Coexisting atrial fibrillation and cancer: time trends and associations with mortality in a nationwide Dutch study

Abstract

Background and Aims

Coexisting atrial fibrillation (AF) and cancer challenge the management of both. The aim of the study is to comprehensively provide the epidemiology of coexisting AF and cancer.

Methods

Using Dutch nationwide statistics, individuals with incident AF (n = 320 139) or cancer (n = 472 745) were identified during the period 2015–19. Dutch inhabitants without a history of AF (n = 320 135) or cancer (n = 472 741) were matched as control cohorts by demographic characteristics. Prevalence of cancer/AF at baseline, 1-year risk of cancer/AF diagnosis, and their time trends were determined. The association of cancer/AF diagnosis with all-cause mortality among those with AF/cancer was estimated by using time-dependent Cox regression.

Results

The rate of prevalence of cancer in the AF cohort was 12.6% (increasing from 11.9% to 13.2%) compared with 5.6% in the controls; 1-year cancer risk was 2.5% (stable over years) compared with 1.8% in the controls [adjusted hazard ratio (aHR) 1.52, 95% confidence interval (CI) 1.46–1.58], which was similar by cancer type. The rate of prevalence of AF in the cancer cohort was 7.5% (increasing from 6.9% to 8.2%) compared with 4.3% in the controls; 1-year AF risk was 2.8% (stable over years) compared with 1.2% in the controls (aHR 2.78, 95% CI 2.69–2.87), but cancers of the oesophagus, lung, stomach, myeloma, and lymphoma were associated with higher hazards of AF than other cancer types. Both cancer diagnosed after incident AF (aHR 7.77, 95% CI 7.45–8.11) and AF diagnosed after incident cancer (aHR 2.55, 95% CI 2.47–2.63) were associated with all-cause mortality, but the strength of the association varied by cancer type.

Conclusions

Atrial fibrillation and cancer were associated bidirectionally and were increasingly coexisting, but AF risk varied by cancer type. Coexisting AF and cancer were negatively associated with survival.

Structured Graphical Abstract

Structured Graphical Abstract.

Prevalence, incidence, time trend, and association with all-cause mortality of coexisting atrial fibrillation and cancer in the Netherlands. AF, atrial fibrillation; aHR, adjusted hazard ratio; CI, confidence interval.

See the editorial comment for this article ‘Cancer begets atrial fibrillation … and vice versa?’, by D. Farmakis and G. Filippatos, https://doi.org/10.1093/eurheartj/ehae301.

Introduction

Atrial fibrillation (AF) and cancer are both prevalent conditions in the general population, representing major health burdens.^1,2 Atrial fibrillation is the most common sustained arrhythmia and is related to unfavourable outcomes such as stroke, heart failure, impaired quality of life, hospitalization, and death,³ while cancer has been the leading cause of death for many years.^4,5 In the past, both conditions were generally viewed as two distinct disease entities, but with cardio-oncology rapidly emerging as a new field, cardiovascular disorders such as arrhythmias are being increasingly recognized and considered in cancer patients.^6–9 The burden of the two conditions is expected to increase with population ageing and improvement in cancer survival,^10–12 and the coexistence of one condition has been shown to make management of the other condition more challenging.^13,14 Studies have shown that cancer patients face an increased risk of AF compared with those without cancer,^15–18 and several underlying mechanisms have been proposed, such as shared risk factors/pathophysiology and side effects of cancer treatment.^19,20 However, several knowledge gaps remain because currently available studies were generally derived from relatively outdated data and limited by a narrow cancer spectrum. It remains unknown how the burden of coexisting AF and cancer has changed in recent years and whether AF affects cancer prognosis.

In addition, the association between AF and cancer seems bidirectional,^12,21 as new-onset AF has also been reported as a predictor of incident cancer,^22,23 while this has been less frequently investigated so far. As far as we know, no large-scale investigation has been performed into both directions (i.e. AF diagnosed after cancer, and cancer diagnosed after AF) within the same population. Given the immediate need for such relevant knowledge, in this study, we aim to comprehensively provide the epidemiology of coexisting AF and cancer bidirectionally, including factors such as prevalence, incidence, time trends, and associations with survival.

Methods

We followed the Strengthening the Reporting of Observational Studies in Epidemiology reporting guideline for cohort studies.

