Aromatase inhibitors and the risk of colorectal cancer in postmenopausal women with breast cancer

Abstract

Background

A large trial of postmenopausal women with breast cancer reported an imbalance in colorectal cancer events with aromatase inhibitors (AIs), compared with tamoxifen in the adjuvant setting. This unexpected signal was observed within 3 years of randomization. To date, no observational studies have examined this important safety question in the natural setting of clinical practice. Thus, the objective of this study was to determine whether AIs, when compared with tamoxifen, are associated with increased risk of colorectal cancer in postmenopausal women with breast cancer.

Patients and methods

Using the UK Clinical Practice Research Datalink, we identified women, at least 55 years of age, with breast cancer newly treated with either AIs or tamoxifen between 1 January 1996 and 30 September 2015, with follow-up until 30 September 2016. High-dimensional propensity score-adjusted Cox proportional hazards models were used to estimate hazard ratios (HRs) with 95% confidence intervals (CIs) of incident colorectal cancer associated with AIs when compared with tamoxifen overall, by cumulative duration of use, and time since initiation. All exposures were lagged by 1 year for latency considerations.

Results

A total of 9701 and 8893 patients initiated AIs and tamoxifen as first-line hormonal therapy (median follow-up of 2.4 and 2.9 years, respectively). Compared with tamoxifen, AIs were not associated with an increased risk of colorectal cancer (incidence rates of 150 per 100 000 person-years in both groups; adjusted HR: 0.90, 95% CI: 0.53–1.52). Similarly, there was no evidence of an association with cumulative duration of use (P-heterogeneity = 0.54), and time since initiation (P-heterogeneity = 0.66).

Conclusions

In this first population-based study, the use of AIs was not associated with an increased risk of colorectal cancer. These findings should provide reassurance to the concerned stakeholders.

Key Message

Randomized controlled trials have reported an imbalance in colorectal cancer events when comparing aromatase inhibitors with tamoxifen in postmenopausal women with breast cancer. In this first observational study using the UK Clinical Practice Research Datalink, we did not find an association between aromatase inhibitors and colorectal cancer in this patient population.

Introduction

Aromatase inhibitors (AIs) are increasingly used upfront or in combination with tamoxifen in treatment of postmenopausal women with hormone receptor-positive breast cancer [1]. A patient-level meta-analysis has indicated that AIs have may be superior to tamoxifen in reducing breast cancer–related mortality and all-cause mortality [2]. In addition, these drugs have been associated with several adverse events including fractures, osteoporosis, arthralgia, and myalgia [3, 4].

There are uncertainties regarding the association between AIs and colorectal cancer. In the long-term follow-up of Arimidex, Tamoxifen, Alone or in Combination (ATAC) randomized controlled trial (RCT), there were imbalances of colorectal cancer events with anastrozole versus tamoxifen after a median follow-up of 33 months (24/3092 versus 19/3094 event, respectively) which persisted after a median follow-up of 120 months (66/3092 versus 44/3094 events, respectively) [5–7]. An increased risk of colorectal cancer was also observed after a median of 7.2 years of follow-up in the IBIS-II DCIS RCT, which randomized postmenopausal women with ductal carcinoma insitu to anastrozole versus tamoxifen (10/1449 versus 5/1489 events, respectively) [8]. However, in IBIS-II trial, which consisted of postmenopausal women at high risk of breast cancer, anastrozole was associated with a statistically significant 72% decreased risk of colorectal cancer compared with placebo after a median of 5 years of follow-up [9].

An association between AIs use and colorectal cancer incidence is biologically plausible as estrogens may exert a protective effect in colorectal carcinogenesis via the β-estrogen receptor [10–12]. To our knowledge, no population-based studies have assessed this important safety concern. Thus, the objective of this study was to determine whether the use of AIs, compared with tamoxifen, is associated with increased risk of colorectal cancer in postmenopausal women with breast cancer in the real-world setting.

