Clinical and genetic definition of serum bilirubin levels for the diagnosis of Gilbert syndrome and hypobilirubinemia.

In 1901, the first description of common benign hyperbilirubinemia, “cholémie simple familiale,” was published by Gilbert and Lereboullet, which included Napoleon Bonaparte and his mother as early proof of genetic origin. 1 Gilbert syndrome (GS) affects ~3%–7% of individuals worldwide and 5%–10% in Europe, and it is associated with reduced morbidity, whereas hypobilirubinemia has the opposite association. 2 – 6 Diagnosis of GS is often made during routine health examinations when mildly elevated levels of serum total bilirubin (TB) are detected. In the absence of specific symptoms, TB is the only measurable phenotypic trait of GS. There is a consensual definition that GS in individuals is determined by the presence of hyperbilirubinemia in the absence of both hemolysis and liver damage, including the fibrosis stage. 7 , 8 Few studies have investigated the prevalence of illnesses in large cohorts of subjects with GS, and none have investigated its association with hypobilirubinemia. The largest study without genetic variants involved 23,925 participants and did not observe more symptoms in possible GS versus controls. 9 Recently, the associations between elevated TB and 19 illnesses with a putative protective signal of TB were analyzed in 61,281 inpatients with genetic variants, without causality proven by Mendelian randomization (MR). 10 The current responses available to individuals with GS from health authorities or charities are insufficient (Supplemental Table S1, http://links.lww.com/HC9/A457 ). Prevalence of symptoms has varied from “most patients with GS have no symptoms” to “1 in 3 people don’t experience any symptoms at all”. Historically, bilirubin ≥1 mg/dl (17.1 μmol/L) has been the standard cutoff for GS. 2 – 4 Sex and age are strongly correlated with bilirubin levels in healthy subjects, suggesting that appropriate personalized bilirubin levels should be defined. 11 , 12 We chose to personalize bilirubin levels (subsequently referred to as “bilirubin centiles”) according to the adjusted distribution: 10th centile defining hypobilirubinemia, 90th centile defining hyperbilirubinemia, and in between representing normobilirubinemia. 13 The exclusion of liver damage requires normal alanine aminotransferase (ALT) and gamma-glutamyl transpeptidase (GGT), 14 , 15 which are not sufficiently sensitive to exclude significant liver fibrosis, in comparison with available noninvasive tests validated in general populations. 7 , 8 Therefore, clinicians cannot accurately diagnose GS or non-benign hyperbilirubinemia using unadjusted liver tests. 2 – 4 The genetic cause of GS is a decrease in uridine-diphosphoglucuronate glucuronosyltransferase family 1 member A1 ( UGT1A1 ), 16 17 18 19 20 the only isoform that significantly contributes to the bilirubin conjugation of . 18 In Europeans, the major genetic variant responsible for GS is a TA insertion in the UGT1A1 gene promoter region, altering the TATA repeat from its usual length of 6 TA repeats. Homozygosity for this short insertion, (TA)7TAA, designated the UGT1A1 *28 rs887829 (ClinVariation ID:12275), defines genotypic GS, the “UGT1A1*28 rs887829 allele homozygosity here named TT genotype; other genotypes that are not UGT1A1*28rs887829 allele homozygotes were named the CT for the heterozygotes and CC for those without UGT1A1*28rs887829 . In Europeans, the TT genotype has variable expressivity and incomplete penetrance in 30%–50% of carriers. 16 17 18 19 20 Therefore, when GS is suspected, the UGT1A genotyping alone is insufficient to indicate GS, which is partly dependent on bilirubin production, geographical origin, the involvement of environmental factors, such as adiposity, and other variants that regulate glucuronidation. 19 20 21 UGT1A1 variants may influence drug-induced toxicities, including numerous medications used in oncology. Genotyping is mandatory for patients with possible GS when drugs that interact with bilirubin metabolism are prescribed as well as the metabolizer state. 22 For the diagnosis of hypobilirubinemia, there is no associated genotype or symptoms. 2 – 6 There is no consensus on the appropriate lower bilirubin level; the most frequently used cutoff is <10.0 μmol/L. 3 – 6 , 23 – 25 Our first aim was to propose new personalized definitions of GS and hyperbilirubinemia. Second, we aimed to assess the prevalence of illnesses possibly associated with GS or hypobilirubinemia. Third, we aimed to assess whether participants with GS or hypobilirubinemia have different overall survival (OS) rates (referred to as “survival”) than those with normobilirubinemia when adjusted for confounders.