Data sources

The study used data accessed from Statistics Netherlands (in Dutch ‘Centraal Bureau voor de Statistiek’, CBS). Statistics Netherlands is a Dutch governmental institution that gathers and links de-identified individual data from various nationwide data sources. In this study, we used nationwide data on household income, personal characteristics (i.e. birth date, sex, and immigration background), diagnoses registered within hospitalizations in Dutch hospitals retrieved from discharge letters, death statistics, and outpatient medication prescriptions. Unless otherwise specified, diseases/conditions (including AF and cancer) were identified by diagnoses registered within hospitalizations. Details about the data sources and coding systems used for variable identification are provided in Supplementary data online, Methods and Table S1.

Study design and study populations

The study used a cohort study design that included an incident AF cohort, an incident cancer cohort, and two corresponding matched control cohorts without AF or cancer, respectively, at baseline (Figure 1).

Figure 1.

Study design and inclusion of the study populations. *Details about the identification of Dutch inhabitants between 2015 and 2019 who were considered eligible for the study are provided in Supplementary data online, Figure S1. ^†A diagnosis of atrial fibrillation (or cancer) was considered incident when there was no previous diagnosis record of atrial fibrillation (or cancer) within the prior 5 years (see Supplementary data online, Figure S2). A total of 42 446 unique individuals from the source population were included in both the incident atrial fibrillation cohort and the incident cancer cohort. ^‡There were 305 999 unique individuals among the matched control cohort without atrial fibrillation history (for the incident atrial fibrillation cohort) and 454 940 unique individuals among the matched control cohort without cancer history (for the incident cancer cohort). A total of 25 221 unique individuals from the source population were included in both the control cohort for the incident atrial fibrillation cohort and the control cohort for the incident cancer cohort. #The follow-up started from the index date of each individual until 1 year after, or when the outcome event (i.e. cancer or atrial fibrillation) was first diagnosed, or until the date of death, whichever came first. For the matched control cohort, they would also be censored when they were diagnosed with atrial fibrillation (or cancer)

Before identifying the specific cohorts, we first identified the Dutch inhabitants between the years 2015 and 2019 who were considered eligible for the study (i.e. the source population). Eligible participants had to be registered in the nationwide data on household income of the year upon entry in the study (i.e. one of the years between 2015 and 2019) and in the nationwide data on personal characteristics of the year upon entry in the study as well as the prior 5 years.

From the source population, by examining data on diagnoses registered within hospitalizations, we identified individuals who had a diagnosis of AF or cancer for the first time between 1 January 2015 and 31 December 2019 as the incident AF cohort or the incident cancer cohort, respectively, after excluding those with diagnosis records of AF or cancer in the prior 5 years. For each individual in either cohort, the admission date of the hospitalization in which AF or cancer was diagnosed for the first time was referred to as the index date (i.e. baseline) of the individual.

For each individual in the cohorts, we randomly sampled an individual from the source population as a control among those who met all of the following criteria: (i) alive on the index date; (ii) with the same age (i.e. an absolute difference in birth date ≤6 months), sex, immigration background, and level of standardized household income; and (iii) without a previous diagnosis record of AF (when matching for the incident AF cohort) or cancer (when matching for the incident cancer cohort) within 5 years before the index date (inclusive). A control would share the same index date of the individual from the incident AF/cancer cohort to whom he/she was matched. The sampling was with replacement.

Details about the identification of the source population and the cohorts are provided in Supplementary data online, Methods and Figures S1 and S2.

Determination of prevalent/incident cancer/atrial fibrillation

For the incident AF cohort and the corresponding matched control cohort (without AF history), we examined data on diagnoses registered within hospitalizations within 5 years before the index dates to determine whether an individual had prevalent cancer (i.e. cancer history) at baseline. If an individual happened to have a diagnosis record of cancer for the first time within the same hospitalization in which the index AF diagnosis was made, the cancer diagnosis would still be categorized as a prevalent diagnosis. For those without prevalent cancer at baseline, we further determined whether they would be diagnosed with (incident) cancer by following them from the index date until the date of death, the admission date of the hospitalization in which cancer was diagnosed for the first time, or 1 year later, whichever came first. For the matched control cohort, subjects would be censored when they were diagnosed with AF during follow-up.

Similarly, in the incident cancer cohort and the corresponding matched control cohort (without cancer history), we determined whether an individual had prevalent AF (i.e. AF history) at baseline or would be diagnosed with (incident) AF within 1 year after the incident cancer diagnosis.

