Association of chocolate consumption with neurological and cardiovascular outcomes in atrial fibrillation: d ata from two Swiss atrial fibrillation cohort studies (Swiss-AF and BEAT-AF)

in


Introduction
Atrial fibrillation (AF) is the most common arrhythmia in European countries. In view of an ageing population, its prevalence is expected to increase further [1]. Atrial fibrillation has been associated with an increased risk of cardiovascular morbidity and death [2]. According to previous investigations, chocolate consumption is linked to a decreased risk of cardiovascular and cerebrovascular disease [3][4][5] and may affect cognitive function [6][7][8][9].
The most important components of chocolate are cocoa flavonoids and methylxanthines [6]. A number of beneficial systemic effects have been attributed to flavonoids, including increased nitric oxide bioavailability [10], antioxidant properties [10,11], antithrombotic mechanisms and anti-inflammatory effects [12]. Both flavonoids and methylxanthines may permeate the blood-brain barrier. Whereas flavonoids may affect brain function via mechanisms such as increased cerebral blood flow [8,13], promotion of cerebral angiogenesis [9,13], or inhibition of neuronal death by apoptosis [9], methylxanthines may act as mild central nervous system stimulants [6].
Chocolate may have the most favourable effect when consumed in moderation [3,4]. Moderate chocolate intake may also be inversely related to the risk of clinically apparent atrial fibrillation [14]. However, data regarding this association are conflicting [15,16]. To date, the effect of chocolate consumption on neurocognitive function, cerebral and cardiovascular outcomes in patients diagnosed with atrial fibrillation has not been studied. With a per capita consumption of 9.9 kg/year, the Swiss population has the highest consumption of this aliment in Europe [17]. The prevalence of atrial fibrillation in Switzerland is 600-699/100,000 persons [18]. The widespread consumption of chocolate in a country with a high prevalence of atrial fibrillation gave rise to searching for associations between chocolate intake and clinical outcomes in a Swiss AF population.
Based on the above-mentioned considerations, the aim of the present study was to evaluate the association between chocolate consumption and 1) neurocognitive function, 2) brain lesions on MRI, and 3) cardiovascular events in patients with atrial fibrillation.

Study design, data sources and participants
Data from two ongoing prospective, observational multicentre cohort studies from Switzerland -the Swiss Atrial Fibrillation Cohort study (Swiss-AF) and the Basel Atrial Fibrillation Cohort study (BEAT-AF) -were included in this study [19,20]. Patients were eligible for participation in Swiss-AF if they had a history of documented atrial fibrillation and if they were aged 65 years or older. An additional 10-15% of patients between 45-65 years of age were aimed to be enrolled to assess atrial fibrillation in individuals who are potentially in the active workforce [19]. In BEAT-AF, patients with atrial fibrillation documented on electrocardiogram were asked to participate [20]. In both cohort studies, patients could be recruited from inand outpatient clinics. Enrolment of patients with acute illnesses was postponed until stabilisation. More detailed in-formation on the underlying cohorts is described elsewhere [19,20]. Both cohort studies use an equivalent set of clinical assessments, including chocolate consumption, patient demographic characteristics, medical history, and medication. During the course of the study, information on clinical events is collected yearly.
For data derived from brain magnetic resonance imaging (MRI) as well as from cognitive functional testing, only data from the Swiss-AF cohort were included. In BEAT-AF, these outcomes were not systematically evaluated. For the analyses regarding brain lesions and cognitive function testing, we used the baseline data (= data of the first visit) from all enrolled Swiss-AF patients with a full data set, i.e., patients that had a baseline MRI as well neurocognitive function testing performed at the initiation visit. The association between chocolate consumption and clinical events was assessed using the data of all Swiss-AF and BEAT-AF patients with available information on baseline chocolate consumption.
The local Ethics Committees approved the study protocols of both registries. All patients gave written informed consent.

Categorisation of chocolate consumption
Chocolate consumption was reported by the patients on a yearly basis via a multiple-choice question with the following nine answer options: never or less than one bar (German: "Riegel") per month, 1-3 bars per month, 1 bar per week, 2-4 bars per week, 5-6 bars per week, 1 bar per day, 2-3 bars per day, 4-5 bars per day, and 6+ bars per day, respectively.
One serving is considered to correspond to approximately 30 g of chocolate [21]. As the size and composition of a chocolate bar is variable and the chocolate content is therefore difficult to determine, we chose to equate a bar of chocolate to one serving. According to two meta-analyses, the quantity of chocolate intake potentially associated with a reduction in the risk of cardiovascular disease ranges from 45g to 180 g/week, which is considered to represent moderate consumption [3,4]. Consequently, we defined three groups of chocolate consumption. Patients were stratified into the no/low chocolate consumption group (No/ Low-Ch) if they stated they ate ≤1 servings (corresponding to ≤30 g/week) of chocolate per week. In the moderate chocolate consumption group (Mod-Ch), the weekly chocolate intake ranged from >1 to 6 servings (>30 g to 180 g per week), and in the high chocolate consumption group (High-Ch), chocolate consumption exceeded 6 servings per week (>180 g per week).

