Abstract
Abstract
Objective
To study the progression of newly diagnosed breast cancer patients towards cardioembolic disorders (CDs) and ischaemic or structural heart disease (ISH) within a multistate framework that includes death as an absorbing state, in order to evaluate long-term cardiovascular burden and to examine whether the observed patterns may reflect short-term acute events or longer-term effects potentially associated with treatment. CD and ISH occurrences following breast cancer treatment may have different underlying biological mechanisms, potentially leading to different survival patterns across patient groups defined by breast cancer characteristics, and we sought to explore this.
Design
Two continuous-time multistate Markov models described patients’ transitions among three states: time of breast cancer diagnosis, first occurrence of CD or ISH, and death, depending on patients’ molecular subtype, tumour stage, age and radiation treatment status.
Participants
Altogether 36 778 patients from the Norwegian Cancer Registry, linked with information on cardiovascular events from the Norwegian Patient Registry, diagnosed earliest in 2006, followed until their first cardiovascular event and/or death or throughout 2020.
Results
HR−/HER2− patients had the highest HR for a first CD event compared with HR+/HER2− patients (HR 1.51, 95% CI 1.39 to 1.65), with no significant differences observed between the remaining groups. For first ISH event, HR−/HER2+, HR+/HER2+ and HR−/HER2− patients had HRs of 1.91 (95% CI 1.61 to 2.27), 1.64 (95% CI 1.43 to 1.88) and 1.56 (95% CI 1.44 to 1.69), respectively, compared with HR+/HER2− patients.
Conclusions
Long-term cardiovascular burden for patients with breast cancer may be more strongly influenced by chronic vascular and myocardial damage, likely driven by treatment-related cardiotoxicity and cumulative exposure compared with acute thromboembolic mechanisms. A multistate Markov modelling approach proved valuable for capturing the dynamics for patients to move between the states breast cancer diagnosis, CD or ISH event and death, and may guide more tailored follow-up and use of cardioprotective strategies.
Keywords: Cardiovascular Disease, Death, Health, Breast tumours, Risk Assessment
STRENGTHS AND LIMITATIONS OF THIS STUDY.
A large representative dataset from a nationwide cancer registry was analysed.
The analysis expands the perspective beyond acute cardiovascular events to consider the cumulative long-term impact of cancer therapies on vascular and myocardial health over a follow-up period of up to 15 years.
The analysis would have been strengthened by access to patients’ relevant health and lifestyle factors, as well as complete treatment trajectories with exact dates throughout the entire observation period, but such data were not available.
Introduction
The survival prognostics for breast cancer (BC) patients have improved over years due to earlier detection and better treatment. This is followed by an increased risk of developing cardiovascular disease (CVD) compared with the general population.1–3 The literature points to a complex and entangled set of factors to explain this; side effects of the cancer treatments, shared risk factors between obtaining BC and CVD, and the long-term impact of cancer therapies. Chemotherapy drugs are known for their cardiotoxicity. This encompasses both the direct impact of the drugs on cardiac function and their indirect effects resulting from alterations in the haemodynamic flow or thrombotic occurrences.4 The drug Trastuzumab, which targets HER2-positive cancer cells making it a more precise treatment with typically fewer side effects compared with chemotherapy, is also associated with an increased risk of heart failure (HF), while the drug Tamoxifen blocks oestrogen in hormone-receptor-positive BC and is associated with an increased risk of stroke. Also, radiation therapy could damage the heart, leading to an increased risk of coronary artery disease, HF and heart attack.
BC-diagnosed patients may experience CVD and/or death, and we describe this progression using a multistate Markov model (MSM) with three states: BC diagnosis, first CVD occurrence and death, commonly referred to as a non-recoverable illness–death model. The CVD occurrence was subdivided into cardioembolic disorders (CD) and ischaemic or structural heart disease (ISH) occurrences, defining the intermediate state in two separate multistate models.
Multistate Markov models are widely used to describe disease progression and event histories among BC patients, including transitions between health states such as recurrence, metastasis and death, as well as for evaluating treatment strategies and outcomes.5 6 Our models naturally account for the competing risks of CVD and death. They provide clinically meaningful measures such as including transition probabilities that enhance the understanding of patients’ disease trajectories. They appropriately capture the temporal ordering of events and the impact of intermediate CVD on subsequent mortality, features that standard single-endpoint survival models cannot accommodate. Taking this approach, we set out to study the long-term cardiovascular burden among BC patients and specifically examine whether this burden was primarily shaped by sustained treatment-related effects or short-term acute events.
