Skip to main content
NCBI home page
NIHPA Author Manuscripts logoLink to NIHPA Author Manuscripts
. Author manuscript; available in PMC: 2023 Feb 1.
Published in final edited form as: Clin Pharmacol Ther. 2021 Oct 24;111(2):435–443. doi: 10.1002/cpt.2441

Identifying potential therapeutic applications and diagnostic harms of increased bilirubin concentrations - a clinical and genetic approach

Jacy T Zanussi 1, Juan Zhao 2, Chad A Dorn 2, Ge Liu 1, QiPing Feng 1, WeiQi Wei 2, Jonathan D Mosley 1,2, C Michael Stein 1, Vivian K Kawai 1
PMCID: PMC8748314  NIHMSID: NIHMS1746934  PMID: 34625956

Abstract

Bilirubin has antioxidant and anti-inflammatory properties in vitro and in animal studies and protects against inflammatory, cardiovascular, and other diseases in observational studies; therefore, bilirubin has potential as a therapeutic agent. However, observational studies could be confounded by many factors. We used a genetic (n=61,281) and clinical (n=234,670) approach to define the association between bilirubin and 19 conditions with a putative protective signal in observational studies. We also tested if individuals with genetically higher bilirubin levels underwent more diagnostic tests. We used a common variant in UGT1A1 (rs6742078) associated with an 26% increase in bilirubin levels in the genetic studies. Carriers of the variant had higher bilirubin levels (P= 2.2×10−16) but there was no significant association with any of the 19 conditions. In a phenome-wide association study (pheWAS) to seek undiscovered genetic associations, the only significant finding was increased risk of “jaundice - not of newborn”. Carriers of the variant allele were more likely to undergo an abdominal ultrasound (OR=1.04 [1.00, 1.08], P=0.03). In contrast, clinically measured bilirubin levels were significantly associated with 15 of the 19 conditions (P< 0.003) and with 431 clinical diagnoses in the pheWAS (P<1×10−5 adjusted for sex, age, and follow-up). With additional adjustment for smoking and body mass index, 7 of 19 conditions and 260 pheWAS diagnoses remained significantly associated with bilirubin levels. In conclusion, bilirubin does not protect against inflammatory or other diseases using a genetic approach; the many putative beneficial associations reported clinically are likely due to confounding.

Keywords: Bilirubin, UGT1A1, pheWAS

Introduction

Bilirubin has potential as a novel therapeutic agent (1); this idea is supported by evidence from in vitro, animal, and epidemiologic studies. Clinical evidence of bilirubin’s anti-inflammatory effects dates back to an observation made in the 1930’s that some patients with rheumatoid arthritis went into remission when they became jaundiced (2). Subsequent studies showed that bilirubin is not only a potent antioxidant (3) and anti-inflammatory (4) agent but also an effective treatment in several animal models of human disease (5–7).

Many epidemiologic and observational studies have reported that higher bilirubin levels may protect against a wide range of clinical conditions including autoimmune inflammatory diseases (rheumatoid arthritis (8) and ulcerative colitis (9)) and many other diseases with an inflammatory component: cardiovascular diseases (CVD)(10), type 2 diabetes (T2D) (11), metabolic syndrome and its components (hypertension (12), obesity (13) and dyslipidemia (14)), chronic kidney disease (CKD) (15), non-alcoholic fatty liver disease (NAFLD) (16), cancers (17) and some neurological disorders (18). However, because bilirubin levels are affected by many factors (1, 19) it is possible that epidemiologic observations of a putative protective effect against such a wide range of conditions represents unmeasured confounding factors. Nevertheless, drugs to induce mild hyperbilirubinemia are under consideration for the treatment of several conditions (1).

Recently, genetic approaches (20) have been used to determine if there is a direct relationship between a biological variable and an outcome (21) because such approaches may be less confounded by clinical and environmental factors. A Mendelian randomization (MR)-type approach is particularly attractive for bilirubin because a common genetic variation resulting in benign hyperbilirubinemia (Gilbert’s syndrome) is present in approximately 5–8% of European ancestry populations (1) and increases average bilirubin concentrations by approximately 0.15 mg/dL (2.607 μmol/L) per allele (22).

In genome-wide association studies (GWAS) a single variant, rs6742078, in a uridine diphosphate-glucuronyltransferase gene (UGT1A1) accounted for the vast majority of genetic variability in bilirubin levels, explaining as much as 19% of variability in an additive model (23). rs6742078, which has a minor allele frequency of 30% in individuals of European ancestry, is in linkage disequilibrium (R2=0.91) with a common functional variant (rs45557732) in the promoter region of UGT1A1 (1). MR approaches using rs6742078 as an instrumental variable have examined the relationship between bilirubin and the prespecified outcomes: stroke (24), T2D (22) and ischemic heart disease (25).

Based on the epidemiologic evidence, people with genetic variation that markedly increases bilirubin concentrations could have wide-ranging clinical benefits; however, they could also suffer adverse consequences resulting from increased diagnostic medical interventions for an abnormal liver function test caused by a benign genetic variation. 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 of their genotype. The potential therapeutic benefits or harms of increased levels of bilirubin have not been defined using unbiased genetic approaches in a large population. The recent availability of large DNA biobanks linked to electronic health records (EHR) provides the resources to answer these questions.

We examined the hypothesis that genetically-predicted levels of bilirubin are associated with: 1) selected clinical conditions previously reported to be associated with clinically measured bilirubin; and 2) previously unknown pleiotropic effects on clinical conditions, particularly inflammatory ones, using an unbiased phenome-wide association (pheWAS) approach. We also examined the hypothesis that individuals carrying alleles associated with higher bilirubin levels undergo more testing for liver disease. Additionally, to illustrate the potential for confounding of clinical studies, we determined the associations between variation in clinically-measured bilirubin levels within the normal range and future illness using a pheWAS approach.

Methods

Data Sources

For the genetic study, we used BioVU, a large DNA biobank at Vanderbilt University Medical Center (VUMC) (26). In brief, BioVU accrues DNA from blood samples obtained during routine clinical care from patients who have consented to have a DNA sample collected. DNA is extracted, de-identified, and linked to a de-identified version of the electronic health record (EHR) at VUMC. For the clinical study we used the VUMC de-identified EHR. Approval for the study was obtained from the Vanderbilt Institutional Review Board.