Clinical measures
Weight and height were directly measured and body mass index (BMI) calculated. Patient history was assessed at baseline. Educational status, smoking status, medical history, and history of oral anticoagulation medication were obtained by questionnaire. Atrial fibrillation type was classified as paroxysmal, persistent, or permanent atrial fibrillation [18].

Outcome measures
The primary interest of our study was to evaluate the association between chocolate consumption and neurocognitive function in a real-world atrial fibrillation population. Given the unique data set on cerebral MRI performed and systematically analysed in a large number of Swiss-AF participants, we additionally intended to search for associations between chocolate intake and brain lesions. Moreover, the assembly of data from both the Swiss-AF and the BEAT-AF cohorts enabled us to evaluate the association between chocolate consumption and clinical outcome measures in a large population of patients suffering from atrial fibrillation. The study was exploratory in nature and aimed to be hypothesis-generating rather than confirmatory. Therefore, no distinct primary endpoint was defined prior to analysis of data.

Cognitive testing
Centrally trained study personnel performed standardised neurocognitive assessments. The tests were provided in paper format in the main national languages of Switzerland (i.e., German, French and Italian) [22]. Neurocognitive testing included the Montreal Cognitive Assessment (Mo-CA) which is a screening test to detect mild cognitive impairment. Patients can obtain a score from 0 to 30 points [23]. The Trail Making Test (TMT) parts A and B and the Semantic Fluency Test (SF) were included for detection of dementia [24,25]. In the TMT, patients connect circled numbers in ascending order by drawing a continuous line (trail) between them. Trails A and B are of different lengths. The test score is defined as the time used divided by the total number of circles correctly connected in that time [22,24]. In the SF test, patients are asked to name as many animals as possible within 60 seconds [25].
The Digit Symbol Substitution Test (DSST) was used to evaluate psychomotor performance [26]. In this test, patients receive a key grid of numbers and matching symbols. The score is the number of correct number-symbol matches achieved within 120 seconds [22,26]. Additionally, we used the cognitive construct score, a factor score developed for the Swiss-AF study that allows for quantification of cognitive function. The test is composed of 17 differently weighted combined items from all of the above-mentioned individual neurocognitive tests [22].

Brain MRI
Brain MRI was performed on 1.5 or 3 Tesla MRI scanners. A standardised protocol was used in all participating centres. The standard protocol did not demand the administration of contrast agent. Brain lesions were assessed as small non-cortical infarcts (SNCIs), large non-cortical or cortical infarcts (LNCCIs), microbleeds (Mb), and white matter lesions (WML) [19]. As 99% of patients presented with WML, we used the Fazekas score as a binary endpoint for WML in the analysis. At least moderate disease was defined as a score ≥2 [27].

Main clinical outcome measures
Main clinical outcome measures were prespecified and included major adverse cardiac events (combined end-point of ischaemic stroke, myocardial infarction and cardiovas-cular death), stroke, major bleeding, myocardial infarction, hospitalisation for heart failure, cardiovascular death, and all-cause death, respectively. If a clinical outcome measure was reported or found in the medical records, additional information was collected from involved hospitals and/or treating physicians. All events were adjudicated by a blinded clinical event committee [19].