These analyses broaden the focus from acute cardiovascular events to the long-term, cumulative impact of cancer therapies on vascular and myocardial health. Understanding this burden is necessary for developing targeted surveillance strategies and cardioprotective interventions, aiding to improve life quality and survival for BC patients. This work follows two prior publications, where we examined temporal patterns in BC survival and cardiovascular outcomes.7 8
Methods and study population
Study population and selection criteria
We considered BC diagnosed women, aged 40 and older at time of diagnosis (earliest diagnosis year was 2006), diagnosed with tumour stage I (localised, small tumour), II (locally advanced tumour) or III (extensive local/regional spread), with data from the nationwide Norwegian Cancer Registry. They were followed until their first CVD and/or death or throughout 2020. CVD events were identified according to International Statistical Classification of Diseases and Related Health Problems, 10th Revision (ICD-10) codes and extracted from the nationwide Norwegian Patient Registry.9 10 We considered patients with valid follow-up timelines who had a unilateral BC event in the year of diagnosis, and with known tumour stage I–III, surgery treatment and radiation status, reducing the original dataset from 45 334 to 36 778 patients (see figure 1 for details).
Figure 1. Flow chart of study population selection.
BC diagnosis and time of death were recorded by year, while CVD events were given by calendar day. For simplicity, we assumed the time of BC diagnosis to be 1 January of the given year. Patients who died within their diagnosis year were assumed, for simplicity, to have survived for half a year.
The aim was to study the risk of subsequent CVD following BC diagnosis, and events occurring within the same year may reflect pre-existing conditions or concurrent diagnoses rather than treatment-related or disease-related effects. We therefore only considered CVD events occurring at least 1 year after diagnosis.
Study variables
Patients diagnosed with BC are classified into four molecular subtypes defined by their hormone receptor (HR) and human epidermal growth factor receptor 2 (HER2) status:
HR positive and HER2 negative (HR+/HER2−).
HR positive and HER2 positive (HR+/HER2+).
HR negative and HER2 positive (HR−/HER2+).
HR negative and HER2 negative (HR−/HER2−).
Other variables from the registry encompassed the patients’ age, tumour stage, radiation treatment, chemotherapy, trastuzumab, zoledronic acid and hormone treatment status.
The first CVD was defined as the earliest occurrence in one of two groups:
CDs: atrial fibrillation or embolic stroke (ES).
ISH diseases: heart attack (myocardial infarction, MCI), angina pectoris (AP), embolism and peripheral vascular disease or HF.
This gave two separate analyses, one focusing on the occurrence of a first CD event and another focusing on the occurrence of a first ISH event. Patients who experienced both CD and ISH events were not part of the analyses. These two groups of cardiovascular outcomes may have different underlying biological mechanisms, risk factor profiles, and potential associations with BC treatments, and could give different survival patterns across patient groups.11 12
Statistical analyses
We applied two continuous-time multistate Markov models to describe patients’ transitions among three states: time of BC diagnosis, first occurrence of a CD or ISH event, and death (a non-recoverable illness-death model). All patients started in state 1, being diagnosed with BC, subsequently moving on to state 2, having a first CD or ISH event, and further on moving to state 3, death (of any cause). Individuals who die without experiencing a CVD move directly from state 1, being diagnosed with BC, to state 3, death. Time was analysed in a clock-forward framework using time since BC diagnosis as the underlying timescale, and we estimated transition intensities between BC, cardiovascular events, and death.13 We considered the transitions between all these three states as dependent on molecular subtype, tumour stage, age and whether the patient received radiation treatment or not. Age was included as a time-varying covariate, given by age at transition between states, while other covariates were treated as fixed. In the statistical analysis, we included molecular subtype, tumour stage, age and radiation treatment status. Chemotherapy, endocrine therapy, trastuzumab and zoledronic acid were not included as explanatory variables because their administration is guided by treatment plans based on tumour characteristics and molecular subtype. We assumed that these treatments were administered according to the Norwegian National Action Programme for Breast Cancer.14 This way we can assess associations between each of these factors and experiencing a first CD or ISH event and compare survival between BC patients with and without CVDs.
Models were fitted by maximum likelihood using the MSM package in R, V.4.2.2, for multistate modelling, with default optimisation settings, with initial values for the transition intensity matrix generated automatically.13 15 95% CIs were obtained from the observed information matrix (asymptotic normal approximation). Convergence was assessed by examining the optimisation status reported by the model together with inspection of parameter estimates. Model description and further details are given in the online supplemental material.
