The association of hypertriglyceridemia with cardiovascular events and pancreatitis: a systematic review and meta-analysis

Background Hypertriglyceridemia may be associated with important complications. The aim of this study is to estimate the magnitude of association and quality of supporting evidence linking hypertriglyceridemia to cardiovascular events and pancreatitis. Methods We conducted a systematic review of multiple electronic bibliographic databases and subsequent meta-analysis using a random effects model. Studies eligible for this review followed patients longitudinally and evaluated quantitatively the association of fasting hypertriglyceridemia with the outcomes of interest. Reviewers working independently and in duplicate reviewed studies and extracted data. Results 35 studies provided data sufficient for meta-analysis. The quality of these observational studies was moderate to low with fair level of multivariable adjustments and adequate exposure and outcome ascertainment. Fasting hypertriglyceridemia was significantly associated with cardiovascular death (odds ratios (OR) 1.80; 95% confidence interval (CI) 1.31-2.49), cardiovascular events (OR, 1.37; 95% CI, 1.23-1.53), myocardial infarction (OR, 1.31; 95% CI, 1.15-1.49), and pancreatitis (OR, 3.96; 95% CI, 1.27-12.34, in one study only). The association with all-cause mortality was not statistically significant. Conclusions The current evidence suggests that fasting hypertriglyceridemia is associated with increased risk of cardiovascular death, MI, cardiovascular events, and possibly acute pancreatitis. Précis: hypertriglyceridemia is associated with increased risk of cardiovascular death, MI, cardiovascular events, and possibly acute pancreatitis


Background
Hypertriglyceridemia is a manifestation of several common metabolic disorders in the western world. A recent cross-sectional study found that over 33% of adults in the United States had hypertriglyceridemia (serum triglyceride levels over 150 mg/dl (1.7 mmol/L)) of whom over 50% had serum triglyceride levels exceeding 200 mg/dl (2.2 mmol/L) [1].
The association of hypertriglyceridemia and clinically important complications such as cardiovascular events and acute pancreatitis has been suggested by several studies. Previous epidemiologic studies demonstrated increase in the risk of cardiovascular events although there has always been significant confounding due to varying levels of adjustments for traditional risk factors and other lipid subfractions [2][3][4]. As for pancreatitis, case series and uncontrolled studies reported that very severely elevated triglyceride levels are associated with lipemic serum, chylomicronemia syndrome, and increased risk of pancreatitis [5][6][7]. Serum triglycerides levels of 1000 mg/dl (11.3 mmol/L) and higher have been observed in 12% to 38% of patients presenting with acute pancreatitis [5]. However, the association with pancreatitis has not been evaluated in controlled studies or with less severe hypertriglyceridemia.
To update the evidence base to the present time (last meta-analysis [2] was performed 6 years ago), we conducted this systematic review and meta-analysis. Our goal was to assess the magnitude of association and the quality of supporting evidence linking hypertriglyceridemia with cardiovascular events, mortality and pancreatitis. We specifically aimed at comparing association measures in studies with varying levels of adjustment for cardiovascular risk factors and to search for controlled studies evaluating the risk of pancreatitis.

Methods
This systematic review was conducted according to a priori established protocol that was commissioned and funded by the Endocrine Society and is reported according to the PRISMA statement (Preferred Reporting Items for Systematic Reviews and Meta-analyses) [8].

Eligibility criteria
Eligible studies were randomized and observational studies that enrolled patients with untreated hypertriglyceridemia and reported a relative association measure between fasting serum triglycerides levels and the outcomes of interest: all-cause mortality, cardiovascular death, cardiovascular events and pancreatitis. We excluded uncontrolled studies and studies of nonfasting hypertriglyceridemia. The detailed search strategy is available in Additional file 1. We also sought recommendations from content expert for potentially relevant studies to be included in the screening process.