Statistical analysis
Baseline patient characteristics by categories of chocolate consumption were described by mean and standard deviation, or absolute and relative frequency, as appropriate. Prevalence of lesions on baseline MRI were presented as absolute and relative frequency, lesion volume and lesion count as median and interquartile range. Results from neurocognitive tests were described by mean and standard deviation. Group comparisons were performed by ANOVA tests (one-way, three groups) for continuous variables and by chi-squared tests for categorical variables.
In all analyses, we considered available data from both the BEAT-AF and Swiss-AF cohort until 13 May 2022.
In all the analyses evaluating the association between chocolate consumption and outcomes, Mod-Ch was used as the reference category and the two other categories were compared to this reference category.
The association between chocolate consumption and cerebral lesions on MRI was examined using multivariable-adjusted mixed-effects logistic regression models, including study centre as a random intercept. In this analysis, we included all patients with a brain MRI reading available at baseline (n = 1727). In patients that presented with cerebral lesions, we also analysed the association between chocolate consumption and lesion volume using multivariableadjusted mixed-effects linear regression models with the lesion volumes (log-transformed due to the skewed distribution and mean-centred) as the dependent variable. LNC-CIs, SNCI, WML and Mb were analysed. These models also included study centre as a random intercept.
In all patients with a brain MRI reading available at baseline, we also assessed the cross-sectional association between chocolate consumption and cognitive function using linear mixed-effects models with study centre as a random intercept. The test scores of the performed neurocognitive tests as well as the calculated cognitive construct factor score were used as continuous outcome variables. Given the previous association between the neurocognitive test scores and the presence as well as the volume of neurological lesions [28], we additionally adjusted the models for neurocognitive outcomes for the presence and volume of LNCCI, the presence of white matter lesions with Fazekas scale ≥2, as well as the volume of white matter lesions. The volumes were set to 0 for patients who did not present with the particular lesions. Given the very low proportion of missing data at baseline, we performed an available case analysis.
We investigated the association between chocolate consumption and clinical events using Cox proportional hazard models with study centre as a stratification factor. All variables in the model were time-updated to account for variations in chocolate consumption and other covariates over time (information updated with each patient visit). In case of missing data occurring at follow-up visits or of missed visits, we performed a simple imputation with the patient's last observation carried forward until the next visit or censoring. In potentially recurring events, only the first event was considered.
For each outcome, two models were constructed: (1) a model adjusted for age and gender, and (2) a model adjusted for age, gender, and additional clinical covariates (the lists of covariates are displayed in the corresponding tables). For each model, the estimates (odds ratio for logistic regression models, coefficient for linear models, and hazard ratios for Cox models) for the fixed effects of the different levels of chocolate consumption along with their corresponding 95% confidence interval (CI) are reported.
As a sensitivity analysis, we investigated the potential effect of the two different cohorts by adding an interaction term of cohort with chocolate consumption. This did not result in a better model fit and therefore it was not considered in the analysis.
All analyses were performed using the statistical software R version 4.2.2. The analytical code is provided as supplementary material (appendix 1).

Ethics statement
The studies involving human participants were reviewed and approved by the Ethics Committee Nordwest-und Zentralschweiz, Switzerland and all local Ethics Committees at the study sites. The project numbers were 2021-00701 for Swiss-AF and EK 331/09 for BEAT-AF (both numbers from the lead Ethics Committee).

Baseline characteristics
Of the 2415 patients enrolled in the Swiss-AF cohort, 1727 (71.5%) patients were included in the analysis for brain lesions on brain MRI and for neurocognitive function test-ing. 672 (27.8%) patients did not undergo brain MRI; the main reason for this was an implanted cardiac device (n = 461; [68.6%]). Other reasons were contraindications for MRI and claustrophobia. 11 (0.5%) patients were excluded from the analysis due to missing MoCA assessment during the baseline visit and in 5 (0.2%) patients we did not have information on chocolate consumption at baseline. The association between chocolate consumption and the prevalence of lesions detected by brain MRI is presented in table 3. In the simple model adjusted for age and gender, as well as in the full model adjusted for additional clinical variables, there was no association found for prevalence of brain lesions on brain MRI and the different groups of chocolate consumption (table 3). Moreover, no significant associations were found between the volume of brain lesions and chocolate consumption (table 3).

Neurocognitive function tests
The results of neurocognitive function tests are shown in table 4. Whereas MoCA and cognitive construct varied between groups, the remaining neurocognitive test results were similar. The association between chocolate consumption and neurocognitive function in multivariate analysis is shown in table 5. In the simple model adjusting for age and gender, Mod-Ch was associated with better TMT-B, DSST, and cognitive construct results compared to No/Low-Ch. However, in the full model of the multivariate analysis, only the association between Mod-Ch and a better cognitive construct score remained significant. Evaluation of neu- rocognitive test results in High-Ch versus Mod-Ch exhibited no significant associations in the two statistical models (table 5).

Baseline characteristics of patients included in the clinical outcome analysis
Of the 4039 patients in the Swiss-AF and the BEAT-AF cohorts, 3931 (97.2%) were included in the analysis. 108 pa-

Association between chocolate consumption and clinical outcomes
During follow-up, a total of 1358 (34.5%) patients experienced one or more clinical events. Major adverse cardiac events occurred in 560, stroke in 210, major bleeding in 392, hospitalisation for heart failure in 632, myocardial infarction in 150, cardiovascular death in 446, and allcause death in 726 patients, respectively. The association between chocolate consumption and clinical outcomes in multivariate analysis is shown in Table 7. No independent association of chocolate consumption with major adverse cardiac events was found. In the simple model adjusting for age and gender as well as following adjustment for various additional clinical parameters, the risk of hospitalisation for heart failure and of all-cause death was higher in No/Low-Ch compared to Mod-Ch (table 7). For the comparison between High-Ch vs. Mod-Ch no corresponding associations were seen. The Kaplan-Meier curves for the end-points hospitalisation for heart failure and all-cause death in the three groups of chocolate consumption are presented in figure 1.