Study design and cohorts:

  1. Genotype cohort. Individuals of European ancestry (see genotype data description) identified through BioVU for whom genotype results were available (irrespective of the availability of bilirubin levels) entered the cohort at the time (t0) of the first clinical encounter in the EHR after age of 18. Diagnoses and laboratory values recorded after the patient entered the cohort were included in the analysis to determine if carriers of rs6742078 (associated with increased bilirubin levels) are at increased risk for 19 pre-selected phenotypes previously reported to be associated with measured bilirubin levels in adults (Table S1) or for previously unreported conditions, and if they undergo increased testing for liver disorders (Table S2).

  2. Clinical cohort. Individuals in the de-identified EHR (irrespective of the availability of genotype) who were recorded as being white were considered to be of European descent and were eligible to enter the cohort at the time (t0) of their first recorded liver function test (LFT) after the age of 18. Subjects were excluded if this qualifying LFT showed any evidence of liver dysfunction defined an aspartate transaminase (AST), alanine transaminase (ALT), gamma-glutamyltransferase (GGT) or alkaline phosphatase value greater than the upper limit of normal or total bilirubin (TB) > 1.2 mg/dL. Only individuals with one year or more of follow-up in the EHR were included in the clinical cohort.

Genotype data:

We extracted genotype information for rs6742078, a SNP in UGT1A1 that explains most of the genetic component of variability in serum bilirubin concentrations in populations of European descent (23). Genotyping was performed using Illumina’s Infinium Expanded Multi-Ethnic Genotyping Array (MEGAEX), and quality control (QC) was performed in PLINK v 1.90β following standard procedures (27), which included reconciling strand flips. Departure from Hardy Weinberg Equilibrium (HWE) were measured using χ2 goodness-of-fit test between the observed and expected genotypes and defined as a P-value<1×10−6(28). Principal components (PCs) were calculated using the SNPRelate package (29). European ancestry was defined using the first 2PCs and individuals who did not segregate with European samples (>4 standard deviations) based on PC analysis were excluded.

Analysis:

Associations with genotype:

To define the relationship between rs6742078 and total bilirubin levels, we extracted bilirubin levels in the genotype cohort after cohort entry and computed the median total bilirubin for each individual. We performed a linear regression analysis between the number of alleles for rs6742078 and the log transformed median total bilirubin levels. We adjusted for sex, year of birth, and length of follow-up. We tested the association between rs6742078 and the 19 pre-selected phenotypes (Table S1) and also performed a pheWAS using all clinical diagnoses in the EHR, we extracted clinical diagnoses from the EHR occurring after cohort entry using the Ninth Revision and Tenth International Classification of Diseases (ICD9/ICD10) codes. We transformed these ICD encounter diagnoses into phecodes, which aggregate one or more related ICD codes into distinct diseases or traits (30, 31). For each phenotype, we defined cases as having two or more counts of a given phecode and controls as individuals without the phecode or any related phecode. To improve power, only phecodes with more than 200 cases were included in the analysis. We performed a multivariable logistic regression for the prespecified phenotypes and for the global pheWAS adjusting for sex, year of birth, length of follow-up in the EHR and 5 PCs. The global pheWAS for rs67402878 was performed as described before (30). Associations were expressed as odds ratio (OR) and 95% confidence interval (95% CI). PheWAS was performed using the R pheWAS package (32). For the pre-specified phenotypes, a Bonferroni-adjusted P-value<0.0026 (i.e., 0.05/19) was considered significant. This type I error would provide more than 80% power to detect an OR of 1.50 for phenotypes with 200 cases and smaller OR for more common phenotypes (e.g. OR of 1.21 for phenotypes with 1000 cases) for a cohort of 50,000 individuals using an additive model. For the global pheWAS, a P-value<1×10−5 was considered significant. PheWAS simulations showed that 200 cases or more provide reasonable power to detect genetic associations for common variants (MAF > 0.01) (33). For validation, we used UK Biobank PheWeb, an open-source web-based tool that includes GWAS data using EHR-derived ICD billing codes from the white British participants of the UK Biobank (34).

To test the association between rs6742078 and increased testing for liver disorders, we studied laboratory and clinical tests that may be ordered for the investigation of liver problems (Table S2) such as liver function tests (LFTs), screening for viral infections, screening for hemolytic disorders, and diagnostic imaging (e.g., abdominal ultrasound). Only tests that were performed after cohort entry were included in the analysis. For LFTs, we compared the number of times the test was ordered across genotype groups; for the other tests we compared whether the test was ever performed. We used logistic regression analysis to assess the effect of genotype on liver tests adjusting by year of birth, sex, length of follow-up in the EHR, and the first 5 PCs and considered P<0.05 significant.

Associations with measured bilirubin concentrations:

We studied the association between measured bilirubin levels and the future occurrence of the 19 pre-specified phenotypes (Table S1) in multivariable logistic regression analyses. To avoid capturing illnesses that may have been related to the performance of the first LFT, we only included phecodes recorded in the EHR one month or more after cohort entry. We performed two logistic regression analyses: the first adjusted for age at cohort entry, sex, and length of follow-up in the EHR; the second additionally adjusted for body mass index (BMI), tobacco use disorder, and smoking status. BMI was calculated as weight in kilograms (kg) divided by the square of height in meters (m). BMIs below 12 or above 70 were considered unreliable and discarded (35). The first reliable BMI after cohort entry was included as a covariate. Smoking status was defined as never/ever smoked and tobacco use disorder was identified by the presence/absence of the phecode 318. We used the same P-value thresholds as for the genetic analysis.

Testing for reverse causation:

To explore if genetically determined reverse causation might explain the associations observed between bilirubin levels and clinical outcomes after adjusting covariates, we performed inverse variance weighted regression (IVWR) meta-analyses. To perform IVWR analysis we extracted GWAS summary statistics for total bilirubin from the UK Biobank repository (Neale lab: http://www.nealelab.is/uk-biobank) and we selected the largest genetic meta-analysis with summary-level data available for individuals of European ancestry for the pre-selected phenotypes (or proxies when the exact phenotype was not available) but avoided datasets overlapping significantly with UK Biobank samples or without significant SNP associations (P<5×10−8).