The investigated cancer types and other baseline characteristics

According to the diagnosis codes, we categorized all cancers into about 30 different types (see Supplementary data online, Table S2). In addition, we identified the following baseline characteristics: age, sex, immigration background, standardized household income, CHA₂DS₂-VASc score, HAS-BLED score [i.e. uncontrolled hypertension (1 score), chronic kidney diseases (1 score), abnormal liver function (1 score), ischemic stroke/transient ischemic attack (1 score), major bleeding (1 score), aged >65 years (1 score), alcohol abuse (1 score), use of antiplatelet agents or nonsteroidal anti-inflammatory drugs (1 score)], and various comorbidities (or medical history), including asthma, chronic obstructive pulmonary disease, other chronic lung diseases, heart failure, myocardial infarction, hypertension, rheumatic mitral stenosis/mechanical heart valves, other valvular heart diseases, peripheral artery diseases, liver diseases, gastro-oesophageal reflux disease, peptic ulcer disease, chronic kidney diseases, anaemia, coagulopathy, diabetes, thyroid diseases, ischaemic stroke, transient ischaemic attack, systemic arterial thromboembolism, Parkinson’s disease, Alzheimer’s disease, autoimmune diseases, systemic connective tissue disorders, venous thromboembolism, and major bleeding. Details about these identifications are provided in Supplementary data online, Methods.

Statistical analysis

Baseline characteristics are presented as mean ± standard deviation or as numbers and percentages. The prevalence of cancer/AF at baseline in the incident AF/cancer cohort or the corresponding control cohort was calculated as the number of individuals with prevalent cancer/AF divided by the number of individuals in the cohort. The prevalence of cancer in the incident AF cohort (and the corresponding control cohort) was also presented per cancer type. Time trends in the prevalence were examined after stratifying the study cohorts by calendar year of the index dates.

After excluding those with prevalent cancer/AF at baseline, the incidence rates of cancer/AF diagnosis in the incident AF/cancer cohort or the corresponding control cohort were calculated as the number of individuals who were diagnosed with cancer/AF during the 1-year follow-up divided by the total amount of observation time. In addition, the 1-year cumulative incidence of cancer/AF diagnosis and the corresponding cumulative incidence curve of a cohort were estimated by the cumulative incidence competing risk method, in which all-cause death was considered as a competing event. For the matched control cohorts, a subject would be censored when the individual was diagnosed with the condition of the comparison cohort (i.e. AF or cancer), which was also treated as a competing event. To compare the risk of cancer or AF between the cohorts, cause-specific Cox regression was used to estimate the hazard ratio (HR) and 95% confidence intervals (CIs). Three pre-specified adjustment models were employed: Model 1, adjusting for age, sex, immigration background, and standardized household income; Model 2, adjusting for Model 1, plus CHA₂DS₂-VASc score and HAS-BLED score; Model 3, adjusting for Model 1, plus the above-mentioned various comorbidities (or medical history), including asthma, chronic obstructive pulmonary disease, other chronic lung diseases, heart failure, myocardial infarction (history), hypertension, rheumatic mitral stenosis/mechanical heart valves, other valvular heart diseases, peripheral artery diseases, liver diseases, gastro-oesophageal reflux disease, peptic ulcer disease, chronic kidney diseases, anaemia, coagulopathy, diabetes, thyroid diseases, ischaemic stroke (history), transient ischaemic attack, systemic arterial thromboembolism, Parkinson’s disease, Alzheimer’s disease, autoimmune diseases, systemic connective tissue disorders, venous thromboembolism, and major bleeding. All these analyses were also repeated after stratifying the study cohorts by cancer type. Time trends in the incidence of cancer/AF diagnosis were examined after stratifying the study cohorts by calendar year of the index dates.

Cancer/AF diagnosed during the 1-year follow-up was further treated as a time-dependent exposure to estimate its association with all-cause mortality in the incident AF/cancer cohort (after excluding those with cancer/AF at baseline), and time-dependent Cox regression (i.e. the Mantel–Byar method) was employed to estimate the HRs and 95% CIs with the above-mentioned adjustment models. The control cohorts were not involved in this analysis.

All statistical analyses were performed using SPSS® Statistics²⁴ and R programme²⁵.

Results

Baseline characteristics of the study cohorts

A total of 320 139 individuals were included as the incident AF cohort, in whom 320 135 subjects were matched as the control cohort without AF history, with a mean age of 74.6 ± 11.9 years and a male proportion of 55.8%. Compared with the control cohort, the incident AF cohort had a higher CHA₂DS₂-VASc score and HAS-BLED score and a higher prevalence of all the investigated comorbidities [hypertension (39.1%), heart failure (21.1%), and diabetes (19.4%)]. With regard to the types of the incident AF (in the incident AF cohort), most (74.6%) were unspecified AF. Details are provided in Supplementary data online, Table S3.