Patient and public involvement
As these are pseudonymised health registry data, the patients were not involved in the analyses. BC patients or the public were not involved in the design or interpretation of this study.
Results
Both models converged. Stability was confirmed by refitting models with alternative initial values, which yielded identical parameter estimates, indicating a stable global optimum. No convergence issues were observed despite a few small covariate strata.
Patient characteristics
The average age was about 60 years at the time of BC diagnosis. For both analyses, the percentage of patients in tumour stage I was 52%, and 37% and 11% in tumour stages II and III, respectively. When focusing on the CD endpoint, we found that the percentage of patients in molecular subtype groups HR+/HER2−, HR+/HER2+, HR-/HER2− and HR−/HER2− were 65%, 7%, 4% and 25%, respectively. When focusing on the ISH endpoint, we found the corresponding distribution to be 64%, 7%, 4% and 25%. In both analyses, 80% received radiation treatment, 59% chemotherapy, 11% trastuzumab, 71% zoledronic acid and 71% hormone treatment.
From table 1, the distribution of patient characteristics and treatments among those transitioning between states differed from that in the overall population in both analyses. For example, in the CD analysis, the proportion of HR+/HER2− tumours was lower among patients transitioning from BC to CD compared with the overall population (57% vs 65%), while HR−/HER2− tumours were more frequent (35% vs 25%). A similar pattern was observed in the ISH analysis (52% vs 64% for HR+/HER2− and 36% vs 25% for HR−/HER2−).
Table 1. Patient characteristics for analysis 1: state 1 (BC diagnosis)—state 2 (first CD event)—state 3 (death) and for analysis 2: state 1 (BC diagnosis)—state 2 (first ISH event)—state 3 (death) and transition matrices for the number of patients moving between the three states.
| Patient characteristics, analysis 1: BC, CD and death (N1=32 095) | Patient characteristics, analysis 2: BC, ISH and death (N2=32 613) | ||||||||
|---|---|---|---|---|---|---|---|---|---|
| All | BC diag. →CD | BC diag. →death | CD →death | All | BC diag. →ISH | BC diag. →death |
CD →death | ||
| Number of patients | 32 095 | 2356 | 3568 | 758 | 32 613 | 2874 | 3568 | 800 | |
| Molecular subtype | HR+/HER2− | 20 720 (65%) | 1351 (57%) | 1531 (43%) | 367 (48%) | 20 852 (64%) | 1483 (52%) | 1531 (43%) | 345 (43%) |
| HR+/HER2+ | 2176 (7%) | 123 (5%) | 180 (5%) | 31 (4%) | 2280 (7%) | 227 (8%) | 180 (5%) | 42 (5%) | |
| HR−/HER2+ | 1137 (4%) | 62 (3%) | 146 (4%) | 22 (3%) | 1217 (4%) | 142 (5%) | 146 (4%) | 33 (4%) | |
| HR−/HER2− | 8062 (25%) | 820 (35%) | 1711 (48%) | 338 (45%) | 8264 (25%) | 1022 (36%) | 1711 (48%) | 380 (48%) | |
| Age | 60.35 (11.5) | 69.18 (11.21) | 66.9 (14.27) | 78.85 (10.62) | 60.20 (11.42) | 65.88 (11.9) | 66.90 (14.27) | 77.36 (11.85) | |
| Tumour stage | I | 16 583 (52%) | 1158 (49%) | 1016 (28%) | 292 (39%) | 16 841 (52%) | 1416 (49%) | 1016 (28%) | 299 (37%) |
| II | 11 906 (37%) | 941 (40%) | 1744 (49%) | 360 (47%) | 12 103 (37%) | 1138 (40%) | 1744 (49%) | 373 (47%) | |
| III | 3606 (11%) | 257 (11%) | 808 (23%) | 106 (14%) | 3669 (11%) | 320 (11%) | 808 (23%) | 128 (16%) | |
| Radiation | 25 577 (80%) | 1508 (64%) | 2089 (59%) | 584 (46%) | 26 024 (80%) | 1955 (68%) | 2089 (59%) | 379 (47%) | |
| Chemotherapy | 18 986 (59%) | 1565 (66%) | 2989 (84%) | 584 (77%) | 19 356 (59%) | 1935 (67%) | 2989 (84%) | 644 (80%) | |
| Trastuzumab | 3313 (10%) | 185 (8%) | 326 (9%) | 53 (7%) | 3497 (11%) | 369 (13%) | 326 (9%) | 75 (9%) | |
| Zoledronic acid | 22 704 (71%) | 2136 (91%) | 2867 (80%) | 722 (95%) | 23 006 (71%) | 2438 (85%) | 2867 (80%) | 747 (93%) | |
| Hormone treatment | 22 896 (71%) | 1474 (63%) | 1711 (48%) | 398 (53%) | 23 132 (71%) | 1710 (59%) | 2867 (48%) | 387 (48%) | |
| Transition matrices (number of patients transiting) | |||||||||
| Analysis 1: BC, CD and death | Analysis 2: BC, ISH and death | ||||||||
| BC | CD | Death | BC | ISH | Death | ||||
| BC | 26 171 | 2356 | 3568 | CD | 26 171 | 2874 | 3568 | ||
| CD | 1598 | 758 | ISH | 2074 | 800 | ||||
All variables except age are summarised by number (percentage), while age is summarised by mean (SD).
