Background
Cardiovascular diseases (CVD), including coronary artery disease, stroke, aortic and peripheral arterial diseases and venous thromboembolism (VTE), are associated with malignant tumour diagnoses and their treatment [
1‐
5]. There is emerging evidence of increased CVD risk in patients with malignant or non-malignant brain tumours. Studies on population-based cancer registry data showed a higher standardised mortality ratio of CVDs in people with brain tumour [
6,
7]. Survivors of malignant brain tumour at 1 year are at the highest risk of stroke compared to people without diagnosed cancer [
8]. Strategies to reduce CVD risks in brain tumour patients may therefore provide an opportunity to improve life and survival after tumour diagnosis. However, a better understanding of CVD risks is needed before prospectively assessing the efficacy of primary and secondary CVD prevention in people with brain tumour.
Brain tumours are a heterogeneous group of tumours comprising over 130 different types [
9]. Survival is different by tumour subtype. For example, 5-year survival is 4% for glioblastoma [
10] and 88% for benign meningiomas [
11]. Each brain tumour subtype also has specific management strategies and, likewise, CVD risks may vary between brain tumour subtypes. Recent population-based studies using cancer registry data in the United Kingdom (UK) and the United States have reported elevated standardised mortality ratio of CVD for people with brain tumours [
6,
7]. However, these findings related only to fatal cardiovascular events and did not account for baseline cardiovascular risk profiles. Short-term incidences of fatal and non-fatal CVDs within 1 year of tumour diagnosis, in contrast to medium- and long-term risks [
8], have not been reported. Patients usually undergo treatment within months of diagnosis, so quantifying CVD risks from time of tumour diagnosis may provide a more comprehensive understanding about the effect of therapies on CVDs.
This study assessed the association between brain tumour diagnosis and CVD by tumour subtypes and estimated the incidences of CVD from tumour diagnosis compared to matched controls leveraging population-based routine healthcare data collected in Wales, UK.
Methods
Patient and public involvement (PPI)
No formal PPI was undertaken for this study. However, the James Lind Alliance priority setting partnership on brain and spinal cord tumour identified long-term physical and cognitive effects of surgery and radiotherapy as one of the top 10 priorities. This study estimated the cardiovascular risks and VTE after brain tumour diagnosis compared to people without any cancer diagnosis. We also present these risks among brain tumour patients stratified by surgery status.
Study design and setting
This is a retrospective matched cohort study based in Wales (UK) using data from the Secure Anonymised Information Linkage (SAIL) Databank. Wales has a population of 3.1 million. The SAIL Databank contains whole-population anonymised individual-level routine healthcare data linkable to disease registries and national statistics registers via pseudonymised unique identifiers. SAIL datasets used in this study included primary care, hospital care, cancer registry and administrative datasets. All datasets have national coverage except the primary care dataset, which covers approximately 75% of the Welsh population.
Selection and matching of patients
We used the cancer registry in SAIL Databank to identify adult patients aged ≥ 18 years with an incident primary intracranial tumour. Patients were eligible if their tumour was diagnosed between 2000 and 2014, and they had primary care data available from at least 1 year before index brain tumour diagnosis. Each patient must have had at least one matched control available, where the date of brain tumour diagnosis was not the date of death. Matched controls were people without a diagnosis of any cancer who had an active registration at a general practitioner practice during 2000 and 2014. We used the cancer registry to confirm no cancer diagnosis in the control population. Controls needed to have primary care data available at least 1 year before study entry, which was the time of tumour diagnosis of the matched patient with a brain tumour. Matching variables included date of birth within 5 years, sex and GP practice at a ratio of one brain tumour patient to up to five controls without replacement.
Outcomes and variables
Code lists for all outcome and data variables are available on Open Science Framework (
https://doi.org/10.17605/OSF.IO/3FMY5). The primary outcome of interest was fatal and non-fatal major vascular events and VTE (deep vein thrombosis and pulmonary embolism). Major vascular events included haemorrhagic stroke (intracerebral haemorrhage and subarachnoid haemorrhage), ischaemic stroke, unspecified stroke, ischaemic heart disease (angina and myocardial infarction) and aortic and peripheral vascular disease. Events were defined as fatal if they were the direct cause of death on the death certificate or death occurred within 30 days of the outcome. We collected time-to-event data from date of study entry to date of event for all outcome variables. Patients may have more than one outcome event.