To ensure the selection of independent SNPs from the exposure and each outcome dataset, we selected SNPs that were significantly associated with the pre-specified phenotypes (P<5×10−8). The SNPs were aligned to the 1000 Genome Project phase 3 and a linkage disequilibrium (LD)-reduced (r2<0.05) set of SNPs with a minor allele frequency>0.05 were selected as instrumental variables (IVs) to estimate the combined effect of these SNPs in the IVWR meta-analysis. The analyses were performed using the Two-Sample Mendelian Randomization R-package and a P-value of 0.05 was considered significant.

Continuous variables are shown as median [interquartile range] and were analyzed using non-parametric tests unless otherwise specified. Categorial variables are shown as frequency (percentage) and analyzed with a chi-squared test.

Results

Genotype cohort

Cohort:

The genotype cohort consisted of 61,281 individuals aged 18 years and older who had genotypes available. The T-allele frequency for rs6742078 was 32%, the same allele frequency reported in 1000 Genomes for Utah residents with European ancestry (CEU), and approximately 10.8% of people were homozygous. Many individuals in the genotype cohort (n=51,114; 82.1%) had at least one LFT test performed after the age of 18. In those, the median bilirubin concentration differed significantly across the 3 genotype groups (P=2.2×10−16) and was higher among individuals who carried a T allele (Table 1). rs6742078 explained approximately 7% of the variability of the log transformed median bilirubin levels in our cohort (P-value < 2.2×10−16). Each T allele increased the median bilirubin levels by a factor of approximately 26%. Other than bilirubin levels, genotype groups did not differ in their clinical characteristics (Table 1).

Table 1:

Clinical characteristics of the genetic and clinical cohorts

Clinical characteristics Genotype cohort Clinical cohort
All GG GT TT P-value*
# individuals 51,114 23,852 (46.67%) 21,763 (42.58%) 5,499 (10.76%) 234,670
Male (%) 22,562 (44.14%) 10,512 (46.59%) 9,691 (42.95%) 2,359 (10.46%) 0.090 108,066 (46.1%)
Age (years)1 51.2 [37.5,62.4] 51.3 [37.8, 62.4] 51.2 [37.2, 62.3] 50.7 [37.4, 62.3] 0.364 50.2 [35.5, 62.9]
Total Bilirubin (mg/dl)2 0.6 [0.4, 0.8] 0.5 [0.3, 0.7] 0.6 [0.4, 0.8] 0.9 [0.6, 1.2] 2.2E-16 0.5 [0.4, 0.7]
ALT or SGTP (U/l)2 22.0 [16.0, 33.0] 22.0 [16.0, 33.0] 22.0 [16.0, 33.0] 22.0 [16.0, 33.0] 0.407 20.0 [15.0, 27.0]
AST or SGOT (U/l)2 23.0 [19.0, 29.0] 23.0 [19.0, 29.0] 23.0 [19.0, 30.0] 23.0 [19.0, 29.0] 0.880 21.0 [18.0, 25.0]
GGT (U/l)2 30.0 [18.0, 57.0] 30.0 [18.0, 53.0] 31.0 [18.0, 65.3] 26.0 [17.0, 52.0] 0.379 20.0 [14.0, 27.0]
Alkaline phosphatase (U/l)2 74.0 [59.0, 73.0] 74.0 [59.0, 93.0] 74.0 [60.0, 93.0] 73.0 [59.0, 93.0] 0.090 70.0 [57.0, 85.0]
Years of follow-up 3 9.8 [4.6, 15.1] 9.9 [4.5, 15.2] 9.7 [4.6, 15.0] 10.0 [4.7, 15.1] 0.197 5.8 [2.9, 9.7]
Open in a new tab

Values are presented as percent or median and interquartile range.

1

Age at cohort entry (for the genotype cohort--age at the first clinical encounter after the age of 18; for the clinical cohort-- age at the first LFT after the age of 18).

2

For the genotype cohort median of LFT values represent the first LFTs measured after the age of 18 (n=51,114); for the clinical cohort values for the qualifying LFT are shown.

3

For both cohorts, follow-up represents the length of clinical data available in the EHR after cohort entry. In the genotype cohort, clinical data was not available for 1,881 individuals after cohort entry.

*

P-values represent the significance of the comparison of the three genotype groups

Association between rs6742078 and clinical disorders:

None of the 19 pre-specified outcomes were significantly associated with rs6742078 genotype (Table 2), and in the global pheWAS the only significant association was “jaundice, not of newborn” (OR=1.36 [1.24,1.49], P=1.75×10−11) (Table S3). We also observed that rs6742078 was associated with “disorders of bilirubin excretion” (OR=3.48 [2.77, 4.37], P=8.89×10−27) but only 157 cases met the case criteria for this phenotype and thus it fell below the pre-specified threshold of 200 cases required to include a phenotype in the analysis.

Table 2:

Association of rs6742078 with selected phenotypes

Phenotype phecode # cases # controls OR [95%CI] P-value
Coronary atherosclerosis 411.4 12924 42542 0.98 [0.95, 1.01] 0.198
Ischemic stroke 433.21 2099 50465 1.09 [1.02, 1.17] 0.008
Peripheral artery disease 443.9 3115 48819 1.00 [0.94, 1.06] 0.918
Metabolic syndrome 277.7 622 58216 1.09 [0.97, 1.22] 0.169
Essential hypertension 401.1 27869 27179 1.01 [0.98, 1.04] 0.491
Obesity 278.1 7777 48948 0.97 [0.93, 1.00] 0.089
Hyperlipidemia 272.13 12077 35035 1.00 [0.97, 1.04] 0.915
Type 2 diabetes 250.2 11869 45371 1.03 [1.00, 1.06] 0.064
Diabetic retinopathy 250.7 1165 54282 1.00 [0.92, 1.09] 0.987
Diabetic nephropathy 250.22 2742 45371 1.03 [0.97, 1.09] 0.384
Chronic kidney disease 585.3 4501 44219 1.00 [0.95, 1.05] 0.894
Non-alcoholic fatty liver disease 571.5 4220 48418 1.06 [1.01, 1.11] 0.015
Rheumatoid arthritis 714.1 2002 55374 0.96 [0.88, 1.03] 0.243
Multiple sclerosis 335 1423 46115 1.00 [0.93, 1.09] 0.926
Parkinson’s disease 332 742 46115 0.92 [0.82,1.03] 0.159
Alzheimer’s disease 290.11 639 47798 1.01 [0.89, 1.14] 0.933
Amyotrophic lateral sclerosis 334.21 * * * *
Optic neuritis 377.3 356 53512 1.02 [0.87, 1.19] 0.810
Schizophrenia 295.1 370 34631 1.01 [0.87, 1.18] 0.870
Open in a new tab

OR [95%CI): odds ratio [95% confidence interval].

