The occurrence of fractures after adjuvant treatment of breast cancer: a DBCG register study

Abstract

Background: Adjuvant treatment in breast cancer patients especially with aromatase inhibitors (AIs) has adverse effects on bone metabolism resulting in an increased occurrence of fractures. In order to demonstrate this occurrence, long-term follow-up studies are necessary. From several national registries in Denmark, it is possible to link data from different sources and analyze this issue.

Methods: A study cohort of 68,842 breast cancer patients prospectively diagnosed and registered in the Danish Breast Cancer Cooperative Group’s database during the period 1995–2012 formed the basis of the analysis. These data were matched with data on all types of fractures from the Danish National Patient Register and vital data from the Danish Civil Registration System.

Results: After data cleaning 66,502 patients were available for analysis and 16,360 of these had incurred 20,341 fractures with 13,182 patients having just one fracture. These fractures were distributed over 214 specific fracture sites. An extended multivariable Cox regression model revealed significant association between the occurrence of fractures and age, menopause, Charlson comorbidity index (CCI) and endocrine therapy such that late menopause and tamoxifen treatment were associated with a lower occurrence and AI treatment, age and CCI were associated with a higher occurrence of fractures.

Conclusion: Before advising adjuvant therapy with AIs fragile patients with chronic diseases should receive special attention in order to reduce the incidence of fractures in this vulnerable group of patients.

Introduction

Breast cancer is the most common cancer among Danish women with a steadily increasing incidence during the last 60 years. Breast cancer mortality has declined as a result of successive improvements in loco-regional and systemic treatment in the same period [1]. In estrogen receptor positive early breast cancer adjuvant endocrine therapy with tamoxifen (TAM), an aromatase inhibitor (AI), or a sequence of these two drugs for 5 to 10 years is considered standard therapy [2]. These drugs inhibit the effects of estrogens on breast tissue: TAM being a partial nonsteroidal estrogen agonist and AIs significantly lowering the plasma estradiol concentration [3,4]. Since the introduction of TAM and AIs the treatment duration with these drugs has increased in favor of their long-term antineoplastic effects, but at the expense of potential long-term side effects, especially on bone metabolism resulting in fractures, especially for the AIs [5] as TAM seems to lower the risk of fractures [6]. Irrespectively of bone mineral density, the addition of adjuvant bisphosphonate, 6-monthly zoledronic acid or daily oral clodronate, prolongs disease-free survival in postmenopausal patients [7].

The aim of this study was to investigate the occurrence of fractures in a cohort of Danish breast cancer patients and possible associations with some patient characteristics, comorbidity and adjuvant, antineoplastic treatment in particular endocrine therapy before the general introduction of adjuvant bisphosphonate treatment.

Methods

Study cohort

The study cohort originated from the Danish Breast Cancer Cooperative Group (DBCG) register in which data on patient and tumor characteristics are collected nationwide and prospectively along with data on given therapy and disease recurrence. Data from all patients registered during the period 1 January 1995 to 31 December 2012 constitute the basis for the present analysis. These data were supplemented and combined with data from the Danish National Patient Register (NPR) [8] and data from the Danish Civil Registration System (CPS) [9]. Since 1995 the NPR contains all somatic diagnoses from hospital admissions and outpatient contacts for each patient together with dates on admission and discharge. Vital and emigration status for each patient were collected from the CPS.

Fracture diagnoses

All fracture diagnoses registered in the NPR were collected for the study cohort. No attempt was made to elucidate the cause of the fractures, that is, whether caused by a high- or low-energy impact. As the same diagnoses in the NPR may be registered more than once only one fracture type was allowed for each patient, that is, even if the same patient had two femoral fractures registered three years apart only the first fracture counted, but the patient could also have had a distal radius fracture that was counted. If the patients had fractures after recurrence of the breast cancer, these fractures were not counted as they might have been pathological and fractures before the breast cancer diagnosis were also excluded.