This was a retrospective analysis of the prospective UK Biobank cohort (UKB-ID 670334), for which measurements and data collection characteristics of the participants are described in detail elsewhere. 6 , 26 – 28 The UKB involved ~500,000 participants in their middle age (40–70 years, 54% women) recruited between 2006 and 2010. The UKB study was approved by the North-West Multicenter Research Ethics Committee (reference 11 /NW/0382). Of the 502,386 cases (Fig. 1 A), 240,426 subjects were excluded due to missing data or non-European ancestry. The remaining 261,960 participants represent the general population of the United Kingdom with European ancestry, referred to in this study as the “general population.” A further 32,802 nonhealthy participants were excluded due to HIV infection, alcohol disorder, nonmetabolic liver disease, significant liver fibrosis, disease with poor prognosis, or existing cancer. Finally, 2 main subsets were separately analyzed. Among the “apparently healthy liver” subset, characteristics were compared according to bilirubin centiles (Table 1 ). This subset of 138,125 participants had nonelevated ALT and GGT, nonsignificant fibrosis using FIB4, 7 no confounders of metabolic syndrome, and C-reactive protein (CRP) <10 IU/L, enabling to respond to the study aims. The remaining 91,033 participants, named the “at risk of NAFLD” subset (Table 2 and Supplemental Table S2, http://links.lww.com/HC9/A458 ), were considered in sensitivity analyses to assess the prognostic value of bilirubin centiles associated with this emerging disease, and they represented 34% of the “general population” subset (Supplemental Table S3, http://links.lww.com/HC9/A459 ). 1B 10th and 90th centiles of total bilirubin levels by age and sex. Abbreviations: ALT, alanine aminotransferase; BMI, body mass index; CC, absence of the allele; CRP, C-reactive protein; CT genotype, heterozygoty; GGT, gamma-glutamyl transpeptidase TT genotype, allele homozygosity. Main characteristics and comparisons of participants in the “apparently healthy liver” subset, (N=138,125) according to bilirubin centiles, adjusted by rs887829 and age Kruskal-Wallis rank sum test; Pearson chi-squared test; Fisher exact test. Bonferroni correction for multiple testing. Abbreviations: ALT, alanine aminotransferase; CC, absence of the allele; CRP, C-reactive protein; CT genotype, heterozygoty; GGT, gamma-glutamyl transpeptidase; IQR, interquartile range; TT genotype, allele homozygosity. We first determined whether previous studies on GS or hypobilirubinemia had used the grouping of 43 morbidities proposed by Barnett et al in addition to the 11 conditions detailed in gallstone and treated dyspepsia main groups, that is, 54 conditions according to the International Classification of Diseases (named illnesses here) prospectively assessed in the UKB (last connection on PubMed March 31, 2023, Supplemental Table S4, http://links.lww.com/HC9/A460 ). 29 30 31 The secondary end points were the prevalence of the frailty phenotype and multimorbidity count. Participants were deemed frail if they met at least 3 of the 5 frailty criteria. 29 For, The survival end point was overall 15 year survival adjusted for confounders of GS and hypobilirubinemia compared with the population with normobilirubinemia. Full methodological details, including UGT1A1 genotyping , are available elsewhere 26 27 28 and on the UK Biobank website ( https://www.ukbiobank.ac.uk/ ). At baseline, all participants provided informed consent for the study and completed a self-administered questionnaire and a computer-assisted interview. Baseline data included the frailty phenotype and morbidity count (Supplemental Table S4, http://links.lww.com/HC9/A460 ). 6 , 26 – 28 Bilirubin was measured by colorimetric assay with a unisex reference interval of 5.1–17.1 µmol/L. ALT, aspartate aminotransferase (AST), and GGT were analyzed by enzymatic rate. FIB4 score was computed using the cutoff of 2.67 for significant fibrosis. 7 Clinical investigations were conducted according to the principles of the Declaration of Helsinki. All authors had access to the study data and reviewed and approved the final manuscript. All analyses were planned before the inspection of the data in accordance with STROBE guidelines. To elucidate the relationships between bilirubin, UGT1A1 genotype, sex, and age, we plotted bilirubin levels against age separately for UGT1A1 genotype (TT, TC, and CC) and sex (Fig. 1 B). According to Figure 1 B, the corresponding median values and centile distributions demonstrate that the usual unisex cutoff of 17.1 µmol/L induced an imbalance between sensitivity and specificity for bilirubin for the diagnosis of hyperbilirubinemia, regardless of the UGT1A1 genotype. When applied to the general population, such an imbalance presents important limitations related to the classical dilemma of balance between sensitivity and positive predictive value. 32 Supplemental Figure S1, http://links.lww.com/HC9/A461 , shows the expected variability according to 95, 90, and 80% limits of the distribution of total bilirubin by age, stratified by genotype and sex in the “General population” subset. We applied the method developed by Ritchie et al to address the issue of reference ranges of serum measurements, which varied according to age and sex. 33 When values were expressed as multiples of age-specific and sex-specific median levels, the resulting distributions fitted a log Gaussian distribution, which can be used to assign an individual’s measurement to the corresponding centile. Ideally, such a population (“apparently healthy liver”) should include healthy individuals without a high risk of false positives or negatives; however, care should be taken to avoid the use of excessively stringent definitions of “normal” or “healthy” that lead to the paradox of normality becoming a rarity. 