Background
Ageing is defined as a gradual decline in the ability to maintain whole-body homeostasis, causing the onset of ageing-related diseases (ARDs) and eventually death [
1]. As the ageing tendency of the population accelerates, the elderlies are becoming an increasingly important subpopulation that merits special attention regarding health and social issues [
2]. Hypertension, a representative ageing syndrome which is common in the middle old-ages (quinquagenarian), is characterized by persistently elevated systemic arterial blood pressure (BP) and may be accompanied by functional or organic damage to the heart, brain, and kidney, which is ranked third as a cause of disability-adjusted life-years and affected over 1/5 adults (26.4%) worldwide [
3‐
5]. Although the age-standardized prevalence of hypertension decreased in high-income countries over the past decade, it has been increasing in low- and middle-income countries [
5]. In China, along with the urbanization, economic growth, and the population ageing, the prevalence of hypertension has been elevated markedly. Some surveys have revealed that 26.6–33.6% of the total population was diagnosed with hypertension [
6,
7], which estimated to cause 23 million deaths per year (increased 89% compared with 1990) [
8]. Besides the high prevalence, the rates of awareness, treatment, and control were also inadequate, which exerts heavy burdens on the public health system in China, even in the entire world [
9]. In view of the severe situations, discovering novel hypertension-related biomarkers and distinguishing high-/low-risk population through biomarker-based models would better facilitate the prevention and control of hypertension [
10].
Bilirubin (C
33H
36N
4O
6), conventionally considered as the ultimate product of heme metabolism, is produced starting from the breakdown of red blood cells to generate carbon monoxide, iron, and biliverdin. Biliverdin is an unstable intermediate and rapidly converts into bilirubin by biliverdin reductases [
11]. In hepatocytes, bilirubin is conjugated with glucuronic acid by uridine-diphosphate glucuronosyltransferase 1A1 (UGT1A1), known as conjugated bilirubin (CB) [
12]. Comparing with unconjugated bilirubin (UCB), CB is soluble, which can be filtered and excreted via kidney but does not cross the brain barrier [
13]. Animal experiments showed bilirubin mediated cytoprotection effects, including anti-oxidation and anti-inflammatory in vivo [
14]. Several epidemiological studies have also shown the inverse correlation between bilirubin and the risk of cardiovascular diseases [
15,
16]. However, bilirubin presents the potential cytotoxic effects either. Toxicological studies manifested that high serum concentration of bilirubin would bound to and deposited on various tissues, and further pleaded to deleterious events such as jaundice, mental disorders, cerebral palsy, brain damage, and even death [
17]. Beyond that, hyperbilirubinemia was associated with a worse world health organization functional class, higher right atrial pressure, higher brain natriuretic peptide, and a larger Doppler right ventricular index [
18]. Hence, the subtle role of this substance in human metabolism keeps unclear.
Until now, some epidemiological studies have aimed to uncover the relationship between bilirubin and BP among children and middle-age adults. However, very limited studies have been conducted among the ageing population, especially in China [
19‐
21]. Given the potential indication of bilirubin towards hypertension, we aimed to examine the association between the baseline levels of bilirubin and the incident risk of hypertension with the adjustments for key covariates on the dataset of the Guankou Ageing Cohort Study (GACS).
Material and methods
Study population
The research protocol of the GACS has been approved by the Medical Ethics Committee of School of Medicine, Xiamen University. All procedures were conducted in accordance with the Declaration of Helsinki, and all patients were required to provide written informed consent prior to participation. No adverse events were reported during or after completion of the study. The GACS was a prospective dynamic cohort study nested in the public health service system in Xiamen. Through tracing the disease process of ARDs (hypertension, diabetes mellitus, and dementia), this cohort study aimed at recognizing risk factors in the ageing process and providing clues to possible pathogenesis. We recruited the ageing inhabitants (≥ 55 years) from a rural area of Jimei District in Xiamen, China. The participants have relatively stable sociological features and can be followed over a long period. Subjects underwent annual comprehensive health check-ups in the Xiamen Guankou hospital from July 1st, 2013, until onset of hypertension, death, or the end of observation (December 31st, 2019). The participants’ lifestyle questionnaires, demographic information, and comprehensive medical check-up data were collected along with the biological samples. The data obtained at the initial medical check-up were served as the baseline information.