*

less than 200 cases for amyotrophic lateral sclerosis

Adjusted for sex, year of birth, length of follow-up in the EHR from cohort entry, and first 5 principal components. P-value threshold for significance is P<0.0026 (0.05/19 phenotypes)

Similarly, in PheWeb none of the 19 pre-specified outcomes were significantly associated with rs6742078. In the PheWeb pheWAS, 6 phenotypes were significantly associated with rs6742078 (P-values≤1×10−5): disorders of bilirubin excretion (P=1.0×10−178), other disorders of metabolism (P=7.7×10−64), cholelithiasis and cholecystitis (P=6.0×10−11), cholelithiasis (P=1.4×10−8), jaundice (not of newborn) (P=1.1×10−6), and other disorders of the liver (P=1.0×10−5).

We observed a departure from Hardy Weinberg Equilibrium (HWE P-value=2.2×10−7) for rs6742078 when all the individuals included in the pheWAS analysis were analyzed (n=61,281). Since this was not a random population but one presenting to hospital, we repeated the analysis excluding those individuals with a phecode for “jaundice (not newborn)” (573.5) and/or for “disorders of bilirubin excretion” (277.4) (n=1576) and then the departure from HWE for rs6742078 was not significant (HWE P-value=0.001).

Tests and procedures in the genotype cohort:

The number of LFT tests and the number of patients who had abdominal or liver ultrasound, or testing for hemolysis, or viral hepatitis for each genotype are shown in Table 3. There was a significant difference in the proportion of individuals tested for viral hepatitis; individuals homozygous for the T allele were more likely to be tested for viral hepatitis compared to the other genotype groups (P=0.03). Individuals homozygous for the T allele were more also more likely to have an ultrasound performed compared to the other genotype groups (OR=1.04 [1.00, 1.08], P=0.03).

Table 3:

Association of rs6742078 with testing for liver disorders

Testing category GG GT TT P-value OR [95%CI]* P-value*
Liver function tests1 8 [3,18] 8 [3,18] 7 [3,18] 0.782 1.04 [0.79, 1.36] 0.800
Hemolysis tests2 3264 (13.7%) 2956 (13.6%) 774 (14.1%) 0.637 1.01 [0.97, 1.05] 0.590
Viral hepatitis tests3 8170 (34.3%) 7333 (33.7%) 1956 (35.6%) 0.029 1.01 [0.98, 1.04] 0.576
Ultrasound4 3392 (14.2%) 3133 (14.4%) 855 (15.6%) 0.04 1.04 [1.00, 1.08] 0.033
Liver Biopsy 808 (3.39%) 703 (3.2%) 196 (3.6%) 0.400 1.00 [0.93, 1.08] 0.973
Open in a new tab
1

alkaline phosphatase (AP), alanine transaminase (ALT), aspartate transaminase (AST), or gamma-glutamyl transferase (GGT);

2

haptoglobin, reticulocyte count;

3

hepatitis A, hepatitis B, hepatitis C;

4

abdominal ultrasound, liver ultrasound.

*

Adjusted by year of birth, sex, first five principal components and length of follow-up in the EHR.

See supplementary table S2 for complete list of tests and ICD and CPT codes for each category. For liver function tests the median (IQR) number of tests performed in each individual is shown; for all other tests the number of patients (percent) who had the test performed is shown.

Clinical cohort

Cohort

There were 359,667 individuals of European descent whose first LFT was recorded after the age of 18 and who met the inclusion criteria, of whom 234,670 had ≥ 1 year of follow-up in the EHR. Of these, 34,648 (14.8%) were also included in the genotype cohort. The median age at cohort entry was 50.2 [35.3, 62.9] years, median length of follow-up in the EHR was 5.80 [2.9, 9.7] years, 61% were women, and the median concentration of plasma bilirubin was 0.5 [0.4, 0.7] mg/dl (Table 1).

Pre-specified phenotypes in the measured bilirubin cohort:

Bilirubin levels at cohort entry were significantly associated with 15 of the 19 pre-specified phenotypes (adjusted for age, sex, and follow-up in the EHR) (Table 4), including coronary atherosclerosis (P=1.3×10−13), ischemic stroke (P=0.002), peripheral artery disease (P=2. 6×10−16), hypertension (P=4.05×10−8), obesity (P=2.2×10−25), hyperlipidemia (P=7.4×10−11), T2D (P=9.8×10−39), diabetic retinopathy (P=3.8×10−4), diabetic nephropathy (P=2.7×10−16), chronic kidney disease (P=8.8×10−23), NAFLD (P=1.2×10−6), rheumatoid arthritis (P=6.4×10−14), multiple sclerosis (P=2.2×10−36), Alzheimer’s disease (P=0.0018), and schizophrenia (P=4.1×10−20). However, when BMI, smoking status, and tobacco use disorder were also included as covariates, the associations for 8 of the 15 previously significant associations were no longer significant. Hypertension, hyperlipidemia, T2D, diabetic nephropathy, chronic kidney disease, rheumatoid arthritis, and multiple sclerosis remained significantly associated with bilirubin (all P< 0.0026).

