The nested case–control study consisted from 1870 individuals, of whom 447 were cases (i.e. they died from any cause, of which 139 were from CVD and 211 from cancer); the 1423 survivors served as controls. Descriptive characteristics of subjects included in the nested case–control study are shown in Table 1. The median plasma bilirubin concentration in the overall sample was 12.8μmol/L (13.5μmol/L in men, 11.6μmol/L in women). Compared to controls subjects who died had significantly lower plasma bilirubin concentrations (Table 1). Clinically important elevations of ALT and GGT activities were seen in 8.6% and 19.5% of subjects, respectively.

The distribution of UGT1A1 genotypes was in agreement with HW equilibrium (p=0.19), and the allele frequencies were similar to the other Caucasian populations (Table 2) [10]. As expected, there was an association between the UGT1A1 genotypes and plasma bilirubin concentrations; 7/7 homozygotes had significantly higher plasma bilirubin concentrations compared to carriers of at least one allele 6 (Table 2). Other biochemical and anthropometrical parameters of interest (ALT, GGT, CRP, urea) were distributed independently of the UGT1A1 repeat polymorphism.

After controlling for age and sex, there was a strong negative association of plasma bilirubin categories with risk of death from all causes and cancer mortality but not from CVD (Table 3). These results remained significant even after adjusting for potential confounders, including age, education, marital status, alcohol intake, smoking, BMI, energy intake, total/HDL cholesterol ratio (model 2). In the adjusted model, the odds ratios in the highest vs. lowest bilirubin categories for death from all causes, cancer and CVD were 0.61 (95% CI: 0.42–0.87), 0.39 (0.24–0.65) and 1.03 (0.58–1.83), respectively (Table 3). After additionally adjusting for elevated plasma ALT, GGT activities, as well as CRP and urea concentrations (model 3), the pattern of results remained similar, although the confidence intervals became wider due to loss of statistical power in the multivariable model. All the studies on the association between bilirubin and clinical outcomes published so far have paid extreme attention to gender with female sex having negligible relationship, whereas males almost always having benefits from higher bilirubin concentrations. For this reason we also conducted gender-specific analyses. In these additional analyses, investigated associations were stronger in men than women (Table 4), although the interactions between bilirubin and sex were not statistically significant.

The associations of the UGT1A1*28 allele (a genetic marker of systemic bilirubin concentrations) with mortality risks showed similar patterns as plasma bilirubin concentrations, although the associations did not reach statistical significance (Table 5). Importantly, the adjustment for plasma bilirubin almost attenuated the (non-significant) associations of the polymorphism with all-cause and cancer mortality; indicating that the effect of genotype was, as hypothesised, driven predominantly by plasma bilirubin concentrations.

This case–control study, nested in a population-based cohort in a Krakow urban population sample, found strong negative associations of plasma bilirubin with all-cause mortality driven predominantly by the effect on cancer but not CVD mortality. These observational associations were supported by the Mendelian randomisation analyses; although non-significant, the patterns of associations of mortality with a genetic marker of raised bilirubin were similar to results of analyses using plasma bilirubin. In theory, with data available on bilirubin, genotypes and mortality outcomes in the same dataset, it would be possible to formally conduct a one sample Mendelian randomisation using a two-stage least squares regression analysis. However, a larger sample size and/or higher number of deaths would be necessary to achieve sufficient statistical power.

There is growing evidence from experimental and clinical studies [3,4] that mildly elevated systemic bilirubin concentrations have potent anti-atherogenic [5], anti-proliferative [7,18], and anti-ageing [19] effects. This association, however, seems to be valid only when the elevated bilirubin is not a consequence of an underlying liver disease [6]. Indeed, elevated liver enzyme activities predict cardiovascular events [20,21], thus underlying liver disease clearly eliminates the potential beneficial effects of bilirubin. It is important to note, that the prevalence of elevated liver enzymes in the general population is as high as 10–20% [22], which is exactly what we found in our current study.

