Statistics from Altmetric.com
Atrial fibrillation (AF) carries a significant burden for both patients and society because it is the most common sustained cardiac arrhythmia, and it is associated with increased cardiovascular morbidity and mortality, mainly due to the higher risk of thromboembolic events.1–3 Classical risk factors include advanced age, congestive heart failure, valvular diseases, left ventricular hypertrophy, arterial hypertension, ischaemic heart disease, male gender, diabetes mellitus, endurance exercise and smoking.4
Caffeine is a major component of some of the most widely consumed beverages, such as coffee and tea.5 The association between caffeine exposure and risk of heart disease has received extensive attention in the literature.6–13 Inconsistencies between results have been attributed to differences in study design and presence of confounders, in particular tobacco smoking. Results from a recent meta-analysis of observational studies and from large cohort studies suggest a mild-to-moderate inverse association between coffee drinking and cardiovascular risk factors and diseases, including global mortality and death due to heart disease.13
The association between caffeine exposure and AF is currently unknown, although a positive association has been suggested in the literature.14 ,15 To further evaluate this putative association, we performed a systematic review and meta-analysis of observational studies.
Randomised controlled trials, prospective or retrospective cohorts and case–control studies evaluating exposure to caffeine (whether as coffee, tea, chocolate or caffeinated beverages) and risk of AF (or atrial flutter) were eligible. Studies addressing the effects of short-term exposure to caffeine (ie, <6 months) as well as studies evaluating caffeine exposure in patients already in AF were excluded. Studies that met inclusion criteria were not excluded a priori on the basis of weakness of design or data quality.
Information sources and search process
Two investigators retrieved potential eligible studies through an electronic search in PubMed, CENTRAL, ISI Web of Knowledge and LILACS, from inception to December 2012. The search strategy for PubMed (in online supplementary information) included free-text words and MeSH (Medical Subject Headings) terms without language restrictions. In addition, we screened and cross-checked identified systematic reviews and meta-analyses evaluating caffeine exposure and heart diseases or cardiovascular risk factors, as well as reference lists of papers found for potential additional studies.
Data extraction, evaluation and synthesis
Titles and abstracts of obtained records were screened by two authors (DC and JC). Doubts and disagreements were solved by consensus. Selected studies were assessed in full-text to determine their appropriateness for inclusion. Study characteristics and results were extracted independently by two authors (DC and JC) into a standardised form. We also extracted the following variables: population characteristics, region of study, study follow-up, caffeine exposure assessment, reference category, outcome assessment and outcome adjustments.
For primary analysis we evaluated the exposure to caffeine at baseline compared with a reference group of non-consumers or with the lowest quintile of intake. Data from different estimates evaluating different levels of exposure compared with a reference were abstracted.
When different risk estimates for the same strata were available in the same publication, we considered for analysis those reflecting the greatest degree of control for potential confounders, or the most comprehensive assessment of caffeine intake, using these criteria sequentially.
The effect measurement estimate chosen was OR because relative estimates are more similar across studies with different designs, populations and lengths of follow-up than absolute effects.18 Studies presenting risk ratio or HR adjusted estimates were considered as OR.
Reporting quality was independently evaluated by two investigators (DC and JC) using a qualitative classification according to risk of bias (high, unclear or low risk). We used a five-item classification system based on MOOSE, QATSO and STROBE adapted from previous published quality assessment instruments.17 ,19–22 The following items were taken into consideration: (1) participants, if the population was adequate and the study reported appropriate inclusion and exclusion criteria; (2) exposure, if caffeine exposure use was adequately assessed through food questionnaires; (3) outcome, if AF was assessed by clinical, ECG methods or through database codes, and not exclusively based on self-report; (4) specific outcome adjustments, for both age and at least one of cardiovascular disease, alcohol intake or smoking; (5) other adjustments.
We used RevMan V.5.1.7 software for statistical analysis (The Nordic Cochrane Centre, The Cochrane Collaboration, 2011) and to derive forest plots showing the results of individual studies and pooled analysis. For primary analysis, we performed random-effects meta-analysis weighted by the inverse variance method to estimate pooled ORs and 95% CIs. For this purpose, in case studies that did not report a single estimate for consumers versus non-consumers, but expressed multiple levels of exposure, we pooled these estimates against the lowest quintile to derive an overall OR for that study, which was then used for pooled analysis.
We used a random-effects model independently of the existence of statistical heterogeneity because we combined studies with different designs and populations. We presented the results stratified according to study design (cohort and case–control) in order to explore differences in the outcome estimate. Statistical heterogeneity was assessed with the I2 test, which measures the percentage of total variation between studies due to heterogeneity.23 If significant heterogeneity was found, we planned to perform a sensitivity analysis excluding studies of poorer quality to explore the impact of the study quality on the results.
