The level of Bilirubin and the risk of ischemic stroke: a systematic review and dose-response meta-analysis of real-world studies CURRENT

Background: Bilirubin, a marker of hepatic and hematological diseases in clinical practice, is not only a waste end-product but also an antioxidant that may protect against diseases associated with oxidative stress. Numerous epidemiological studies have shown an inverse relationship between the serum total bilirubin (TB) level and the risk of ischemic stroke (IS). However, markedly elevated TB levels may exert neurotoxic effects. Based on this, we conducted a dose-response meta-analysis to quantify the relationship between blood TB and IS as well as between TB and all types of stroke (AS) in the physiological range of bilirubin. Methods: PubMed, Embase, Web of Science, and Cochrane Central databases were searched up to March 2019. Additional studies were identified by reviewing references and contacting authors. Categorical and dose-response meta-analyses were performed to quantify the relationship between TB and IS. The primary outcome was ischemic stroke, and the secondary outcome was all types of stroke. Results: Nine observational studies (seven publications) involving 110,032 participants and 3710 stroke cases were included for analysis. The average OR of IS for every 1 µmol/L increment in TB level was 0.978 (95%CI: 0.957–0.999). The summary OR of AS for every 1 µmol/L increment in TB level was 0.974 (95%CI: 0.956–0.992). Subgroup analysis based on gender showed a negative dose-response relationship between the circulating TB level and IS or AS in males, but not in females. Conclusions: The present study indicates a negative dose-response relationship between the circulating TB level and the risk of IS or AS within physiologic range of serum TB in males. Moderately elevated blood TB levels were associated with a diminished prevalence of IS. Every 1 µmol/L increment in serum TB level was associated with a 2.2% decrease in the risk of IS and a 2.6% decrease in the risk of AS. Large-scaled prospective studies TB: Total bilirubin; CVD: Cardiovascular disease; CI: Confidence interval; HR: Hazard ratio; OR: Odds ratio; RR: Relative risk; RHE: Routine health examination; SCI: Silent cerebral infarction; SIVD: Subcortical ischemic vascular disease; SCOUT: Sibutramine Cardiovascular Outcomes Trial; NFS: Non-fatal stroke; NSHDSC: The Northern Sweden Health and Disease Study; NHANES: The National Health and Nutrition Examination Survey

are needed to confirm the conclusion of the current analysis.

Background
Stroke is a leading cause of death and long-term severe disability worldwide [1]. More than two thirds of stroke deaths worldwide occur in developing countries [2]. In China, stroke has been an enormous health issue as the most common cause of death in recent years [3]. Identifying and managing risk factors for stroke is highly important for the general population in order to improve their prognosis after stroke. A number of serum biomarkers have been intensively studied in clinical trials of stroke and summarized according to their pathophysiological importance in stroke pathogenesis [4]. Among them, circulating total bilirubin (TB), a breakdown product of normal heme catabolism [5], is proved to be one of the strongest endogenous antioxidants in mammalian tissues [6].
Accumulating evidence has shown that bilirubin harbors anti-inflammatory [7], neuroprotective [8], and inhibiting platelet activation [9] effects. These properties entitle bilirubin to prevent or delay the formation of thrombi in ischemic stroke (IS), and to protect neurons from ischemic injury, suggesting that elevated bilirubin levels may be linked to a lower rate of morbidity and mortality of stroke patients. Recently, research has shifted to focus on its role in reducing the risk of developing strokes [10][11][12][13][14][15][16][17][18] though some of them showed only moderately positive or null effect [17,18]. A large prospective study showed that every 1 µmol/L increment in bilirubin concentration was associated with a 2% reduction in the hazard ratio in a multivariate-adjusted model for IS and 1% reduction in the hazard ratio for all types of stroke (AS). However, the participants are only Korean [19]. The National Health and Nutrition Examination Survey involving 13,214 adult participants reported that a 1.71 µmol/L increment in the bilirubin level was associated with a 9% reduction in the odds ratio for stroke. However, this is only a cross-sectional study [20]. Subsequently, a meta-analysis of 12 population-based prospective studies involving 9378 incident cardiovascular disease (CVD) cases showed that every 1-SD increase in the TB level was associated with a 7% reduction in the relative risk of CVD, but stroke was not the end point [11]. Zhong et al. reported an inverse relationship between the level of circulating TB and the risk of IS or AS in males through meta-analysis which included 11 observational studies involving 5,060 stroke cases among 131450 subjects, but it is not a dose-response analysis [21].
At present, it is sufficiently clear that low serum bilirubin level was independently associated with an increased risk of incident CVDs in the general population, but a markedly elevated TB level may exert neurotoxic effects [22,23] and the dose-response relationship between the circulating TB level and the risk of IS/AS has not been confirmed.
We, therefore, explored the exact dose-response correlation between the level of blood TB and the incidence of IS within the physiological range of TB.

