Cognitive effects of adjuvant endocrine therapy in older women treated for early-stage breast cancer: a 1-year longitudinal study

Patients were recruited from medical and radiation oncology clinics at two BC centers in large university-affiliated teaching hospitals in Toronto, Canada, from July 2015 to March 2016. The study was approved by the research ethics boards at both hospitals (Sunnybrook Health Sciences Centre Research Ethics Board ID 130-2015; Mount Sinai Hospital Research Ethics Board ID 15-0249-E). Written informed consent was obtained from all individual participants included in the study.

Potential participants were identified according to a consecutive sampling procedure. Daily clinic lists were reviewed to identify newly diagnosed patients at each site, and patient eligibility was assessed by review of medical records and discussion with treating oncologists. Women were eligible if they were postmenopausal, age 60+, chemotherapy-naïve, and prescribed TAM or an AI for the adjuvant treatment of early-stage HR+ BC. HR positivity was defined as > 1% positive cell staining for estrogen and/or progesterone receptors, and “early-stage” disease was defined as T1-2, N0-1, M0 using the TNM staging system [14]. Patients were proficient in English with at least a grade five education, capable of reading large print and hearing normal conversation (with or without corrective aids), and able to provide informed consent.

To prevent potential confounding effects on cognitive performance, patients were excluded if they met any of the following criteria: (1) treatment with chemotherapy (for previous or current disease); (2) previous history of BC or other cancers, except non-melanoma skin cancers or cancers treated for cure over 5 years ago; (3) previous history of ET; (4) ET initiation > 14 days prior to baseline study assessment; (5) use of estrogen or hormone replacement therapy (ERT/HRT) within 4 weeks of baseline assessment; (6) diagnosis of major psychiatric disorder other than depression within the last 10 years; (7) acute illness or delirium; (8) history of neurologic disorder (including dementia, stroke, Parkinson’s disease, etc.); or (9) signs of dementia on the Memory Impairment Screen (i.e., scores < 4) [15].

A prospective repeated-measures design was used to control for pre-treatment cognitive performance. Participants completed a 2-hour baseline assessment of neuropsychological function and mood after surgery but within 14 days of starting ET (T1), and again after 1 year of treatment (T2). The battery of valid and reliable tests was individually administered in a private room at each assessment. Tests were administered by a trained psychometrist (EAU) with the oversight of a neuropsychologist (MCT). ET adherence was monitored monthly throughout the follow-up period. A full description of the measures used is provided below.

The Shapiro-Wilk test was used to assess whether verbal memory change scores were normally distributed, and a paired t test (or Wilcoxin signed-rank tests when data were not normally distributed) was used to assess our primary outcome measure of change in verbal memory z-score from T1 to T2. An effect size (Cohen’s d) was calculated to describe the magnitude of change in this domain. If significant change was observed from T1 to T2, multivariable regression analyses were conducted to explore the extent to which change in performance (i.e., T2 z-score – T1 z-score) was associated with relevant independent variables. Based on our sample size, a maximum of four independent variables could be included in the regression models without overfitting [24]. Age at T1, years of education, duration of retest interval, and change in measures of mood/psychological distress were selected a priori based on their potential to influence change in neuropsychological performance. Multicollinearity between independent variables was examined based on inspection of variance inflation factors, and residual plots were examined to assess departure from the normality assumption. To minimize bias, these independent variables were entered simultaneously into the regression model and kept in the model regardless of whether or not they were found to be significantly associated with the dependent variable [24]. Because various aspects of mood/psychological distress were measured using the POMS-sf, which may differ in their effects on cognition, we ran three separate regression models which differed only in the POMS-sf change score that was entered as the measure of mood/psychological distress: one that used the POMS-sf TMD score (an aggregate measure of global psychological distress which encompasses depression, anxiety, fatigue, and confusion), one that used the depression subscale score, and one that used the anxiety subscale score. Secondary exploratory analyses examined change from T1 to T2 in the other four cognitive domain z-scores, using the same analyses described above for the verbal memory domain. Analyses were conducted using SAS version 9.2, and a two-sided alpha level of 0.05 was used to assess statistical significance.