We planned to conduct three subgroup analyses: (1) to explore the effect of the level of caffeine exposure on AF risk; (2) to explore the effect of the source of caffeine on AF risk; (3) to explore the effect of the length of follow-up. For the first analysis, we considered three levels of caffeine intake: low, moderate and high. The highest category of exposure reported in each study was considered for the group of high caffeine intake, independently of the cut-off value used in that study. We considered low intake to be <350 mg, moderate intake 350–699 mg and high intake ≥700 mg. Many factors contribute to the caffeine content of beverages and food and differences exist in caffeine content of a given caffeine source (in particular coffee) between, and even within, countries.24 ,25 For the purpose of this study, when cups of coffee were provided as a measure of caffeine intake, we considered each cup to have an amount of caffeine according to the geographic region of the study; UK/Northern Europe 140 mg; Southern Europe 50 mg; USA 85 mg.25 When a study’s interval for caffeine consumption crossed two of the above mentioned categories, we considered for analysis the mean value of caffeine intake.
For the second subgroup analysis, we considered only studies that took coffee consumption as the sole source of caffeine exposure because this beverage was the main source of caffeine.
In the third subgroup evaluation, we assessed pooled estimates according to follow-up, using the value of 10 years as the threshold (<10 vs ≥10 years).
Publication bias was assessed through visual inspection of funnel plot asymmetry and the Peters regression test.26
An electronic database search performed in December of 2012 yielded a total of 266 published references. Following our inclusion and exclusion criteria, we were able to include seven studies for analysis. Online supplementary figure S1 shows the detailed results of the search strategy. The main reasons for excluding studies were the type of publication (eg, basic science studies and case reports) and failure to report exposure to caffeine or caffeine-containing products. Eleven studies were excluded in the late stages of the selection process. These included patients already in AF (n=3), studies assessing acute effects of caffeine mostly within 72 h (n=3), abscence of AF incidence data (n=1), and review papers (n=4). Reviews were analysed to retrieve any missing study not identified by the electronic search.27–30
Description of studies
Seven studies fulfilled the inclusion criteria. One study had a case–control design31 and six were cohort studies.32–37 The case–control study had a retrospective design, five cohort studies were prospective, and the remaining one was unclear.32 Four studies reported data exclusively for coffee,31 ,32 ,34 ,37 whereas the other three referred to overall caffeine-containing beverages/food.33 ,35 ,36 Three studies were conducted in the USA and four in Europe (three in Scandinavia and one in Italy). The lengths of the studies were considered to be reasonable, with a mean follow-up ranging from 4 to 25 years in the cohort studies.
Overall, 115 993 patients were included in these seven studies (ranging from 232 to 57 053 patients). The mean age of participants at baseline varied between 51 and 62 years-old. Table 1 shows the main study characteristics.
The reporting quality of the included studies was globally acceptable (see online supplementary figure S2). All studies showed correct reporting of participants’ inclusion/exclusion criteria and correctly fulfilled the outcome assessment criterion. The method of caffeine exposure ascertainment was unclear in the study of Wilhelmsen and colleagues.32 Neither the Wilhelmsen nor the Mattioli study reported their estimates adjusted for several potential confounders.31 ,32 Wilhelmsen et al only reported the age-adjusted estimate.32 For the Mattioli study, we had to derive the crude OR from the raw data.31
Risk of AF
There was no significant association between caffeine exposure and AF risk (OR 0.92; 95% CI 0.82 to 1.04; I2 72%). We found similar results when considering pooled results from cohort studies (OR 0.91; 95% CI 0.81 to 1.02; I2 73%) and from the single case–control study (OR 1.36; 95% CI 0.84 to 2.19), which had a small weight in the overall pooled analysis (4.7%). Figure 1 shows the forest plot with individual study results and pooled analysis.
Significant heterogeneity existed between the results of the studies (I2=72%). Sensitivity analysis after exclusion of studies of poorer quality (studies in which the caffeine exposure assessment method was unclear32 and/or that failed to report AF risk estimates adjusted for multiple counfounders31 ,32) explained about half the heterogeneity among results (decrease of I2 from 72% to 39%). Moreover, the pooled estimate became significant with a 13% reduction in the odds of AF among caffeine consumers (OR 0.87; 95% CI 0.80 to 0.94; I2=39%).
Analysis according to the level of caffeine exposure, showed that people with a low intake may be at lower risk of AF (OR 0.85; 95% CI 0.78 to 0.92; I2 0%). No differences existed for the other categories (figure 2).