Data collection
Before recruitment, all investigators and staff members accepted specific trainings to be familiar with the aims of the study as well as the application methods of equipment. The standardized questionnaire developed by the School of Public Health, Xiamen University, was used to determine the information regarding of marriage status, lifestyle factors (e.g., smoking, drinking and physical exercise). All questionnaires were collected through computer-assisted face-to-face interviews during clinic visits under the guidance of investigators. After at least 10-min relaxation, three consecutive BP readings were obtained with a five-minute interval and the average was used as BP value. Subjects with BP variation beyond 10 mm Hg among the measurements were required to take a repeated measurement in three days later. Suspected hypertensive patients were asked to take further consultation until confirmation. Waist circumference (WC), which is the horizontal circumference of the mid-point line between the lowest rib and the upper edge of the iliac crest (about 1 cm to the upper edge of navel), was measured with errors of 0.5 cm. Body height and weight were measured when the subjects were taken in light clothing and without shoes (with errors 0.5 cm and 0.1 kg, respectively). Body mass index (BMI) was computed as weight in kg divided by height in m2. A 5-mL of venous blood was drawn from each subject was drawn into sodium citrate anticoagulant tube after overnight fasting (about 12 h) and then sent immediately to biochemistry laboratory in an ice cooler for the further processing and analyses.
Definitions and participant classifications
Hypertension was defined as sustainably systolic blood pressure (SBP) ≥ 140 mm Hg and/or diastolic blood pressure (DBP) ≥ 90 mm Hg taken in clinic or with a history of taking antihypertensive medications. Cases of incident hypertension were defined as those who had no baseline hypertension and were diagnosed during follow-up. Prehypertension was defined as SBP 120 to 139 mm Hg and DBP 80 to 89 mm Hg without antihypertensive medication. Stage 1 hypertension was defined as SBP 140 to 159 mm Hg and DBP 90 to 99 mm Hg, Stage 2 as 160 to 179 mm Hg and 100 to 109 mm Hg, and Stage 3 as ≥ 180 mm Hg and ≥ 110 mm Hg [
22]. Four subsets of participants in GACS were initially categorized based on the STB, CB, and UCB quartiles at baseline to examine the overall relationship between hypertension and bilirubin. Type 2 diabetes mellitus (T2DM) was defined as fasting plasma glucose at least 7.0 mmol/L or if the subjects being on antidiabetic agents currently [
23]. And hyperuricemia was diagnosed as serum uric acid ≥ 420 μmol/L (7.0 mg/dL) in males and serum uric acid ≥ 360 μmol/L (6.0 mg/dL) in females [
24].
Statistical analysis
Multiple imputation by chained equations (MICE) was implemented to fill out the missing covariate values prior to statistical analysis [
25]. Kolmogorov–Smirnov test and Levene’s test were performed to assess the normality and homogeneity of variance, respectively. All data are displayed as mean standard deviation (± SD), median (lower and upper quartiles), and frequencies (percentages) depending on the type of data. Student’s t-test or one-way analysis of variance (ANOVA), and Manne-Whitney U test or Kruskal–Wallis test, and Pearson’s Chi-square test were performed for comparing normally distributed continuous variables, uneven distributed variables, and categorical variables, respectively. The incidence density of hypertension was calculated as the total event number divided by the sum of follow-ups (per 100 person-years).