A total of 472 745 individuals were included as the incident cancer cohort, and 472 741 subjects were matched as the control cohort without a cancer history, with a mean age of 67.3 ± 13.8 years and a male proportion of 50.6%. Compared with the control cohort, the incident cancer cohort had a slightly higher CHA₂DS₂-VASc score and HAS-BLED score and a higher prevalence of all the investigated comorbidities [hypertension (18.0%), diabetes (10.8%), and anaemia (8.2%)]. With regard to the types of the incident cancer (in the incident cancer cohort), cancers of the breast (21.5% among the female), prostate (13.9% among the male), and colon/rectum (11.0%) were the most frequent (see Supplementary data online, Table S3).

Prevalence of cancer in the incident atrial fibrillation cohort and prevalence of atrial fibrillation in the incident cancer cohort

As presented in Figure 2 and Supplementary data online, Table S3, 12.6% of individuals in the incident AF cohort had prevalent cancer at baseline, among which cancers of ill-defined or multiple sites (4.3%), prostate (2.7%, among the male), breast (2.3%, among the female), colon/rectum (2.1%), and lung (2.0%) were the most prevalent types. The control cohort without AF history had a similar distribution of prevalent cancer types, but the prevalence (overall 5.6%) was generally lower than that in the incident AF cohort, particularly for cancers of the lung (0.3% vs. 2.0%), leukaemia (0.1% vs. 0.6%), oesophagus (0.1% vs. 0.5%), lymphoma (0.2% vs. 0.8%), and myeloma (0.1% vs. 0.4%).

Figure 2.

The prevalence of cancer among the incident atrial fibrillation cohort and the prevalence of atrial fibrillation among the incident cancer cohort vs. that in the control cohorts. For readability, the names of cancer types in the figure were shortened. Detailed descriptions can be found in Supplementary data online, Table S2. CNS, central nervous system

In the incident cancer cohort, 7.5% had AF history at baseline, while in the matched control cohort without cancer history, this was only 4.3%.

Incidence of cancer diagnosis in the incident atrial fibrillation cohort and incidence of atrial fibrillation diagnosis in the incident cancer cohort

After excluding individuals (n = 40 224) with prevalent cancer from the incident AF cohort, 2.54% (95% CI 2.48%–2.60%) was diagnosed with cancer within 1 year, while this was 1.80% (95% CI 1.75%–1.84%) in the control cohort without AF history, leading to an adjusted HR (aHR, by Model 3) of 1.52 (95% CI 1.46–1.58). Cancer was diagnosed more frequently in the first 3 months after the incident AF diagnosis than in the later months (Figure 3 and Supplementary data online, Table S4), and such a time course was consistently observed for most cancer types (see Supplementary data online, Figures S3–S32). The most frequently diagnosed cancer types were the same between the two cohorts (with similar relative risk estimates, Figure 4), namely cancers of ill-defined or multiple sites, prostate (among the male), breast (among the female), colon/rectum, and lung.

Figure 3.

Cumulative incidence curves for cancer diagnosed after incident atrial fibrillation or atrial fibrillation diagnosed after incident cancer vs. that in the control cohorts. Individuals who had a history of cancer (or atrial fibrillation) at baseline were excluded from this analysis. The cumulative incidence curves were plotted using the cumulative incidence competing risk method, in which all-cause death was considered as a competing event. The control cohorts would also be censored when atrial fibrillation (or cancer) was diagnosed during the 1-year follow-up, which was also considered as a competing event

Figure 4.

The one-year cumulative incidence of cancer diagnosed after incident atrial fibrillation vs. that in the control cohort. For readability, the names of cancer types in the figure were shortened. Detailed descriptions can be found in Supplementary data online, Table S2. Individuals who had a history of cancer at baseline were excluded from this analysis. The cumulative incidence was estimated by the cumulative incidence competing risk method, in which all-cause death was considered as a competing event. The control cohort would also be censored when atrial fibrillation was diagnosed during the 1-year follow-up, which was also considered as a competing event. The hazard ratios refer to the hazard of cancer diagnosis (overall or of a specific type) in the incident atrial fibrillation cohort compared with that in the control cohort, which were estimated by cause-specific Cox regression, after adjusting for age, sex, immigration background, standardized household income, and various comorbidities (or medical history). CNS, central nervous system; CI, confidence interval; HR, hazard ratio

