Lipid levels and risk of new‐onset atrial fibrillation: A systematic review and dose‐response meta‐analysis

Abstract Lipid levels are closely associated with health, but whether lipid levels are associated with atrial fibrillation (AF) remains controversial. We thought that blood lipid levels may influence new‐onset AF. Here, we used a meta‐analysis to examine the overall association between lipid levels and new‐onset AF. PubMed and EMBASE databases were searched up to 20 December 2019. We conducted a systematic review and quantitative meta‐analysis of prospective studies to clarify the association between lipid levels and the risk of new‐onset AF. Sixteen articles with data on 4 032 638 participants and 42 825 cases of AF were included in this meta‐analysis. The summary relative risk (RR) for a 1 mmol/L increment in total cholesterol (TC) was 0.95 (95% CI 0.93‐0.96, I2 = 74.6%, n = 13). Subgroup analyses showed that follow‐up time is a source of heterogeneity; for low‐density lipoprotein cholesterol (LDL‐C), RR was 0.95 (95% CI 0.92‐0.97, I2 = 71.5%, n = 10). Subgroup analyses indicated that adjusting for heart failure explains the source of heterogeneity; for high‐density lipoprotein cholesterol (HDL‐C), RR was 0.97 (95% CI 0.96‐0.99, I2 = 26.1%, n = 11); for triglycerides (TGs), RR was 1.00 (95% CI 0.96‐1.03, I2 = 81.1%, n = 8). Subgroup analysis showed that gender, age, follow‐up time, and adjustment for heart failure are sources of heterogeneity. Higher levels of TC, LDL‐C, and HDL‐C were associated with lower risk of new‐onset AF. TG levels were not associated with new‐onset AF in all subjects.


| INTRODUCTION
Atrial fibrillation (AF) is the most common type of sustained cardiac arrhythmia and currently affects over 2.3 million American adults, a number that is expected to more than double in the next five decades. 1 AF is associated with increased risk of heart failure, stroke, and death from cardiovascular disease. Major risk factors for AF include age, white race, European, obesity, hypertension, lack of physical activity, sedentary lifestyle, smoking, alcohol intake, diabetes mellitus, and obstructive sleep apnea. Many of these predictors are also risk factors for coronary heart disease (CHD). Hyperlipidemia is a major contributor to the development of atherosclerosis and CHD.
Higher levels of low-density lipoprotein cholesterol (LDL-C) and lower levels of high-density lipoprotein cholesterol (HDL-C) have been consistently associated with increased risk of CHD. Lowering of LDL-C and total cholesterol (TC) with statins reduces the risk of coronary events. However, a large randomized clinical trial (ALLHAT) showed no relationship between use of pravastatin and incidence of AF. These data are consistent with a meta-analysis in which an analysis of randomized controlled trials showed no significant effect of statins on the incidence of AF. 2 Since hyperlipidemia is a risk factor for other cardiac conditions, it seems likely that hyperlipidemia would also be a risk factor for AF. There is, however, a "cholesterol paradox" in AF 3,4 and the association between lipid levels and the risk of new-onset AF is less clear.
Many recent epidemiological studies have explored association between lipid levels and the risk of new-onset AF, some studies showed no significant association, [5][6][7][8][9][10][11] and some studies were associated with lower risk. 4,[12][13][14][15][16][17] Given the increasing prevalence of AF globally, establishing the association between lipid levels and newonset AF is of major public health importance. For these reasons, we conducted a systematic review and meta-analysis of prospective studies exploring the link between lipid levels and risk of new-onset AF in order to clarify the direction and strength of the association.

| METHODS
We followed the preferred reporting protocol for systematic reviews and meta-analyses set out in the PRISMA Statement 18 (CRD42020162579).

| Inclusion and exclusion criteria
We used the following inclusion criteria: (a) the paper evaluated associations of lipid levels with the incidence of AF, (b) the paper provided adjusted relative risk estimates (hazard ratio, risk ratio) with 95% confidence intervals (CIs), and for dose-response analyses, provided a quantitative measure of exposure and the total number of cases and person-years or continuous risk estimate, and (c) when multiple publications were available from the same study, we used the study with the largest number of cases. We excluded the literature using the following criteria: (a) when multiple publications were available from the same study, we used the study with the largest number of cases; (b) when the publication was a meeting abstract with insufficient data available online; and (c) when the publication described a meta-analysis, review, case-control study, or cross-sectional study. We reviewed all relevant studies and identified 64 published articles that discussed the association between blood lipid levels and incidence of AF; 16 articles met our inclusion criteria.

| Data abstraction
Data abstraction was carried out independently by two authors (Yisong Yao and Yangyang Wang) and disagreements were resolved through discussion. The following information was abstracted: first author's last name, publication year, country where the study was conducted, study period, sample size, number of cases/controls, exposure variables, exposure levels, adjusted relative risk estimates (hazard ratio, risk ratio), and 95% CIs for the highest vs the lowest level of the exposure variable in the publication.