The Kaplan–Meier log-rank analyses and Cox proportional hazards regression models were applied after adjusting for potential key confounders to calculate the hazard rations (HR) and 95% confidence intervals (
CI) for studying the relationship between hypertensive incidence and bilirubin levels. As the levels of fasting plasma glucose (FPG), STB, CB, and serum creatinine (SCr) in the population did not meet normal distribution, different types of data transformation methods were performed before the above-mentioned analyses. These data were converted by base 10 logarithmic, Box-Cox [
26], and square root (SQRT) transformation to obtain normal distributions. Among them, the calculation formula of Box-Cox transformation is shown as follows:
$$y (w, \lambda ) =\left\{\begin{array}{c}\frac{{w}^{\lambda }-1}{\lambda }, \lambda \ne 0,\\ \mathrm{ln}w, \lambda =0.\end{array}\right.$$
where
y represents the novel variable obtained after Box-Cox transformation,
w is the original continuous dependent variable, and
λ is the transformation parameter to be identified [
27].
A multivariable Cox model with restricted cubic spine (RCS) with 4 knots were further constructed to check the bilirubin-hypertension dose–response associations to avoid the inappropriate linearity assumptions [
28], as the RCS model is a smoothly joined sum of polynomial functions that do not assume linearity of the relationship. The threshold was determined as the identification of the risk function inflexion point. The 95%
CI was derived by bootstrap resampling. The calculation formula of multivariable Cox proportional hazards regression is described below [
29]:
$$ h\left( {t,X} \right) \, = h_{0} \left( t \right){\text{ exp }}(\beta_{{1}} X_{{1}} + \beta_{{2}} X_{{2}} + \cdots + \beta_{{\text{m}}} X_{{\text{m}}} ) $$
The above formula gave the underlying hazard at time t for subject i with covariates (explanatory variables) Xi.
To evaluate if the model of bilirubin combined with other statistically significant risk factors could improve sensitivity to distinguish the high and low risk of hypertensive group, we calculated the area under the ROC curve (AUC) for each model by using pROC package [
30]. The goodness of model fits was evaluated by the ANOVA and Akaike information criterion (AIC). And the distinguishing ability of hypertensive or non-hypertensive population was further determined by subsequent unsupervised clustering analysis based on multiple factor analysis (MFA) algorithm [
31]. Two-tailed probability values < 0.05 were considered being statistically significant at 0.05 level. We performed all analyses using R software version 4.0.5 (R foundation, Vienna, Austria) for Windows and SPSS software version 26.0 (IBM SPSS Inc., Chicago, IL, USA).
Patient and public involvement
Neither patient was involved in setting the research question or the outcome measures, nor of them have involved in developing the plans for recruitment, design, or implementation of the study. No patient was asked of advice on interpreting or writing up of the results.
Discussion
Hypertension represents a primary risk factor of cardiovascular and cerebrovascular diseases linked with endothelial dysfunction, which is the leading cause of mortality in the world. The population ageing, rapid urbanization, as well as the changes of environment and lifestyle count great significance for hypertension preventing and controlling [
35‐
37]. In addition, more attention should be paid to the molecular risk factors, which may indicate the pathology and strengthen prevention of hypertension. Bilirubin could be served as one of the molecular factors that has been associated with hypertension in few studies, however, the previous conclusions are still controversial.
In our present work, we analyzed the relationship between bilirubin concentration and hypertension risk according to the data obtained from the GACS. The bidirectional effects of the different bilirubin species were observed along their concentration changes. The lower levels of STB and UCB showed the protective effects towards hypertension, while the opposite effects were observed at the higher levels. The inverse associations have been observed between serum CB and hypertension. The validity of our research is bolstered by previous research that has reached the similar conclusions.