Table 4:

Association of measured serum bilirubin with selected phenotypes

Adjusted by age, sex, and follow-up in EHR Adjusted by age, sex, BMI, follow-up in EHR, smoking status, and tobacco use disorder
Phenotype phecode # cases # controls OR [95%CI] P-value # cases # controls OR [95%CI] P-value
Coronary atherosclerosis 411.4 22525 189476 0.78 [0.73, 0.83] 1.27E-13 17919 135746 1.00 [0.93, 1.08] 0.947
Ischemic stroke 433.21 3571 208486 0.80 [0.69, 0.92] 0.0018 2848 151122 0.99 [0.84, 1.16] 0.858
Peripheral artery disease 443.9 6037 205647 0.62 [0.55, 0.70] 2.56E-16 4882 148621 0.86 [0.75,0.97] 0.018
Metabolic syndrome 277.7 1513 224510 0.81 [0.65, 1.01] 0.057 1345 163186 0.96 [0.75, 1.21] 0.708
Essential hypertension 401.1 50873 142810 0.87 [0.83, 0.92] 4.05E-08 40866 97251 1.18 [1.12, 1.26] 1.26E-08
Obesity 278.1 16082 202983 0.69 [0.64, 0.74] 2.18E-25 14104 145021 0.99 [0.91, 1.07] 0.781
Hyperlipidemia 272.13 36096 153056 1.19 [1.13, 1.26] 7.42E-11 31991 103431 1.25 [1.18, 1.32] 1.04E-12
Type 2 diabetes 250.2 24883 189684 0.68 [0.64, 0.72] 9.81E-39 19619 135789 0.86 [0.81, 0.93] 2.88E-05
Diabetic retinopathy 250.7 2944 211074 0.75 [0.64, 0.88] 3.76E-04 2416 152525 0.95 [0.80, 1.13] 0.552
Diabetic nephropathy 250.22 6356 189684 0.63 [0.57, 0.71] 2.74E-16 5397 135789 0.77 [0.68, 0.87] 3.49E-05
Chronic kidney disease 585.3 8198 194393 0.61 [0.56, 0.68] 8.76E-23 6528 140054 0.77 [0.69,0.86] 3.00E-06
Non-alcoholic fatty liver disease 571.5 7837 208326 0.79 [0.71, 0.87] 1.22E-06 6991 150249 0.91 [0.82, 1.01] 0.072
Rheumatoid arthritis 714.1 6272 213323 0.65 [0.58, 0.73] 6.39E-14 5035 154070 0.77 [0.68, 0.88] 5.78E-05
Multiple sclerosis 335 2700 196429 0.32 [0.26, 0.38] 2.24E-36 2334 141812 0.31 [0.25, 0.37] 9.17E-33
Parkinson’s disease 332 1726 196429 0.96 [0.78, 1.17] 0.665 1389 141812 0.86 [0.68, 1.08] 0.181
Alzheimer’s disease 290.11 1718 201761 0.71 [0.58, 0.88] 0.0018 1143 146616 0.91 [0.71, 1.18] 0.481
Amyotrophic lateral sclerosis 334.21 208 196429 1.02 [0.58, 1.80] 0.945 * * * *
Optic neuritis 377.3 736 208713 0.71 [0.51, 0.97] 0.031 614 149592 0.75 [0.53, 1.07] 0.114
Schizophrenia 295.1 1188 152598 0.29[0.23, 0.38] 4.07E-20 767 106139 0.53 [0.39, 0.73] 1.1E-04
Open in a new tab

OR[95%CI]: odds ratio [95% confidence interval]. Smoking status was categorized as never smoker and any type of smoker (former, current daily smoker, current some day smoker) and tobacco user disorder was defined as the presence of the respective phecode (318).

*

only 154 cases for amyotrophic lateral sclerosis.

P-value threshold for significance is P<0.0026 (0.05/19 phenotypes)

Testing for reverse causation:

Among the seven pre-selected disorders that remained associated with clinically measured bilirubin, we found GWAS summary statistics for five: LDL cholesterol (as proxy for hyperlipidemia), T2D, chronic kidney disease, rheumatoid arthritis, and multiple sclerosis (Table S4). In the IVWR meta-analysis, genetic predisposition for these 5 conditions were not associated with bilirubin (P>0.05, Table S4)

PheWAS in the measured bilirubin cohort:

In the pheWAS adjusted for age, sex and length of follow-up in the EHR, 431 phenotypes were significantly associated (P≤1×10−5) with measured bilirubin; “tobacco use disorder” was the top associated phenotype (P=4.8×10−139) (Table S5). When BMI, smoking status, and tobacco use disorder were added to the model, the number of phenotypes significantly associated with bilirubin decreased to 260 with epilepsy and seizure disorders now appearing as the top association (P=4.66×10−66, Table S6)

Discussion

Using EHRs, we studied the effects of bilirubin elevations using genetic and clinical information from more than 60,000 and 200,000 patients, respectively. A common genetic variant associated with large increases in bilirubin was not significantly associated with any of the pre-specified clinical phenotypes previously reported to be associated with bilirubin levels, and in a pheWAS analysis the only significant associations were with “disorders of bilirubin metabolism” and “jaundice – not of newborn”. In contrast, clinically measured bilirubin was significantly associated with most of the pre-specified clinical phenotypes and in a pheWAS analysis was significantly associated with 431 phenotypes. Thus, the principal finding of the study was that genetically elevated bilirubin levels were not associated with protection from any disease whereas higher clinically measured bilirubin levels were associated with apparent protection against several hundred conditions.

In vitro bilirubin is a potent antioxidant (3) and reduces cellular inflammatory responses to stimuli such as oxidized LDL and TNF alpha (4); in animal studies it reduced neointima formation after balloon injury (36), reduced the size of experimental myocardial infarction (5), and reduced cytokine expression and improved disease in a model of colitis (6). Moreover, bilirubin delivered by nanoparticles ameliorated lung inflammation in a murine model (7). Such experimental findings have raised interest in the therapeutic potential of manipulating bilirubin levels pharmacologically as a treatment for a range of conditions.

Evidence for the therapeutic potential of bilirubin in humans comes from the many observational and epidemiologic studies which reported that higher bilirubin levels may protect against a range of conditions including autoimmune inflammatory diseases (rheumatoid arthritis (8) and ulcerative colitis(9)), cardiovascular disease (10), stroke (37), cancer (17), T2D (11), metabolic syndrome and its components (12, 13), and many others (8). However, not all studies have been concordant (38, 39) and bilirubin levels are affected by many factors including age, sex (40), ethnicity (41), BMI (42), alcohol intake (43), diet (44), exercise (45), smoking (46), and some drugs (1). It is difficult to control for all these factors in observational studies and therefore concern has remained that the associations observed with clinically measured bilirubin levels are due to unmeasured confounding.

