Growth differentiation factor‐15 and the risk of cardiovascular diseases and all‐cause mortality: A meta‐analysis of prospective studies

Background and Aim Previous studies have documented that the association between growth differentiation factor‐15 (GDF‐15) the risk of patients with cardiovascular diseases (CVDs). In this meta‐analysis, our main objective is to explore the associations between GDF‐15 and the risk of CVD or all‐cause mortality. Methods PubMed and ISI Web of Science (up to January 2018) electronic databases were browsed for eligible studies. The studies provided relevant data depicted as hazard ratio (HR) with 95% confidence interval (CI), with regard to the association between GDF‐15 levels and subsequent risk of CVDs or all‐cause mortality. A random‐effect model was applied to pool the HR and 95% CI. Results Thirty‐one prospective studies met the eligibility criteria involving 53 706 subjects with 7020 adverse outcome events. It was concluded that GDF‐15 levels were associated with an incremental risk of CVDs or all‐cause mortality. Highest GDF‐15 category was associated with greater risk of cardiovascular mortality (HR, 2.66; 95% CI, 1.69‐3.63), all‐cause mortality (HR, 2.52; 95% CI, 2.06‐2.97), and complex adverse outcome (HR, 1.81; 95% CI, 1.42‐2.21). As each log‐unit increment in GDF‐15 concentration, the corresponding risk of adverse events also escalated, cardiovascular mortality (HR, 2.11; 95% CI, 1.57‐2.66), all‐cause mortality (HR, 2.70; 95% CI, 2.29‐3.12), and complex adverse outcome (HR, 1.96; 95% CI, 1.64‐2.29). Conclusions Judging from the results of the data analysis, GDF‐15 levels may increase the risk of CVDs or all‐cause mortality.


| INTRODUCTION
Growth differentiation factor-15 (GDF-15), first named as macrophage inhibitory cytokine-1 (MIC-1),was a stress-response member of transforming growth factor-β cytokine superfamily. It was found that GDF-15 messenger RNA (mRNA) expressing increased during macrophage activation. 1 Normally GDF-15 is weakly expressed in most tissues under physiological conditions but its expression level may sharply upregulate in response to ischemia-reperfusion injury, reactive oxygen species, and mechanical stretch, possibly mediated through pro-inflammatory cytokine and oxidative stress dependent signaling pathways. 2,3 Moreover, it had suggested that elevated GDF-15 was a cardioprotective cytokine when exposed to cardiovascular injury in an animal model. 3 In humans, increased GDF-15 had been observed in atherosclerotic plaque macrophages. 4 To date, cardiovascular diseases (CVDs), a leading cause of mortality worldwide, have brought heavy burden to social healthcare and individuals. Thus, intensive investigation has been focused on controlling the risk factors aimed at lowering CVD risks. Plenty of clinical research has been conducted to explore the relationship between GDF-15 levels and CVDs. These experiments conclusively demonstrate that GDF-15 levels link to the adverse cardiovascular events across a spectrum of CVD conditions including heart failure (HF), chest pain, acute coronary syndromes (ACS), stable ischemic heart disease, stroke, and atrial fibrillation. [5][6][7][8] The potential ability of GDF-15 may attribute to the earlier diagnosis, risk stratification and prognosis assessment.
However, no systematic review and meta-analysis have analyzed the available data pertaining to the association between GDF-15 levels and CVDs or all-cause mortality. Hence, we perform a meta-analysis for the purpose of qualitatively and quantitatively assessing the relationship between GDF-15 levels and CVDs or all-cause mortality.

| Literature Search
Following the Preferred Reporting Items for Systematic Reviews and Meta-Analysis guideline, we searched PubMed and ISI Web of Science databases from January 1, 1950 to December 31, 2017 for the terms "growth differentiation factor-15" or "macrophage inhibitory cytokine 1"or "placental transforming growth factor beta" or "nonsteroidal anti-inflammatory drug-activated gene-1" or "prostatederived factor" and "cardiovascular disease" or "coronary heart disease" or "ischemic heart disease" or "myocardial infarction" or "heart failure" or "stroke" or "all-cause mortality"or "acute coronary syndrome" or "troponin." The retrieval process was independently completed by two authors (S. Xie and L. Liu). In addition, we also retrieved the reference list of the selected studies and recent reviews for obtaining further information. The literature search was restricted to human studies and published in English language.