After excluding individuals (n = 35 483) with prevalent AF from the incident cancer cohort, 2.84% (95% CI 2.79%–2.89%) was diagnosed with AF within 1 year, which was 1.19% (95% CI 1.16%–1.22%) in the control cohort without cancer history, with an aHR of 2.78 (95% CI 2.69–2.87). Atrial fibrillation was also diagnosed more frequently in the early stage after the incident cancer diagnosis than in the later stage, particularly in the first 3 months (Figure 3 and Supplementary data online, Table S5). When stratifying by cancer type, such a time course of AF diagnosis was consistently observed for most cancer types (see Supplementary data online, Figures S3–S35), but for individuals with cancer of the oesophagus, AF appeared to be diagnosed most frequently in the third–sixth month after the incident cancer diagnosis (see Supplementary data online, Figure S4). Although individuals with any of the investigated cancer types generally had a higher risk of AF diagnosis than the control cohort (Figure 5), several cancer types, including cancer of the oesophagus (aHR 9.63, 95% CI 8.89–10.43), lung (aHR 6.27, 95% CI 5.92–6.64), stomach (aHR 5.73, 95% CI 5.07–6.47), myeloma (aHR 5.18, 95% CI 4.65–5.77), and lymphoma (aHR 4.24, 95% CI 3.91–4.59), yielded higher relative risk estimates of AF diagnosis than the other cancer types.

Figure 5.

The one-year cumulative incidence of atrial fibrillation diagnosed after incident cancer vs. that in the control cohort. For readability, the names of cancer types in the figure were shortened. Detailed descriptions can be found in Supplementary data online, Table S2. Individuals who had a history of atrial fibrillation at baseline were excluded from this analysis. The cumulative incidence was estimated by the cumulative incidence competing risk method, in which all-cause death was considered as a competing event. The control cohort would also be censored when cancer was diagnosed during the 1-year follow-up, which was also considered as a competing event. The hazard ratios refer to the hazard of atrial fibrillation diagnosis in the incident cancer cohort (overall or of a specific type) compared with that in the control cohort, which were estimated by cause-specific Cox regression, after adjusting for age, sex, immigration background, standardized household income, and various comorbidities (or medical history). For female- or male-specific cancer (as indicated), only the female or male individuals in the control cohort were included as the reference group. CNS, central nervous system; CI, confidence interval; HR, hazard ratio

Details about baseline characteristics of the study cohorts after excluding those with prevalent cancer/AF and the results of incidence analysis are provided in Supplementary data online, Tables S6–S8.

Time trends in coexisting atrial fibrillation and cancer

When the study cohorts were stratified by calendar year of the index dates, the prevalence of cancer upon incident AF diagnosis increased from 11.9% in 2015 to 13.2% in 2019, while such a trend was also observed in the matched control cohort without AF history (i.e. from 5.4% to 5.7%). Similarly, the increasing prevalence of AF upon incident cancer diagnosis was found as well as in the control cohort without cancer history (i.e. from 6.9% to 8.2% and from 3.9% to 4.6%, respectively). The incidence of cancer (or AF) diagnosed after incident AF (or cancer), however, remained constant for years (Figure 6). Details about baseline characteristics of the study cohorts in different calendar years and the results of time trend analyses are provided in Supplementary data online, Tables S9–S18.

Figure 6.

Time trends in coexisting atrial fibrillation and cancer. The hazard ratios refer to the hazard of cancer/atrial fibrillation diagnosis in the incident atrial fibrillation/cancer cohort diagnosed in different calendar years compared with that of the same cohort diagnosed in 2015, which were estimated by cause-specific Cox regression, after adjusting for age, sex, immigration background, standardized household income, and various comorbidities (or medical history). CI, confidence interval; HR, hazard ratio

Association between cancer/atrial fibrillation diagnosed after incident atrial fibrillation/cancer and all-cause mortality

As presented in Figure 7 and Supplementary data online, Table S19, for individuals in the incident AF cohort who had no cancer history at baseline, cancer diagnosed within 1 year (as a time-dependent exposure) was associated with increased all-cause mortality (aHR 7.77, 95% CI 7.45–8.11). The association was the strongest for cancers of the brain (aHR 25.99), other lymphoid/haematopoietic tissue (aHR 16.18), mesothelial/soft tissue (aHR 7.49), ill-defined or multiple sites (aHR 7.10), and lymphoma (aHR 7.08).

Figure 7.