BC, breast cancer; CD, cardioembolic disorders (atrial fibrillation, embolic stroke); ISH, ischaemic and structural heart disease (heart attack, angina pectoris, embolism and peripheral vascular disease, heart failure).
Patients transitioning from BC diagnosis to death in both analyses were more likely to have more advanced disease: in the CD analysis, tumour stages II and III accounted for 49% and 23% (vs 37% and 11% overall), with a similar distribution observed in the ISH analysis (49% and 23%, also vs 37% and 11% overall).
With respect to treatment, the distribution also differed for patients transitioning to CD or ISH, with a lower proportion receiving radiation therapy (eg, 64% vs 80% overall in the CD analysis; 68% vs 80% in the ISH analysis) and hormone treatment, but a higher proportion receiving chemotherapy (66% vs 59% in the CD analysis; 67% vs 59% in the ISH analysis). The mean follow-up times were 6.87 and 6.78 years in analyses 1 and 2, respectively. Mean follow-up was calculated as the average observed time from BC diagnosis until death or the end of the observation period.
Transition probabilities between states
Among the 36 778 patients 1809 experienced both CD and ISH events. Altogether, 2356 patients experienced at least one CD event and 2874 experienced at least one ISH event. To present short- and long-term risks for someone who had not experienced a CD or ISH event, to still not having experienced a CD or ISH event, have a first CD or ISH event or die, table 2 displays the 1-year and 5-year transition probability matrices for the three states. These estimated probabilities are marginal (population-average) probabilities derived from the fitted multistate models and are therefore not specific to any individual patient’s covariate profile. After 1 year, there is a 1% chance for a BC diagnosed patient to experience a first CD event and a 1.4% chance to experience an ISH event. After 5 years these chances have increased to 4.5% and 6.3%, respectively.
Table 2. 1-year and 5-year transition probability matrices for the three states for analysis 1: state 1 (BC diagnosis)—state 2 (first CD event)—state 3 (death) and for analysis 2: state 1 (BC diagnosis)—state 2 (first ISH event)—state 3 (death), with 95% CIs.
| Analysis 1: BC, CD and death | Analysis 2: BC, ISH and death | ||||||
|---|---|---|---|---|---|---|---|
| 1 year | BC | CD | Death | BC | ISH | Death | |
| BC | 0.98 (0.98 to 0.98) | 0.01 (0.010 to 0.011) |
0.01 (0.01 to 0.01) |
BC | 0.98 (0.98 to 0.98) |
0.01 (0.01 to 0.01) |
0.010 (0.01 to 0.01) |
| CD | 0.97 (0.96 to 0.97) |
0.03 (0.03 to 0.04) |
ISH | 0.97 (0.96 to 0.97) |
0.03 (0.03 to 0.04) |
||
| 5 years | BC | CD | Death | 5 years | BC | ISH | Death |
| BC | 0.90 (0.90 to 0.91) | 0.04 (0.04 to 0.05) |
0.05 (0.05 to 0.05) |
BC | 0.89 (0.88 to 0.89) |
0.06 (0.06 to 0.07) |
0.05 (0.05 to 0.05) |
| CD | 0.84 (0.82 to 0.86) |
0.16 (0.14 to 0.18) |
ISH | 0.85 (0.83 to 0.87) |
0.15 (0.13 to 0.17) |
||
BC, breast cancer; CD, cardioembolic disorder; ISH, ischaemic and structural heart.