Methods

Data source

This study was conducted using the UK Clinical Practice Research Datalink (CPRD). The CPRD is a large primary care database comprised of more than 14 million individuals enrolled across 700 general practices [13]. The CPRD contains data on anthropometric measures, such as body mass index (BMI), clinical diagnoses and procedures, lifestyle variables (such as smoking status), and prescriptions written by general practitioners (GPs). The patient population in CPRD has been shown to be broadly representative of the UK population in terms of age, sex, ethnicity, and BMI [13]. Medical diagnoses and procedures are captured using Read code classification and prescriptions are based on UK Prescription Pricing Authority Dictionary [13]. Diagnoses recorded in the CPRD have high validity, with a median positive predictive value of 88.6% for all diagnoses [14].

The study protocol (16_227RA) was approved by the Independent Scientific Advisory Committee of the CPRD and by the Research Ethics Board of the Jewish General Hospital, Montreal, Canada.

Study population

We identified a cohort of women, at least 55 years of age, newly diagnosed with breast cancer between 1 January 1996 [the year the first third-generation AI (letrozole) was approved in UK] [15] and 30 September 2015. All patients were required to have at least 1 year of medical history in the CPRD before first diagnosis of breast cancer. Patients with evidence of metastatic disease at any time before and including the date of breast cancer diagnosis were excluded. We also excluded patients with history of any endocrine therapy (AIs or tamoxifen) before breast cancer diagnosis, as these likely represented patients with prevalent disease.

Using the cohort defined above, we identified all patients newly treated with either a third-generation AI (anastrozole, letrozole, exemestane) or tamoxifen; cohort entry was defined as the date of first prescription of either endocrine therapy. For each exposure cohort, we excluded patients with previous diagnoses of colorectal cancer, familial adenomatous polyposis, cystic fibrosis, or Lynch syndrome at any time before cohort entry. Patients initiating treatment on AIs or tamoxifen were required to have at least 1-year follow-up after cohort entry. Exposure was lagged by 1 year to account for delays in diagnosis of colorectal cancer, allow a biologically plausible latency period, and minimize biases related to detection bias and reverse causality (where exposure might be initiated or terminated at early signs or symptoms of the outcome).

All patients meeting the study inclusion criteria were followed starting 1 year after initiation of the endocrine therapy (i.e. person-time at risk) until an incident diagnosis of colorectal cancer (defined in detail below), switch to a different endocrine therapy (e.g. tamoxifen to AI, and vice versa), death from any cause, end of registration with the general practice, or end of the study period (30 September 2016), whichever occurred first.

Outcome ascertainment

We identified all incident colorectal cancer events occurring during the follow-up period based on Read codes (supplementary Table S1, available at Annals of Oncology online). Diagnoses of colorectal cancer in the CPRD have been previously validated, with a positivity predictive value of 98%, sensitivity of 92%, and specificity of 98%, when compared with diagnoses in the UK National Cancer Data Repository [16]. Moreover, colorectal cancer diagnoses recorded by primary care physicians were shown to have a high positive predictive value and low false-negative rate [17].

High-dimensional propensity scores

To minimize confounding, models were adjusted on high-dimensional propensity scores (HDPSs), a seven-step algorithm that empirically selects variables from different data dimensions along with predefined variables [18]. For this study, the HDPS empirically selected 500 variables measured at cohort entry from five data dimensions, including drug prescriptions, diagnoses, procedures, disease history and referrals, and administrative information [18]. The predefined variables included year of cohort entry (categorical), age (modeled flexibly as restricted cubic spline with three interior knots), alcohol-related disorders, smoking status (ever, never, unknown), BMI (< 25, 25–30, ≥30.0 kg/m², unknown), and use of statins, aspirin, and other non-steroidal anti-inflammatory drugs (NSAIDs). The propensity score also included diagnoses of inflammatory bowel disease (ulcerative colitis and Crohn’s disease), history of polyps, type 2 diabetes, cholecystectomy, and previous diagnoses of other cancers (other than non-melanoma skin cancer), all measured at any time before cohort entry. Patients in non-overlapping regions of the propensity score distributions were trimmed from the analysis.

Statistical analysis

Descriptive statistics were used to summarize the characteristics of the cohorts. Crude incidence rates of colorectal cancer and 95% confidence intervals (CIs) based on the Poisson distribution were calculated for each exposure group. Cox proportional hazards models were used to estimate hazard ratios (HRs) and 95% CIs of incident colorectal cancer associated with use of AIs, compared with use of tamoxifen. All models were adjusted for HDPS modeled as a continuous variable using a restricted cubic spline with three interior knots to account for a possible non-linear relationship between the propensity score and the outcome. This was considered the primary analysis.