Discussion
In this large cohort study of patients with atrial fibrillation, we observed no association between chocolate consumption and the prevalence or volume of vascular brain lesions on brain MRI. On the other hand, No/Low-Ch, when compared to Mod-Ch, was independently associated with low-   er neurocognitive function as assessed by the cognitive construct score. Moreover, whereas major adverse cardiac events, comprising ischaemic stroke, myocardial infarction and cardiovascular death, appeared to be unrelated to chocolate consumption, No/Low-Ch was associated with an increased risk of hospitalisation for heart failure and allcause mortality when compared to Mod-Ch.
It is generally accepted that a healthy lifestyle has a positive effect on health in patients with atrial fibrillation and lifestyle modification is considered an important therapeutic intervention in these patients according to current atrial fibrillation guidelines [18]. Chocolate, when consumed in moderation, may potentially represent a nutritional contribution to the well-being of atrial fibrillation patients.
Of note, in our study, BMI was lower in patients with higher chocolate consumption, a finding that has previously been observed by others [14]. This association is somewhat counter-intuitive. Potentially, it might indicate a generally healthier lifestyle in patients consuming more chocolate. Alternatively, the finding may have been influenced by disease, as obese patients, particularly those with diabetes mellitus, are generally advised to reduce their consumption of high-sugar foods, including chocolate. Moreover, as chocolate intakes were self-reported by patients, underreporting by obese subjects due to social desirability bias could have added to this finding. Interestingly, according to a previous study in a population with no history of cardiovascular disease, frequent chocolate consumption may independently be linked to lower BMI. The observed The simple model is adjusted for age and gender. The full model is additionally adjusted for educational status, BMI, smoking status, arterial hypertension, history of diabetes, history of myocardial infarction, history of hospitalisation for acute heart failure, history of renal failure, atrial fibrillation type, oral anticoagulant medication, antithrombotic medication, history of coronary artery disease, history of major bleeding, and history of stroke or transient ischaemic attack.
No/Low-Ch: no or low chocolate consumption; Mod-Ch: moderate chocolate consumption; High-Ch: high chocolate consumption; HR: hazard ratio; CI: confidence interval; MACE: major adverse cardiovascular events association was not explained by calorie intake, activity, or other potential confounders [29]. These results are intriguing, but in line with preclinical data in animals [30]. To account for the differences in baseline characteristics observed among the three groups of chocolate consumption, we corrected for these potential confounders in multivariate analyses.
The beneficial systemic effects of chocolate are thought to be primarily mediated by cocoa flavonoids, a group of polyphenols that may improve endothelial function, decrease platelet reactivity, decrease sympathetic tone and lower blood pressure [10,31]. Additionally, oxidative stress reduction and anti-inflammatory effects have been attributed to chocolate intake [11,12,14].
Cardiovascular health is closely linked to cognitive performance [8]. Moreover, components of chocolate such as flavonoids and methylxanthines may cross the blood-brain barrier and therefore exhibit direct cerebral effects [6,9]. Flavonoids may increase central blood flow, promote angiogenesis, inhibit neuronal cell destruction by neurotoxicants, and interact with cellular and molecular signalling cascades in regions involved in learning and memory [8,9,13]. Methylxanthines may act as mild central nervous system stimulants and lead to expression of neurotrophins that influence neurocognitive function [6].
Published data indicate that moderate consumption of chocolate may have a positive effect on cardiovascular health [3,4]. However, the link between chocolate consumption and atrial fibrillation is less well-established. To date, research in the field has mainly focused on the risk of atrial fibrillation and results have been inconsistent. Although in the Danish Diet, Cancer and Health Study, an inverse association between moderate chocolate intake and the incidence of atrial fibrillation was observed [14], other studies did not find a similar link between chocolate consumption and the risk of atrial fibrillation occurrence [15,16].
To the best of our knowledge, our study is the first to evaluate the potential associations between chocolate consumption and neurocognitive function, vascular brain lesions, and clinical outcome in patients with an established diagnosis of atrial fibrillation.
Whereas findings on brain MRI appeared to be unrelated to chocolate intake, better neurocognitive function as assessed by the cognitive construct score was independently associated with moderate chocolate consumption. The cognitive construct score, as a summary measure of the common aspects of the four neurocognitive tests used in our study, has previously been shown to reveal good psychometric properties and to increase measurement sensitivity when applied to the Swiss-AF population [22]. This may explain the fact that a significant independent association between neurocognitive function and chocolate consumption was only found when the cognitive construct score was used. However, the absence of independent associations between chocolate intake and any of the four tests performed with the patients may call into question the clinical relevance of our finding.
In line with the observed link between the cognitive construct score and moderate chocolate consumption, a positive association between cognitive performance and im-proved memory has previously been attributed to habitual chocolate intake in patients without atrial fibrillation [6][7][8][9].
Interestingly, it has also been recognised that the highest number of Nobel Prize winners can be found in countries with the highest chocolate consumption [32]. However, this potential link is rather speculative and has not been elaborated in detail. It cannot be excluded that the association might simply be due to the fact that in highly developed countries, where the luxury good chocolate is more frequently consumed, more research can be afforded and conducted and with this, the likelihood of citizens receiving Nobel Prizes may rise.
In our study, a decreased risk of hospitalisation for heart failure could be seen for Mod-Ch compared to No/Low-Ch, when assessed in multivariate analysis. This observation is in accordance with a previously found association between moderate chocolate consumption and a lower risk of heart failure [5]. Improvement of endothelial function with activation of nitric oxide, reduction in sympathetic tone, lowering of blood pressure and anti-inflammatory properties are potential effects of flavonoids that may contribute to a positive influence on heart failure [10,12,31,33].
Unlike for heart failure hospitalisation, we did not find a link between chocolate consumption and the risk of myocardial infarction or cardiovascular death. This contrasts with previously published data implying a beneficial effect of chocolate consumption on the risk of cardiac events, including myocardial infarction and cardiac death [3,4,34,35]. The discrepancy in findings regarding cardiovascular outcomes is difficult to explain. A potential factor might be a variability in the consumed concentration of plantderived flavonoids. Whereas protective effects on the cardiovascular system have been attributed to high amounts of flavonoids ingested either directly via daily cocoa consumption or by intake of dark chocolate [10,33,34], we had no information on the type of chocolate consumed in our study.
In contrast to the lack of an association with cardiovascular mortality, moderate chocolate consumption was independently related to a decreased risk of all-cause mortality in our study. Given the large difference in event rates of the two endpoints, this disparity in findings, at least in part, may have been related to a statistical power issue. A true dissociation between cardiovascular and all-cause mortality would imply the presence of relevant effects of chocolate consumption on extra-cardiovascular systems. For example, this might be mediated by anti-inflammatory actions or decreased genotoxicity due to antioxidant properties [11,12,36,37].

