Anticoagulants and Breast Cancer Survival: A Nationwide Cohort Study

It is well known that venous thromboembolism (VTE) is more common in patients with advanced cancer (1), including breast cancer, and that VTE is associated with a worse prognosis (2). In addition to their anticlotting properties in vivo, anticoagulants exhibit direct antitumor properties in breast cancer cells in vitro (3, 4).

In recent experimental studies, the expression of protease activated receptors (PAR) 1 and 2 has been associated with breast cancer histology and tumor size (5). PAR1 signaling is driven by thrombin; which is not only a crucial element in the coagulation cascade, but it also promotes breast cancer cell growth and invasion (6, 7). The PAR2 [coagulation factor II (thrombin) receptor-like 1] pathway, which is dependent on tissue factor, is another important element in the coagulation cascade and has been crucial in evoking angiogenesis during in vivo breast cancer experiments (8, 9), possibly involving insulin-like growth factor (10). Thus, it is tempting to speculate that anticoagulants may limit breast cancer growth, with this effect being mediated through the PAR signaling pathways. And, indeed, direct thrombin inhibitors have been shown to reduce breast cancer metastasis (3, 4, 11). In addition, studies suggest that warfarin exerts antiadhesion properties in breast cancer cells; thus, warfarin has been further postulated to have direct antimetastatic properties, especially when combined with cimetidine (12). Furthermore, heparins have been reported to delay tumor growth in breast cancer independent of their anticoagulant properties (13). In particular, low-molecular weight heparin (LMWH) administration is associated with improved survival in patients with T3, T4, and M1 breast cancer (14) as classified in the international tumor–node–metastasis (TNM) classification of malignant tumors (TNM-classification by Union of International Cancer Control. Available at: https://www.uicc.org/resources/tnm); however, the benefit for M1 patients has been challenged (15).

Few observational studies examine the association between breast cancer–related mortality and anticoagulant use. To the authors' knowledge, only one previous study has been published on this topic, and it was limited to warfarin use (16). That study found no risk association for prediagnostic warfarin use, but an increased risk of breast cancer–related death was associated with postdiagnostic warfarin use in subjects who had not previously received the drug [HR 1.36; 95% confidence interval (CI), 1.12–1.64]. Notably, however, although the cohort included 16,523 patients with breast cancer, only 400 participants (2.4%) had used warfarin. Because experimental studies with a variety of different anticoagulant drugs show promising results, and because observational studies are limited, we conducted a retrospective cohort study to explore the association between breast cancer survival and anticoagulant drug use in Finland between 1995 and 2015.

All diagnoses of female breast cancer in Finland between 1995 and 2013 were obtained from the national comprehensive Finnish Cancer Registry (as described in previous publications; refs. 17, 18). Information on each patient's primary therapy was also obtained from the Finnish Cancer Registry. After excluding 4,504 cases of carcinoma in situ with no later diagnosis of invasive carcinoma, a total of 73,170 women were left in the final cohort.

Death-related information was obtained from the national death certificate registry of Statistics Finland, which assigns official causes of death based on mandatory death certificates. The available information included the death date and the immediate, primary, and contributory causes of death. Only deaths with breast cancer (ICD-10: C50) listed as the primary cause were regarded as breast cancer–related deaths.

A mammography screening program was started in Finland in 1987 and became government mandated in 1992. Between 1992 and 2009, the program had an average national coverage rate of 86.7% (18); invitations to participate were sent at 2-year intervals (19). By linking our study cohort to the national Mass Screening Registry maintained by the Finnish Cancer Registry, we were able to obtain data on cohort members' participation in the mammography screening program and the number of screening rounds attended by each.

We obtained information on conditions that are major indications for anticoagulant use from the national Care Register for Health Care (HILMO), which is maintained by the National Institute for Health and Welfare; the relevant conditions included pulmonary embolism (ICD-10: I26.0, I26.9), VTE (ICD-10: I82.0–82.9), and atrial fibrillation (ICD-10: I48). HILMO covers all public hospitals in Finland and records all diagnoses from in- and outpatient visits. We examined HILMO data for diagnoses made between 1995 and 2013. Diagnoses made during primary care visits are not reported in this database. Charlson comorbidity scores (20) were calculated on the basis of diagnoses recorded in the HILMO database during the follow-up timeframe.

