxifen [36]. Identifying and targeting such factors may ultimately promote treatment adherence, and thereby reduce breast cancer recurrence and mortality among the growing population of survivors.

Acknowledgments

We are indebted to Arianne Dorval for proofreading the English manuscript.

Conflict of interest

None.

Copyright information

© Springer Science+Business Media New York 2013