| Quality assessment
Quality assessment of the publications was carried out using the Newcastle-Ottawa Quality Assessment Scale (NOS), which has a ninepoint scale (four for quality of selection, two for comparability, and three for quality of outcome and adequacy of follow-up). The literature was divided into high quality (score ≥8) and low quality (score <8) Table 1).

| Statistical analysis
All of our results were analyzed using the Stata 14.0 software. We calculated summary relative risks (RRs) and 95% CIs for 1 mmol/L increments in TC, LDL-C, HDL-C, and TGs, using a random effects model or a fixed effects model, which takes into account heterogeneity between studies. RRs were pooled using a fixed effect model if I 2 was lower than 50%, otherwise the random effect model was used. Publication bias was estimated using Egger's test and Begg's test. Subgroup analyses were completed using the characteristics of studies to find sources of heterogeneity. To verify the stability of our results, a sensitivity analysis was performed, in which one study at a time was removed and the remaining studies analyzed to evaluate whether the result could have been affected markedly by a single study.
We used the method described by Orsini et al to analyze doseresponses of lipid levels and calculated study-specific slopes (linear trends) and 95% CIs from the natural logarithms of the reported RRs and CIs across categories of each lipid level. 19 The mean lipid level in each category was assigned to the corresponding RR for each study and, for studies that reported exposures in ranges, we calculated the average of the upper and the lower cutoff points and used this as a midpoint. When the lowest or highest category was open-ended or had an extreme range, we used the width of the adjacent interval to calculate an upper or lower cutoff value. For studies that reported continuous risk estimates per 10 mg/dL or per 1.17 mmol/L, these risk estimates were converted to a risk estimate per 1 mmol/L lipid by taking the natural logarithm (ln) of the RR (95% CI), dividing the ln (RR, 95% CI) by the increment reported, multiplying by 3.9 and then back transforming to a nonlogarithmic scale before inclusion in the metaanalysis. A potential nonlinear dose-response relationship of TC, LDL-C, HDL-C, and TGs with risk of new-onset AF was examined using fractional polynomial models. 19 We determined the best-fitting second order fractional polynomial regression model, defined as the one with the lowest deviance. A likelihood ratio test was used to assess the difference between nonlinear and linear models to test for nonlinearity. 19 Since the nonlinear dose-response analysis requires that data are reported for at least three of the categories TC, LDL-C, HDL-C, and TGs, studies that reported only a continuous risk estimate and not categorical data were excluded from the analysis.

| RESULTS
The processes for retrieving and filtering articles, together with outcomes, are shown in Figure 1. After removal of duplicates, a preliminary screen of our search terms in the PubMed and EMBASE databases identified 1678 citations. We were able to exclude 1614 citations that did meet our criteria by reading the title or abstract, leaving 64 articles. After reading the full texts of these 64 articles, 48 were excluded (Supplementary Table 2). Four articles were excluded because they were a duplicate report on the same cohort population and 28 articles were excluded because insufficient information was available online. Two reviews, one meta-analysis, two case-control studies, and 11 cross-sectional studies were also excluded, leaving 16 articles 4-16,20-22 for inclusion in our systematic review ( Figure 1).
The characteristics of these 16 publications are shown in Table 1. The age range or mean age for each study is provided; the lowest and highest ages of participants across all studies were 18 and 96 years, respectively. The duration of follow-up was between 3.28 and 40 years. One study included only men; fifteen cohort studies included both men and women. Four studies were carried out in Asia and 12 studies were carried out in the United States or Europe.