Secondary and sensitivity analyses

We conducted three secondary analyses by cumulative duration and time since initiation of AI and incidence of colorectal cancer, and by drug molecule (described in supplementary material, available at Annals of Oncology online). In addition, we conducted four sensitivity analyses to assess the robustness of the findings (described in supplementary material, available at Annals of Oncology online). All analyses were conducted with SAS version 9.4 (SAS Institute, Cary, NC).

Results

A total of 9701 patients were newly treated with AIs and 8893 patients newly treated with tamoxifen met the inclusion criteria (supplementary Figure S1, available at Annals of Oncology online). AI users generated 29 935 person-years of follow-up (median 2.4 years), and 46 incident colorectal cancer events [incidence rate: 150 (95% CI: 110–210) per 100 000 person-years]. Tamoxifen users generated 37 401 person-years of follow (median 2.9 years), and 57 incident colorectal cancer events [150 (95% CI: 120–200) per 100 000 person-years]. Among AI users, 5408 (55.8%) patients initiated treatment on anastrozole, 4248 (43.8%) on letrozole, and 45 (0.5%) patients on exemestane.

Table 1 presents baseline characteristics of AI and tamoxifen users. Compared with tamoxifen users, AI users were older, more likely to have alcohol-related disorders, had higher BMIs, and were more likely to have ever smoked. They were also more likely to have used prescription drugs, and had a higher prevalence of polyps and type 2 diabetes.

Table 1.

Baseline demographic and clinical characteristics of the cohort stratified by patients initiating treatment of aromatase inhibitors and tamoxifen

Characteristic	Aromatase inhibitors	Tamoxifen
Total	9701 (52.2)	8893 (47.8)
Age, years (mean, SD)	70.8 (9.8)	68.6 (9.5)
Alcohol-related disorders, n (%)	739 (7.6)	425 (4.8)
Smoking status, n (%)
Ever	3706 (38.2)	3072 (34.5)
Never	5864 (60.5)	5302 (59.6)
Unknown	131 (1.4)	519 (5.8)
Body mass index, n (%)
<25 kg/m2	2947 (30.4)	3046 (34.3)
25–30 kg/m2	3156 (32.5)	2748 (30.9)
≥30.0 kg/m2	2839 (29.3)	1942 (21.8)
Unknown	759 (7.8)	1157(13.0)
Statins, n (%)	3559 (36.7)	1723 (19.4)
Aspirin, n (%)	3004 (31.0)	1912 (21.5)
Non-steroidal anti-inflammatory drugs, n (%)	5441 (56.1)	4829 (54.3)
Inflammatory bowel disease, n (%)	111 (1.1)	86 (1.0)
Polyps, n (%)	252 (2.6)	141 (1.6)
Type 2 diabetes, n (%)	1122 (11.6)	681 (7.7)
Cholecystectomy, n (%)	777 (8.0)	549 (6.2)
Other cancers, n (%)	1777 (18.3)	1417 (15.9)

SD, standard deviation.

In primary analysis, the use of AIs was not associated with an overall increased risk of colorectal cancer, when compared with use of tamoxifen (HDPS-adjusted HR: 0.90, 95% CI: 0.53–1.52) (Table 2). In secondary analyses, there was no apparent association in terms of cumulative duration of use, time since initiation, and by AIs drug molecule (Table 2).

Table 2.

Crude and adjusted HRs for the association between the use of AI and the risk of colorectal cancer among patients with breast cancer