Data items included the following: demographics (age at study entry, sex, Welsh Index of Multiple Deprivation [WIMD] quintile, year of study entry), medical history (previous stroke, VTE, ischaemic heart disease (IHD), hypertension, hyperlipidaemia, diabetes mellitus) and medications (antihypertensives, antiplatelet, anticoagulants, lipid-lowering drugs). For people with brain tumours, we also collected tumour behaviour (malignant or non-malignant), tumour types (glioblastoma and World Health Organisation [WHO] grade 1 non-malignant meningioma) and surgery status. WIMD is a postcode-based measure of relative deprivation in Wales published by the Welsh Government. Lower WIMD represents higher deprivation. Participants were recorded as being on a medication if there were at least 6 prescriptions of that class of medication before the date of study entry. For medical history of hyperlipidaemia, we combined those with coding of hyperlipidaemia with those on lipid lowering drugs. Diabetes status used both coding of diabetes and antidiabetic drugs. Tumour behavioural code in the International Classification of Diseases for Oncology Third Edition (ICD-O-3) determined malignancy where /3 denoted malignant tumours and /0–1 denoted non-malignant tumours. The binary surgery status was determined by data from cancer registry, procedural codes and histological diagnosis as surgery is the only means for histological diagnosis. The end of follow-up is defined as date of death, date of deregistration from primary care or end of study period (31 December 2018) for those alive and registered with primary care.
Data sources and measurement
Data sources used in this study included primary care general practitioner dataset, Welsh demographic service, Patient Episode Database for Wales, Welsh Cancer Intelligence and Surveillance Unit (cancer registry) and Annual District Death Extract. We used Clinical Terms Version 2 READ codes to extract information from primary care data. Diagnoses using the International Classification of Diseases version 10 (ICD-10) codes and procedures using the Office of Population Censuses and Surveys Classification of Surgical Operations (OPCS-4) codes were used to define variables and outcomes in hospital data. Where available, we used code lists from related publications to define our variables (
https://doi.org/10.17605/OSF.IO/3FMY5). All outcomes and variables were captured using both primary care and hospital data. There was no formal sample size calculation, but the study design allowed detection of a hazard ratio of 1.20 with 80% power and 5% type 1 error in subgroup analyses. We planned to use all data available.
Statistical methods
In all analyses, follow-up started from study entry to the earliest occurrence of death, date of deregistration from primary care, end of study period or first occurrence of outcome of interest. We consider malignant and non-malignant tumours separately because of their different prognosis. Additionally, our analysis period was split into two: within 1 year of study entry and 1 year after study entry. This was because the rates of CVD events in patients with brain tumour were higher in the first year after brain tumour diagnosis. Therefore, our analyses were stratified by tumour behaviour and time periods.
We calculated crude incidences of each outcome for patients with brain tumours and their matched controls separately and generated the corresponding 95% confidence intervals (CI) using the Collett exact method. Cox regression was our primary statistical model to assess the association between brain tumour diagnosis and different CVDs. We initially fitted models including brain tumour diagnosis and the matching variables to account for the study design [
12] and then adjusted for risk factors listed in Table
1. To compare incidence trends of CVD in brain tumour patients with their controls by age, we fitted flexible parametric survival models with four degrees of freedom using the same covariates in the fully adjusted Cox model and then generated predicted incidences at ages 50, 65 and 75 years.