34 A simplified table of bilirubin centiles is presented in Table 3 and detailed in Supplemental Table S5, http://links.lww.com/HC9/A462 . Main characteristics and comparisons of participants for subsets “apparently healthy liver” and “at risk of MAFLD” Wilcoxon rank sum test; Pearson chi-squared test. Bonferroni correction for multiple testing. Abbreviations: ALT, alanine aminotransferase; AST, aspartate aminotransferase; BMI, body mass index; CC, absence of the allele; CT genotype, heterozygoty; GGT, gamma-glutamyl transpeptidase; IQR, interquartile range; TT genotype, allele homozygosity. Simplified reference interval for hypobilirubinemia, normobilirubinemia and hyperbilirubinemia (µmol/l and mg/dl) adjusted by rs887829 , sex and age Notes: Low (10th centile) and high (90th centile) values for bilirubin in « apparently healthy liver » subset (n=138,125), per sex, age group, and genotype. Subjects below low are defined as hypobilirubinemia, between low and high as normobilirubinemia, and above high as hyperbilirubinemia. The extensive version of this table is available as Supplemental Table S5. Abbreviations: CC, absence of the allele; CT genotype, heterozygoty; TT genotype, allele homozygosity Univariate and multivariate logistic regressions were computed for each illness against all confounders and expressed as OR separately for women and men and for GS and hypobilirubinemia. We first assessed univariate correlations between bilirubin centiles, UGT1A1 genotyping, and age, with 17 confounders associated with bilirubin in the literature. 2 – 4 The confounders were analyzed in 4 groups. In addition to age, 1 group referred to environmental factors: tobacco consumption, alcohol consumption, physical activity (walking at a brisk pace), and Townsend Deprivation Index. A second group referred to metabolism: body mass index, total cholesterol (referred to as “cholesterol”), triglycerides, LDL, and apolipoprotein-A1 (apoA1). A third group referred to inflammation: CRP, albumin, and platelets. A fourth group comprised biomarkers of liver damage: ALT, AST, and GGT. The final group referred to biomarkers of hemolysis, which is a natural source of bilirubin: hemoglobin and reticulocytes (Supplemental Figure S2, http://links.lww.com/HC9/A463 ). After 4 regression analyses in each group (round 1), the remaining significant confounders were analyzed in a final regression analysis (round 2). The characteristics were compared according to sex and bilirubin centiles using a Q-test and Bonferroni correction, considering the number of comparisons. Significant differences between illnesses were defined when the p -value was < 0.005 and p < 0.05 for OR comparison. We summarized the literature analyzing causal factors by MR (Table 4 ) detailed references (Supplemental Table S6, http://links.lww.com/HC9/A464 ). Long-term morbidity (n=54 illnesses) and confounders of bilirubin centiles adjusted by rs887829 and age Notes: Summary of risk in hyperbilirubinemia and hypobilirubinemia in the present study in “apparently healthy liver” subset of a general population and in the literature. Causal was identified in a single illness, cholelithiasis-gallstone, 34 were not causal, and 19 not tested. Zanussi et al explored the association between TB and 19 illnesses using Mendelian randomization (MR) in 16,281 participants in a hospital-based biobank, not a general population. Abbreviations: AF, Atrial fibrillation; UKB, UK Biobank; CHS, Coronary Heart Disease; COPD, Chronic obstructive pulmonary disease; CVD, Cardio vascular disease; MR, Mendelian randomization; TIA, Transient Ischemic Attack. To assess the effect of bilirubin on illnesses and confounders, we performed two-sample Mendelian randomizations (TSMR) using published databases (Table 5 ). TSMR is a method for strengthening causal inference in observational studies using genetic variants ( UGT1A1 genotyping) and genome-wide association studies associated with exposure (bilirubin) as instrumental variables. TSMR, using bilirubin (after inverse rank normalized transformation) as the exposure, was performed sequentially for the illnesses (outcomes) and confounders (Supplemental Table S3, http://links.lww.com/HC9/A459 ). The selected variants in the genome-wide association studies databases were those associated with Europeans from the UKB. For each TSMR, the effect and the p -value were reported; p -value <0.05 demonstrated a causal effect. Two-sample Mendelian randomization. Notes: Assessing the causal effect of bilirubin on self-reported illnesses and confounders. The exposure and the outcome database selected in the GWAS databases are the one for the European subjects from the UK Biobank and with available data. Results display the OR with 95% CI, number of SNPs involved, p -value using inverse variance weighted method, and the sample size in the UK Biobank (source GWAS). Abbreviation: GWAS, genome-wide association studies. Kaplan-Meier curves were compared using first the log-rank test. Then, the primary end point of adjusted 15-year survival was assessed using the 17 confounders and multivariate Cox proportional hazards analyses. The proportional hazard assumption was evaluated by visual inspection of curves and using the Schoenfeld residual plots versus time. Survminer libraries and R software were used. The intra-participant variability of bilirubin was assessed at 6 years, by sex, and by UGT1A1 genotype using repeated ANOVA with Bonferroni adjustment. The correlations between confounders were analyzed using Pearson correlation coefficient (Supplemental Figure S2, http://links.lww.com/HC9/A463 ).