Literature search
Articles on blood TB levels and stroke in adults (≥18 years), published up to March 2019, were systematically searched in databases of PubMed, Embase, Cochrane Central, and Web of Science. Detailed search strategies were shown inthe Supplementary material.
Language was limited to English. In addition, references cited by relevant articles as well as those listed in published meta-analyses were manually searched to identify additional eligible studies. This study was registered at PROSPERO (https://www.crd.york.ac.uk/PROSPERO/[CRD42017075988]), and the PRISMA and the MOOSE guidelines for meta-analyses of observational studies were followed [24,25].

Study selection
The inclusion criteria were: 1) prospective cohort, case-cohort, or cross-sectional design; 2) studies investigating the relationship between serum TB and incidence of IS/AS; 3) estimates of adjusted RR (HR, OR) with 95% CIs were included; 4) For the dose-response analysis, a quantitative measure of blood TB levels and the total number of cases and person-years or the number of events had to be provided. Abstracts, animal studies, in vitro studies, comments, letters, editorials, ecological studies, gray literature, unpublished results or information, and randomized controlled trials were excluded. The primary outcome of this study was ischemic stroke, and the secondary outcome was stroke (defined as all types of stroke).

Data extraction
From each included study, data including the last name of the first author, year of publication, country where the study was conducted, study name, study type, follow-up period, sex, age, the number of stroke cases and the total number of study participants, bilirubin levels, adjusted ORs, RRs, or HRs of IS/AS (with corresponding 95% CIs), and variables adjusted for in the analysis were extracted. Two authors (X Wang and Y Zhou) extracted the data into a standardized form, and a third author (P Zhong) checked them for accuracy.

Quality assessment
Biases within each individual study were evaluated by two independent reviewers (X Wang and Y Zhou) using the validated ROBINS-I tool which provides a comprehensive and structured approach to assess non-randomized studies of interventions [26]. This tool covered domains of pre-intervention bias due to confounding factors and selection of participants, intervention bias in classifying interventions, and post-intervention bias due to deviation from intended interventions, missing data, measurement of outcomes, and selection of reported results. An overall judgment was reached after assessing each of the domains and classified as low, moderate, serious, critical risk of bias, or no information on which to base a judgment on the risk of bias. Disagreements were resolved by consensus or by a third-party adjudication (X Ye) as necessary.

Statistical analysis
Statistical analysis was performed using the Review Manager (version 5.1 for Windows, Cochrane Collaboration, Oxford, UK, 2011) and STATA 11.0 (Stata Corporation, Lakeway, Texas, USA). The ORs and associated 95% CIs were considered as the effect sizes.
Summary ORs and 95% CIs for the highest versus the lowest levels of serum TB and for the dose-response analysis were calculated. The average of the natural logarithm of ORs was estimated, and the OR from each study was weighed by the inverse of its variance. A previously described method [27] was used for the dose-response analysis and study specific slopes (linear trends) and 95% confidence intervals from the natural logs of the odds ratios and confidence intervals across categories of blood bilirubin levels were computed. It required that the distribution of cases and person-years or non-cases, and the ORs with their variance estimates were known for at least three quantitative categories of bilirubin levels. If a study did not report the distribution of cases or personyears but the total number of cases or person-years, then the results were analyzed by quantiles (and could be approximated). The median or mean level of bilirubin in each category was assigned to the corresponding OR for each study. If the study reported the bilirubin level by ranges, the midpoint was estimated by calculating the average of the lower and upper bounds in each category. If the highest or lowest category was open ended, the length of the open-ended interval was assumed to be the same as that of the adjacent interval. Studies that reported results separately for men and women were deemed as two separate studies. The dose-response results in the forest plots were presented for every 1 µmol/L increment for IS or AS. A potential nonlinear dose-response relationship between circulating TB levels and all types of stroke was examined using fractional polynomial models [28]. A likelihood ratio test was used to assess the difference in nonlinearity between the nonlinear and linear models [29]. Generalized least squares for trend estimation and the restricted cubic spline regression model were applied in the dose-response analysis.
Heterogeneity between studies was evaluated using χ 2 test and I 2 statistic, and between subgroups using meta-regression analyses. Heterogeneity was considered to be present when P <0.1. In the presence of statistical heterogeneity, a random effects model was used for the analysis; otherwise, a fixed effects model was employed. The publication bias was assessed using the Egger's [30] and Begg's tests [31]. Publication bias was indicated when P was <0.1. Sensitivity analysis was carried out by excluding one study at a time to explore whether the result was driven by a large study or by a study with an extreme result. A two-tailed P value less than 0.05 was considered statistically significant.