To obtain information on anticoagulant use, the study cohort was linked to the national medication reimbursement database maintained by the Social Insurance Institution (SII) of Finland. As a part of the national health insurance program that covers all Finnish citizens, SII provides reimbursement for physician-prescribed drugs (21). In Finland, all citizens of every age are eligible for medication reimbursements. The Government-directed Pharmaceutical Pricing Board (HILA) decides which medications are eligible for reimbursement; most physician-prescribed drugs used for the treatment of disease are covered. During the study period, drug reimbursements ranged from 35% to 100%, depending on condition severity (Finnish Statistics on Medicines. Available at: https://www.kela.fi/documents/10180/1889281/SLT+2013_net.pdf/0758ba68-1886-4b69-bdb6-b7566e9daa2c). The reimbursement system began in 1970 (History of Medication reimbursements in Finland. Available at: https://www.kela.fi/web/en/operations-history). The study cohort was linked to the reimbursement database using the unique personal identity code assigned to all residents of Finland. All anticoagulant drugs are physician prescribed in Finland; thus, all reimbursements in outpatient settings are recorded in the database. However, drugs used during inpatient hospital visits are not recorded in the database.

All 14 anticoagulant drugs used in an outpatient setting during 1995–2015 were identified on the basis of their Anatomical–Therapeutic–Chemical codes. The identified drugs included two vitamin K antagonists [warfarin (B01AA03) and phenindione (B01AA02)], three heparins [dalteparin (B01AB04), enoxaparin (B01AB06), and tinzaparin (B01AB10)], four platelet aggregation inhibitors [clopidogrel (B01AC04), dipyridamole (B01AC30), iloprost (B01AC11), and ticlopidine (B01AC05)], two direct thrombin inhibitors [dabigatran (B01AE07) and ximelagatran (B01AE05)], and three factor Xa inhibitors [rivaroxaban (B01AX06), apixaban (B01AF02), and fondaparinux (B01AX05)]. We also obtained information on the use of cholesterol-lowering drugs, antihypertensive drugs, and antidiabetic drugs (Supplementary Table S1) that may influence breast cancer survival (22–24).

Cox proportional hazard regressions were used to calculate HRs and 95% CIs for breast cancer–related death. The follow-up timeframe started at breast cancer diagnosis and continued until death, emigration, or January 1, 2016 (whichever came first).

During analysis, we used two different model adjustments. First, the Cox regression model was adjusted for age only. Then, the Cox regression model was adjusted for age, breast cancer spreading at diagnosis (localized, locally advanced, or metastatic; information available for 85.6% of cases), primary treatment (available for 99.6% of cases), Charlson comorbidity score, number of mammography screening rounds attended, diabetes diagnosis, antihypertensive drug use, statin use, hormonal treatment, tumor histology (ductal invasive, lobular invasive, or other invasive; available for 99.9% of cases), and indications for anticoagulant use.

We performed four separate analyses related to the pre- and post-diagnostic use of anticoagulants: (i) users of any anticoagulant drug compared with nonusers; (ii) warfarin users compared with anticoagulant nonusers; (iii) LMWH users compared with anticoagulant nonusers; and (iv) warfarin users compared with users of other anticoagulants.

We calculated each woman's total annual anticoagulant use; this amount was calculated for each calendar year and separately for the pre- and postdiagnostic periods. Doses between different anticoagulants were standardized by dividing the annual total milligram amount by the standard defined daily dose (DDD) listed by the World Health Organization (WHO ATC/DDD index 2016. Available at: http://www.whocc.no/atc_ddd_index/). Each year with a registered anticoagulant purchase was considered a year of usage, regardless of the purchase amount. The intensity of a woman's anticoagulant use was estimated by dividing the sum of all her annual doses by the number of years she used anticoagulants.