Limitations
In the present study, we did not have information on the type of chocolate or the cocoa concentration consumed. However, it cannot be assumed that patients only ever eat the same type of chocolate and our data may represent a real-life setting in this regard. In Switzerland, the most frequently consumed chocolate is milk chocolate. Therefore, the amount of flavonoids consumed in our study may have been lower than if patients had eaten only dark chocolate. Despite this limitation, we observed significant associations between chocolate consumption and important clin- ical endpoints in multivariate analyses. Whether the associations would have been more pronounced if only dark chocolate was used remains speculative.
Our study was exploratory in nature. Therefore, a high number of tests on associations between chocolate consumption and outcome measures were performed, no clear primary endpoint was defined, and no adjustment for multiple comparisons was done. Consequently, given an alpha level of 5%, we cannot exclude that some of the findings may have resulted by chance and therefore, the data should be interpreted with caution. However, as discussed above, similar associations were found in other studies investigating different populations.
Another limitation of our study is adherent to its nonrandomised, observational design. Although we performed multivariate analyses adjusting for multiple co-variables, the possibility of residual confounding due to selection bias, including Berkson's bias, or due to unmeasured potentially influencing factors, as well as the possibility of reverse causality, cannot be excluded. This must be taken into account when interpreting the results. Due to its exploratory nature, our work is hypothesis-generating rather than confirmatory.

Conclusion
Based on our findings, Mod-Ch consumption may potentially be beneficial in patients with atrial fibrillation. No/ Low-Ch, when compared to Mod-Ch, was associated with lower cognitive function as assessed by the cognitive construct factor score, higher risk of heart failure hospitalisation and increased all-cause mortality, while there were no associations of chocolate consumption with MRI findings and major adverse cardiac events in a real-world atrial fibrillation population.