The association of cancer/atrial fibrillation diagnosed after incident atrial fibrillation/cancer with all-cause mortality. For readability, the names of cancer types in the figure were shortened. Detailed descriptions can be found in Supplementary data online, Table S2. Individuals who had a history of cancer (or atrial fibrillation) at baseline were excluded from this analysis. The individuals were followed for 1 year after the incident atrial fibrillation (or cancer) diagnosis or until all-cause mortality, whichever came first, while cancer (or atrial fibrillation) diagnosed during the follow-up was treated as a time-dependent exposure to estimate its association with all-cause mortality by multivariable Cox regression, with adjustment for time-fixed covariates (identified at baseline), including age, sex, immigration background, standardized household income, and various comorbidities (or medical history). For the analyses of developing non–sex-specific cancer among the incident atrial fibrillation cohort, the individuals were also censored when sex-specific cancer was diagnosed and vice versa (as indicated). CNS, central nervous system; HR, hazard ratio; CI, confidence interval

For individuals in the incident cancer cohort without AF history at baseline (Figure 7 and Supplementary data online, Table S20), AF diagnosed within 1 year was also associated with increased all-cause mortality (aHR 2.55, 95% CI 2.47–2.63), and the strongest associations were observed in those with cancers of endocrine glands (aHR 10.12), other skin cancer (aHR 7.74), melanoma (aHR 5.55), bone/cartilage (aHR 5.30), and other respiratory/intra-thoracic organs (aHR 4.90).

Discussion

In the current study, we thoroughly examined the epidemiology of coexisting AF and cancer. Our main findings included the following: (i) AF and cancer commonly coexist, and the prevalence of having one condition among those with the other has increased in recent years; (ii) compared with the general population, having one condition is associated with an increased risk of having the other, and the risks are the highest during the first 3 months; (iii) the types of cancer that are more likely to be diagnosed after AF are in line with the most frequent cancer types in the general population, while individuals with some types of cancer are noticeably more likely to be diagnosed with AF than other cancer types; and (iv) newly diagnosed cancer is associated with increased all-cause mortality among AF patients, and vice versa, but the strength of the association with mortality varies by cancer type (Structured Graphical Abstract).

Implications of the findings

Since our investigation was based on a recent (2015–19) and unselected nationwide population, which covered both directions (i.e. cancer diagnosed after AF and AF diagnosed after cancer) and various cancer types, the findings present recent and comprehensive epidemiological knowledge about coexisting AF and cancer, including prevalence, incidence, time trends, and associations with survival. These detailed statistics about both absolute and relative risk as well as the time course cover the complete cancer spectrum, pointing out what cancer types should receive more attention in clinical management and research regarding the issue of comorbid AF and vice versa. Results of the time trend analysis indicate an increasing burden of coexisting AF and cancer, which has not been reported before and might warrant awareness. The investigations into multiple cancer types, together with the inclusion of the matched control cohorts from the general population, reveal that the distribution of cancer types diagnosed in AF patients is similar to the general population, whereas in the opposite direction, AF was diagnosed more frequently after some cancer types than others, which deepens the understanding of the bidirectional associations between AF and cancer. Last but not least, the associations we observed between coexisting AF and cancer and survival raise the question of whether preventing and better managing one condition would benefit the prognosis of those with the other condition. In short, the abundant epidemiological information from our study will help fulfil the immediate need of the rapidly emerging field of cardio-oncology and at the same time feeds new and relevant research questions.

Atrial fibrillation after cancer

The growing recognition of the link between cardiovascular diseases and cancer may stem from the combination of the ageing population, improved cancer management that prolongs survival, and the introduction of novel cancer therapies that bring cardiovascular toxicity.^7,21,26,27 As a result, existing epidemiological investigations mainly focused on the occurrence of AF in cancer patients, either for a specific cancer type or treatment^28–36 or multiple cancer types.^15–18 A meta-analysis found cancer patients had a 47% higher risk of developing AF compared with those without cancer, especially in the first 3 months.³⁷ What we observed is overall consistent with these findings, but what we added to this topic is that we included a more recent study population (2015–19) and investigated more granular cancer types. This makes our findings better reflect recent practice, which is relevant given the rapid advances in cancer management. In addition, our study cohorts were defined under strict and the same criteria, in which each individual by design had a complete 1-year follow-up, and many covariates were included for comparisons with the general population. However, it should be noted that the increased risk of AF in cancer patients cannot be simply interpreted as a causal relationship, since residual confounding could not be ruled out in the observational study design we used.

We found there was an increasing trend in the prevalence of AF among cancer patients over the years, but the incidence of newly diagnosed AF remained stable. Since we also observed an increasing trend in AF prevalence in the general population, ageing might partly explain the increase in AF prevalence in cancer patients. It is worth mentioning that due to data limitation, we could not distinguish which condition actually occurred first when both AF and cancer were diagnosed for the first time within the same hospitalization. When this was the case, we would always classify AF as a prevalent AF, and therefore, we might have underestimated the incidence of AF, particularly for AF that actually occurred immediately after the incident cancer diagnosis. However, since these cases only accounted for 2.2% of the incident cancer cohort, this should have had limited impact, and moreover, it would only suggest that the true comorbidity burden of coexisting AF and cancer is actually more substantial than we observed.