Four studies (three cohort studies and one case–control study) reported caffeine exposure based on coffee consumption estimates (figure 3).31 ,32 ,34 ,37 Results were similar to overall caffeine exposure (OR 0.94; 95% CI 0.72 to 1.22; I2=85%).
We performed an aggregation of studies according to mean/median follow-up, taking 10 years as the threshold. We found a statistically significant association between caffeine exposure and AF protection (OR 0.88; 95% CI 0.78 to 0.98; I2=49%) in the subgroup of studies with <10 years of follow-up. The pooled estimate of the two studies with ≥10 years of follow-up did not show any significant association with increased or decreased risk of incident AF (OR 0.99; 95% CI 0.71 to 1.36; I2=92%).32 ,37 There were no differences between the estimates of the group with <10 years follow-up and the group with ≥10 years of follow-up (p=0.51).
Visual inspection of funnel plots shows symmetry, suggesting that publication bias was not a major drawback of our review (see online supplementary figure S3). However, direct interpretation of funnel plots with a small number of studies is not recommended and often inconclusive.38 A funnel plot is shown in online supplementary figure S3. The Peters regression test did not show any significant publication bias (p=0.134).
The main finding of this systematic review is that the best evidence available does not support the hypothesis that caffeine exposure increases the risk of AF.
Our study had the merit of overviewing studies and giving an estimate with increased statistical power through meta-analysis. Heterogeneity was significant, and many factors could have contributed to this result. We included both cohort and case–control studies, and follow-up was different among the studies. Nevertheless, the most important source of heterogeneity can be attributed to differences in the estimates reference category and outcome adjustments to confounders. In fact, sensitivity analysis after exclusion of studies of poorer quality showed significantly reduced heterogeneity and suggests an inverse association between caffeine exposure and AF risk. Three of the seven studies reported estimates of caffeine/coffee consumption compared with the first quintile. The mean/median caffeine intake in these cases was about 22–23 mg/day, but the Danish Diet, Cancer, and Health Study first quintile of caffeine consumption was 10 times higher (248 mg/day) because of the high average daily caffeine intake in Northern Europe.33 The Mattioli and Wilhelmsen studies did not report either minimum or other adjustments in outcome estimates.31 ,32
Myers has previously reviewed the clinical data on the association between caffeine exposure and risk of cardiac arrhythmias. He found no increase in the frequency or severity of cardiac arrhythmias in healthy subjects, or in patients with ischaemic heart disease or known ventricular arrhythmias.30 Klatsky et al37 also failed to detect any increased risk of global arrhythmias. In fact, a small yet significant risk reduction was found in this study for arrhythmias among coffee drinkers (HR 0.97 per cup per day; 95% CI 0.95 to 0.99). The acute effect of caffeine on heart rate has been studied in randomised controlled trials, and the results point towards a decreased heart rate, particularly in the subset of individuals who are not caffeine-naïve.39 ,40 Previous ECG studies showed that caffeine exposure (by means of coffee consumption) had no effect on the analysis of the signal-averaged P-wave from atrial ECG.41 ,42
Previous systematic reviews have addressed the question whether caffeine exposure (mainly based on coffee) increase the incidence of conditions considered to be AF risk factors. Zhang and colleagues showed that coffee intake did not increase the risk of arterial hypertension.43 Similarly, coffee intake does not appear to be associated with coronary heart disease, another important risk factor for AF.9 Evidence exists for a reduced risk of heart failure among coffee consumers.11
In this study, we found that subjects in the low-dose caffeine exposure category may benefit from a reduction in AF risk. We can only speculate whether this finding is a sketch of a J-shape curve, a well-known epidemiological phenomenon,44 ,45 because of the relatively small number of studies included. This putative dose–response effect was in fact documented in a previous meta-analysis for coffee intake and heart failure incidence.11
Our review has limitations inherent to the included studies themselves and to meta-analytical methodology.
Pooling data of studies with different designs that evaluated different populations should also be considered a potential limitation. Nevertheless, it increases the power and external validity of the data obtained. Furthermore, pooled results of studies of acceptable quality provided estimates with low heterogeneity and suggestive of a protective effect.
In most studies, caffeine exposure is estimated on the basis of coffee consumption. Although caffeine is one of the major components of coffee, other substances such as chlorogenic acid, cafestol, kahweol, flavonoids, melanoidins and quinide can contribute to the pleiotropic effects of coffee.46 ,47 The association in such studies may have been biased.
There are no data to support the hypothesis that long-term caffeine exposure is associated with an increased risk of AF. Conversely, the exposure to low doses of caffeine may offer a small protective effect against AF.
This web only file has been produced by the BMJ Publishing Group from an electronic file supplied by the author(s) and has not been edited for content.
Files in this Data Supplement:
If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.