In the “apparently healthy liver“ subset, the 10th and 90th centiles of bilirubin, stratified by sex, age, and UGT1A1 genotype (Fig. 1 B), allowed us to readjust the GS prevalence among women and identify several significant differences in illness prevalence between GS, normobilirubinemia, and hypobilirubinemia (Figure 1 , Table 2 , and Supplemental Table S2, http://links.lww.com/HC9/A458 ). Almost all these differences were explained by confounders (Supplemental Table S6 references, http://links.lww.com/HC9/A464 ) as well as the differences in OS observed after (Figure 2 A) and before adjustments (Figure 2 B). Adjusted and nonadjusted survivals. (A) Adjusted survivals in the « apparently healthy liver» subset. Adjusted Cox curve in women (left) and men (right) for overall survival according to bilirubin centiles; hyperbilirubinemia (green curve) and hypobilirubinemia (red curve) against the reference normobilirubinemia group (blue). The curves are adjusted for confounders (age, smoking status, alcohol intake, walking pace, deprivation index, CRP, albumin, PLT, and cholesterol). The table shows the HR and 95% CI for bilirubin centiles for the multivariate Cox regression. Abbreviation: PLT, platelets. The centile distributions of bilirubin according to age, separately for UGT1A1 genotype and sex, graphically in females (upper lines), demonstrated the lack of sensitivity of the usual cutoff (horizontal dotted black line) for hyperbilirubinemia. In females with the TT genotype, the median bilirubin level was 17 μmol/L at 40 years of age, which decreased to a plateau of 14 μmol/L between 60 and 70 years of age (Fig. 1 B). The hypobilirubinemia cutoff varied from 4.6 to 9.4 µmol/L, and the hyperbilirubinemia cutoff ranged from 9.0 to 26.2 μmol/L (Supplemental Table S5, http://links.lww.com/HC9/A462 ). The characteristics of bilirubin centiles were compared in the 3 subsets: “apparently healthy liver” (Table 2 ), “at risk of NAFLD” (Table 2 and Supplemental Table S2, http://links.lww.com/HC9/A458 ), and “general population” (Supplemental Table 3, http://links.lww.com/HC9/A459 ). Associations between bilirubin centiles and confounders were described in Table 4 and Supplemental Table S6 (Table 4 and Supplemental Table S6, http://links.lww.com/HC9/A464 ) For 9 hepatobiliary or digestive illnesses and 13 other illnesses, the correlations were not significant for causality. For depression and thyroid disorders, we did not observe significant causality, in contrast with previous MRs. Aging resulted in an expected significant ( P <0.001) decrease in TB in 7249 participants with repeated measurements at 6 years, which was demonstrated for the first time according to UGT1A1 genotype in both sexes. The median decrease ranged from −0.24 μmol/L in CC women to −0.97 μmol/L in TT men (Figure 2 C). In women (Supplemental Table S7, http://links.lww.com/HC9/A465 ), for GS, the confounders with the most significant ORs (all P <0.001) were current smoking (negative, −), daily or almost daily alcohol consumption (positive, +), deprivation index (−), brisk walking pace (+), triglycerides (−), apoA1 (+), CRP (−), albumin (+), AST (+), and reticulocytes (+). Inverse results were observed for hypobilirubinemia. In men (Supplemental Table S8, http://links.lww.com/HC9/A466 ), the same results were observed for ORs, with the exception of alcohol intake 3 to 4 times a week having the highest OR for regression for both GS and hypobilirubinemia, and there was a lack of significant association between hypobilirubinemia and body mass index and cholesterol. Detailed analyses of round 1 are detailed in Supplemental Table S9, http://links.lww.com/HC9/A467 , for women and in Supplemental Table S10, http://links.lww.com/HC9/A468 , for men, permitted to select non-colinear confounders. Round 2 allowed us to identify the most significant independent factors. In women (Supplemental Table S11, http://links.lww.com/HC9/A469 ) and men (Supplemental Table S12, http://links.lww.com/HC9/A470 ), for GS, these