Results
The initial search yielded 422 studies from PubMed, 1471 from Embase, 10 from Cochrane Central, and 539 from Web of Science. After excluding of duplicates and irrelevant studies, 46 potentially relevant full-text publications were selected. 5 studies with categories of blood bilirubin levels <3, 5 studies with no original data, 1 case report, 1 conference literature, 1 letter, 4 review studies, 1 study with low quality and no clear data relationship, 6 studies with a comparison of unreasonable outcomes, and 15 irrelevant studies were excluded. A manual search of references of these studies did not yield any new eligible studies. Results of the study-selection process were shown in Fig. 1. Finally, nine observational studies (seven publications) [10, 13-16, 19, 20] were selected for analysis, including 2 prospective cohort studies [14,19], 1 case-control study [10], and 4 cross-sectional studies [13,15,16,20], involving 110,032 participants with 3710 stroke cases. The ranges of blood TB levels and general characteristics were shown in Table 1.
Included studies were assessed for bias using the ROBINS-I tool. All studies in analysis had a moderate risk of bias (Supplementary Table Ⅰ).. This meant that all included observational studies were judged to provide sound evidence as non-randomized studies but could not be considered comparable to well-performed randomized trials.
Correlation between the level of circulating total bilirubin and ischemic stroke Five observational studies (4 publications) [10,15,16,19] explored the association between TB and IS for the highest versus the lowest levels of serum TB and for the doseresponse analysis, including two prospective cohort studies (1 publications) [19], 1 casecontrol study [10], and two cross-sectional studies [15,16] involving 2312 ischemic stroke cases among 81,632 participants ( Table 1).. As Kimm 2009 separately reported results for men and women, we deemed them as two studies as mentioned previously.

Highest versus lowest bilirubin levels
Five observational studies (4 publications) [10,15,16,19] were included in the analysis of highest versus lowest levels of circulating TB and the risk of IS. There was mild heterogeneity between studies (I 2 = 39.6%, P = 0.157), and thus the fixed effects model was used for this analysis. The summary odds ratio for the highest bilirubin level to the lowest bilirubin level for IS was 0.729 (95%CI: 0.621-0.855) ( Supplementary Fig. Ⅰ showing a significant difference.

Dose-response analysis
In the dose-response meta-analysis investigating the relationship between TB and IS, the summary OR for every 1 µmol/L increment of the TB level was 0.978 (95%CI: 0.957-0.999,  Table Ⅲ and Supplementary Fig. Ⅱ).. The publication bias was not evident as shown by the Egger's test (P = 0.620) or Begg's test (P = 0.624). A negative linear relationship between blood TB levels and the risk of IS was indicated (P = 0.392, Fig. 3)..

Subgroup and meta-regression analyses
Subgroup analysis of two prospective cohort studies (1 publication) [19] showed that the total OR for every 1 µmol/L increment of the TB level for IS was 0.

Association of blood total bilirubin level with all types of stroke
In nine observational studies (7 publications) [10, 13-16, 19, 20], the correlation between the level of bilirubin and the risk of AS from the highest to the lowest levels of serum TB and the dose-response analysis were investigated. There were 2 prospective cohort studies [14,19], 1 case-control study [10], and 4 cross-sectional studies [13,15,16,20]

Highest versus lowest bilirubin levels
Nine observational studies (7 publications) [10, 13-16, 19, 20] were included in the analysis of highest versus lowest serum TB levels and the risk of AS. There was moderate heterogeneity (I 2 = 45.3%, P = 0.067) between these 9 studies, and thus the random effects model was used. The summary odds ratio for the highest bilirubin to the lowest bilirubin for AS was 0.721 (95%CI: 0.606-0.858) (Supplementary Fig. Ⅲ)..