Table 1
Characteristics of 6800 brain tumour patients and their 33,785 age, sex and GP practice-matched controls by tumour behaviour
Number of participants | 40,585 | 6800 | 33,785 | 17,140 | 2869 | 14,271 | 23,445 | 3931 | 19,514 |
Age (median; IQR) | 63 (49, 75) | 63 (49, 74) | 63.0 (49, 75) | 64 (52, 74) | 64 (52, 74) | 64 (52, 74) | 62 (40, 76) | 62 (47, 75) | 62 (47, 76) |
Age-group |
18–50 years | 10,366 (25.5) | 1751 (25.8) | 8615 (25.5) | 3631 (21.2) | 610 (21.3) | 3021 (21.2) | 6735 (28.7) | 1141 (29.0) | 5594 (28.7) |
50–54 years | 3078 (7.6) | 525 (7.7) | 2553 (7.6) | 1295 (7.6) | 226 (7.9) | 1069 (7.5) | 1783 (7.6) | 299 (7.6) | 1484 (7.6) |
55–59 years | 3716 (9.2) | 653 (9.6) | 3063 (9.1) | 1661 (9.7) | 286 (10.0) | 1375 (9.6) | 2055 (8.8) | 367 (9.3) | 1688 (8.7) |
60–64 years | 4188 (10.3) | 693 (10.2) | 3495 (10.3) | 2022 (11.8) | 336 (11.7) | 1686 (11.8) | 2166 (9.2) | 357 (9.1) | 1809 (9.3) |
65–69 years | 4543 (11.2) | 801 (11.8) | 3742 (11.1) | 2297 (13.4) | 419 (14.6) | 1878 (13.2) | 2246 (9.6) | 382 (9.7) | 1864 (9.6) |
70–74 years | 4118 (10.1) | 684 (10.1) | 3434 (10.2) | 2063 (12.0) | 333 (11.6) | 1730 (12.1) | 2055 (8.8) | 351 (8.9) | 1704 (8.7) |
75–79 years | 4051 (10.0) | 643 (9.5) | 3408 (10.1) | 1868 (10.9) | 293 (10.2) | 1575 (11.0) | 2183 (9.3) | 350 (8.9) | 1833 (9.4) |
80–84 years | 3474 (8.6) | 568 (8.4) | 2906 (8.6) | 1425 (8.3) | 238 (8.3) | 1187 (8.3) | 2049 (8.7) | 330 (8.4) | 1719 (8.8) |
85 + years | 3051 (7.5) | 482 (7.1) | 2569 (7.6) | 878 (5.1) | 128 (4.5) | 750 (5.3) | 2173 (9.3) | 354 (9.0) | 1819 (9.3) |
Sex |
Male | 18,847 (46.4) | 3159 (46.5) | 15,688 (46.4) | 9854 (57.5) | 1650 (57.5) | 8204 (57.5) | 8993 (38.4) | 1509 (38.4) | 7484 (38.4) |
Female | 21,738 (53.6) | 3641 (53.5) | 18,097 (53.6) | 7286 (42.5) | 1219 (42.5) | 6067 (42.5) | 14,452 (61.6) | 2422 (61.6) | 12,030 (61.6) |
Year of study entry |
2000–2004 | 10,318 (25.4) | 1730 (25.4) | 8588 (25.4) | 4942 (28.8) | 828 (28.9) | 4114 (28.8) | 5376 (22.9) | 902 (22.9) | 4474 (22.9) |
2005–2009 | 13,619 (33.6) | 2280 (33.5) | 11,339 (33.6) | 5633 (32.9) | 942 (32.8) | 4691 (32.9) | 7986 (34.1) | 1338 (34.0) | 6648 (34.1) |
2010–2014 | 16,648 (41.0) | 2790 (41.0) | 13,858 (41.0) | 6565 (38.3) | 1099 (38.3) | 5466 (38.3) | 10,083 (43.0) | 1691 (43.0) | 8392 (43.0) |
WIMD |
I (most deprived) | 8293 (20.7) | 1318 (19.6) | 6975 (20.9) | 3413 (20.1) | 528 (18.6) | 2885 (20.5) | 4880 (21.0) | 790 (20.3) | 4090 (21.2) |
II | 8062 (20.1) | 1313 (19.5) | 6749 (20.2) | 3339 (19.7) | 529 (18.6) | 2810 (19.9) | 4723 (20.4) | 784 (20.1) | 3939 (20.4) |
III | 8290 (20.7) | 1395 (20.7) | 6895 (20.6) | 3543 (20.9) | 586 (20.6) | 2957 (21.0) | 4747 (20.5) | 809 (20.7) | 3938 (20.4) |
IV | 7500 (18.7) | 1302 (19.3) | 6198 (18.6) | 3200 (18.9) | 572 (20.2) | 2628 (18.6) | 4300 (18.5) | 730 (18.7) | 3570 (18.5) |
V (least deprived) | 8000 (19.9) | 1410 (20.9) | 6590 (19.7) | 3446 (20.3) | 623 (22.0) | 2823 (20.0) | 4554 (19.6) | 787 (20.2) | 3767 (19.5) |
Unknown | 440 | 62 | 378 | 199 | 31 | 168 | 241 | 31 | 210 |
Past medical history |
Hypertension | 12,710 (31.3) | 2185 (32.1) | 10,525 (31.2) | 5423 (31.6) | 876 (30.5) | 4547 (31.9) | 7287 (31.1) | 1309 (33.3) | 5978 (30.6) |
Diabetes mellitus | 4454 (11.0) | 676 (9.9) | 3778 (11.2) | 1973 (11.5) | 254 (8.9) | 1719 (12.0) | 2481 (10.6) | 422 (10.7) | 2059 (10.6) |
Hyperlipidaemia | 10,759 (26.5) | 1772 (26.1) | 8987 (26.6) | 4683 (27.3) | 717 (25.0) | 3966 (27.8) | 6076 (25.9) | 1055 (26.8) | 5021 (25.7) |
Heavy alcohol use | 1123 (2.8) | 123 (1.8) | 1000 (3.0) | 519 (3.0) | 64 (2.2) | 455 (3.2) | 604 (2.6) | 59 (1.5) | 545 (2.8) |
Major vascular events | 5914 (14.6) | 966 (14.2) | 4948 (14.6) | 2695 (15.7) | 444 (15.5) | 2251 (15.8) | 3219 (13.7) | 522 (13.3) | 2697 (13.8) |