main confounders (ORs all P <0.001) were smoking status (−), TC (−), triglycerides (−), LDL (+), apoA1 (+), CRP (−), albumin (+), AST (+), hemoglobin (+), and reticulocytes (+), with inverse results in hypobilirubinemia. Unexpected discordances were observed according to sex; alcohol intake was not significant in GS men, but positive in GS women, and brisk pace was not significant in GS women but negative in GS men. For the first time, 28 out of 54 illnesses with significant differences were identified in both GS and hypobilirubinemia compared with normobilirubinemia ( P <0.05) (Figs. 1 C, 1 D). Cholelithiasis in men and jaundice of unknown cause in women (Fig. 1 C) were the only 2 conditions out of the 10 detailed biliary and digestive illnesses that were more frequent in GS than normobilirubinemia. Two factors were different in GS: higher risk for cholecystitis/bile duct disease/obstruction and lower risk for hiatus hernia in women (Fig. 1 C). In hypobilirubinemia, only gastroesophageal reflux risk was lower for both sexes (Fig. 1 D). Regarding nonbiliary digestive conditions (Fig. 1 E), 3 illnesses were more frequent in GS for both sexes: atrial fibrillation, coronary heart disease, and hypertension. Heart failure was more frequent only in men, and glaucoma was more frequent only in women. Four illnesses were less frequent in both sexes: schizophrenia, obstructive disease, asthma, and depression. Four illnesses were less frequent only in women: epilepsy, connective tissue disease, painful conditions, and endometriosis. Four illnesses were less frequent only in men: diabetes, sinusitis, psoriasis/eczema, and osteoporosis. For hypobilirubinemia, most results were similar and reversed (Fig. 1 F). Only 18%/18% (women/men) of participants declared no illness (Table 1 ). Three illnesses were declared at a frequency above 10%: hypertension (24%/32%), painful conditions (24%/22%), and asthma (15%/15%). The remaining 20 illnesses had a frequency of 5%–10%. Frailty was significantly less frequent in GS compared with normobilirubinemia in women (GS: 70/7,741 [0.9%]; normobilirubinemia: 905/(15%) (2.7%) and men (GS: 50/6179 [0.8%]; normobilirubinemia: 573/48,988 [1.2%]) ( P <0.001); the inverse was observed for hypobilirubinemia. Similarly, the multimorbidity counts were lower in GS compared with normobilirubinemia in both women and men and the inverse was observed for hypobilirubinemia (Table 1 ). The analyses for multivariate round 1 are detailed in Supplemental Table S9, http://links.lww.com/HC9/A467 for women and Supplemental Table S10, http://links.lww.com/HC9/A468 for men. Multivariate round 2 is presented in Supplemental Table S11, http://links.lww.com/HC9/A469 for women and Supplemental Table S12, http://links.lww.com/HC9/A470 for men. Eleven confounding factors were independently associated with GS and hypobilirubinemia. The deprivation index was not associated with GS in men or women, nor was it associated with hypobilirubinemia in men. Furthermore, in men, body mass index and GGT were not associated with hypobilirubinemia. We found a significant causal association between bilirubin centiles and cholelithiasis/gallstone illness. The TSMR assessing the causal effect of bilirubin on illnesses is summarized in Table 5 . No other significant causal association between bilirubin centiles and illnesses was found. There was no significant difference between OS once adjusted for the significant 11-factor model, which combined bilirubin centiles (including three confounders: sex, age, and UGT1A1 ) and 8 other confounders: smoking, alcohol intake, walking pace, deprivation, CRP, platelets, albumin, and cholesterol (Fig. 3 A). In women with GS (hyperbilirubinemia), the OS HRs were nonsignificant for normobilirubinemia (HR: 1.10, 95% CI [0.94–1.22], P =0.32) or for hypobilirubinemia (HR: 1.10, 95% CI [0.98–1.23], P =0.09). The differences were nonsignificant in men with GS (HR: 1.06, 95% CI [0.95–1.18], P =0.30) and hypobilirubinemia (HR: 0.98, 95% CI [0.89–1.08], P =0.71). Adjusted survival and Kaplan-Meier in the « apparently healthy liver » subset. (A) Adjusted Cox curve in women (left) and men (right) for overall survival according to bilirubin centiles; hyperbilirubinemia (green curve) and hypobilirubinemia (red curve) against the reference normobilirubinemia group (blue). The curves are adjusted for confounders (age, smoking status, alcohol intake, walking pace, deprivation index, CRP, albumin, PLT, and cholesterol). The table shows the HR and 95% CI for bilirubin centiles for the multivariate Cox regression. (B). Kaplan–Meier curve and 95% CI in women (left) and men (right) for overall survival according to bilirubin centiles; hyperbilirubinemia (green curve) and hypobilirubinemia (red curve) against the reference normobilirubinemia group (blue). The table shows the HR and 95% CI for bilirubin centiles for the univariate Cox regression. The horizontal time scale is displayed in years of exposure to the risk. All figures use the same scale to ease visual comparison. Cumulative number of events are shown in the lower left table for women and the lower right table for men. Abbreviation: PLT, platelets. According to a nonadjusted model (Fig. 3 B), there was considerably lower OS in hypobilirubinemia for both sexes compared with normobilirubinemia in women (HR: 1.43, 95% CI [1.28–1.59]) and men (HR: 1.38, 95% CI [1.27–1.51]) ( P <0.001). There were no significant differences in OS between GS and normobilirubinemia for women. The main causes of mortality are presented in Supplemental Table S13, http://links.lww.com/HC9/A471 . In both sexes, the most striking difference was the higher mortality associated with pulmonary cancer in hypobilirubinemia compared with normal and hyperbilirubinemia. Analyses against confounders are detailed for univariate regression in Supplemental Table S14, http://links.lww.com/HC9/A472 , for women and Supplemental Table S15, http://links.lww.com/HC9/A473 , for men and multivariate round 1 in Supplemental Table S16, http://links.lww.com/HC9/A474 , for women and Supplemental Table S17, http://links.lww.com/HC9/A475 , for men. Multivariate round 2 is presented in Supplemental Table S18, http://links.lww.com/HC9/A476 , for women and Supplemental Table S19, http://links.lww.com/HC9/A477 , for men. Compared with the “apparently healthy liver” subset, in the “at risk of NAFLD” subset, there was a significant association between a deprivation index and hyperbilirubinemia for both sexes; in men only, there was no association between CRP and hyperbilirubinemia or between apoA1 and hypobilirubinemia. Illnesses had varying prevalence according to bilirubin centile. In hyperbilirubinemia, more cholelithiasis was observed in men, but also in women (Supplemental Figure S3A, http://links.lww.com/HC9/A478 ), and in hypobilirubinemia, more hiatus hernias were observed in women and for men (Supplemental Figure S3B, http://links.lww.com/HC9/A478 ). In contrast with the “apparently healthy liver” subset, women with normobilirubinemia (Fig. 2 D, Fig. 2 E blue lines) had lower risk of death (HR=1) than women with hypobilirubinemia (HR=1.18, 95% CI [1.07–1.33], P <0.001) or hyperbilirubinemia (HR=1.16, 95% CI [1.01–1.33], 0.03), suggesting a U-shaped curve of bilirubinemia risk. In men, there was a higher HR (1.51, 95% CI [1.40–1.62]) in hypobilirubinemia compared with normobilirubinemia, but no difference for hyperbilirubinemia (Fig. 2 E). Interestingly, univariate comparisons indicated a significantly increased risk of death in hypobilirubinemia (Fig. 2 E). The main causes of mortality are indicated in Supplemental Table S13, http://links.lww.com/HC9/A471 . Despite the small sample size, the most striking results in the causes of death were the ranking of COVID-19 in participants in the “at risk of NAFLD” group, where COVID-19 was in the top 6 causes of death, including 2 in the second place in women, in comparison with “apparently healthy liver” group, where COVID-19 never appeared in the top 4 causes of death.