Dose-response analysis
Nine observational studies (7 publications) [10, 13-16, 19, 20] were included to investigate the exact dose-response relationship between TB and all types of stroke. The summary OR for every 1 µmol/L increment of the TB level for AS was 0.974 (95%CI: 0.956-0.992, Fig. 4 Publication bias was not observed as shown by the Egger's test (P = 0.101) or Begg's test (P = 0.404). A negative linear dose-response relationship was observed between the blood TB level and the incidence of AS (P = 0.315, Supplementary Fig. Ⅴ)..
Stratified analysis based on gender showed that the OR was 0.982 (95%CI: 0.966-0.998) for males and 0.981 (95%CI: 0.951-1.012) for females, indicating a significant difference for males and no difference for females. Significant difference was also observed when the average age of patients was <60 years, but not when the average age was ≥60 years.
Moreover, subgroup analysis showed a significant difference in studies with a sample size ≥2000 participants, but not in those with <2000 participants. In meta-regression analyses of stroke outcomes, no significant heterogeneity between subgroups was observed (All P values >0.1) (shown in Supplementary Table Ⅳ)..

Discussion
Based on available data from observational studies, the present meta-analysis demonstrated a negative linear dose-response relationship between the serum TB level and the risk of IS or AS after adjusting for potential confounding factors. Every 1 µmol/L increment of the circulating TB level was associated with a 2.2% decreased risk of IS and a 2.6% decreased risk of AS.
Bilirubin, an end-product of heme catabolism, has been considered as a potentially toxic agent at high levels in the human body. However, it is now recognized as a natural antioxidant, especially against the oxidation of lipids. This property is believed to diminish plaque formation, thereby protecting against atherosclerosis and the subsequent macrovascular diseases [32]. A number of studies have demonstrated that serum TB was negatively correlated with the intima-media thickness of the carotid artery (IMT) [33][34][35][36][37].
In humans, a low (<7 µmol/L) TB concentration has been shown to be a risk factor for systemic diseases associated with increased oxidative stress, such as CVD, diabetes, metabolic syndrome, certain cancers, autoimmune, and neuropsychiatric diseases [38]. Thus, increasing blood TB level has a number of health benefits. Previous studies showed that a mild-to-moderate increase in the bilirubin level might be a promising strategy for preventing cardiovascular events [39][40][41]. The result of the present study coincided with previous reports that bilirubin might be an independent predictor of stroke incidence [19,20]. The negative association between serum TB and the incidence of IS may be explained by the mechanisms of bilirubin-mediated inhibition of lipid oxidation, immune reactions, inflammatory processes, cell migration and proliferation, as well as apoptosis [22,23].