Yasuko Takeda et al. selected 37 patients with pulmonary arterial hypertension (PAH) as the research objects and manifested that elevated serum bilirubin is a risk factor for death in patients with PAH [
18]. Beyond that, according to the research of 37,544 newborns (31,819 term and 5725 preterm births) from the U.S. Collaborative Perinatal Project conducted from 1959 to 1965, Huan Yu et al. proposed that neonatal serum bilirubin levels at 48 h after birth were positively associated with childhood blood pressure/hypertension in the preterm infants at 7 years [
19]. However, other experts offered diverse perspectives that high serum bilirubin may decrease hypertensive risk. For example, Lina Wang et al. analyzed data from the National Health and Nutrition Examination Surveys (NHANES) 1999–2012 (N = 31,069) and demonstrated that SBP decreased progressively up to − 2.5 mmHg (
P < 0.001) and the prevalence of hypertension was up to 25% lower (
P < 0.001) in those with bilirubin ≥ 1.0 mg/dL-the highest two deciles-compared with those with 0.1–0.4 mg/dL-the lowest decile. The author supposed that the fundamental mechanism was high serum bilirubin level might inactivate and inhibit the synthesis of reactive oxygen species in vascular cells to decrease the risk of hypertension [
20]. Ho Jun Chin et al. also proposed that bilirubin concentration was negatively correlated with hypertension incident risk among normotensive Korean adults [
21]. Such protective effects were mainly attributed to the fact that bilirubin has significant antioxidant properties, such as preventing vitamin A and polyunsaturated fatty acids from oxidation [
17]. Meanwhile, bilirubin was supposed to be a potent substance of scavenging hydrogen peroxide radicals and therefore fulfills the anti-oxidative function. However, when exploring the effects of bilirubin in some ageing-related diseases, such as cardiovascular and metabolic disorders, some large-scale cross-sectional and cohort studies have suggested the potential protective effects of bilirubin [
38,
39]. We would like to offer a perspective view that STB and UCB mainly exerted both harmful and preventive influence to hypertension, which have the patterns similar to hormesis effect [
40]. We also observed that CB was weakly and negatively correlated with hypertension, which was partially in agreement with those of previous studies [
41]. The metabolism cross-talk of STB, UCB and CB may well illustrate the role of bilirubin, which suggested that different bilirubin metabolites should be treated dialectically when assessing their risk of hypertension.
The pathological mechanisms of hypertension remain extremely complicated. The partial characteristic of hypertension is mild symptoms of inflammation, and the main mechanisms to pathogenesis involving in the up-regulation of the sympathetic nervous system and the increased renin–angiotensin–aldosterone system (RAAS)-activity [
42]. CB and UCB are the two major species of bilirubin, which display unique chemical properties, and UCB constitutes larger proportion than CB. As a lipophilic molecule, UCB is able to bind to human neurons, which is enriched in phospholipids. While the excitability of sympathetic nervous system exhibited a positive relationship with the bilirubin deposited in the central nervous system, which matters in the hypertensive pathogenesis [
19,
43]. In contrast, CB presented the inverse effect as UCB, which can be plausibly interpreted by its chemical properties of water-soluble and urinary ready excretion. Therefore, phase-II conjugation is important to balance serum CB and avoid UCB accumulation. The breakthrough increasing of UCB may add hypertension risk. Anyway, how the adverse action of bilirubin initiating and/or progressing hypertension remains unclear.
Because of the bidirectional effect, bilirubin appears not to be suitable as an ideal hypertension biomarker as the AUC of crude model is relatively low (AUC
Base 0.56, 95%
CI 0.53–0.59), but this result hinted at the significance of subsequent researches. Regarding of the balance of cost and predicative performance, Model 1 showed a good practical potential to discriminate the high-hypertension risks. In addition, the 7-year hypertension incident rate in the GACS was 6.29 per 100 person-years (5.97 and 6.51 per 100 person-years in men and women, respectively), which was relatively lower in the similar researches conducted in German [
44] and Korea [
45]. The possible reasons might be as follows: (1) The participants in the GACS were relatively younger than those in other researches. The CARLA study, conducted among Germany general population, was comprised of 1779 subjects with a mean age of 64.9 (SD = 10.2) years for men and 63.8 (SD = 9.9) years for women at baseline. The positive association between increasing age and incidence of hypertension was in agreement with many literatures, which could be attributed to lower-level physical activity, differences in dietary intake, the age-dependent hardening of the vascular system and worsening of kidney function [
46,
47]; (2) The participants in the GACS were shown to be more “slim” than other cohorts. The male and female participants showed the mean BMI values of 22.13 (SD = 0.10) and 22.88 (SD = 0.10) in the GACS, which were relatively lower than Germany and Korean population involved. BMI is a well-recognized risk factor for hypertension [
48], and in a meta-analysis, the mean SBP and DBP reductions associated with an average weight loss of 5.1 kg were 4.4 and 3.6 mmHg, respectively [
49]; (3) Another possible reason might be attributed to the annual routine health examination of the subjects, which may be the relatively healthy part of the population have been involved.