| Inclusion and exclusion criteria
Studies were eligible for inclusion if they met the following criteria: (a) prospective cohort study, (b) follow-up duration of at least 3 months, (c) reported at least one of the interesting outcomes: major cardiovascular endpoints (cardiovascular death, myocardial infarction, stroke, and HF) or all-cause mortality, (d) provided relative risk (RR) or hazard ratio (HR) with 95% confidence interval (CI) for GDF-15 levels comparing the highest levels to the lowest or definite increases for risk factor as a continuous variable (data after logarithmic transformation). Studies were excluded if the study design was review articles, case-controlled studies ,retrospective cohort studies, commentaries, editorials, or case report; studies concerning ecological ,animal, or cell culture, genetic variation in GDF-15 relevant genes were not selected; we also excluded the study population based on non-adult.

| Data extraction
Two investigators (S. Xie and L. Liu) performed the relevant data extraction with discrepancies reconciled through deliberation with a third investigator (L. Lu).The information was extracted as follows: authors, publication year, sample size, characteristics of the baseline population, mean age of the participants, study location, mean levels of GDF-15, detective method, follow-up, study endpoints, total number of related events, RR or HR with 95% CI, covariates adjusted for in multivariable analyses. The data from different sub-cohort of the same study was extracted separately.

| Assessment of study quality
According to the Newcastle-Ottawa quality assessment scale (for cohort study), the assessment of including study quality was performed following three aspects: participants selection, comparability of groups, and ascertainment of outcome. A study can be awarded a maximum of one score for each numbered item within the selection.
A maximum of two scores can be given for comparability and selection. Higher scores of studies represented better quality.

| Statistical analyses
Various studies reported the results in different patterns and presented the effect sizes for comparison between groups or for a given unit of increase in GDF-15 levels. To display the results more concise and understandable, we assessed the association between GDF-15 categorical level and the risk of CVD mortality or all-cause mortality or complex adverse outcome (composite of death or cardiovascular events), by comparing the highest category of GDF-15 with the lowest. We also analyzed GDF-15 level as a continuous variable to evaluate the risk of CVD mortality or allcause mortality or complex adverse outcome based on per log-unit increase of GDF-15. Pooled HR with 95% CI was presented as an effect size for estimating the association between GDF-15 levels and the risk of CVD mortality or all-cause mortality or complex adverse outcome. In the process of handling the data, the SD of log-GDF-15 was estimated from the Framingham Heart Study, Framingham, MA. 9 A random-effect model was applied to pool the data across studies. 10 For assessing the extent of divergence, the heterogeneity of trials was examined by Q-statistic and quantified by the I 2 statistic. It was considered that a relatively larger extent of I 2 represented greater heterogeneity. We undertook metaregression analysis to explore the possible reason resulted in heterogeneity. According to the different characteristics of including studies, subgroup analyses were performed, respectively, stratified by the sample size (≤1000 or >1000), duration of follow-up (≤5 years or>5 years), assay method (enzyme-linked immunosorbent assay, radioimmunoassay, or others), whether to adjust the variable, and baseline population (general population, coronary heart disease, others). Sensitivity analysis was performed to appraise an excessive estimate of a single study by way of eliminating each study individually. An estimation of potential publication bias was performed by both Egger's linear regression test. All statistical analysis was performed with software package Stata version 12.0 (STATA Corp LP, Texas). P < 0.05 was identified with a statistical significance.

| Identification of studies
The procedures of literature retrieval and selection were present in Figure 1. We initially retrieved 886 relevant publications from the PubMed and ISI Web of Science electronic databases. The majority of these were excluded after screening the titles or abstracts, because of editorials/reviews/case reports/cross-sectional design or not related.