The findings addressed the 3 aims of the study. These findings must be discussed in the context of the literature, the study limitations, and the expected consequences for patients and health care services. We agree with Vitek that due to the many confounding factors affecting bilirubin serum concentration levels, it is hard to establish reliable decision limits and that data on women were scarce. 4 The response to the first aim was a proposal of personalized bilirubin centiles according to sex, age, and UGT1A1 genotyping, using the 10th centile for hypobilirubinemia, 90th centile for hyperbilirubinemia, and values in between for normobilirubinemia—as simple decision limits that was already used by the National Institutes of Health for liver markers. 13 This choice permitted a better balance between sensitivity and specificity, especially in women, than the unisex cutoff. 32 For women, the results demonstrated that the existing GS cutoff was not sufficiently sensitive to assess the prognosis and illnesses associated with bilirubin centiles. GS is described as being more frequent in males than females (Supplemental Table S1, http://links.lww.com/HC9/A457 ), which is not true according to the age distribution of bilirubinemia (Figure 1 B). A population study illustrated this risk of underestimation of GS in women. Among participants referred for genetic testing after incidental discovery of hyperbilirubinemia using the usual cutoff, a total of 1191—1150 males and only 41 females had the UGT1A1rs8175347,rs4148323 genotyping. 35 Another implication for women with the UGT1A1 genotyping is the increased risk of toxicity associated with chemotherapies, such as irinotecan. The largest study on colorectal cancer (n=1362) observed a higher confounder-adjusted risk of irinotecan-induced severe neutropenia and more severe diarrhea in women compared with men. 36 The new definitions should improve studies on the effectiveness of drugs interacting with bilirubin metabolism. The Food and Drug Administration recommendations concerning pharmacogenetic associations, seven drugs were affected by bilirubin according to UGT1A1 -genotyping variants and metabolizer status. Three drugs require therapeutic management: belinostat (28/28 variant, poor metabolizers), irinotecan (1/6, 1/28 intermediate metabolizers; or 6/6, 6/28, 28/28 poor metabolizers), and sacituzumab govitecan-hziy (28/28 poor metabolizers). For nilotinib and pazopanib (both 28/28 poor metabolizers), there is a potential impact on safety and response, and for dolutegravir (poor metabolizers) and raletegravir (28/28 poor metabolizers), there is an impact only on pharmacokinetic properties. 37 Regarding hypobilirubinemia, there is no rationale for using a unisex cutoff. Several mechanisms could contribute to low bilirubin levels in women despite the same prevalence of the UGT1A1 genotype (Table 1 ), such as combinations of mutations, polymorphisms, estrogen, and the degradation of heme and nutrients. 23 24 25 For the first time, our findings allow the adjustment of personalized hypobilirubinemia definitions in the Europeans, with a range of 4.7–9.2 μmol/L in women and 5.6–12.1 μmol/L in men (Table 3 ). The previously recommended cutoffs of hypobilirubinemia without adjustment were 5.0 μmol/L for women 24 and 7 μmol/L for men. 2 The genetic association of UGT1A1 and several other variants with cholelithiasis/gallstones was established in 2010 but was never assessed prospectively in a large general population. For the first time, the prevalence of cholelithiasis in men with GS was assessed at 20% of European adults together with a 20% increase of the risk of complications, significantly more frequent than in men with normobilirubinemia, after adjusting for confounders. Furthermore, our TSMR validated the causality pathway, including UGT1A1 variants observed in 3 studies. 38 39 40 41 Among the remaining 34 illnesses evaluated by TSMR, none had a causal association with TB. A nonsignificant causality association is not proof of an absence of causality; however, our results were concordant with previous MR and multivariate confounder analyses for 11 common illnesses (Table S4 Supplemental references, http://links.lww.com/HC9/A460 ). The most frequent confounders for TB are now well identified and include sex, age, smoking, alcohol consumption, and drugs interacting with the UGT1A1 genotype. For 2 illnesses, thyroid disorder 44 and depression, 45 we did not find the significant associations reported by others (Table 4 ). These discordances might be explained by the lower power of our study, with a smaller panel of genetic instrumental variables and a smaller number of participants in our database. However, the association might be inverse due to the prescription of antidepressant drugs interacting with bilirubin metabolism. 37 Recent associations with gut metagenomes in patients with inflammatory bowel disease or irritable bowel syndrome suggest a potential causal role of bilirubin (Table 4 ). The prevalence of the frailty phenotype, estimated for the first time in GS, was very low in both sexes (<1%), significantly less than in normobilirubinemia, and contrasting with the prevalence of the pre-fail status (33% in GS vs. 34% in normobilirubinemia) (Table 1 and Supplemental Table S2, http://links.lww.com/HC9/A458 ). These differences may be explained by a greater proportion of participants younger than 65 years of age and the exclusion of nonhealthy participants in our study compared with previous studies. 