Prediagnostic anticoagulant use was analyzed as a time-fixed variable. Cumulative use from 1995 onward was stratified by tertiles. Post-diagnostic use was analyzed as a time-dependent variable. Both, user status and cumulative use were updated separately for each year of follow-up, beginning from the year of breast cancer diagnosis. Women were categorized as nonusers until their first recorded anticoagulant purchase; they maintained their user status for each year with recorded anticoagulant purchases. Women who discontinued anticoagulant use were categorized as previous users. Women remained categorized as previous users unless anticoagulant use was restarted at another point during the follow-up timeframe; their status was then changed back to current user. User status was updated annually. Similarly, cumulative amount, use duration, and average annual dose were updated for each follow-up year. In the main analysis, we included user status and cumulative duration of use. Cumulative amounts and usage intensities were included in additional analyses.

In the analysis comparing warfarin users to anticoagulant nonusers, women were categorized as warfarin users each year with recorded purchases of warfarin, even if other anticoagulants were also used during that year. A similar methodology was used to compare LMWH users with anticoagulant nonusers. In the analysis comparing warfarin users to users of other anticoagulants, women were considered users of other anticoagulant drugs if they had recorded purchases of an anticoagulant other than warfarin.

The long-term impact of postdiagnostic anticoagulant use was assessed in a lag-time analysis. More specifically, anticoagulant exposure was lagged 1–3 years forward from the actual year of usage (i.e., its effect was ignored for the first 1 to 3 years following initial exposure). This approach is crucial because anticoagulants are used to treat breast cancer symptoms, and thus anticoagulant use may be associated with an elevated risk of breast cancer–related death in the short term. This phenomenon is known as protopathic bias (25).

We performed subgroup analyses for postdiagnostic anticoagulant use stratified by clinical and background characteristics. In the subgroup analyses, Charlson comorbidity scores were stratified into three groups (0 points, 1 point, and 2 or more points). We tested the statistical significance of each background variable's effect modification; this was done by adding an interaction term between anticoagulant use and the tested background variable to the Cox regression analysis and determining whether or not inclusion of the interaction term improved model fit. A P < 0.05 was considered statistically significant.

To evaluate the impact of advanced disease on breast cancer survival, we performed a sensitivity analysis in which patients with metastases or unknown spreading at diagnosis were excluded. The role that the timing of breast cancer diagnosis plays in survival was determined by including diagnosis year in the Cox regression model during sensitivity analysis.

For 5,412 women in the Tampere University Hospital area, we also obtained information on hormone receptor status [estrogen receptor (ER) and progesterone receptor (PR)] and HER2 receptor status during a woman's first breast cancer diagnosis between 1995 and 2013. Adjusting for the same control variables included in the main analysis, we conducted multivariate analyses relating anticoagulant use, warfarin use, and LMWH use to breast cancer survival; this allowed us to evaluate the prognostic value of these factors.

All statistical analyses were carried out using IBM SPSS Statistics 22.0. All statistical tests were two-sided.

At a median follow-up of 5.8 years after breast cancer diagnosis, a total of 22,520 (31%) women had died; for 10,900 (15%) of those women, breast cancer was the primary cause of death. A total of 25,622 (35%) women had used anticoagulants between 1995 and 2015. Of those women, 10,594 (41%) had used warfarin, and 14,224 (56%) had been prescribed LMWH. Compared with nonusers, women who used anticoagulants were older at breast cancer diagnosis (median age 66 and 59 years, respectively). The median age at breast cancer diagnosis among warfarin users was 72. The pattern of tumor spreading and tumor histology at diagnosis were similar in all groups. Curative-intent surgery was more common among anticoagulant nonusers than users (66.8% and 33.3%, respectively). The distribution of each background variable is presented in Table 1.

When comparing prediagnostic users of any anticoagulant to nonusers, there was no significant risk association for breast cancer survival (multivariable HR = 1.00; 95% CI, 0.93–1.07; Table 2). Short-term use (≤1 year) was associated with a statistically significant risk decrease in both the age- and multivariable-adjusted analyses. In the analysis comparing prediagnostic warfarin users to anticoagulant nonusers, there was no statistically significant risk association (overall multivariable HR = 0.97; 95% CI, 0.89–1.06; Table 2). When comparing prediagnostic LMWH users to anticoagulant nonusers, there was a statistically significant risk decrease (multivariable HR = 0.71; 95% CI, 0.59–0.85). This risk decrease was limited to short-term LMWH users. Table 2 shows results subdivided by the prediagnostic use of different anticoagulants. Results stratified by the amount and intensity of anticoagulant use are presented in Supplementary Table S2. The results were similar to those stratified by duration of use.