Study identification and data extraction
Reviewers working independently and in duplicate assessed each abstract for eligibility. Disagreements yielded an automatic inclusion into the following level of screening. Included studied were retrieved and full text screening commenced in duplicate as well. Disagreements in this level were resolved by discussion and consensus. Online reference management system was used to conduct this review and it reported a real-time chance-adjusted agreement (kappa) statistic to evaluate the agreement among reviewers. Kappa averaged 0.80. Two reviewers working independently and in duplicate extracted baseline and outcome data and assessed the quality of included study. A third reviewer compared the reviewer's data and resolved inconsistencies by referring to the full text article.

Quality
Using the Newcastle-Ottawa scale, [9] reviewers assessed the quality of included observational studies (and control arms of RCT, considered as observational cohorts) by determining outcome ascertainment, adjustment for confounders, proportion of patients lost to follow-up as well as sample selection. We used the GRADE approach in evaluating the evidence yielded from included studies [10].

Statistical analysis
We pooled the relative association measures of relevant complications from included studies and analyzed the data using the random-effects model described by DerSimonian and Laird [11]. Heterogeneity in results across studies was measured using the I 2 statistic, which estimates the proportion of variation in results across studies that is not due to chance. An I 2 of 50% or more indicates large inconsistency between studies. Meta-analysis was completed using Comprehensive Meta-analysis (CMA) version 2.2 (Biostat Inc., Englewood, NJ).

Subgroup analyses and publication bias
A priori hypotheses were designed to explain betweenstudy inconsistencies in results. These analyses sought an interaction with whether triglycerides levels were adjusted for other lipid fractions or not; whether the underlying metabolic disorder was diabetes vs. not; and whether the association differed between men and women. Publication bias was evaluated by assessing the symmetry of funnel plots and using Egger's regression test. In this regression, the size of the treatment effect is captured by the slope of the regression line and bias is captured by the intercept [12].

Search results and included studies
Electronic search yielded 760 potentially eligible studies. Following screening, 60 studies met inclusion criteria, of which 35 reported data sufficient for metaanalysis [ Figure 1].

Methodological quality and risk of bias
Included studies had a fair methodological quality ( Table 2) with follow-up period reported by 85% of studies averaging 114 months; 58% of studies reported loss to follow-up of participants that ranged 0% to 27%. Adjustment for potential confounders was reported in 90% of studies and the outcome ascertainment method was reported in all studies. Cohort selection was random in 18% of the studies.

Meta-analysis
The total number of included studies was 35 enrolling 927,218 patients who suffered 132,460 deaths and 72,654 cardiac events; respectively. Hypertriglyceridemia was significantly associated with cardiovascular death, cardiovascular events, myocardial infarction, and pancreatitis; with odds ratios (95% confidence interval) of It is worth noting that the association with acute pancreatitis was estimated by only one eligible study that included 129 patients with severe hypertriglyceridemia (119 with type IV phenotypes and 10 with type V phenotypes according to Fredrickson's classification) of whom 26 suffered acute pancreatitis [33]. In this study, subjects in the third tertile of TG had a 4.0-fold increased risk (95% confidence interval, 1. 3-12.3) compared with the first tertile and those diagnosed with dyslipidemia at a younger age also had increased risk.
All analyses were associated with important heterogeneity (I 2 > 50%) that our planned subgroup analyses could only partially explain ( Table 3). The association of hypertriglyceridemia with mortality and cardiovascular mortality seemed to be stronger in women. These findings are consistent with a previous meta-analysis published in 1996. Hokanson and Austin estimated adjusted relative risks for incident cardiovascular events of 1.14 (95% Cl 1.05-1.28) in men and 1.37 (95% Cl 1.13-1.66) in women. The association with cardiovascular events was somewhat stronger in patients with diabetes although this effect was not statistically significant. Hence, there were no other significant subgroup interactions to explain heterogeneity (based on the level of adjustment for lipids subfractions, sex or the presence of diabetes).
There was no evidence of publication bias (P value for Eggers test > 0.05 for all outcomes) although these analyses were underpowered to detect this problem and the presence of heterogeneity further limits the ability to detect publication bias.

Discussion
We conducted a systematic review and meta-analysis and documented an association between fasting hypertriglyceridemia and the risk of several cardiovascular adverse events and with pancreatitis.

