Search strategy

We searched nine electronic databases from inception to July 2018 (eMethods 2). One author (NJ) screened all the abstracts and titles, and two authors screened the full texts (NJ, PA and KT (in pairs)). We manually checked the references of the full-text papers to ensure literature saturation.

Eligibility criteria

We included randomised controlled trials (RCTs) and cohort studies of weight change (measured by kg, BMI or percentage weight change) in adults who reported the incidence of paroxysmal, persistent or permanent AF as an outcome. There was no restriction by language or setting.

We excluded case–control studies, case reports and case series, as they could not assess the incidence of AF, and conference abstracts as they provided insufficient information to appraise the study. We excluded studies of bariatric surgery, as the rapid, large volume of weight loss along with nutritional and physiological differences induced by the surgery may not reflect the association between weight change and AF in the general population.

For each study, two authors (NJ and KT) independently extracted data, contacting authors as necessary.

Assessment of risk of bias and quality of the evidence

Two authors (NJ and KT) independently assessed the risk of bias using the Newcastle-Ottawa Scale (NOS) for non-randomised studies.14 Two authors (NJ and PA) independently assessed the risk of bias and strength of evidence across the studies, using the Grading of Recommendations Assessment, Development and Evaluation (GRADE) framework headings.15 A full GRADE assessment was precluded because not all studies reported the absolute number of people with incident AF in each weight change category. For each assessment, disagreements were discussed and resolved.

Data analysis and synthesis

We log-transformed the reported HRs across the weight change categories and calculated study-specific slopes (linear trends) using the method of Greenland and Longnecker and the Stata glst command of Orsini et al.16 17 Where possible, we undertook random effects meta-analysis based on the DerSimonian and Laird method to compute the summary HRs with 95% CIs of incident AF.18 We assessed statistical heterogeneity using the I-squared statistic and calculated approximate 95% prediction intervals to estimate the likely effect in an individual setting using the methods of Higgins et al.19 20 Where there were insufficient data to pool, we report results descriptively.

We could not carry out our prespecified subgroup analyses (by intentionality of weight loss, initial weight, comorbidity and gender) and instead, we subgrouped by population type and duration of follow-up in a post hoc analysis. We assessed whether the results were sensitive to the choice of model, quality of the studies, whether data were estimated from the presented results and the heterogeneity of the patient populations. To do so, with separate analyses, we carried out fixed-effects analysis and excluded individual studies, studies with poor quality data and those for which we estimated data. There were insufficient data to carry out meta-regression or assess publication bias (exploratory contour plots are included in the online supplementary file (eMethods 3). All analyses were carried out using Stata V.14.2.

Study characteristics

From 5407 records, we screened 45 full texts, of which 10 met the eligibility criteria (figure 1). Nine were cohort studies (table 1). In the single RCT, Alonso et al randomised participants with type 2 diabetes (T2DM) to an intensive weight loss intervention or a diabetes support and education programme.21 We included the combined data for these two groups, reported in a multivariate analysis adjusted for a range of important covariates including age, blood pressure and intervention arm (eTable 1).

• Download figure
• Open in new tab
• Download powerpoint

Figure 1

Preferred Reporting Items for Systematic Reviews and Meta-analyses flow diagram of study selection. AF, atrial fibrillation.

In general, participants were middle-aged people attending routine health checks, but in two studies all participants had T2DM.21 22 AF incidence rates were higher in studies of people with diabetes or cardiovascular risk factors compared with population surveys of healthy people (table 1). In four studies, most participants had a healthy weight at baseline,23–26 whereas in six most were overweight or obese. Only the RCT of the weight loss programme assessed whether participants intended to lose weight.21 Follow-up varied between 4.6 and 35 years. Six studies reported the incidence of AF during follow-up after the period of measured weight change, whereas four reported the association between weight change and AF incidence across the same period (table 2). Nine studies used linked electronic health records to capture incident cases of AF (table 1). In four studies, this was triangulated with ECGs at study follow-up visits. One study relied on a single follow-up visit ECG to capture incident cases.27 In one study, AF was detected by 7 day Holter monitoring.28

Weight gain and risk of incident AF

Percentage weight gain

Pooled analysis from six populations (five studies) showed that for a 5% gain in weight, the incidence of AF increased by 13% (HR 1.13, 95% CI 1.04 to 1.23, with 95% prediction interval 0.88 to 1.44) (figure 2), with moderately high statistical heterogeneity. Excluding the study by Rosengren et al reduced the I² statistic from 70% to 33%.29 Most participants were overweight or obese at the outset in five of the six populations.

• Download figure
• Open in new tab
• Download powerpoint

Figure 2

Risk of 5% weight gain on incidence of atrial fibrillation.

Sensitivity analysis generally produced consistent results (eTable 2). Outcome data were converted to percentage weight gain for two studies included in the pooled analysis.22 23 Excluding these studies did not change the point estimate for incident AF but the result was no longer statistically significant (eTable 2). In both studies of general populations and in the single study of a population with T2DM,22 a 5% gain in weight was associated with a significant risk of AF, and the same result applied for both studies with follow-up less than and greater than 10 years (eTable 2).

Supplementary file 5

table 2

[heartjnl-2019-314931supp006.docx]

Weight loss and risk of incident AF

Percentage weight loss

In the RCT, the intervention group achieved a mean percentage weight loss of 6.0%, compared with 3.5% in the control group, and the HR for the onset of AF was 0.99 (95%CI 0.77 to 1.28).21 As results were reported by trial arm, those who did not lose weight were included in the analysis and therefore the results may underestimate any association between weight loss and AF.