Comorbidity

Information on comorbidity for each patient was extracted from the NPR and used to construct the Charlson Comorbidity Index (CCI) [10]. For the present study, the CCI was obtained using ICD-8 and ICD-10 diagnoses up to 10 years before the date for the breast cancer diagnosis.

Statistical analyses

The cumulative incidence of new fractures after the breast cancer diagnosis was estimated, considering recurrence and deaths as competing events. Patients were censored (a) at date of death, (b) at recurrence regardless of the location, (c) at last registered visit date, or (d) when followed for the whole observation period which ended 31 December 2016.

As many patients in the study cohort had several fractures and because patients with a prior fracture are at an increased risk of new fractures, the data were analyzed using regression models for recurrent events [11]. The extended Cox regression model development followed the suggestions described as the purposeful selection method along with the assessment of model adequacy [12].

The following covariates were analyzed: age and menopause (women >55 years at diagnosis and all patients treated with AI alone were defined as postmenopausal), CCI, chemotherapy and endocrine therapy. Antineoplastic treatments were coded as categorical covariates, for example, TAM: no/yes, without regard to the total treatment duration.

All analyses were done using R, version 3.4.1 [13] and the survival package [14]. The study was given file number: 2012-58-0004 by the Danish Data Protection Authority.

Results

The initial study cohort comprised 68,842 patients. These patients have been described meticulously in a previous publication [1]. The number of patients with missing follow-up dates was 2340 leaving 66,502 (96.6%) patients for the final analyses. Of these 47,302 (71.1%), patients were classified as postmenopausal. During the follow-up period, 27,561 (41.4%) patients died. Median follow-up time was 5.9 years with an interquartile range of 3.7–9.2 years and range 0.01–21.9 years. Table 1 shows information about the patients receiving adjuvant endocrine therapy or no endocrine therapy.

Table 1. Baseline information on endocrine treatment, age and CCI.

Treatment	None	Tamoxifen and aromatase inhibitor	Aromatase inhibitor	Tamoxifen	Other	Total
Number of patients	36,412	11,454	9875	8466	295	66,502
Age (mean; sd) years	64.0; 14.3	59.57; 10.3	65.6; 9.0	55.1; 12.3	53.0; 10.3
CCI
0	27,655	9666	7388	7235	271	52,215
1	4683	1196	1519	781	20	8199
2	2420	419	591	287	3	3720
3+	1654	173	377	163	1	2368

The NPR contained 56,759 fracture codes corresponding to the study cohort. After removal of patients with duplicated identification number, contact date and coded diagnosis 20,651 different fracture diagnoses remained. Further 310 patients had a fracture diagnosis before the breast cancer diagnosis leaving 20,341 fractures for analysis. These fractures were distributed over 214 specific fracture sites as coded in the NPR. Table 2 shows the number of fractures classified in 10 major groups. It is seen that most fractures were located at typical sites for osteoporotic fractures: femur, forearm, and upper arm. Recurrent fractures were frequent: 2541 patients had two fractures, 504 patients had three, 119 patients had four, 13 patients had five and four patients had six fractures.

Table 2. The distribution of fractures in 10 major groups.

Location	Face	Cervical spine	Thoracic spine	Lumbar spine	Upper arm	Forearm	Hand	Femur	Lower leg	Foot	Total
Num-ber of patients	557	99	625	1219	3059	5226	1825	4060	2212	1459	20341

During 564,977 person-years (p-y) of observation (range, 0.01 to 21.9 years per subject), 16,363 women experienced 20,341 different fractures (crude incidence, 36 fractures per 1000 p-y; 95% confidence interval, 34–39). In the premenopausal group, 4513 fractures were observed compared to 15,624 fractures in postmenopausal women.