In contrast to the associations reported with clinically measured bilirubin levels, using a genetic approach we found that bilirubin was not significantly associated with any of the clinical phenotypes for which it was postulated to be protective. Moreover, in a global pheWAS that examined associations across a wide range of a clinical conditions the only significant findings were with diagnoses denoting elevated bilirubin levels: “disorders of bilirubin excretion” and “jaundice-not of newborn”. These findings were confirmed in a different population not derived from a tertiary hospital; in the UK Biobank, the top significant association between the instrumental variable (rs6742078) was disorders of bilirubin excretion, followed by other diagnoses directly related to elevated bilirubin levels.

In keeping with the published observational studies, in the clinical component of our study we found that 14 of 19 pre-specified phenotypes, and 431 of 1412 pheWAS phenotypes were significantly associated with measured bilirubin levels. The contrast between the findings of the clinical and genetic approaches is striking and the most likely explanation is confounding. We did not set out to perform a perfectly controlled observational epidemiologic study of conditions associated with bilirubin levels; rather, the goal of the clinical study was to illustrate that many of the previously reported associations were also present in our patient population, as well as to identify additional clinical associations. Since the methodologic adjustments required to minimize confounding would potentially differ for many phenotypes of interest, it would be difficult to perform a well-controlled epidemiologic study, considering that we identified hundreds of associated phenotypes. However, we did use broad approaches to reduce confounding. First, we only studied individuals whose liver function tests were normal at cohort entry. This would help to exclude individuals with subclinical liver disease. Second, we included only individuals of European ancestry and adjusted for age, sex, and length of follow-up in the VUMC system, and then additionally for smoking and BMI. The results of the approach were striking: the number of significant clinical associations was reduced markedly by even coarse adjustment for factors such as smoking and BMI.

Although smoking is known to decrease bilirubin levels (46), the strong confounding effects of smoking on epidemiologic studies of bilirubin may be underappreciated. Indeed, the most significant clinical associations we found were an apparent protective effect of higher bilirubin levels on tobacco use disorder and several smoking related diseases such as chronic obstructive airways disease, emphysema, pneumonia, and cancer. All these putative protective associations were likely due to the confounding effects of smoking on bilirubin levels and were not significant after adjustment. Interestingly, after adjustment for smoking, the top association was a putative protective effect of higher bilirubin levels against epilepsy with recurrent seizures, and higher bilirubin levels were also highly statistically significantly for a putative protective effect against several other epilepsy-related phenotypes. The likely reason for this apparent protective effect of measured bilirubin levels against seizures is that several commonly used anti-epileptic drugs induce hepatic enzymes and decrease bilirubin levels (47). Thus, the higher bilirubin levels in patients not taking anti-epileptic drugs (because they did not have epilepsy) compared to those taking them would lead to the erroneous conclusion that higher bilirubin levels may protected against epilepsy. This is also the likely explanation for the apparent protective effect of bilirubin against bipolar disorder, a condition for which antiepileptic drugs are frequently used as mood stabilizers.

It is also possible that genetic predisposition for certain disorders might influence bilirubin levels (reverse causation). We found no relationship between the genetic drivers of bilirubin and a genetic predisposition to five of the preselected phenotypes that remained significantly associated with measured bilirubin after adjustment for age, sex, BMI, follow-up length, smoking, and tobacco related disorders. This is concordant with the idea that measured bilirubin levels are affected by many factors and the observed association with a particular disease may be related to factors or than that disease itself or to a subtype of the disease; for example, the association between type-2 diabetes and measured bilirubin levels could be driven by patients with acute illness or those with hepatic steatosis.

Besides confounding and reverse causation, another potential reason for the differences in the findings of the genetic and clinical approaches that should be considered is a weak genetic instrumental variable. If the relationship between an instrumental variable and a risk factor is weak then true associations might not be detected. Weak instruments can be detected by an F-statistic less than 10 (48). The instrument variable we used is strong; it explained 19.5% of variability in bilirubin concentrations and had an F-statistic >100 (16, 24). The strong genetic effect was clearly visible in the actual bilirubin concentrations observed in the different genotypes: compared to individuals without a variant allele, the median bilirubin concentrations were 20% higher in heterozygotes and 80% higher in homozygotes. Also, the same instrumental variable has been used in other MR studies. For example, using published genetic data from large consortia, Kunutsor and collaborators did not find evidence of an association between rs6742078 and several cardiovascular outcomes including triglycerides, BMI, waist-hip ratio, blood pressure and coronary disease (49). Similarly, MR studies did not suggest a causal role for bilirubin in stroke (24), hypertension (50), or NAFLD (16); a protective effect for T2D was reported in one study (22) but not in another larger study (51).

The instrumental variable we used is a marker for Gilbert’s syndrome, a common condition also known as benign hyperbilirubinemia. Individuals with Gilbert’s syndrome may have episodes of jaundice caused by fasting, exercise, dehydration, and concomitant illness.(1) These elevated bilirubin levels, although benign, could trigger unnecessary diagnostic tests and if this were the case it would argue for preemptive genotyping to reduce such testing. We found that abdominal ultrasound was performed more often in patients with the genotype associated with higher bilirubin (TT) but the magnitude of the increase was clinically insignificant.

Our study had several limitations. We only studied individuals of European ancestry because the instrumental variable for bilirubin is best characterized in this population; however, despite this restriction we had a large sample size in both the clinical and genetic studies. We are underpowered for some phenotypes with small number of cases (e.g., amyotrophic lateral sclerosis). We did not examine every phenotype previously reported to be associated with bilirubin; some phenotypes, such as response to cancer treatment, are not easily ascertained in the EHR. Also, although the genetic approach does not provide supporting evidence, we cannot conclude that every clinical association noted is spurious--to do this would require a dedicated epidemiologic study for each phenotype of interest and is not feasible because many of the variables that can alter bilirubin levels such as diet, alcohol, smoking and exercise (1) are difficult to quantify accurately and vary over time.