Gilbert's syndrome (GS) is a benign hereditary disorder of bilirubin conjugation
characterized by an isolated, elevated blood level of unconjugated bilirubin [42]. Normally (allowing for differing laboratory ranges), the blood TB level is <17µmol/L and clinical jaundice becomes apparent at levels >40µmol/L. However, Gilbert's syndrome is considered a phenotypic effect rather than a serious disorder because of its profound beneficial effects. Studies have shown that GS could apparently protect against chronic inflammatory disorders including CVD [43], atherosclerosis [44], cancer and all-cause mortality [9], thus supporting the concept that mildly elevated unconjugated bilirubin could be protective against the development of future CVDs. Large prospective cohort studies are needed to explore the risk of IS/AS in phenotypic Gilbert syndrome defined as total bilirubin >17µmol/L. Stroke is categorized into two types: ischemic and hemorrhagic. The present metaanalysis found that the serum TB level was negatively correlated to the incidence of IS or AS. In our analysis, only one study investigated the relationship between serum TB and hemorrhagic stroke (HS) and did not find a significant relationship between them [19].
Pathophysiology of IS is different from that of HS. A stronger relationship between TB levels and the risk of IS than between TB levels and HS may support the antiatherogenic property of bilirubin as suggested in a previous study [20]. Moreover, IS is an etiologically heterogeneous disease and can also be categorized into different clinical subtypes. Lin et al conducted a study to explore the potential use of serum TB as a biomarker to differentiate cardioembolic stroke from other types of stroke [45]. In this study, 628 consecutive patients had IS whose etiology was classified into large artery atherosclerosis, cardioembolism, small vessel occlusion, other determined etiology, and undetermined etiology according to the TOAST criteria [46]. It was found that serum TB was an independent predictor of cardioembolic stroke. In a cross-sectional study, the association between TB and silent cerebral infarction (SCI) was evaluated among 2865 subjects undergoing medical checkup. It was found that participants with a low level of TB had a higher prevalence of SCI. This suggests that TB may be a novel biomarker for SCI [47]. Further studies are needed to investigate the relationship between the serum TB level and the risk of different subtypes of IS.
In this study, a negative dose-response relationship was observed between the serum TB level and the incidence of IS or AS in males, but not in females. Vítek et al and Kunutsor et al also demonstrated an inverse relationship between circulating TB and carotid artery IMT [37] as well as CVD [11] in males, respectively. The difference between men and women might be attributed to lifestyle factors, including tobacco, alcohol, or diet. Men have higher levels of stored iron than women [48], and high dietary intake of heme iron has been associated with an increased risk of coronary heart diseases [49]. The potential effect of iron load on the heme oxygenase-1 and bilirubin pathway should not be excluded [50].
The present study has a few strengths. Firstly, the greatest strength for this study was that it quantified the relationship between blood TB levels and the risk of IS or AS by conducting linear and nonlinear dose-response analyses. This study demonstrated the protective effect of bilirubin on IS or AS, in particular in men, despite the limitations of the meta-analyzed data. Secondly, more than 110 thousand participants from different countries were included in the present meta-analysis and the follow-up period of included studies was as long as 9.4 years. Therefore, the present study had an adequate statistical power to clarify the dose-response relationship between circulating TB and the risk of IS/AS and to detect moderate reductions in the risk. Finally, the validated ROBINS-I tool was used to assess the study quality. This is a new tool to evaluate the risk of bias in estimating the comparative effectiveness of interventions from non-randomized studies [26]. It is particularly useful for those undertaking systematic reviews on non-randomized studies. All included studies had adjusted for potential confounding factors, and the methodological assessment was satisfactory.
However, our study also has some limitations. Firstly, available resources for our metaanalysis are insufficient for us to draw an exclusive conclusion. In the present analysis, 4 cross-sectional studies were included to assess the relationship between bilirubin levels and all types of stroke. Though there is a negative relationship between them, the causal relationship could not be confirmed. In addition, only one study on HS was included and our conclusion on the relationship between serum TB levels and the risk of HS is not convincing. Secondly, serum TB consists of direct and indirect bilirubin. Few studies have focused on the relationship between different types of bilirubin and stroke. Thirdly, the analysis included cohort, case-control, and cross-sectional studies. This inconsistency in study design might lead to bias in our conclusion. Finally, studies have found that markedly elevated TB levels may exert neurotoxic effects under some circumstances [22,23]. Thus, results of this analysis should be interpreted with caution in the absence of definitive evidence.

Conclusions
In conclusion, the present study supports a negative dose-response relationship between circulating TB and the risk of IS or AS within physiologic range of serum TB in males.
Moderately elevated blood TB levels were associated with a diminished incidence of IS.
Every 1 µmol/L increment of the blood TB level was associated with a 2.2% decreased risk of IS and a 2.6% decreased risk of AS. Large-scaled prospective cohortstudies are needed to confirm the conclusion of the current analysis.

Availability of data and materials
All data generated or analyzed during this study are included in this article.

Competing interests
The authors declare that they have no competing interests. Authors' contributions XW contributed to study design, literature search, data collection, data analysis, figure preparation, data interpretation, and article writing. YZ contributed to literature search, data collection, data interpretation, and article writing. XFY contributed to data collection, data analysis, figure preparation, data interpretation, and article writing. FCL contributed to data collection, data interpretation, and article writing. XZ contributed to data collection, data interpretation, and article writing. DHW contributed to study design, administrative support, and article writing. PZ contributed to study design, administrative support, and article writing.  Dose-response analyses between the blood total bilirubin level and the risk of ischemic stroke. The summary OR for ischemic stroke for every 1 µmol/L increment of the TB level was 0.978 (0.957-0.999).

Figure 3
Linear dose-response relationship analysis between bilirubin levels and the risk of ischemic stroke. The solid and short dashed lines represent the estimated OR and its 95% CI (per 1µmol/L increment).

Figure 4
Dose-response analyses between the blood total bilirubin level and the risk of stroke. The summary OR for stroke for every 1 µmol/L increment in the TB level was 0.974 (0.956-0.992).

Supplementary Files
This is a list of supplementary files associated with the primary manuscript. Click to download.
Supplementary material.doc