When comparing breast cancer survival in postdiagnostic anticoagulant users and nonusers, the risk was significantly increased for both current users (HR = 1.41; 95% CI, 1.33–1.49) and previous users (HR = 1.30; 95% CI, 1.22–1.39; Table 3). With the exception of those using anticoagulants for 4 years or more (multivariable HR = 1.06; 95% 0.96–1.19), the risk associations were similar in the analyses stratified by duration of anticoagulant use. In the lag-time analysis, the increased risk was attenuated but remained significant in all cases, except among short-term users and previous users.

When comparing warfarin users with anticoagulant nonusers, the risk of breast cancer–related death was elevated among previous users and current users (HR = 1.58; 95% CI, 1.45–1.72 and HR = 1.10; 95% CI, 1.02–1.19, respectively; Table 3). However, the risk increase disappeared for previous users in the lag-time analyses. In contrast, the risk of breast cancer–related death remained practically unchanged for current warfarin users.

When compared with anticoagulant nonusers, current LMWH users were at a significantly higher risk of breast cancer–related death (HR = 2.62; 95% CI, 2.42–2.83); but previous LMWH use was not associated with any risk increase. The risk increase was limited to long-term users (≥ 2 years; HR = 3.39; 95% CI, 3.00–3.82). When LMWH use was lagged for 3 years in the lag-time analysis, the observed risk increases were attenuated and disappeared. The results for post-diagnostic use of any anticoagulant stratified by use amount and intensity are presented in Supplementary Table S3.

After a Bonferroni correction for multiple testing, age at diagnosis, number of mammography screens, tumor spreading at diagnosis, primary breast cancer treatment, antihypertensive use, and indications for anticoagulant use all significantly modified the breast cancer–related mortality associations for anticoagulants in general (Fig. 1). The risk increase among anticoagulant users was only significant for participants diagnosed with breast cancer at age 62 or younger (P_interaction < 0.001). Having metastatic disease at diagnosis, nonsurgical primary treatment (P < 0.001 for each), and histology other than ductal or lobular carcinoma (P = 0.002) removed the risk association. Simultaneous use of antihypertensive drugs and anticoagulants ameliorated the risk increase (P < 0.001), except among LMWH users; in LMWH users, the risk increase was stronger when antihypertensive drugs had been used simultaneously (P = 0.002). When patients had not received a mammography screening, there was no elevated risk associated with anticoagulant use (P < 0.001). Furthermore, the risk of breast cancer–related death was higher among anticoagulant users with a recorded diagnosis of atrial fibrillation (P = 0.002). When comparing women who had initiated anticoagulant use before breast cancer diagnosis with those who used anticoagulants after diagnosis, the risk of breast cancer–related death was significantly lower (P = 0.034; Fig. 1). This effect modification was generally similar for all anticoagulant subtypes (not shown).

After excluding patients with metastatic disease or unknown spreading at diagnosis, the risk of breast cancer–related death increased slightly among anticoagulant users (multivariable HR = 1.67; 95% CI, 1.56–1.79) compared with those in the main analysis. A similar association was observed for other anticoagulant user categories, except for previous LMWH users. However, the effect was attenuated or missing for some categories in the lag-time analyses (Table 4). Including the timing of breast cancer diagnosis (year of diagnosis) in the sensitivity analysis of the Cox regression had virtually no effect on results (Supplementary Table S4).

Among 5,412 women with data on ER, PR, and HER2 status, the risk of breast cancer–related death was elevated among anticoagulant users compared with nonusers. The risk was also elevated in LMWH users compared with anticoagulant nonusers independent of ER or PR status; however, this elevation was not evident in patients receiving warfarin. On the other hand, among women with triple-negative disease, anticoagulant use (both overall and for each subtype) was associated with a significantly elevated risk of breast cancer–related death (Supplementary Table S5).