HRs for transitions between states
From the two fitted multistate models we estimated the HRs for transitions between states, and the results are displayed in table 3.
Table 3. Estimated HRs from multistate model with independent variables, where a baseline patient had molecular subtype HR+/HER2−, were in tumour state I and without radiation treatment.
| Analysis 1: BC, CD and death | Analysis 2: BC, ISH and death | |||
|---|---|---|---|---|
| Covariate | Transition | HR (95% CI) | HR (95% CI) | |
| Molecular subtype | HR+/HER2+ | BC diagnosis–CVD* | 1.15 (0.96 to 1.37) | 1.64 (1.43 to 1.88) |
| HR-/HER2+ | BC diagnosis–CVD | 1.13 (0.89 to 1.45) | 1.91 (1.61 to 2.27) | |
| HR-/HER2− | BC diagnosis–CVD | 1.51 (1.39 to 1.65) | 1.56 (1.44 to 1.69) | |
| HR+/HER2+ | BC diagnosis–death | 0.84 (0.67 to 1.06) | 0.77 (0.61 to 0.98) | |
| HR-/HER2+ | BC diagnosis–death | 1.38 (1.10 to 1.72) | 1.31 (1.04 to 1.64) | |
| HR-/HER2− | BC diagnosis–death | 2.14 (1.95 to 2.35) | 2.11 (1.92 to 2.32) | |
| HR+/HER2+ | CVD–death | 1.31 (0.99 to 1.73) | 1.12 (0.87 to 1.44) | |
| HR-/HER2+ | CVD–death | 1.71 (1.17 to 2.50) | 1.04 (0.76 to 1.43) | |
| HR-/HER2− | CVD–death | 1.22 (1.07 to 1.38) | 1.26 (1.11 to 1.43) | |
| Age | BC diagnosis–CVD | 1.11 (1.11 to 1.11) | 1.08 (1.08 to 1.09) | |
| BC diagnosis–death | 1.03 (1.03 to 1.04) | 1.03 (1.03 to 1.04) | ||
| CVD–death | 1.08 (1.07 to 1.09) | 1.08 (1.08 to 1.09) | ||
| Tumour stage | II | BC diagnosis–CVD | 1.03 (0.94 to 1.12) | 1.13 (1.05 to 1.23) |
| III | BC diagnosis–CVD | 1.50 (1.31 to 1.72) | 1.52 (1.34 to 1.72) | |
| II | BC diagnosis–death | 2.79 (2.50 to 3.11) | 2.71 (2.42 to 3.03) | |
| III | BC diagnosis–death | 7.71 (6.79 to 8.76) | 7.56 (6.65 to 8.60) | |
| II | CVD–death | 1.14 (1.00 to 1.30) | 1.42 (1.25 to 1.62) | |
| III | CVD–death | 1.83 (1.50 to 2.23) | 2.26 (1.85 to 2.76) | |
| Radiation | BC diagnosis–CVD | 0.67 (0.62 to 0.74) | 0.65 (0.60 to 0.71) | |
| BC diagnosis–death | 0.50 (0.45 to 0.55) | 0.50 (0.45 to 0.55) | ||
| CVD–death | 0.69 (0.60 to 0.79) | 0.61 (0.53 to 0.71) |
CVD refers to either CD (fourth column) or ISH (fifth column).
BC, breast cancer; CD, cardioembolic disorder; CVD, cardiovascular disease; ISH, ischaemic and structural heart.
Table 3 shows that HR–/HER2– patients had the highest hazard for a first CD event compared with the HR+/HER2– patients, HR: 1.51 with 95% CI 1.34 to 1.65. There was no significant difference between the HR+/HER2+, HR–/HER2+ and the HR+/HER2– patients. Table 3 also shows that patients in all three molecular subtype groups had a higher HR for a first ISH event compared with the HR+/HER2– patients, with HR–/HER2+patients having the highest HR (HR: 1.91 with 95% CI 1.61 to 2.27), the HR+/HER2+ patients the second highest (HR: 1.64 with 95% CI 1.43 to 1.88) and the HR–/HER2– patients the third (HR: 1.56 with 95% CI 1.44 to 1.67).