Figure 1 shows the cumulative fracture incidence curves for patients receiving endocrine treatment or no endocrine treatment. Initially, the curves follow each other closely, but after about 5 years the curve for the AI-treated patients tends to increase above the non-treated patients. The curve for the TAM + AI-treated patients coincides with the curve for the TAM-treated patients until about 8 years when this curve starts to increase and crosses the curve for non-treated patients after about 12 years of observation. The curve for the TAM-treated patients remains well below the non-treated patients along with the curve for patients treated with other endocrine modalities for the whole observation period.

Figure 1. The cumulative fracture incidence curves for endocrine treated breast cancer patients.

The extended Cox model chosen was the Prentice, Williams and Peterson (PWP) model which analyzes ordered multiple events by stratification, based on the prior number of events during the follow-up period [11], because it was reasonable to assume that the occurrence of the first fracture increases the likelihood of a new fracture [15].

Table 3 shows separate analyses for pre- and postmenopausal women. In premenopausal women, only age and CCI were significantly associated with fracture occurrence compared to postmenopausal women where age, CCI and endocrine treatment were significantly associated with fracture occurrence.

Table 3. Summary of the multivariable PWP regression analyzes for pre- and postmenopausal women.

Covariate	Relative risk (95% confidence interval)	p value
Premenopausal women (n = 19,200)
Age (years)	1.015 (1.010–1.021)	<.0001
CCI
No comorbidity	1.0
Category 1	1.40 (1.25–1.56)	<.0001
Category 2	1.45 (1.18–1.77)	<.0001
Category 3+	1.93 (1.46–2.56)	<.0001
Endocrine therapy
None	1.0
TAM + AI	1.00 (0.93–1.08)	.99
TAM	0.95 (0.88–1.03)	.22
Other	1.06 (0.86–1.31)	.59
Postmenopausal women (n = 47,302)
Age (years)	1.031 (1.029–1.033)	<.0001
CCI
No comorbidity	1.0
Category 1	1.25 (1.20–1.31)	<.0001
Category 2	1.26 (1.18–1.34)	<.0001
Category 3+	1.49 (1.38–1.62)	<.0001
Endocrine therapy
None	1.0
AI	1.07 (1.02–1.12)	.0066
TAM + AI	0.89 (0.85–0.93)	<.0001
TAM	0.88 (0.83–0.93)	<.0001
Other	0.72 (0.54–0.96)	.025

Analyzing all patients together the following covariates remained in the final model: age, menopause, CCI, and endocrine therapy, (Table 4). Chemotherapy either coded as a multinomial or dichotomized covariate was not associated with the occurrence of fractures in any of the analyses. Statistically the proportional hazards assumption was violated for covariates age and endocrine therapy, specifically for patients treated with AIs and for patients in the combined group TAM + AI. This was also assumed from the cumulative incidence plot. However, when assessed graphically the violations were not severe. As a consequence of this and due to the very large sample size with many events, it was surmised that the nonproportionality would make no difference to the interpretation of the data [14].

Table 4. Summary of the final multivariable PWP regression analysis, all patients (n = 66,502).

Covariate	Relative risk (95% confidence interval)	p value
Age (years)	1.031 (1.028–1.032)	<.0001
Menopause
Premenopause	1.0
Postmenopause	0.82 (0.78–0.87)	<.0001
CCI
No comorbidity	1.0
Category 1	1.28 (1.23–1.34)	<.0001
Category 2	1.30 (1.22–1.37)	<.0001
Category 3+	1.55 (1.43–1.68)	<.0001
Endocrine therapy
None	1.0
AI	1.10 (1.05–1.15)	<.0001
TAM + AI	0.90 (0.87–0.94)	<.0001
TAM	0.88 (0.84–0.92)	<.0001
Other	0.84 (0.70–1.01)	.060

Discussion

In a nationwide population-based study, we have demonstrated a highly significant and clinically important association between adjuvant AI treatment and an increased risk of fractures. In contrast, the risk of fractures was decreased in patients treated with TAM alone as compared to endocrine untreated patients, suggesting a protective effect regarding fractures. These results are in accordance with other studies reporting on the occurrence of fractures in breast cancer patients treated with adjuvant TAM and AIs [5,6,16,17].