Limitations, strengths and comparison with other reports
The main limitation of association studies is the observational nature of the existing evidence. Therefore, confounders (particularly, baseline risk of patients for developing cardiovascular disease and the effect of other lipid subfractions abnormalities) threaten the validity of results. In meta-analyses of observational studies, the ability to adjust for confounding is limited by the level of adjustment conducted in the original studies. We attempted to evaluate confounding by conducting subgroup analysis; however, this analysis was underpowered. Other limitations pertain to heterogeneity of the metaanalytic estimates, publication bias (which remains likely in the context of observational studies that do not require prospective registration) and reporting bias (which is also likely considering that several studies met the eligibility criteria for this review but did not report the outcomes of interest) [48]. It was unclear in most studies if enrolled patients did not have some of the outcomes pre-existent at baseline and it was also unclear if patients were treated with drugs that can affect TG level (both of these elements lower the confidence in the observed associations). We only found one controlled study that evaluated the association with acute pancreatitis. Studies included in review (35) Studies with insufficient data for analysis (n=25) Figure 1 Study selection process.           The overall confidence in the estimated magnitude of associations is low [10]considering the described methodological limitations in evaluating the association with cardiovascular events; and imprecision (small number of events) in evaluating the association with pancreatitis.
The strengths of this study stems from the comprehensive literature search that spans across multiple databases and duplicate appraisal and study selection. Our results are consistent with previous evidence synthesis reports about the association of hypertriglyceridemia with cardiovascular events. We estimated increased odds by 37% (odds ratio of 1.37). Hokanson and Austin [3] estimated adjusted relative risks of 1.14 (95% Cl 1.05-1.28) in men and 1.37 (95% Cl 1.13-1.66) in women. Sarwar et al. [2] reported odds ratio of 1.73 in prospective cohort studies published prior to 2006. A systematic review by Labreuche et al. [49] demonstrated that baseline triglyceride levels in randomized trials is associated with increased stroke risk (adjusted RR, 1.05 per 10-mg/ dL (0.1 mmol/L) increase; 95% CI, 1.03-1.07). To our knowledge, this is the first systematic review that sought to identify controlled studies evaluating the association with pancreatitis.

Implications
The associations demonstrated between hypertriglyceridemia and cardiovascular risk should not necessarily translate into a recommendation for treatment. It is plausible that the benefits of lowering triglycerides do not merely depend on how much the level is lowered, but rather on how it is lowered (i.e., lifestyle interventions vs. pharmacological therapy). Therefore, randomized trials of the different approaches with patientimportant outcomes [50] used as primary endpoints are needed for making policy and clinical decisions.
Several systematic reviews and meta-analyses [49,[51][52][53][54] have summarized the evidence from randomized trials of fibrate therapy and demonstrated that fibrate therapy reduced the risk of vascular events (RR 0.75, 95% CI 0.65 to 0.86) in patients with high triglyceride levels or atherogenic dyslipidemia (low HDL cholesterol combined with high triglyceride level) although all-cause mortality and non cardiovascular mortality were both significantly increased in clofibrate trials. Meta-analyses [55,56] of niacin therapy demonstrate significant reduction in the risk of major coronary events (25% reduction in relative odds; 95% CI 13, 35), stroke (26%; 95% CI 8, 41) and any Odds ratio and 95% CI Odds Lower Upper ratio limit limit p-Value    15,37). However, contemporary trials in the statin era have failed to substantiate these findings with fenofibrate among patients with diabetes [57] and with niacin in high risk patients [58]. Also, to our knowledge, there are no trials assessing the value of triglyceride lowering to reduce the risk of pancreatitis. Thus, lifestyle changes should remain the mainstay of therapy. Treatment of the underlying metabolic disorder (e.g., insulin resistance) should also be an essential and first step in the management plan of hypertriglyceridemia.

Conclusions
The current evidence suggests that hypertriglyceridemia is associated with increased risk of cardiovascular death, MI, cardiovascular events, and acute pancreatitis. The strength of inference is limited by the unexplained inconsistency of results and high risk of confounding and publication bias.

Additional material
Additional file 1: Search Strategy.