Transition probabilities between states over years for patient groups
From the fitted models presented in table 3, we estimated the probabilities of transition between the BC diagnosis, a first CD or ISH event and death states, for patients in each of the four molecular subtypes, in each of the tumour stages and 48, 60 and 70 years old, all given radiation treatment and other treatments assumed to have been given according to guidelines.14 Estimated probabilities of transition between the BC diagnosis and CD and ISH states are displayed in figures 2 and 3, respectively. Overall, the transition probabilities increased with age and were generally higher for patients with stage III disease. Differences between molecular subtypes were observed across all age and stage groups. For transition between the BC diagnosis and the CD state, the highest probabilities were generally observed for patients with the HR–/HER2– subtype, whereas for transition between the BC diagnosis and the ISH state, the highest probabilities were generally observed for patients with the HR–/HER2+ subtype.
Figure 2. Estimated probabilities of transition between the BC diagnosis and CD states, for patients with a given molecular subtype (all four), tumour stage (all three) and age (48, 60 and 70 years old). BC, breast cancer; CD, cardioembolic disorder.
Discussion
The proportions of patients receiving radiation, hormone therapy and chemotherapy were similar across the two analyses (table 1). As expected, mortality risk increased over time, with similar patterns observed for both analyses (table 2).
In the MSM analyses (table 1 and figures 2 and 3), HRs for a first ISH event were consistently higher across molecular subtypes differing from the HR+/HER2− baseline compared with those for a first CD event, except for the HR−/HER2− subtype where the HRs were similar. This was also true for tumour stage II patients compared with the tumour stage I baseline, while for tumour stage III patients the HRs were similar.
Figure 3. Estimated probabilities of transition between the BC diagnosis and ISH states, for patients with a given molecular subtype (all four), tumour stage (I–III) and age (48, 60 and 70 years old). BC, breast cancer; ISH, ischaemic or structural heart.
The ISH group includes MCI and HF, which are directly related to the heart and particularly sensitive to cardiotoxicity from BC treatments. Even though the risk for a CD and ISH event is elevated in BC patients, ISH conditions are often more prevalent and persistent, largely due to treatment-related vascular damage, shared metabolic risk factors for BC such as hypertension, diabetes, obesity and smoking, and chronic structural changes in the heart.16 These differences are expressed in the HRs for the patient’s characteristics and are further discussed in the next sections.
Comparing the risk for CD and ISH for the four molecular subgroups
Comparing HRs among patients in different molecular subtypes, we found HR+/HER2+ and HR–/HER2+ patients, relative to the HR+/HER2− patients, to have higher (and significant) HRs for a first ISH event compared with a first CD event, which could be explained by the different treatments given to patients in the molecular subgroups. Both the HR+/HER2+ and HR–/HER2+ patients received trastuzumab treatment, a key HER2-targeted agent used in the treatment of HER2+ patients, which is characterised by a potential well-known cardiotoxicity risk.17 Furthermore, these HER2+ patients received chemotherapy alongside trastuzumab, resulting in both direct and cumulative cardiotoxic effects that contribute to long-term myocardial damage, giving a particular risk for ISH events.
The comparable and significantly elevated risks of developing a first CD (HR: 1.51 with 95% CI 1.34 to 1.65) and ISH (HR: 1.56 with 95% CI 1.44 to 1.66) among HR−/HER2− patients, relative to HR+/HER2− patients, could be explained by the general cytotoxic effects from chemotherapy, particularly by the addition of taxanes, more commonly given to HR−/HER2− patients.18 HR−/HER2− patients did not receive trastuzumab and therefore their risk was lower for developing a first ISH event compared with the HR+/HER2+ and HR–/HER2+ patients who had a somewhat higher hazard due to the combination of trastuzumab and chemotherapy treatment. This is illustrated in figure 2 for patients 48, 60 and 75 years of age across all tumour stages. This displays the relatively high probability for a first CD event for HR−/HER2− patients, consistent over the 15 years of follow-up, except for the oldest patients in tumour stage III. In figure 3, however, those with highest probability for a first ISH event were HR−/HER2+patients followed closely by HR+/HER2+ patients. This finding is also consistent over the ages, tumour stages and over the 15-year follow-up time.
Comparing the risk for CD and ISH for patients diagnosed in tumour stage I–III
The higher HR for a first ISH compared with a first CD event for tumour stage II versus tumour stage I patients may reflect the progressive nature of structural heart changes and ischaemic risk, due to more intensive treatment for stage II patients. Also, as previously noted, risk factors for ISH events share many risk factors for BC. Risk factors for CD events could, on the other hand, be more influenced by acute triggers, coagulation status, and arrhythmogenic potential, which may not be directly linked to long-term cancer therapy effects.