We conclude that higher clinically measured bilirubin levels are associated with a reduced incidence of many illnesses, but that this is largely due to confounding since a genetic approach found no evidence for a protective effect of bilirubin against any condition. Therapeutic strategies to increase bilirubin levels for the treatment or prevention of disease appear unlikely to be successful.

Supplementary Material

tS1-6

Study Highlights.

  • What is the current knowledge?

    Observational studies suggest that bilirubin levels may protect against several diseases. However, it is unclear if these findings reflect causality or rather are the result of confounding or reverse causality.

  • What question did this study address?

    Using clinical and genetic approaches, we explored the association between bilirubin levels and clinical conditions using a hospital-based biobank.

  • What does this study add to our knowledge?

    Our findings suggest that most of the putative beneficial clinical associations previously reported for bilirubin are likely due to confounding.

  • How might this change clinical pharmacology or translational science?

    Interventional strategies, such as randomized clinical trials are difficult and expensive to implement. To improve the success in drug development, we can use genomics to prioritize target interventions by selecting disease mechanisms with strong genetic support or with evidence of causal inference.

Funding:

The dataset(s) used for the analyses described were obtained from Vanderbilt University Medical Center’s BioVU which is supported by numerous sources: institutional funding, private agencies, and federal grants. These include the NIH funded Shared Instrumentation Grant S10RR025141; and CTSA grants UL1TR002243, UL1TR000445, and UL1RR024975. Genomic data are also supported by investigator-led projects that include U01HG004798, R01NS032830, RC2GM092618, P50GM115305, U01HG006378, U19HL065962, R01HD074711; and additional funding sources listed at https://victr.vanderbilt.edu/pub/biovu/. VKK was supported by NIH/NIGMS K23GM117395, CMS was supported by R35GM131770 (NIGMS), and JDM was supported by NIH/NIGMS R01 GM130791 (NIGMS).