Details of participant recruitment and attrition are shown in Fig. 1. Of the 626 patients aged 60+ that were screened, 82 (13%) met study criteria and were eligible to participate. Forty-six (56%) of these patients consented to the study and completed T1 assessments. There was no significant age difference between eligible patients who declined participation in the study (mean age = 70.81 years) and those who consented (mean age = 68.07) (t = 1.66, p = 0.11). No further demographic or medical information was collected for patients who declined study participation. Despite attempts to arrange T1 assessments prior to ET initiation, 25 (54%) participants had initiated ET prior to T1. Among these participants, the average number of days on ET at T1 was 7.84 (SD = 3.95, range: 1–14). T1 summary z-scores on our primary outcome measure of verbal memory did not differ significantly between participants who had (mean = 0.07, SD = 0.94) and had not (mean = 0.32, SD = 0.86) initiated ET at T1 (t = 0.95, p = 0.34). Forty-two (91%) participants returned to complete T2 assessments, which took place an average of 382.64 days after T1 assessments (SD = 36.10; range 318–475). Reasons for declining retest were not related to cognitive functioning or risk factors for cognitive decline (Fig. 1). Due to the small number of participants who declined retest (n = 4), no statistical comparisons were made between those who did and did not return for follow-up.

Participants underwent menopause at an average age of 51.02 years (SD = 4.19). There was no change in medical comorbidities or in the use of psychotropic medications (antidepressants, benzodiazepines, or non-benzodiazepine sedatives, anxiolytics, or hypnotics) within the study sample from T1 to T2. At T1, 37 (88%) participants had been prescribed an AI (23 letrozole and 14 anastrozole) and five (12%) had been prescribed TAM. Three participants underwent a switch in their prescribed ET agent over the course of the year (one from letrozole to TAM, one from letrozole to exemestane, and one from anastrozole to letrozole). In each case, the reason for the switch was musculoskeletal side effects. With the threshold for ET adherence set as an intake of ≥ 80% of prescribed doses, all 42 participants who returned for their T2 assessments were adherent throughout the study period. Additional demographic and clinical characteristics of the study sample are presented in Table 1.

Mean TMD, depression, and anxiety scores are presented in Table 1. There was no statistically significant change in these scores from T1 to T2 (Table 1).

A significant change in the verbal memory domain, t(41), p < 0.01 was observed, with participants performing more poorly at T2 than at T1 (Table 2). The effect size of this difference was small to moderate in magnitude (Cohen’s d = − 0.40). Regression analyses revealed that change in verbal memory was not significantly associated with age, education, length of retest interval, or change in psychological distress, regardless of whether POMS-sf TMD score, depression subscale score, or anxiety subscale score was entered into the model (Supplementary Table 1). There was no evidence of multicollinearity between independent variables in any of the regression analyses (all variance inflation factor values < 1.5), and based on the residual analysis, we observed no departure from the normality assumption.

There were no significant changes from T1 to T2 found in our secondary analyses of the other four domain z-scores (Table 2).

A limitation of the current study is its lack of a comparison group. The ideal comparison group would consist of women with HR+ BC who did not undergo ET; however, given that ET is standard of care for women with this disease, recruiting such a group is not feasible. An alternative option could have been to include a group of women without BC that underwent the same testing procedures as our ET group, which would have allowed us to control for measurement error, regression towards the mean, and practice effects using reliable change methods [38]. Given that practice-related improvements in performance are common with repeated administration of neuropsychological tests, it is possible that even greater decline might have been observed had we been able to control for normal practice effects in our analysis. Second, our sample size was relatively small and this precluded us from examining potential risk factors for ET-related changes in verbal memory (e.g., body mass index, medical comorbidities, physical activity, smoking history, alcohol consumption), or from comparing the cognitive effects of the different ETs used by participants. Given evidence that differences may exist both between (i.e., TAM vs. AI) and within (i.e., non-steroidal vs. steroidal AIs) ET classes [5, 10, 25, 39, 40], and since current treatment guidelines permit a choice between ETs, the question of whether ETs differ in their cognitive effects is clinically relevant and should be a focus of future research. Finally, because our sample consisted of well-educated women who were predominantly Caucasian, our findings may not be generalizable to the overall population of older women with BC and should be replicated in a more representative sample.