| Heterogeneity test
Significant heterogeneity was found in our meta-analysis. TC, LDL-C, and TGs were used for subgroup analyses. Subgroup analyses may be the source of heterogeneity.
In the case of TC, two factors need to be taken into account.
First, age should be taken into account since this has a significant impact on AF; heterogeneity was noticeably different among the age F I G U R E 3 Forest plot for LDL-C and risk of new-onset AF, per 1 mmol/L LDL-C increase. AF, atrial fibrillation; CI, confidence interval; LDL-C, low-density lipoprotein cholesterol F I G U R E 4 Forest plot for HDL-C and risk of new-onset AF, per 1 mmol/L HDL-C increase. AF, atrial fibrillation; CI, confidence interval; HDL-C, high-density lipoprotein cholesterol subgroups in our study. Greater heterogeneity was found in the subgroups age < 50 years (I 2 = 78.80%, P < .001), age < 60 years (I 2 = 67.50%, P = .005), and age < 70 years (I 2 = 84.50%, P < .001); the subgroup age > 70 years showed less heterogeneity (P = .71, I 2 = 0.0%). Second, follow-up time may also influence heterogeneity.
In the case of LDL-C, two factors must also be taken into account.
First, gender should be taken into account since this has an obvious impact on AF. Less heterogeneity (P = .19, I 2 = 40.50%) was found in the subgroup women; higher heterogeneity was found in the subgroups men (P < .001, I 2 = 79.50%) and men and women (P = .11, I 2 = 54.2%).
Second, heart failure should also be taken into account since this has a significant impact on AF. Less heterogeneity (P = .10, I 2 = 43.80%) was found in the subgroup that was adjusted for heart failure and higher heterogeneity (P < .001, I 2 = 84.10%) was found in the subgroup that was not adjusted for heart failure (Supplementary Figure 6).
In the case of TGs, four factors should be taken into account.
First, gender should be taken into account since this has a significant impact on AF. Less heterogeneity was found in the subgroups women (P = .249, I 2 = 28%) and men (P = .557, I 2 = 0%); higher heterogeneity was found in the subgroup men and women (P = .015, I 2 = 59.8%,).
The summary RR was 0.93 (95% CI 0.92-0.95, I 2 = 0, n = 3) per 1 mmol/L increment in men. Second, age has a significant impact on AF. Less heterogeneity was found in the subgroup age 51 to 60 years (P = .807, I 2 = 0%,) and the subgroup age ≥ 70 years (P = .23, I 2 = 30.4%). Third, follow-up time has a significant impact on AF. Less heterogeneity was found in the subgroup follow-up time < 5 years (P = .364 I 2 = 7.5%) and the subgroup follow-up time > 10 years (P = .372, I 2 = 0%). Fourth, heart failure has a significant impact on AF. Less heterogeneity was found in the subgroup that was adjusted for heart failure (P = .806, I 2 = 0%) and higher heterogeneity was found in the subgroup that was not adjusted for heart failure (P < .001, I 2 = 76.8%) (Supplementary Figure 7).
Finally, we performed sensitivity analysis. Since all the results were statistically significant, we had achieved a relativity stable outcome.

| Major outcomes
To our knowledge, this is the first meta-analysis of lipid levels and the risk of new-onset AF. There was a 5% decrease in RR per 1 mmol/L increase in TC, a 5% decrease in RR per 1 mmol/L increase in LDL-C, and a 3% decreases in RR per 1 mmol/L increase in HDL-C. No association was observed between TGs and new-onset AF in all subjects, but subgroup analysis found that the summary RR was 0.93 (95% CI 0.92-0.95, I 2 = 0, n = 3) per 1 mmol/L increment in men. The doseresponse curve between TC and AF was U-shaped and that between LDL-C and AF had a reversed spoon shape; a linear association was observed between HDL-C and AF.

| Possible biological mechanisms
Several mechanisms may explain the negative correlation between blood lipids and new-onset AF. that is attributable to multiple changes at each step in the pathway. 33 TG levels are known to increase in inflammation and several cytokines are known to increase TG levels. Higher levels of TG reflect the level of inflammation within the host, but in our study, there was no association between TG and new-onset AF. In our study, we found a nonlinear (U-shaped) association between TC and new-onset AF. We also found a nonlinear (reverse spoon-shaped) association between concentrations of LDL-C and the risk of new-onset AF. Our study suggests that as a risk factor for new-onset AF, neither "the higher, the better" nor "the lower, the better" is correct in terms of cholesterol concentration. [36][37][38]

| Previous studies
Although most of the studies included here found that high cholesterol levels were associated with lower risk of new-onset AF, the dose-response curve was not clear and the best concentration range was unknown. Our quantitative meta-analysis found a non-linear relationship between both TC and LDL-C and new-onset AF. A negative relationship between HDL-C and new-onset AF was also found.

| Limitations
In our meta-analysis, although higher TC, LDL-C, and HDL-C were associated with lower risk of new-onset AF, there are still some problems. First, our meta-analysis showed an obvious heterogeneity between studies, which may affect the reliability of the results of the meta-analysis and means that careful interpretation is needed. Second, there was publication bias with TC, which may affect the reliability of the results. Third, the heterogeneity was partially improved after subgroup analysis of follow-up years, study quality, lipidlowering therapy, valvular atrial fibrillation, and diabetes. We believe that despite significant heterogeneity among studies, higher TC, LDL-C, and HDL-C were associated with lower risk of new-onset AF. Our results provide an epidemiological basis for the underlying trials, but follow-up studies on the relationship between new-onset AF and cholesterol are needed.

| CONCLUSIONS
Our meta-analysis suggests that higher levels of TC, LDL-C, and HDL-C were associated with a lower risk of new-onset AF and that TG levels were not associated with new-onset AF across the entire study population. TC concentrations in the range 232 to 238 mg/dL (6.009-6.164 mmol/L) were associated with lower risk of new-onset AF, LDL-C concentrations in the range 133 to 149 mg/dL