With respect to the incidence of AF by cancer type, we found that individuals with cancers of the oesophagus, lung, stomach, myeloma, and lymphoma faced the highest risk of AF. This finding was also generally consistent with other studies, although the exact magnitude of the risk might differ. For example, in the study by Yun et al.,¹⁵ myeloma and oesophageal cancers were found to be strongly associated with AF, but stomach cancer showed the lowest association with AF among the investigated solid cancers. This could be due to the difference in cancer epidemiology (e.g. regarding cancer grade/stage/treatment upon initial diagnosis) between regions/countries, which we could not further examine due to lack of such data. The findings from a Danish study¹⁶ somewhat supported the above speculation, as the incidence of AF after lung cancer was the highest (among the cancer types they investigated), followed by upper gastrointestinal cancer, which is consistent with what we found.

With regard to the time course, although AF tended to be diagnosed more frequently in the first 3 months after the incident cancer diagnosis, there were variations between cancer types, which might be explained by differences in anticancer treatment trajectory. For example, among patients with cancer of the oesophagus, we found that AF was frequently diagnosed in the first 6 months, particularly during the third—sixth month. In future studies, it would be interesting to explore what causes this pattern (e.g. surgical treatment performed after pre-operative chemoradiotherapy.³⁸) Of note, the HRs reported in our study should be interpreted as weighted average of the true HRs over the 1-year follow-up period.³⁹

Another naturally raised question is what might explain the association between cancer and AF. Our study cannot provide an answer directly, but several potential explanations have been proposed previously,^19,20 including common risk factors/pathophysiology, cancer treatment that may induce AF (e.g. radiation therapy, chemotherapy, major transthoracic surgery, and targeted therapy), paraneoplastic effects, and detection bias [i.e. more electrocardiogram (ECG) examinations are performed after a cancer diagnosis, leading to more AF diagnoses]. It is very likely that multiple mechanisms contribute to the increased AF risk in cancer patients, either directly or indirectly or causally or non-causally.

Previous studies have shown that AF was associated with adverse outcome events in various types of cancer,^28,40–45 but as far as we know, no study investigated this association across multiple cancer types in the same source population. We found that in cancer patients, newly diagnosed AF was associated with increased all-cause mortality after adjusting for the various baseline characteristics, but the association seemed to vary by cancer type. This again raises more research questions to explore in the future, but it should also be realized that our findings cannot be interpreted in a causal way, as AF itself might be caused by advanced cancer therapeutics used for advanced cancer.^46,47

Cancer after atrial fibrillation

Evidence about the opposite direction of the association, namely the occurrence of cancer after AF, however, is very limited.²² According to a recent meta-analysis,²³ which included five cohort studies^48–52 and a case–control study,⁵³ new-onset AF was associated with a 24% increase in cancer risk during the initial 90 days but not after, and the association was found only for lung cancer but not for colorectal cancer and breast cancer. Besides these studies, there were only three other studies that investigated the association of developing cancer after AF: Müller et al.⁵⁴ found that AF tended to be one of the diseases diagnosed preceding colon cancer; Ostenfeld et al.⁵⁵ reported that 2.5% of incident AF patients were diagnosed with cancer within 3 months; Kahr et al.⁵⁶ found that participants undergoing screening colonoscopy who had AF showed a higher burden of colorectal cancer. Compared with these findings, we found that incident AF was associated with an overall 52% increase in cancer risk and that cancer was diagnosed the most frequently in the first 3 months after the incident AF diagnosis. The strengths of our study when investigating AF diagnosed after cancer also applied to this investigation (i.e. cancer diagnosed after AF), including the well-defined study cohorts, the investigation into time trends, the coverage of different cancer types, and the adjustment for various baseline characteristics.

Since we found the types of cancer that are more likely to be diagnosed after AF were generally in line with that in the general population, our findings do not seem to support AF as a cause of new-onset cancer (instead, e.g. detection bias), although again our data and study design were unable to examine a causal relationship. There are several potential mechanisms, including radiation exposure during AF ablation (which was, however, not supported by a further investigation⁵⁷), anticoagulant-related bleeding leading to early diagnosis of cancer,^58–61 incidental imaging findings for AF ablation,^62,63 and potential cancer risk carried by some medications used for AF.⁶⁴ A recent Mendelian randomization study did not support a causal role of AF in increasing cancer risk.⁶⁵ These speculations should be interpreted with caution and remain to be confirmed.