In our study, the prevalence of T2DM and hyperuricemia were not significantly different between genders. He et al. used real-world data to estimate the changing tendencies in the prevalence of T2DM in Xiamen City from 2014 to 2019. Interestingly, the overall prevalence among the male and female adults were 4.18% and 5.52% in Xiamen, and T2DM prevalence exhibited an increasing trend with advancing age regardless of gender [
50]. In our work, female participants were younger than males, which could be one possible reason to explain the prevalence of T2DM were not significantly different between genders. The discrepancy of gender-specific hyperuricemia prevalence is also existed, which might be attributed to the lifestyles, dietary habits, regional economic level, and individual living standards [
51,
52]. Besides, women in post-menopausal could lead to estrogen deficiency. Estrogens may promote renal clearance of serum urate and its deficiency can result in changes in the endocrine system and increase of metabolic diseases [
53]. However, the underlying mechanisms that lead to the gender-specific prevalence of hyperuricemia requires further exploration.
In our risk discrimination analysis, the predictive performance of model exhibited a stepwise increase with the processive inclusion of significant associated factors and relevant biological variables, especially in Model 4. A large body of previous studies proposed smoking was a risk factor for hypertension [
54‐
56], thus, the ability of risk discrimination in Model 4 improved drastically. However, there were very few differences of the predictive performance between Model 1 and Model 3, which hinted variables of age, gender, BMI, and WC could be considered as the main parameters in the risk discrimination model. Moreover, as shown in Crude Model, the AUC was relatively low (AUC
Base 0.56, 95% CI 0.53–0.59), which suggested STB appears not to be suitable as an ideal hypertension biomarker because of the bidirectional effect. Regarding of the balance of cost and predicative performance, Model 1 showed a good practical potential to screen the subjects with high-hypertension risks.
The current analyses had several strength and limitations. The first strength was that our research was based on a large prospective study, the GACS with high-quality design, and the tracing of diseases made it possible to recognize potential protective and harmful factors as well as their modes of action on ARDs. The data of physiology, lifestyle factors, socioeconomics, and biological samples during the long-term follow-up increased our confidence to infer the underlying mechanisms of certain questions raised from the observation. And the treatment regimens based on bilirubin in clinical practice may be available when it is verified by the future studies. As for limitations, one of the limitations was hypertensive state is unstable, which can be affected by environmental and psychological causes, and the case might inevitably include the false positive confirmation. The standard definition of hypertension clinical diagnosis was based on the three BP measurements in at least two different occasions [
57]. The BP readings of our subjects were measured only in the same hospital. It is also difficult to require subjects to use 24 h blood pressure monitoring equipment during the follow-up. The second limitation was hypertension subtypes (primary and secondary hypertension) were not well discriminated because of technical and cost limits during follow-up, which might introduce bias into further analysis. At last, the state of hyperuricemia in our research was relied on anamnesis and medical records of participants, rather than dynamic monitoring data, which might pose a potential risk of bias towards our conclusions.
All in all, this study provides the valuable clues of the associations between bilirubin and hypertension, even though the underlying mechanisms remain ambiguous to date. Advancing theories and methodologies is urgently called for to uncover the bilirubin-hypertension association from the more refined perspective [
58].
Publisher's Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.