The course for the TAM + AI-treated patients is particularly interesting, because the patients in this group were first treated with TAM for an intended 5-year period followed by intended 5-year treatment with AI. This suggests an initial osteoprotective effect of TAM that is later dissolved due to the succeeding treatment with AI. The curve for the TAM-treated patients suggests that TAM provides a long-lasting effect on bone metabolism similar to the suggested long-lasting antineoplastic effect [18]. The curve for the last ‘other’ group similarly suggests an osteoprotective effect. However, the sample size for this group was small and so the curve shape should be interpreted with caution. This reservation is supported by the nonsignificant association in the multivariable model. Endocrine therapy was entered in the statistical models as a categorical covariate. However, as the endocrine therapy has been extended considerably during the study period it would be highly relevant to include the treatment duration as a covariate to clarify a possible dose response association.

The incidence of osteoporosis and the risk of later fractures increase with aging thus making age a known risk factor. In the present study, menopause at 55 years seemed to have a protective potential. This complies with early menopause being a known risk factor for osteoporosis, sustaining fragility fractures and increased mortality [19]. A late menopause postpones the period with accelerated bone loss associated with the menopause preceding the unavoidable bone loss associated with aging. An age of 55 years as a cut point for menopause was selected for comparability with other studies [20].

The study demonstrated a very strong association with increasing comorbidity as assessed by the CCI. This emphasizes the importance of taking into consideration other chronic diseases as for example diabetes mellitus [21] besides the breast cancer when an adjuvant treatment is advised. As an aid, an online service for health care professionals for calculating a fracture risk score, FRAX exists [22]. However, only some of the chronic disorders incorporated in CCI are included in FRAX such as diseases causing secondary osteoporosis.

Nevertheless, our study combined data from three registers and the available data were not intended for analyzing the occurrence of fractures. Other limitations include lack of reliable information on possibly important covariates like falling tendency, height, weight, smoking habits, alcohol use, osteoporosis, drug use, vitamin D status and bone mineral density. Other important confounders may not have been identified, measured and registered and may thus have had an unaccountable impact on the results.

It was not possible to clarify the causes of each fracture due to the large number and some surely have been caused by factors, for example, high-energy impact or focal bone pathology unrelated to the main question in this study. Fractures that occurred early in the study period were counted. Some of these were probably also unrelated to the main question in the study. On the other hand, it is well known that many fractures in the spine are not diagnosed [23], so the number of fractures may actually have been larger. Fractures in the forearm, femur, upper arm, lower leg, and hand were the most frequent in the study. These fractures are likely to cause pain and disability necessitating professional health care. For this reason, they are probably not underdiagnosed to the same extent as vertebral fractures. Counting just one different fracture location in each patient may also have underestimated the true number of fractures in the study cohort.

The concomitant use of antiosteoporotic treatment in the study period may have had an impact on the number of registered fractures possibly lowering the occurrence. Even so, it was already known during the study period that AIs had unwanted side effects [24] and TAM had beneficial effects on bone metabolism [25]. As a consequence, patients planed for AI treatment are recommended a bone density measurement as a routine. In light of such knowledge, one would expect the AI-treated patients to be treated with anti-osteoporotic drugs with a higher probability than TAM-treated patients and this would tend to reduce the occurrence of fractures in the AI-treated patients. However, judged by the study outcome this does not seem to be the case thus favoring our interpretation of the results.

In conclusion, our study supports previous studies that treatment with AIs in postmenopausal women is accompanied by an increased occurrence of fractures and that TAM has an osteoprotective effect resulting in a decrease in fracture occurrence in postmenopausal women. When advising adjuvant therapy with AIs especially fragile patients with chronic diseases should receive special attention including regular bone density measurements in order to reduce the occurrence of fractures in this particularly vulnerable group of patients.

Disclosure statement

The authors have no conflicts of interest to report. The authors alone are completely responsible for the content, analysis, and writing of the paper.