Nevertheless, the increasing time trend over the years we observed in the prevalence of cancer among incident AF patients is relevant to know, since AF patients require long-term management,³ while comorbid cancer has been shown to challenge AF management (particularly about anticoagulation).^66–69 This AF subpopulation (i.e. with cancer) has attracted increasing attention,⁸ but there are still many knowledge gaps. For example, the CHA₂DS₂-VASc and HAS-BLED scores seem to show suboptimal performance in AF patients with cancer,^70,71 which challenges treating these patients optimally. Taking this relevance into account, even if AF is non-causally associated with cancer, it might be worthwhile to examine whether there is a role for cancer screening among individuals with newly diagnosed AF, although cost-effectiveness and potential harms should be also considered.

Limitations

There are several other limitations in our study. First, we only used routinely collected data, which is likely to have introduced some misclassification. This limitation especially applies to our identification of AF and cancer, for which only data on diagnoses made within hospitalization were used. Due to the lack of data on outpatient diagnoses of AF or cancer, we might have underestimated the actual burden of coexisting AF and cancer, but the prevalence and incidence we observed were still rather substantial. The date of an incident diagnosis of AF/cancer we defined in the study was the admission date of the hospitalization in which the diagnosis of AF/cancer was first made. Therefore, our findings about the time course of cancer/AF diagnosed after incident AF/cancer might differ from studies that identified the index dates in a different way (e.g. via outpatient diagnoses). Ideally, screening for AF with better approaches (such as ECG, wearable photoplethysmography-enabled device)⁷² is needed to precisely evaluate the burden of AF among cancer patients, which should be considered in further investigations. Second, since no data on cancer stage/grade/treatment were available, we could not further examine whether the increased AF risk after cancer was related to such cancer characteristics. This is relevant to explore in future studies to provide insights into mechanisms that might explain the observed cancer–AF association. Third, we limited the follow-up to 1 year only, and we could not distinguish whether a previous cancer diagnosis was cured or active when analysing the incident AF cohort. In addition, we only investigated the association of coexisting AF and cancer with all-cause mortality, without examining other relevant outcome events (e.g. progression-free survival, ischaemic stroke, etc.). Last but not least, the source population of our study was the unselected Dutch population, which should be highly generalizable to western populations, but the epidemiology of coexisting AF and cancer might differ in a population with different age/sex distributions, ethnic backgrounds, or different healthcare systems. These remain relevant research directions for the future.

Conclusions

There is an important bidirectional association between AF and cancer, and AF risk varies by cancer type. The burden of coexisting AF and cancer has increased in recent years, and having one condition is negatively associated with the survival of patients with the other condition. Awareness of this comorbidity burden should be raised in both AF and cancer patient populations, and further explorations of the underlying mechanisms and optimal management are warranted.

Supplementary data

Supplementary data are available at European Heart Journal online.

Declarations

Disclosure of Interest

Q.C. was supported by the Chinese Government Scholarship (No. 201906380148) for his PhD study at the Leiden University Medical Center between September 2019 and September 2023 and received travel awards from the International Society on Thrombosis and Haemostasis. S.A.T. received a teaching grant from Biosense Webster, payment for lectures for the Dutch Cardiovascular Research Institution and EHRA preparatory course, and payment as an expert in a personal injury lawsuit; he is a member of the EHRA certification committee, the chair of the EHRA Core Curriculum writing committee, and a member of the supervisory board of the Dutch Society for Cardiology. H.H.V. received reimbursement of costs for State-of-the-Art ISTH 2024, and he is the editor-in-chief of Thrombosis Research. F.A.K. received research support from Bayer, BMS, BSCI, AstraZeneca, MSD, Leo Pharma, Actelion, Farm-X, the Netherlands Organization for Health Research and Development, the Dutch Thrombosis Foundation, the Dutch Heart Foundation, and the Horizon Europe programme. All the other authors declare no disclosure of interest for this contribution.

Data Availability

The study was conducted by the authors using non-public microdata from Statistics Netherlands, but these data cannot be shared directly by the authors. Under certain conditions, these microdata are accessible for statistical and scientific research. For further information, visit microdata@cbs.nl.

Funding

All authors declare no funding for this contribution.

Ethical Approval

The study complied with the Declaration of Helsinki and received an approval from the Scientific Committee of the Department of Clinical Epidemiology of the Leiden University Medical Center (No. A181) with a waiver of participant consent due to the use of pre-existing, de-identified data only.

Pre-registered Clinical Trial Number

Not applicable.