We also compared the risk of breast cancer death between warfarin-only and LMWH-only users and found no risk association in general (HR = 0.94; 95% CI, 0.85–1.05). However, when women were stratified by treatment duration, those who used these medications for 2 years or less had a 1.3-fold risk of breast cancer–related death. In contrast, use of these medications for 3 years or more was associated with 0.6-fold risk of breast cancer–related death (not shown).

In an additional sensitivity analysis, we restricted the cohort to those who underwent curative-intent surgery. In this analysis, the risk of breast cancer–related death was elevated 1.7-fold in anticoagulant users compared with nonusers. The risk increase was attenuated in the lag-time analysis (not shown).

According to the literature, there are several specific pathways that may link the coagulation cascade to breast cancer growth and progression (e.g., via angiogenesis, which is vital for tumor cells; refs. 5–10). Several anticoagulant subtypes (i.e., warfarin, direct thrombin inhibitors, and LMWH) are postulated to improve breast cancer survival by affecting the growth and spread of tumors (3, 4, 11–14). Therefore, there has been increasing interest in clarifying the association between anticoagulant use and breast cancer survival. In this epidemiologic study, we did not observe a positive effect of anticoagulant use on survival; this was true when examining anticoagulants in general and by subtype.

To the authors' knowledge, only one previous observational study (15) has evaluated the association between breast cancer survival and anticoagulant use. That study was limited to warfarin users; moreover, although the study included a relatively large number of breast cancer cases, its number of warfarin users was low (400). That study reported an approximately 1.4-fold increase in breast cancer–related mortality among warfarin users. While that result is similar to those described here, that study did not examine other anticoagulant drugs. Furthermore, that study did not evaluate the impact that the timing of anticoagulant use has on the survival of patients with breast cancer.

Our study revealed that, although postdiagnostic anticoagulant use was associated with an increased risk of breast cancer–related death, using anticoagulants for more than 4 years eliminated that risk. The risk association was particularly elevated among women who used LMWH for 2 years or more (3.4-fold risk of breast cancer–related death). However, the increase in risk appeared limited to active LMWH users; it did not persist when LMWH use was terminated. Using warfarin for 2 years or less after breast cancer diagnosis was also associated with an increased risk of breast cancer–related death. The risk increase was more significant for previous warfarin use than for current use. In a supplemental analysis, the risk of breast cancer–related death was significantly lower in warfarin users compared with women being treated with other anticoagulant drugs. However, this likely reflects the high risk associated with LMWH use. For LMWH, the risk increase was most likely attributable to reverse causation by terminal-phase breast cancer because LMWH is the drug of choice for all VTE treatment, including VTEs caused by advanced cancer. Because the risk association between anticoagulant use and breast cancer–related death was attenuated or eliminated in lag-time analyses, the relationship is not causal. In addition, no risk association was observed between prediagnostic anticoagulant use and breast cancer–related death. Furthermore, women initiating anticoagulant therapy after breast cancer diagnosis were at a higher risk of dying from breast cancer than those who became anticoagulant users prior to diagnosis.

In the analysis evaluating the risk associated with prediagnostic anticoagulant use, we found that low-dose, short-term use of LMWH was associated with a reduced risk of breast cancer–related death. This association is unlikely to be causal because no such effect was observed with long-term use. Therefore, our study does not support the theory postulated following previous in vivo and in vitro experiments that LMWH use improves survival (13, 14).

In lag-time analyses evaluating the impact that the timing of anticoagulant use plays, the risk increase observed in the main analysis was attenuated but remained elevated. For LMWH users, the timing of anticoagulant use reduced the risk of breast cancer–related death; however, among women who had used LMWH for 2 years or more, the risk of death from breast cancer remained nearly double that of nonusers. Risk associations were greatly modified by having an indication for anticoagulant use. Unfortunately, however, this finding is less conclusive because data on indications for anticoagulant use in primary healthcare settings were not available. The increased risk of breast cancer–related death seems limited to short-term use because thromboembolic events are more common in patients with advanced cancer (1) and are associated with a poorer prognosis (2). However, thrombosis could be a long-term risk factor because the risk increase did not disappear in the lag-time analysis.