In a pooled analysis from seven populations (five studies), there was no statistically significant association between 5% weight loss and incidence of AF (HR 1.04, 95% CI 0.94 to 1.30 with a prediction interval 0.77 to 1.42) (figure 3). In all the studies, most participants were overweight or obese at baseline. Grundvold et al reported no significant difference in risk of AF between those who were obese at baseline and lost weight compared with the whole population in the only such subgroup analysis.22

• Download figure
• Open in new tab
• Download powerpoint

Figure 3

Risk of 5% weight loss on incidence of atrial fibrillation.

There was high statistical heterogeneity (I²=73%). Excluding the study by Huxley et al 30 dropped the I² to 0% and resulted in a minor but significant reduction in incidence of AF (HR 0.95, 95% CI 0.90 to 0.99) (eTable 3). In both studies of general populations and in the single study of a population with T2DM, there was no significant association between weight loss and AF. In studies with follow-up less than 10 years, there was a significant reduction in risk of AF and in the study populations with follow-up more than 10 years, there was a significant increase in risk (eTable 3).

Supplementary file 6

table 3

[heartjnl-2019-314931supp007.docx]

Rosengren et al found no evidence that losing more than 4% body weight from age 20 years to midlife was associated with a significant change in the incidence of AF compared with individuals whose weight only changed by between −4% and 4% in the same period (HR 1.00, 95% CI 0.72 to 1.27).29 These results could not be pooled.

Quality assessment

The NOS scores ranged from 5 to 7 reflecting a moderate to low risk of bias (table 3). Based on the GRADE framework, there was moderate certainty in the 5% weight gain summary finding and very low certainty in the summary estimate of 5% weight loss. Further details of the quality assessments are included online (eResults).

Main findings

Weight gain was associated with an increased risk of AF in our pooled and unpooled analyses. Weight loss was not associated with an altered risk of incident AF in our pooled analysis. One unpooled study reported a significant association between weight loss and reduced AF incidence, another showed no association, and a third showed that those who dropped weight below the obesity threshold were no longer at an elevated risk of AF.24 26 29 In both pooled analyses, the majority of people were overweight or obese at baseline, although there were significant numbers of people with a normal BMI in most studies. Both results had high heterogeneity. There was moderate certainty in the weight gain summary finding, but only very low certainty in the weight loss finding.

Comparison with other studies

Our results suggest that weight gain itself in addition to BMI status may be linked to future risk of AF. Three systematic reviews have showed that obesity increases the risk of incident AF.7–9 One also analysed weight gain and AF, reporting no significant association between a 5% increase in bodyweight and AF incidence, based on two studies (relative risk of AF 1.08, 95% CI 0.97 to 1.19).7 Our review expands on this result, adding three further studies in the pooled analysis.

Strengths

This is the first systematic review to assess the risk of incident AF in relation to weight loss and extends our understanding of the association between weight gain and AF risk. We analysed data from 108 996 people across 10 individual studies, including seven in the pooled analyses. Most studies were conducted within the past 10 years across a range of economically developed countries and recruited middle-aged participants with a relatively low burden of comorbidity, providing contemporary, generalisable results.

Limitations

Pooling data from people who were healthy weight with those who were obese or overweight at baseline may have diluted the summary estimates and thus underestimate the effect of weight loss on the subsequent risk of AF. Most studies had significant numbers of people who had a normal BMI at baseline and there were insufficient data to perform the planned subgroup analysis by baseline weight. The weight gain studies which did split by baseline weight status reported very similar estimates of AF risk across groups. Some studies reported AF incidence during follow-up after the period when weight change was measured and it is possible that further weight change may have occurred that altered the subsequent risk of AF. There was inconsistency in the results of our analysis of weight loss measured by percentage change and change in BMI, and we were unable to summarise the overall effect as it was not possible to pool all these data. Studies presented data differently, which we sought to overcome by making conservative approximations to convert data into a common form to enable pooling from more studies.

Some studies relied on periodic ECG assessments, which may miss cases of paroxysmal AF and so underestimate true incidence, but this is unlikely to have biased differences by weight change status. Many studies attempted to triangulate their case detection with other electronic health records. Two studies calculated weight change using a recalled weight as the baseline measure, but these results were consistent with other studies.29

Nine of our 10 studies were observational. We included results from multivariate analyses but relevant covariates such as physical activity and history of cardiovascular disease were sometimes excluded.

Policy implications

Weight gain of 5% was associated with an increase in the risk of AF of 13% (95% CI 4% to 23%), with the likely effect in an individual setting varying between 18% reduction and a 44% increase in risk. Our results suggest that the risk of incident AF continues to rise in relation to weight gain across all categories of weight status. This evidence supports public health interventions that promote maintaining a healthy weight and preventing weight gain to reduce the burden of AF. Given the paucity of data and high heterogeneity, our conclusions are only tentative.

Whether intentional weight loss can reduce the risk of developing AF remains uncertain and may be related to the amount of weight people lose.31 Large amounts of weight loss achieved through bariatric surgery has been reported to reduce the risk of AF by 29% compared with no weight change (HR 0.71, 95% CI 0.60 to 0.83) over long-term follow-up.32

Unintentional weight loss may occur as a result of diseases that are independent risk factors for AF, such as cancer, hyperthyroidism or muscle wasting conditions as suggested by one of the included studies, which reported weight loss being associated with a significant increased risk of developing AF.30

Future research

Obesity is the second highest population attributable risk factor for incident AF behind hypertension, accounting for over 10% of all cases.33 The link between obesity and incident AF has been established, but more research is needed to understand the degree of increased risk associated with weight gain, whether non-surgical weight loss can reduce the risk of incident AF and if this depends on baseline weight status, intentionality and amount of weight loss.