Exposure	Events	Person-years	Incidence rate (95% CI)a	Crude HR	Adjusted HR (95% CI)b
Tamoxifen	57	37 401	150 (120–200)	1.00 [Reference]	1.00 [Reference]
Aromatase inhibitors	46	29 935	150 (110–210)	1.13 (0.75–1.69)	0.90 (0.53–1.52)
Duration of aromatase inhibitor use, years
≤2	16	11 203	140 (80–230)	1.03 (0.53–1.99)	0.78 (0.37–1.67)
2.1–4	20	11 067	180 (110–280)	1.60 (0.87–2.97)	1.28 (0.63–2.60)
>4	10	7665	130 (60–240)	0.84 (0.41–1.72)	0.72 (0.33–1.57)
					P-heterogeneity: 0.54
Time since first aromatase inhibitor use, years
≤2	13	8447	150 (80–260)	1.04 (0.47–2.27)	0.78 (0.33–1.85)
2.1–4	19	11 070	170 (100–270)	1.70 (0.81–3.57)	1.32 (0.58–3.01)
>4	14	10 418	130 (70–230)	0.87 (0.46–1.67)	0.75 (0.37–1.55)
					P-heterogeneity: 0.66
Aromatase inhibitors type
Anastrozole	31	19 964	160 (110–220)	1.13 (0.72–1.76)	0.93 (0.54–1.60)
Other aromatase inhibitors	15	9971	150 (80–250)	1.14 (0.63–2.05)	0.82 (0.40–1.67)

^aPer 100 000 person-years.

^bAdjusted for high-dimensional propensity score as a restricted cubic spline with three-interior knots.

HR, hazard ratio; CI, confidence interval.

The results of the sensitivity analyses are summarized in supplementary Figure S2 (and supplementary Tables S2–S6, available at Annals of Oncology online). Overall, these results were consistent with the primary analysis. The use of time-varying exposure definition to account for switching between AIs and tamoxifen led to consistent results. Similarly, lengthening the exposure lag period to 2 and 3 years led to HRs close to the null value (adjusted HR: 1.05, 95% CI: 0.57–1.93 and adjusted HR: 0.81, 95% CI: 0.41–1.62, respectively). Consistent findings were also observed when accounting for competing risks (subdistribution HR: 0.84, 95% CI: 0.51–1.39), and stratifying the exposure cohorts by colorectal cancer screening age groups (<60 years, adjusted HR: 0.80, 95% CI: 0.19–3.33 versus ≥ 60 years, adjusted HR: 0.93, 95% CI: 0.54–1.60).

Discussion

To our knowledge, this is the first population-based study to examine the association between AIs and the risk of colorectal cancer in setting of clinical practice. In this study, when compared with tamoxifen, the use of AIs was not associated with an increased risk of colorectal cancer overall, by measures of duration, or by AI drug molecule. These findings remained consistent across sensitivity analyses.

RCTs have produced mixed findings regarding the association between AIs and risk of colorectal cancer. In ATAC trial, postmenopausal women with breast cancer were randomized to 5 years of treatment with anastrozole (n = 3125), tamoxifen (n = 3116), or combination of these drugs (discontinued after first interim analysis) [5, 6]. After a median of 33 months, there were imbalances in colorectal cancer events between the anastrozole and tamoxifen arms [24/3092 (0.8%) versus 19/3094 (0.6%)] [7]. These imbalances persisted after a median follow-up of 120 months, with 66/3092 (2.1%) versus 44/3094 (1.4%) events in the anastrozole and tamoxifen arms, respectively [5]. Similarly, in Tamoxifen Exemestane Adjuvant Multinational trial, there were imbalances reported in colorectal cancer events at median 9.8 years when comparing 5 years of upfront adjuvant treatment with exemestane to sequential treatment with tamoxifen and exemestane [52/3075 (2%) versus 40/3045 (1%), respectively] [19]. The IBIS-II DCIS RCT, which consisted of postmenopausal women with diagnosis of ductal carcinoma insitu, also showed imbalances in colorectal cancer events when comparing anastrozole to tamoxifen after a median of 7.2 years (anastrozole: 10/1449, tamoxifen: 5/1489, odds ratio: 2.06, 95% CI: 0.64–7.71) [8]. In contrast, a reversed effect was observed in the IBIS-II trial, which consisted of postmenopausal women at high risk of breast cancer. After a median follow-up of 5 years, anastrozole was associated with a statistically significant decreased risk of colorectal cancer, although based on few events (anastrozole: 3/1920 versus placebo: 11/1944, risk ratio: 0.28, 95% CI: 0.08–0.99) [9]. Finally, two meta-analyses of RCTs did not find an association between AIs and a composite end point of all second primary cancers in the adjuvant and extended adjuvant setting, although the authors did not specifically examine colorectal cancer [3, 20]. None of the aforementioned RCTs were designed or powered to assess colorectal cancer as an outcome.