| GDF-15 and the risk of cardiovascular mortality
Eight studies reported cardiovascular mortality as outcomes. Four studies handling the data as a categorical variable demonstrated that the pooled HR for highest GDF-15 category vs lowest was 2.39 (95% CI, 1.36-3.41) ( Figure 2A). Moreover, when regarded GDF-15 level as a continuous variable, the pooled HR for cardiovascular mortality from six studies was 2.11 (95% CI, 1.57-2.66) per log-unit ng/L increment using a random effect model. ( Figure 2B). Figure 2C showed the pooled HR for all-cause mortality comparing the highest GDF-15 category with lowest of GDF-15 level (HR, 2.54; 95% CI, 2.07-3.01). When GDF-15 concentration increased one logunit ng/L, 2.7-fold of the risk of all-cause mortality correspondingly varied (HR, 2.70; 95% CI, 2.29-3.12 ( Figure 2D).

| GDF-15 and the risk of complex adverse outcome
We observed that GDF-15 concentration was associated with an increased risk of complex adverse outcome (HR, 1.80; 95% CI, 1.38-2.23), when pooling data from six studies reporting the estimates as categorical

| Heterogeneity measurement
As the heterogeneity inspection tools, meta-regression, analysis and sensitivity analysis were carried out aimed at exploring the potential sources of heterogeneity on all-cause mortality regarding GDF-15 as a categorical variable or continuous variable in our study. The detailed results were displayed in the Figure 3A, B. Meta-regression analysis was executed to exploit the heterogeneity stratified by such feasible causes, sample size, baseline population duration of follow-up, assay method, and adjustment. In evidence, the aforementioned four factors were unable to strike significant heterogeneity, but whether to adjust the confounding variables may be one of the sources of   heterogeneity. We conducted a sensitivity analysis based on all-cause mortality. In the sensitivity analyses, the results suggested that the pooled HRs or 95% CI did not reflect significant difference when omitting one of the studies from the analysis. (Figure S1 and S2, Supporting Information).

| Publication bias
Egger test showed that no publication bias was observed (coefficient   an indicator for the prognosis of major bleeding and death. 33 Owing to the incidence trend of ischemic heart disease, the application of GDF-15 on early diagnosis and prognosis of disease is widely studied.
In an animal model, it was confirmed that GDF-15 played a cardiopro- provided more valuable information for risk stratification and prognosis in CVDs. 11 An epidemiological research demonstrates that GDF-15 levels are associated with carotid artery intima-media thickness, plaque burden, and endothelial dysfunction in elderly individuals, which may provide insight into disparate mechanism of GDF-15 pathophysiology. 40 Some studies included in our meta-analysis demonstrate that GDF-15 has performed well both in identifying stable and unstable CAD and in evaluating the prognosis, and likewise, independent of traditional clinical risk biomarkers, such as troponin T, NT-proBNP, and hs-CRP. 6,13,15 To find the potential heterogeneity, we did metaregression analysis and sensitivity analysis to identify conceivable sources of the discrepancy from the studies that reported all-cause mortality as endpoints. Several probable aspects were filtered for the subgroup analyses. Paradoxically, the results of meta-regression analysis and sensitive analysis was failed to elucidate that where the heterogeneity stems from genuinely. It was inexplicit that which was the initiator of the difference. Furthermore, the results of the sensitivity analysis also indicated that the elimination of any study did not make a significant alteration in the pooled HR.
In conclusion, it can be noted that the present study has stated the association between GDF-15 and adverse prognosis of CVDs and all-cause mortality. A large number of research need to be done until GDF-15 could contribute to clinical and bring more valuable information for clinician.

| LIMITATION
We delimit the studies published in the English language as one of the eligible criteria, which may be a limiting factor. First based on prospective study data, the result of our meta-analysis is subject to potential bias. Although potential confounders, such as age, sex, and body mass index have been adjusted in most studies, residual confounding cannot be precluded. Second, some studies of small sample size resulting in extremely strong associations may have an impact on pooled estimates. Third, the characteristics of baseline participants differ across included studies which cause the initial production of heterogeneity.