Venous thromboembolism and mortality in breast cancer: cohort study with systematic review and meta-analysis

Based on data from a large cohort of women with breast cancer representative of the United Kingdom, the risk of mortality was more than doubled in the time following a VTE event, reflecting the high short-term mortality following a thromboembolic event. In contrast, using the landmark approach which assigned women as being a VTE or non-VTE case for the entire follow-up period, VTE exerted no increased risk of mortality once important covariates such as stage of disease and a measure of overall health status was taken into account. When our data were pooled with those from seven additional studies (including two which are currently unpublished), the pooled hazard ratio was 2.35 (2.17–2.55) for studies using a TVC analysis and 1.69 (1.12–2.55) for those using an nTVA, the latter of which contained substantial heterogeneity. The hazard ratio we report for TVC analysis is comparable to that reported by Posch et al. more recently of 2.98 (2.36–3.77) using a multi-state model applied to data from the Vienna Cancer and Thrombosis Study which considered all cancer types rather than breast cancer specifically [39]. Sub-group analyses reported higher HRs in studies which did not adjust for key covariates, whereas the timing of VTE diagnosis in relation to the cancer diagnosis did not have an appreciable impact on the magnitude of the hazard ratios observed.

To our knowledge, this is the first attempt to systematically evaluate all available data exploring whether or not among women with breast cancer, the risk of mortality is raised following development of a VTE. Our systematic review was strengthened by inclusion of two established databases (MEDLINE and EMBASE) with carefully selected search terms. Furthermore, through obtaining additional data for studies originally published in the form of conference abstracts, we were able to include data which is currently unpublished in our synthesis of the evidence. Thirdly, by inclusion of our data from the CPRD we were able to include data in the overall synthesis which has the strength of utilizing recently linked primary, secondary and cancer registration data from a large representative sample of women from the UK. Our two distinct approaches to analysis, enabled us to assess the effect of a VTE on short-term and long-term mortality separately.

Limitations of our work include the fact the methods of meta-analysis employed in our systematic review relied on survival data being presented both separately for breast cancer patients in studies where patients with a mixture of cancer types were reported, and also in an appropriate numerical form so that hazard ratios (and standard errors or confidence intervals) from these could be obtained. As such there were several potentially relevant studies which have been conducted but which we were unable to include. Our systematic review also contained a high degree of heterogeneity, meaning that it was not possible for us to determine the “true” degree which developing a VTE has on subsequent mortality. Instead effect sizes would be influenced by characteristics of the study population (age, tumour characteristics and treatment modalities), methods for establishing VTE (including whether methods such as a Doppler scan were used to confirm the diagnosis) and duration of follow-up. In part, we were successful in elucidating specific reasons for this heterogeneity, namely that our finding in the CPRD that effect sizes were attenuated considerably after adjustment for key covariates was also demonstrated within one of the papers included in our systematic review [36]. However, even in sub-group analyses whereby data were stratified by factors, which we anticipated, would account of heterogeneity of results between studies, considerable residual heterogeneity remained in many instances (as indicated by the I-square statistic). Finally, our findings could be influenced by the potential for publication bias as is inherent with any systematic review. However, in the present review no obvious differences were found in the magnitude of the effect size between the five studies currently published and three presently unpublished.

Differences in methodological quality of original studies represent another potential source of heterogeneity in reviews of observational studies as addressed by the sub-group analyses described above. Similarly, methodological deficiencies in some or all of the component studies could bias estimates of the pooled result. Many of the source studies relied on routinely collected administrative data for determining VTE status in study participants. Misclassification of VTE events could attenuate the magnitude of an association between VTE and survival. In the CPRD, our algorithm for defining VTE was previously shown to have positive predictive value of 84% when compared with more detailed investigations of patient records [20]. However, this algorithm has not been validated specifically in cancer patients and would not capture anticoagulant prescriptions emanating from secondary care. In studies which did not use a TVC approach, the complex nature of the chronology between diagnosis of VTE, diagnosis of cancer and subsequent outcome could influence findings. For example, it is common for studies to start follow-up at the time of cancer diagnosis. If VTE occurs after this date then there would be a period of guaranteed follow-up time between the cancer and VTE dates, which would create a favourable impression of survival in this “exposed” group and thus weaken any true association (immortal time bias). Whilst we attempted to stratify results by timing of VTE and cancer, this information could not always be adequately established from the original study reports.