Venous thromboembolism | 964 (2.4) | 174 (2.6) | 790 (2.3) | 385 (2.2) | 64 (2.2) | 321 (2.2) | 579 (2.5) | 110 (2.8) | 469 (2.4) |
Medication |
Antihypertensive drug(s) | 13,524 (33.3) | 2251 (33.1) | 11,273 (33.4) | 5845 (34.1) | 931 (32.5) | 4914 (34.4) | 7679 (32.8) | 1320 (33.6) | 6359 (32.6) |
Antiplatelet drug(s) | 8230 (20.3) | 1249 (18.4) | 6981 (20.7) | 3544 (20.7) | 483 (16.8) | 3061 (21.4) | 4686 (20.0) | 766 (19.5) | 3920 (20.1) |
Anticoagulant drug(s) | 1903 (4.7) | 260 (3.8) | 1643 (4.9) | 831 (4.8) | 94 (3.3) | 737 (5.2) | 1072 (4.6) | 166 (4.2) | 906 (4.6) |
We performed the same Cox models in subgroups of matched cohorts with glioblastoma and non-malignant meningioma for assessing whether incidence trends were reproducible in more homogeneous tumour groups. We did not examine effect modification because of anticipated low power in our dataset with relative few outcome events. Our primary multivariable analysis did not include body mass index (BMI) or smoking status because of relatively elevated levels of missingness (Additional File
1: Figure S1). The levels of missing for cases and controls were 20% and 13% in smoking status and 9% and 3% in BMI; these were higher (~ 40%) at the beginning of the study period. We therefore performed sensitivity analyses of multivariable Cox models that included BMI and smoking status to assess confounding effects of these variables. Separately, we repeated our models considering a stroke diagnosis occurring after 14 days of tumour diagnosis as valid. This was because brain tumour presentation can mimic stroke and hospital coding may not distinguish between them. We also performed adjusted competing risk analyses as sensitivity analyses to assess subhazard ratios and cumulative incidences in the presence of competing risk from death. We used complete case analysis throughout. In each time-to-event analysis, failure time was from study entry to the first occurrence of the specific outcome event. We presented the proportion of patients with major vascular events and VTE after tumour diagnosis stratified by tumour subgroups and antiplatelet status. No statistical analyses were applied to the outcome data stratified by tumour subgroups and antiplatelet status because this was exploratory and likely to be underpowered. Statistical analyses were performed in R version 4.1.1 using packages ‘
epiR’ (v2.0.19), ‘
survival’ (v3.2–13), ‘
rstpm2’ (v1.5.1) and ‘
cmprsk’ (v2.2–10).
Ethics and reporting
This project has received approval from the Information Governance Review Panel in SAIL Databank (project number: 0918). We used the STROBE checklist when writing our report.
Discussion
Comparing 6800 people diagnosed brain tumour to their matched controls from the general population, we showed that both malignant and non-malignant brain tumours were associated with a higher risk of major vascular events and VTE within the first year of tumour diagnosis. Stroke was the major vascular event with the highest increased risks. These were consistent in patients with glioblastoma or meningioma. The incidence of major vascular events was highest within first 2 months of tumour diagnosis. Incidence of VTE showed a bimodal distribution with peaks at 1 month and 5 months.
Our findings of increased stroke risks in brain tumour patients are consistent with studies in England and the US using population-based data [
3,
6,
8]. It is plausible that in the context of a confined intracranial space, a space-occupying lesion (tumour) increases intracranial pressure and reduces cerebral blood flow. Surgery, chemotherapy and radiotherapy can also affect cerebral vasculature [
13], together with the hypercoagulability state caused by tumours [
14,
15]. These effects could in turn manifest as an ischaemic stroke. While our data and that of others [
3,
6,
8] could not provide evidence for this mechanism, further investigation into the location of cerebral ischaemia and clinical syndrome in addition to baseline stroke risk profile would help to clarify this.