Radiation therapy and CVD hazard
Radiation therapy was linked to a reduced hazard of both CVD and death. This finding should be interpreted with caution, as it may reflect residual confounding. Patients selected for radiation therapy may have been relatively healthier, typically with stage I tumours and fewer comorbidities, characteristics that are also associated with a lower risk of CVD and mortality. In addition, patients receiving radiation therapy may have undergone closer clinical follow-up, potentially contributing to earlier detection and management of health conditions. Because detailed information on comorbidities, lifestyle factors, and treatment selection was unavailable, we cannot exclude the possibility that unmeasured confounding contributed to the observed associations.
Long-term implications
Short- and longer-term transition probabilities (table 2 and figures 2 and 3) indicated an elevated risk for both a first CD and ISH event following BC diagnosis. This could be due to a combination of delayed treatment-related cardiotoxicity, ageing-related vulnerability, and under-recognised long-term effects. Therapies such as anthracyclines and trastuzumab can cause progressive cardiac damage that may not manifest until later. Hence, this suggests that the risk for cardiovascular events after BC treatment is not confined to the treatment period and highlights the need to integrate long-term cardiovascular monitoring as part of the routine follow-up.
Limitations
Information on the patients’ relevant health and lifestyle factors was not available. Unfortunately, we had no detailed information on patients’ treatments over time. Also, we had no information whether patients entered a higher tumour stage over time. Although treatment protocols on BC are relatively standardised, variations can exist between patients. The analyses would have been more interesting if we had access to the patients given treatment combinations, sequence of therapies, number of treatment lines and further treatment information at given dates. In summary, the lack of detailed patient information is a limitation. On the other hand, having nationwide patient data is a clear strength. We believe our analyses and findings are of interest, but they should, of course, be interpreted within the boundaries these data impose.
Only the year of diagnosis was available for analysis, and not the exact date, perhaps for pseudonymisation purposes. We acknowledge that setting the first day of the year as the time of diagnosis may introduce a modest overestimation of survival time. Similarly, only the year of death was available; assuming survival of half a year is a standard approximation in such settings and is unlikely to introduce systematic bias, although it remains a limitation of the analyses.
Nowadays, the increasing use of immune checkpoint inhibitors, such as pembrolizumab, for triple-negative BC patients also introduces further risk of cardiotoxicity. These therapies were not widely used during the study period, but future studies should also address this.
Another limitation arises from the decision to perform separate analyses of first CD and first ISH event. Because the two outcome groups may reflect different biological mechanisms, risk factor profiles, treatment associations and prognostic trajectories, patients who experienced both CD and ISH events during follow-up were excluded and may be viewed as a distinct, clinically more complex subgroup which was not the focus of our analyses. We acknowledge that this approach may introduce some selection related to survival and follow-up duration, since patients who survive longer have greater opportunity to experience both event types and therefore are not part of the analyses. Consequently, the results should be interpreted as applying to patients whose observed CVD trajectory remained within one of the two predefined outcome categories during follow-up.
Conclusions
Our findings suggest that the long-term cardiovascular burden for BC patients may be more strongly influenced by chronic vascular and myocardial damage, likely driven by treatment-related cardiotoxicity and cumulative exposure compared with acute thromboembolic mechanisms. An MSM approach proved valuable for capturing the dynamics for patients to move between the states BC diagnosis, a first CD or ISH and death, and may guide more tailored follow-up and use of cardioprotective strategies.
Supplementary material
Acknowledgements
We thank the Norwegian Cancer Registry and the Norwegian Patient Registry for providing the data for these analyses. We thank the referees for their insightful and valuable comments. Co-author MK was employed at the University of Oslo, Department of Pharmacology during the research and making of this paper, and is now retired from the University of Oslo. We are grateful to the Academic Editor and the four reviewers for their valuable and insightful comments, which helped improve the manuscript.
The Foundation Dam had no role in study design, data collection and analysis, decision to publish or preparation of the manuscript.
Footnotes
Funding: The study was financially supported by Foundation Dam, project number 389462.
Prepublication history and additional supplemental material for this paper are available online. To view these files, please visit the journal online (https://doi.org/10.1136/bmjopen-2026-117647).
Provenance and peer review: Not commissioned; externally peer reviewed.
Patient consent for publication: Not applicable.
Data availability free text: The data are Norwegian public health registry data, only available on application. Under no circumstances are we permitted to make the data publicly available online. This applies as a rule to all Norwegian health registry data. Applications for access to health registry data for research, analysis and quality improvement must be submitted through the national portal: https://helsedata.no/en/.