The current knowledge regarding effect of estrogen on colorectal cancer is mixed. Estrogen may have a protective role in colorectal carcinogenesis via activation of pro-apoptotic pathways through the β-estrogen receptor [10–12]. Consistent with this hypothesis, age-standardized incidence of colorectal cancer was shown to be lower in females compared with males in the developed countries [21]. Some observational studies have reported that exogenous estrogen may decrease colorectal cancer incidence in postmenopausal women [22, 23]. However, these findings conflict with Women’s Health Initiative (WHI) RCT, where a decreased risk of colorectal cancer was observed for exogenous progesterone and estrogen and not exogenous estrogen alone in postmenopausal women with prior hysterectomy [24]. At a median follow-up of 13 years, there was a non-significant increased risk of colorectal cancer (HR: 1.13, 95% CI: 0.85–1.51) and colorectal cancer mortality (HR: 1.46, 95% CI: 0.86–2.46) when comparing exogenous estrogen only to placebo [24, 25]. In contrast to the WHI RCT, a meta-analysis has shown that majority of observational studies have indicated a protective effect of hormone replacement therapy on colorectal cancer incidence in postmenopausal women [25]. However, most of these studies could not disentangle exogenous progesterone from estrogen.

Overall, the findings of our study suggest that AIs are not associated with an increased risk of colorectal cancer, when compared with tamoxifen. To our knowledge, one observational study reported an increased risk of colorectal cancer with tamoxifen, compared with no hormone therapy, when follow-up was restricted to at least 5 years post-breast cancer diagnosis (risk ratio: 1.47, 95% CI: 1.00–2.15) [26]. However, no imbalances have been observed in placebo-controlled tamoxifen RCTs [27, 28]. While it is not possible to rule out an increased risk of colorectal cancer with tamoxifen, our findings should be interpreted from context of clinical equipoise as tamoxifen remains the clinically meaningful alternative in management of estrogen-receptor positive breast cancer [29].

Our study has several strengths. To our knowledge, this is the first observational study to examine the association between AIs and colorectal cancer incidence. We used a lagged exposure definition to exclude prevalent colorectal cancer events and allow a minimum latency period. The new user, active comparator design reduced confounding at the design stage while eliminating prevalent-user bias [30, 31]. Colorectal cancer diagnosis has been shown to have high validity with almost complete concordance with the UK National Cancer Repository (98% positive predictive value) [14, 16, 17, 32]. Thus, we do not anticipate significant outcome misclassification. Finally, our results remained consistent across secondary and sensitivity analyses that considered different sources of bias.

Our study has some limitations. The study population was limited to 9701 AIs and 8893 tamoxifen patients, with median follow-ups of 2.4 and 2.9 years after 1-year treatment lag, respectively. Nevertheless, our study had twice the patient population of the ATAC trial, and some patients were followed for up to 14 years. Prescriptions in the CPRD represent those issued by GPs, and thus misclassification of exposure is possible if patients did not comply with the treatment regimen, or if they were treated by specialists. However, approximately 77% of postmenopausal women newly diagnosed with breast cancer initiated treatment on AI or tamoxifen. This is consistent with the prevalence of hormone-receptor positive breast cancer reported in other studies [33, 34]. Further, GPs in the UK are extensively involved in management and treatment of patients with breast cancer including administration of endocrine therapy to postmenopausal women with hormone-receptor positive breast cancer [35]. Given the observational nature of the study, there remains the possibility of residual confounding. To minimize this potential bias, our models were adjusted for HDPS, which included over 500 potential confounders. The CPRD does not capture breast cancer grade, stage, and other tumor-related characteristics. However, these variables have not been previously associated with incidence of colorectal cancer. Finally, our study was not powered to examine the association with specific AIs, although more than half of AI users initiated treatment on anastrozole.

Conclusions

In this population-based study, the use of AIs was not associated with increased risk of colorectal cancer. While additional studies with longer follow-up are needed to confirm our findings, the results of this first population-based study should provide some reassurance with respect to the colorectal cancer safety of AIs to women with breast cancer.