According to our subgroup analysis, the risk of breast cancer–related death was particularly high for those who had a local or locally advanced disease. An elevated risk of breast cancer–related death was also observed among those diagnosed with atrial fibrillation. Because the risk of cancer-induced thrombosis can be assumed to be lowest for those with atrial fibrillation, this finding supports the theory that thrombosis promotes cancer progression and worsens patient prognosis (rather than the reverse). On the other hand, compared with exclusively postdiagnostic anticoagulant users, those individuals who were both prediagnostic and postdiagnostic users had a lower risk of breast cancer–related death. In either case, the use of anticoagulants was not associated with an elevated risk of breast cancer–related death.

This study has several strengths. The sample includes all breast cancer cases occurring in a nationwide population over a period of 19 years (n = 73,170), thereby minimizing selection bias. We believe that this is the largest and most comprehensive epidemiologic study on breast cancer survival in anticoagulant users. Moreover, we had detailed information on tumor spread and histology at the time of diagnosis. We also had comprehensive data on anticoagulant use, including the timing of anticoagulant use. Furthermore, it was possible to stratify cohort members according to anticoagulant subtype, usage amount, duration of use, and intensity of use. The analyses were adjusted for multiple background variables to minimize bias. In a subpopulation of our cohort, we were able to evaluate the effect that important prognostic factors, such as ER, PR, and HER2 status, have on breast cancer survival.

This study also had some limitations. We had no information on smoking habits, but smoking has been associated with worse breast cancer prognoses (26). We had no data on hormone replacement therapy or body mass index. Similarly, clinicians report only the most severe cases of obesity to health registries; as such, information related to obesity was limited. In addition, we did not have data on socioeconomic factors that may influence breast cancer survival (27). We did not have data on anticoagulant use prior to 1995, which limits our ability to evaluate long-term risk associations. Moreover, we did not have information on indications for anticoagulant use identified during primary healthcare visits; without this data we could not fully evaluate the effect that indications for anticoagulant use have on breast cancer survival. Furthermore, our data did not include anticoagulants used during hospital inpatient visits; therefore, some exposure misclassification is possible. However, we consider this to be a minor concern because indications for anticoagulant use other than thromboprophylaxis would continue after hospital discharge; thus, those indications would be captured in our anticoagulant use data. Finally, we did not have data on mammography screenings performed outside the national screening program, and this could be a source of healthy user bias.

For clinicians, the primary takeaway of this study is that anticoagulants provide no benefit for cancer control, even though they are useful in the treatment of thrombosis. Our study does not find any support for the prophylactic use of such potentially dangerous medications. Moreover, because our results suggest that anticoagulant use is correlated with cancer-related death, epidemiologists should adjust for anticoagulant use in future studies. Despite promising results from previous in vitro and in vivo studies, bioscience researchers should be aware that the positive effects of anticoagulant use on cancer survival are not evident at the population level.

General anticoagulant use, warfarin use, and LMWH use confer no clinical benefits against breast cancer. In fact, the risk of breast cancer–related death is increased for postdiagnostic anticoagulant users. Among prediagnostic users, the low-dose, short-term use of LMWH is associated with improved survival; however, other patterns of LMWH use predict a risk of death similar to that of nonusers. It is possible that the association between thrombosis and cancer masks some potential survival benefit. Future studies should focus on determining whether the administration of newer oral anticoagulants is associated with breast cancer survival; direct thrombin inhibitors, such as dabigatran, have been reported to reduce breast cancer metastasis (7–9).

T.J. Murtola is a consultant for Astellas, Janssen-Cilag, and Ferring and reports receiving speakers bureau honoraria from Astellas. No potential conflicts of interest were disclosed by the other authors.

Conception and design: P.T. Kinnunen, M. Artama, T.J. Murtola

Development of methodology: P.T. Kinnunen, M. Artama, E. Pukkala, T.J. Murtola

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): P.T. Kinnunen, E. Pukkala, T.J. Murtola

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): P.T. Kinnunen, M.O. Murto, E. Pukkala, T.J. Murtola

Writing, review, and/or revision of the manuscript: P.T. Kinnunen, M.O. Murto, M. Artama, E. Pukkala, K. Visvanathan, T.J. Murtola

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): P.T. Kinnunen

Study supervision: T.J. Murtola

T.J. Murtola has received a grant from Pirkanmaa Hospital District (grant numbers 50640 and 9S067).

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.