Patient and public involvement: Patients and/or the public were not involved in the design, or conduct, or reporting, or dissemination plans of this research.
Data availability statement
No data are available.
References
- 1.Armenian SH, Xu L, Ky B, et al. Cardiovascular Disease Among Survivors of Adult-Onset Cancer: A Community-Based Retrospective Cohort Study. J Clin Oncol. 2016;34:1122–30. doi: 10.1200/JCO.2015.64.0409. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 2.Ramin C, Schaeffer ML, Zheng Z, et al. All-Cause and Cardiovascular Disease Mortality Among Breast Cancer Survivors in CLUE II, a Long-Standing Community-Based Cohort. JNCI . 2021;113:137–45. doi: 10.1093/jnci/djaa096. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 3.Ades F, Zardavas D, Pinto AC, et al. Cardiotoxicity of systemic agents used in breast cancer. Breast. 2014;23:317–28. doi: 10.1016/j.breast.2014.04.002. [DOI] [PubMed] [Google Scholar]
- 4.Albini A, Pennesi G, Donatelli F, et al. Cardiotoxicity of Anticancer Drugs: The Need for Cardio-Oncology and Cardio-Oncological Prevention. JNCI. 2010;102:14–25. doi: 10.1093/jnci/djp440. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 5.Vaidya A, Vaidya P, Both B, et al. Health economics of targeted intraoperative radiotherapy (TARGIT-IORT) for early breast cancer: a cost-effectiveness analysis in the United Kingdom. BMJ Open. 2017;7:e014944. doi: 10.1136/bmjopen-2016-014944. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 6.Cheung LC, Albert PS, Das S, et al. Multistate models for the natural history of cancer progression. Br J Cancer. 2022;127:1279–88. doi: 10.1038/s41416-022-01904-5. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 7.Tvete IF, Klemp M. Investigating Time-Varying Predictor Effects on Cardiovascular Outcomes in Breast Cancer Survivors. J Breast Cancer. 2026;29:68–80. doi: 10.4048/jbc.2025.0151. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 8.Tvete IF, Klemp M. Temporal dynamics of prognostic factors in breast cancer survival. PLoS One. 2026;21:e0349115. doi: 10.1371/journal.pone.0349115. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 9.Cancer registry of Norway. https://www.fhi.no/en/cancer/ n.d. Available.
- 10.The Norwegian patient registry. https://helsedata.no/en/forvaltere/norwegian-institute-of-public-health/norwegian-patient-registry-npr/ n.d. Available.
- 11.Camen S, Csengeri D, Geelhoed B, et al. Risk Factors, Subsequent Disease Onset, and Prognostic Impact of Myocardial Infarction and Atrial Fibrillation. J Am Heart Assoc. 2022;11:e024299. doi: 10.1161/JAHA.121.024299. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 12.Kamel H, Healey JS. Cardioembolic Stroke. Circ Res. 2017;120:514–26. doi: 10.1161/CIRCRESAHA.116.308407. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 13.Jackson C. Multi-state models for panel data: the msm package for R. J Stat Softw. 2011:1–29. doi: 10.18637/jss.v038.i08. [DOI] [Google Scholar]
- 14.National action program with guidelines for diagnosis, treatment and follow-up of patients with breast cancer. https://www.helsedirektoratet.no/retningslinjer/brystkreft-handlingsprogram n.d. Available.
- 15.R software: rstudio team RStudio: integrated development for R. RStudio, PBC, Boston, MA. 2020. http://www.rstudio.com/ Available.
- 16.Leoce NM, Jin Z, Kehm RD, et al. Modeling risks of cardiovascular and cancer mortality following a diagnosis of loco-regional breast cancer. Breast Cancer Res. 2021;23:91. doi: 10.1186/s13058-021-01469-w. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 17.Dauccia C, Agostinetto E, Arecco L, et al. Cardiovascular toxicity of breast cancer treatments: from understanding to enhancing survivorship care. ESMO Open. 2025;10:105128. doi: 10.1016/j.esmoop.2025.105128. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 18.Fitrianti AE, Wardani NO, Astuti A, et al. Cardiotoxicity in Breast Cancer Therapy: Risks, Mechanisms, and Prevention Strategies. Med Sci (Basel) 2025;13:130. doi: 10.3390/medsci13030130. [DOI] [PMC free article] [PubMed] [Google Scholar]