A population matched-cohort study in England reported 906 1-year survivors of malignant central nervous system tumours had a higher hazard of stroke compared to their matched controls (HR 4.42, 95% CI 2.54–7.72) [
8]. This study observed more CVD outcome events than our study, which would increase their power of detecting the association. Their outcome definition (
https://datacompass.lshtm.ac.uk/id/eprint/1113/) differed from ours because they included subdural haematoma (ICD-10 code: I62.0, READ codes: G621., G622., G623., 70,170) and extradural haemorrhage (ICD-10 code: I62.1, READ codes: G620., 70,320). Non-traumatic extradural haemorrhage is rare. In the atraumatic setting, most subdural haematomas are subacute or chronic, which do not have shared cardiovascular risk factors and clinical management with stroke [
16]. Including subdural haematoma would increase the number of outcomes for classified as stroke because the incidence per 100,000 person-year of chronic subdural haematoma is relatively high at ~ 48 [
17] compared to the incidence of stroke of ~ 200 for those aged > 65 [
18]. This difference in outcome definition may explain the association between brain tumour diagnosis and stroke in 1-year survivors observed in the English study. Furthermore, our study had relatively few outcome events and would be underpowered to detect this association. Future population-based studies with standardised coding of stroke and adequately powered cohort can address this uncertainty.
Advances in treatment for certain cancer types have improved the survival of many patients in high-income countries [
19]. Together with extended longevity, the risk of heart diseases associated with the cancer state and cancer treatment has fuelled the growth of cardio-oncology [
20]. Our previous study of brain tumours using population mortality data in the United States and Wales (UK) suggested higher mortality from heart diseases [
7]. However, findings from our current study do not suggest an elevated risk of fatal and non-fatal IHD (Fig.
2). This different observation may result from the matched design of this study. The propensity for patients with a brain tumour to survive a cardiac event may be lower compared to matched controls because of chemotherapy, radiotherapy and frailty. The data may also be affected by survival bias, increasing the risks in those without a brain tumour. Prospective studies with detailed clinical characteristics and treatment data can investigate this association further.
Risks of major vascular events and VTE can be reduced with antithrombotic drugs. We only presented descriptive statistics on CVD events by antiplatelet status because of few patients (Table
2). Higher proportions of patients on antiplatelets at the time of brain tumour diagnosis had major vascular events, which was expected as they are more likely to have a higher risk profile. But whether antiplatelet drugs were discontinued after surgery needs to be clarified in patients receiving surgery. Detailed treatment and follow-up data would be needed to consider whether antiplatelet drugs can reduce the risks of major vascular events as primary prevention. Whether other markers of risk could clarify the aetiology of major vascular events for brain tumour patients would be of interest in future research. For example, imaging markers such as small vessel disease [
21] and cerebral microbleeds [
22] may have additional prognostic value for estimating future risk, which warrants further investigation. The elevated VTE risk is complex because cancer-associated hypercoagulable state, reduced mobility, surgery and chemotherapy can contribute to VTE. A better understanding of these factors can inform stratified care for those at higher risk since there is evidence for using antithrombotic drugs to prevent VTE in cancer patients, including brain tumour patients [
23].
Strengths and limitations
Strengths of this analysis include the availability of medical comorbidities associated with CVD. We demonstrated that risks observed for all non-malignant tumours can be applied to meningiomas; the risk profiles of malignant tumours and glioblastoma are different, suggesting that tumour subtypes should be considered separately. Our results add to the existing literature by describing the risk of CVDs from the time of brain tumour diagnosis. We were unable to include body mass index and smoking status in our primary analyses due to low-quality data availability. However, given similar effect estimates when restricted to the subpopulation with these data, it is unlikely to be a key confounder for the associations observed. This interpretation is supported by data from the English matched-cohort study [
8], which found little confounding effect by these covariates [
8]. We did not report chemotherapy and radiotherapy for our cohort because of insufficient quality. Treatment data routinely collected into the cancer registry lacks specificity about agents, doses and cycles. Data linkage with detailed treatment data may offer more clinically relevant analyses. Post-operative VTE prophylaxis may differ in regimen, timing and duration, which could affect our CVD estimates for surgical patients. Analyses incorporating this data could clarify the effect of VTE prophylaxis and the risk of bleeding. Lastly, routine healthcare data can provide a large dataset but may be subjected to misclassification and may lack relevant clinical details such as BMI, alcohol history and physical activity.
Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article's Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit
http://creativecommons.org/licenses/by/4.0/. The Creative Commons Public Domain Dedication waiver (
http://creativecommons.org/publicdomain/zero/1.0/) applies to the data made available in this article, unless otherwise stated in a credit line to the data.
Publisher’s Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.