The potential for immortal time bias was avoided in both of the approaches to analysis we adopted for the CPRD data. The use of a time-varying covariate analysis incorporates changes in exposure status throughout the follow-up period and thus is sensitive to picking up changes in risk of outcome which occur shortly after a change in exposure status [40]. This approach is supported by the recent EPIPHANY study findings which reported fatality percentages following a pulmonary embolism of 14% at 30 days and 27% at 90 days follow-up in 1033 cancer patients [41]. Therefore, the Landmark approach excludes a relatively high percentage of all VTE-related deaths and is more appropriate for assessing mortality longer term in patients who survive the initial event. The analysis also has some more favourable statistical properties as the alternative approach used (landmark analysis) does not include VTE events which occur after 6 months in addition to the exclusion of the 6-months following cancer diagnosis from the follow-up time However, this approach does have limitations especially when using routine healthcare data, as in the case of mortality as an outcome, acute medical events are more likely to get diagnosed in the intensive period of medical consultation which is known to take place in the weeks prior to death. In this particular context, however, the key advantage of the landmark approach is that it allows us to interpret how a VTE event occurring relatively soon after diagnosis (when the risk of VTE is highest) influences mortality longer term for which the clinical implications may be more apparent.

Finally we were unable to clearly establish whether factors such as cancer stage and underlying health status may have influenced the extent to which a VTE is associated with the risk of mortality. Whilst HRs were larger for women with local disease at the time of diagnosis, given that the risk of mortality was considerably higher in women with metastatic disease (314 deaths in 1200 person-years of follow-up) than in women with local disease (807 deaths in 32,000 person-years of follow-up) this is likely to be due to the issue of scale dependence whereby there is the potential for VTE to have a greater impact on a measure of relative association (such as the hazard ratio) in subgroups where the underlying risk of an outcome event is lower [42].

There are several mechanisms via which a VTE may exert a detrimental impact on cancer survival. There is an immediate impact due to the known high short-term fatality resulting from a thrombotic event which among all patients is estimated to be around 1% following a DVT and over 20% following a pulmonary embolism [41, 43]. Pooled results from two studies from the US and UK which would capture this short-term effect through incorporating VTE as a time-varying covariate indicate a greater than 2-fold of risk of mortality following a VTE. Compliance with existing clinical guidelines on primary prevention of VTE in cancer patients which advise targeting of prophylaxis in selected patients undergoing cancer surgery along with some patients in the outpatient setting [44‐46]. However, it should be noted that in the Khorana score women with breast cancer may not be recommended for primary prophylaxis as these tend to score poorly on cancer type, anaemia and thrombocytosis. We have previously shown with this cohort that VTE events in women with breast cancer are likely to occur either during or immediately following chemotherapy or in the first month following surgery [6].

Cancer patients are at increased risk of bleeding from anticoagulation, with an estimated 2-fold increased risk for major bleeding compared to non-cancer patients [47]. Unsurprisingly, major and minor bleeding increases the hazard of death by over two-fold [48]. In addition, cancer patients are at 2–3 fold increased risk of recurrent VTE [47, 49‐51]. However, based on the data from the current study, in the case where a woman with breast cancer is fortunate enough to survive her initial thrombotic event, the influence on long term prognosis is more difficult to establish, with a suggestion from this current study that mortality is not raised at all once cancer stage and underlying health status are taken into account. Guidelines from the UK National Institute of Health and Care Excellence (NICE) along with equivalent guidelines from other countries advise that cancer patients who develop VTE should receive at least 6 months of anticoagulation and in some instances treatment should continue indefinitely [52]. It is plausible to suggest that if adherence to these guidelines is good, then this could at least in part explain the relatively promising prognosis for women with breast cancer who survive their VTE, with prophylactic anticoagulation successfully mitigating against recurrent VTE (a likely cause of mortality). However the current NICE guidelines were not as robust in the era covered by the CPRD data and studies included in our meta-analysis. A move from vitamin-K antagonists to low molecular weight heparins in recent years because of greater efficacy in preventing recurrent VTE may further negate the negative survival impact of recurrent VTE [53]. More contemporary data reporting rates of VTE recurrence in cancer patients from the last decade as well as those with specific types of cancer are needed.

A further explanation for the detrimental impact of VTE on cancer survival relates to complex mechanisms underlying the symbiotic relationship between coagulation and tumour factors. Coagulation parameters are understood to play an important role in tumour progression and metastases, with changes in the haemostatic system evident in cancer patients even in the absence of a VTE [3]. It is hypothesized that VTE, even at the subclinical level of biochemical hypercoagulability, may have a role in promoting cancer growth and metastases and be associated with a more aggressive tumour behavior [54]. This has led researchers over many decades to explore the antineoplastic effects of anticoagulants and whether they could improve cancer survival even in the absence of a VTE. Overviews of the most recent randomized trial data comprising cancer patients without indication for anticoagulation (usually cancer outpatients) found no evidence of both oral anticoagulation (warfarin) [25] and low molecular weight heparin [50] on mortality at 12 months. However, evidence from the LMWH review indicates that this intervention does have modest (16%) reduction in longer term mortality, in line with previous evidence that the beneficial effects of LMWH occur after 12 months and also in patients with less advanced disease [51]. It is possible that if barriers to adherence with long term LMWH use could be overcome then there is the potential for a greater reduction mortality risks in cancer patients both with and without a previous VTE [55]. The current consensus is that future research in this area should focus on patients with specific cancer types rather than heterogeneous groups of tumours [3, 51].