The beneficial effects of bilirubin on mortality have been shown in several epidemiological studies. The first study, in 2001, a 10-year follow-up retrospective investigation of more than 11,000 Belgian men and women [23], reported a significant negative association of bilirubin with all-cause and cancer mortality in men but not in women; surprisingly, no association was found with CVD mortality. Another large study by Fulks et al., included almost 2 million insurance applicants, found an negative association between all-cause mortality and bilirubin levels [24]. Serum bilirubin levels had a strong and independent protective impact on the association with all-cause and CVD mortality in a study by Ajja et al. in a male population [25]. In a prospective study by Horsfall et al. [26] in a statin-treated population of approximately 130,000 participants, a negative relationship between bilirubin concentrations and all-cause mortality risk was observed, whereas the relationship between bilirubin and CVD/myocardial infarction events was U-shaped [26]. In a second report, the same group analysed mortality rates in participants diagnosed with Gilbert syndrome; these subjects had half the mortality rate compared to controls [27]. Protective effects of bilirubin on overall mortality were reported also from a prospective study on haemodialysed patients by Chen et al. [28]. Importantly, subjects with Gilbert syndrome genotype (UGT1A1*28 homozygosity) had one fourth the risk for all-cause mortality as compared to the wild-type allele carriers. A significant negative correlation between serum bilirubin and all-cause mortality was observed also in NHANES study on elderly population [29], although no detailed analysis was provided. Consistently, serum bilirubin concentrations below 10.5μmol/L were found to be a risk factor for cardiovascular mortality in a small Serbian older population [30].

On the other hand, no survival advantage of higher serum bilirubin concentrations was observed among older persons by Boland et al. [31] and, higher baseline bilirubin levels predicted slightly increased mortality in the Ludwigshafen Risk and Cardiovascular Health study (although confusingly, a UGT1A1*28 gene variant predicted lower overall mortality [32]. No significant associations between serum bilirubin and either all-cause, CVD or non-CVD mortality were observed in Diabetes Heart Study [33]. Similarly, bilirubin was not associated with cancer mortality in the EPIC-Heidelberg cohort [34]. Overall, although not completely consistent, most available studies suggest a beneficial effect of bilirubin on all-cause mortality, driven predominantly by the effect on cancer, but in some studies also on CVD.

Our results confirm data from previous studies since they are consistent with the majority of published reports. In our study, the association with all-cause mortality was driven mainly by cancer mortality, while that with cardiovascular mortality was not observed. One of the reasons for this observation may lie in the relatively young age of the examined population (61.6±6.5 years), in which cancer was found a leading mortality cause.

The association between UGT1A1 promoter gene variations and mortality rates was not significant (most likely due to low penetrance of UGT1A1*28 mutation [10]) but the direction of the association also suggests protective effects. The observation that the adjustment for bilirubin virtually eliminated the (non-significant) effect of the genotype is important because it confirms that bilirubin is (1) strongly associated with the genotype, and (2) responsible for the protective effect of the UGT1A1*28 gene variation.

Our findings are further supported by clinical studies of cancer morbidity. For example, in a large Chinese study on patients with colorectal cancer, the prevalence of Gilbert syndrome genotype was very low, reaching only 1.4% [35]. Similarly low prevalence of this Gilbert syndrome genotype (4.7%) was found in a study of colon cancer by Liu et al. [36]. Serum bilirubin concentrations were reported to be negatively associated with non-dermatological malignancies, and this effect was the most pronounced for colorectal cancer; subjects with Gilbert syndrome phenotype had 4-times lower risk compared to individuals with normal bilirubin levels [18]. Similarly, in our previous study low serum bilirubin levels were associated with an increased risk of colorectal cancer in both males and females, whereas UGT1A1*28 allele carrier status was proved to be protective against the development of colorectal cancer in males [7].

The main limitation of our study is the relatively small number of events, particularly deaths from specific causes. The average follow-up time of 6.4 years can also be considered short for the development of chronic diseases and may be partly responsible for some of the non-significant findings. Further limitation is the fact that site-specific cancer rates could not be assessed. Giving the large variation within this disease group in terms of risk factor profile, future analyses should explore the link between bilirubin and specific types of cancer. On the other hand, the prospective design minimises the reverse causation bias, and the extensive set of data on participants allows effective control for confounding. In addition, the combination of observational and Mendelian randomisation design, and the consistency of these two sets of findings, supports the validity of our findings.

In conclusion, our study is consistent with earlier reports that plasma bilirubin concentrations is negatively related to total mortality; in our data, this association was mostly driven by the effect on a cancer, rather than CVD. These findings were further supported by the (non-significant) pattern of associations between mortality and the UGT1A1 promoter gene polymorphism which was largely explained by plasma bilirubin. Larger epidemiological studies, presumably on older populations, are needed to confirm our as well as previously published data on lacking relationship between bilirubin and CVD mortality, and to reveal the reasons for this phenomenon.AbbreviationsCVD

cardiovascular diseases