29 30 31 In contrast, the multimorbidity count revealed that 82% of the participants with normobilirubinemia in the “apparently healthy liver” subset declared at least 1 morbidity at baseline. This is a major finding for improving communication with anxious individuals after a diagnosis of GS, who have a significantly lower multimorbidity count than individuals with normobilirubinemia (Supplemental Table S1, http://links.lww.com/HC9/A457 ). Our choice of defining hyperbilirubinemia and hypobilirubinemia stratified by age, sex, and UGT1A1 genotype allowed us to interpret the confounders on the pathway from bilirubin to illness. For individuals with GS, the findings will facilitate the construction of personalized risk scores combining genetic and environmental factors for early diagnosis and prevention of cholelithiasis in men. 38 The response to the third aim was that the prognostic value of TB published in the literature and initially retrieved in our univariate results disappeared after adjustment for confounders in both females and males. These results are original due to the power of the cohort but are certainly not definitive. Another original result was observed in the “at risk of NAFLD” subset, as participants with normobilirubinemia had higher adjusted survival (Figure 2 D) than those with hyperbilirubinemia, which is visible in nonadjusted curves (Figure 2 E). These participants were selected for having a low risk of significant liver fibrosis by FIB4; therefore, it is prudent to assess the absence of significant fibrosis with nonelevated transaminases using more accurate, noninvasive biomarkers. To date, contrasting results have been published in various cohort studies investigating an association between genetic UGT1A variants and NAFLD. A likely explanation for the inconsistent data reported involves the presence of UGT1A polymorphisms found in isoforms other than UGT1A1 . 42 These findings on definition, symptoms, and survival must be discussed in consideration of the methodological limitations and the expected consequences for patients with newly identified phenotypes. The UKB has limitations, including the middle-aged European participants with better survival than the general population of the United Kingdom, 29 which decreases the generalizability of the results. As the reference intervals start at the age of 40 years, it remains unclear if this will be useful in younger subjects. It could be the ones needing advice and or reassurance when first diagnosed with having GS. We limited the number of confounders to 17 to balance the risk of missing an independent confounder and the risk of overfitting. We did not include several behavioral confounders: the time of the day of blood sampling, fasting hours, and coffee consumption. 43 However, we tested the intra-subject TB variability using repeated samples (n=7249) and a sample that was small but representative of the general population. The decrease in TB with age was expected and underlined the risk of a false-negative diagnosis of GS when the UGT1A1 genotype is not assessed in both sexes in older individuals (Fig. 2 C). Many illnesses were self-reported and monitored by trained staff; other items were assessed directly by trained employees, such as grip strength. 30 We used the simple methodology outlined by Ritchie et al 33 following the reference distributions for the 10th-90th centiles adjusted for age and sex, and not narrower intervals, in maximizing the balance between the sensitivity and specificity of bilirubin centiles for establishing nonlinear relationships with survivals and confounders. We focus on total bilirubin and not on unconjugated bilirubin, also available in the UK Biobank. Indeed, the Pearson correlation in 118,796 participants was =0.997, and the linear regression showed no pragmatic interest in large population as unconjugated bilirubin is a less widespread biomarker (Supplemental Figure S4, http://links.lww.com/HC9/A479 ). We did not analyze the variants in populations with more genetic heterogeneity, such as Asian populations, nor did we compare recessive and dominant genes. The variant rs887829 in the UGT1A1 gene has been reported to explain at least 30%–50% of the variance in bilirubin levels, and it is in nearly complete linkage disequilibrium with the genetic polymorphisms that underlie GS in Europeans. We did not study the possible association of survival with other loci, such as the solute carrier organic anion transporter family member 1B1, which is also strongly associated with bilirubin levels. 46 Even after MR, genetic analysis was not robust enough to detect nonlinear effects or to quantitatively estimate causal effects; therefore, further studies are required to investigate to what extent bilirubin centiles are beneficial for predicting survival through different related functions. 21 The finding that participants who never or only occasionally drank alcohol was more likely to have the frailty genotype may be explained by abstainer bias, suggesting that these subjects drank no alcohol because they had poorer health and might have been following advice to abstain. 29 The differences in alcohol-bilirubin correlations observed between GS and hypobilirubinemia suggest more complex correlations between confounders than the U-shape (Table 5 ). FIB4 was the fibrosis score computed with the data available in the UKB but has been shown to be a poor predictor of fibrosis in patients with NAFLD 46 and has worse performance in older patients, which has the potential to further skew the data over time. 15 As stated in other studies, the identification of either benefits or harms would be a powerful argument for pre-emptive genotyping so that these individuals could enjoy the benefits and avoid the harms associated with their genotype. 10 , 38 Even if confounders ruled out direct and causal relationships between bilirubin centiles, the UGT1A1 genotype, illnesses, and survival, the findings of this study should improve the awareness of GS and hypobilirubinemia among individuals in European populations. The frequencies of most illnesses in GS were less than or similar in normobilirubinemia. In conclusion, 2 consequences of these findings are expected, the first being a better awareness of the benefits and risks associated with GS, especially in women needing chemotherapy for whom GS is underestimated. Second, the findings support the construction of a personalized cholelithiasis risk score in men with GS.