Footnotes

Conflict of Interest: All authors declared no competing interests for this work

References

  • (1).Vitek L, Bellarosa C & Tiribelli C Induction of Mild Hyperbilirubinemia: Hype or Real Therapeutic Opportunity? Clin Pharmacol Ther 106, 568–75 (2019). [DOI] [PubMed] [Google Scholar]
  • (2).Hench PS Effect of Jaundice on Rheumatoid Arthritis. Br Med J 2, 394–8 (1938). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • (3).Stocker R, Yamamoto Y, McDonagh AF, Glazer AN & Ames BN Bilirubin is an antioxidant of possible physiological importance. Science 235, 1043–6 (1987). [DOI] [PubMed] [Google Scholar]
  • (4).Kawamura K et al. Bilirubin from heme oxygenase-1 attenuates vascular endothelial activation and dysfunction. Arterioscler Thromb Vasc Biol 25, 155–60 (2005). [DOI] [PubMed] [Google Scholar]
  • (5).Ben-Amotz R, Bonagura J, Velayutham M, Hamlin R, Burns P & Adin C Intraperitoneal bilirubin administration decreases infarct area in a rat coronary ischemia/reperfusion model. Front Physiol 5, 53 (2014). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • (6).Zheng JD et al. Unconjugated bilirubin alleviates experimental ulcerative colitis by regulating intestinal barrier function and immune inflammation. World J Gastroenterol 25, 1865–78 (2019). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • (7).Kim DE, Lee Y, Kim M, Lee S, Jon S & Lee SH Bilirubin nanoparticles ameliorate allergic lung inflammation in a mouse model of asthma. Biomaterials 140, 37–44 (2017). [DOI] [PubMed] [Google Scholar]
  • (8).Fischman D, Valluri A, Gorrepati VS, Murphy ME, Peters I & Cheriyath P Bilirubin as a Protective Factor for Rheumatoid Arthritis: An NHANES Study of 2003 – 2006 Data. J Clin Med Res 2, 256–60 (2010). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • (9).Schieffer KM, Bruffy SM, Rauscher R, Koltun WA, Yochum GS & Gallagher CJ Reduced total serum bilirubin levels are associated with ulcerative colitis. PLoS One 12, e0179267 (2017). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • (10).Kunutsor SK, Bakker SJ, Gansevoort RT, Chowdhury R & Dullaart RP Circulating total bilirubin and risk of incident cardiovascular disease in the general population. Arterioscler Thromb Vasc Biol 35, 716–24 (2015). [DOI] [PubMed] [Google Scholar]
  • (11).Cheriyath P et al. High Total Bilirubin as a Protective Factor for Diabetes Mellitus: An Analysis of NHANES Data From 1999 – 2006. J Clin Med Res 2, 201–6 (2010). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • (12).Wang L & Bautista LE Serum bilirubin and the risk of hypertension. Int J Epidemiol 44, 142–52 (2015). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • (13).Lin LY et al. Serum bilirubin is inversely associated with insulin resistance and metabolic syndrome among children and adolescents. Atherosclerosis 203, 563–8 (2009). [DOI] [PubMed] [Google Scholar]
  • (14).Seyed Khoei N et al. Mild hyperbilirubinaemia as an endogenous mitigator of overweight and obesity: Implications for improved metabolic health. Atherosclerosis 269, 306–11 (2018). [DOI] [PubMed] [Google Scholar]
  • (15).Ryu S et al. Higher serum direct bilirubin levels were associated with a lower risk of incident chronic kidney disease in middle aged Korean men. PLoS One 9, e75178 (2014). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • (16).Kunutsor SK, Frysz M, Verweij N, Kieneker LM, Bakker SJL & Dullaart RPF Circulating total bilirubin and risk of non-alcoholic fatty liver disease in the PREVEND study: observational findings and a Mendelian randomization study. Eur J Epidemiol 35, 123–37 (2020). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • (17).Horsfall LJ et al. Serum bilirubin and risk of respiratory disease and death. JAMA 305, 691–7 (2011). [DOI] [PubMed] [Google Scholar]
  • (18).Gazzin S, Vitek L, Watchko J, Shapiro SM & Tiribelli C A Novel Perspective on the Biology of Bilirubin in Health and Disease. Trends Mol Med 22, 758–68 (2016). [DOI] [PubMed] [Google Scholar]
  • (19).Fujiwara R, Haag M, Schaeffeler E, Nies AT, Zanger UM & Schwab M Systemic regulation of bilirubin homeostasis: Potential benefits of hyperbilirubinemia. Hepatology 67, 1609–19 (2018). [DOI] [PubMed] [Google Scholar]
  • (20).Smith GD & Ebrahim S ‘Mendelian randomization’: can genetic epidemiology contribute to understanding environmental determinants of disease? Int J Epidemiol 32, 1–22 (2003). [DOI] [PubMed] [Google Scholar]
  • (21).Voight BF et al. Plasma HDL cholesterol and risk of myocardial infarction: a mendelian randomisation study. Lancet 380, 572–80 (2012). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • (22).Abbasi A et al. Bilirubin as a potential causal factor in type 2 diabetes risk: a Mendelian randomization study. Diabetes 64, 1459–69 (2015). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • (23).Johnson AD et al. Genome-wide association meta-analysis for total serum bilirubin levels. Hum Mol Genet 18, 2700–10 (2009). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • (24).Lee SJ, Jee YH, Jung KJ, Hong S, Shin ES & Jee SH Bilirubin and Stroke Risk Using a Mendelian Randomization Design. Stroke 48, 1154–60 (2017). [DOI] [PubMed] [Google Scholar]
  • (25).Stender S, Frikke-Schmidt R, Nordestgaard BG, Grande P & Tybjaerg-Hansen A Genetically elevated bilirubin and risk of ischaemic heart disease: three Mendelian randomization studies and a meta-analysis. J Intern Med 273, 59–68 (2013). [DOI] [PubMed] [Google Scholar]
  • (26).Roden DM et al. Development of a large-scale de-identified DNA biobank to enable personalized medicine. Clin Pharmacol Ther 84, 362–9 (2008). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • (27).Purcell S et al. PLINK: a tool set for whole-genome association and population-based linkage analyses. Am J Hum Genet 81, 559–75 (2007). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • (28).Reed E, Nunez S, Kulp D, Qian J, Reilly MP & Foulkes AS A guide to genome-wide association analysis and post-analytic interrogation. Stat Med 34, 3769–92 (2015). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • (29).Zheng X, Levine D, Shen J, Gogarten SM, Laurie C & Weir BS A high-performance computing toolset for relatedness and principal component analysis of SNP data. Bioinformatics 28, 3326–8 (2012). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • (30).Wei WQ et al. Evaluating phecodes, clinical classification software, and ICD-9-CM codes for phenome-wide association studies in the electronic health record. PLoS One 12, e0175508 (2017). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • (31).Wu P et al. Mapping ICD-10 and ICD-10-CM Codes to Phecodes: Workflow Development and Initial Evaluation. JMIR Med Inform 7, e14325 (2019). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • (32).Carroll RJ, Bastarache L & Denny JC R PheWAS: data analysis and plotting tools for phenome-wide association studies in the R environment. Bioinformatics 30, 2375–6 (2014). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • (33).Verma A et al. Human-Disease Phenotype Map Derived from PheWAS across 38,682 Individuals. Am J Hum Genet 104, 55–64 (2019). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • (34).Gagliano Taliun SA et al. Exploring and visualizing large-scale genetic associations by using PheWeb. Nat Genet 52, 550–2 (2020). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • (35).Cheng FW et al. Body mass index and all-cause mortality among older adults. Obesity (Silver Spring) 24, 2232–9 (2016). [DOI] [PubMed] [Google Scholar]
  • (36).Ollinger R et al. Bilirubin: a natural inhibitor of vascular smooth muscle cell proliferation. Circulation 112, 1030–9 (2005). [DOI] [PubMed] [Google Scholar]
  • (37).Perlstein TS, Pande RL, Creager MA, Weuve J & Beckman JA Serum total bilirubin level, prevalent stroke, and stroke outcomes: NHANES 1999–2004. Am J Med 121, 781–8 e1 (2008). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • (38).Troughton JA et al. Bilirubin and coronary heart disease risk in the Prospective Epidemiological Study of Myocardial Infarction (PRIME). Eur J Cardiovasc Prev Rehabil 14, 79–84 (2007). [DOI] [PubMed] [Google Scholar]
  • (39).Pineda S et al. Association of serum bilirubin with ischemic stroke outcomes. J Stroke Cerebrovasc Dis 17, 147–52 (2008). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • (40).Rosenthal P, Pincus M & Fink D Sex- and age-related differences in bilirubin concentrations in serum. Clin Chem 30, 1380–2 (1984). [PubMed] [Google Scholar]
  • (41).Zucker SD, Horn PS & Sherman KE Serum bilirubin levels in the U.S. population: gender effect and inverse correlation with colorectal cancer. Hepatology 40, 827–35 (2004). [DOI] [PubMed] [Google Scholar]
  • (42).Jenko-Praznikar Z, Petelin A, Jurdana M & Ziberna L Serum bilirubin levels are lower in overweight asymptomatic middle-aged adults: an early indicator of metabolic syndrome? Metabolism 62, 976–85 (2013). [DOI] [PubMed] [Google Scholar]
  • (43).Chan-Yeung M, Ferreira P, Frohlich J, Schulzer M & Tan F The effects of age, smoking, and alcohol on routine laboratory tests. Am J Clin Pathol 75, 320–6 (1981). [DOI] [PubMed] [Google Scholar]
  • (44).Loprinzi PD & Mahoney SE Association between flavonoid-rich fruit and vegetable consumption and total serum bilirubin. Angiology 66, 286–90 (2015). [DOI] [PubMed] [Google Scholar]
  • (45).Swift DL, Johannsen NM, Earnest CP, Blair SN & Church TS Effect of different doses of aerobic exercise training on total bilirubin levels. Med Sci Sports Exerc 44, 569–74 (2012). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • (46).Van Hoydonck PG, Temme EH & Schouten EG Serum bilirubin concentration in a Belgian population: the association with smoking status and type of cigarettes. Int J Epidemiol 30, 1465–72 (2001). [DOI] [PubMed] [Google Scholar]
  • (47).Gough H, Goggin T, Crowley M & Callaghan N Serum bilirubin levels with antiepileptic drugs. Epilepsia 30, 597–602 (1989). [DOI] [PubMed] [Google Scholar]
  • (48).Davies NM, Holmes MV & Davey Smith G Reading Mendelian randomisation studies: a guide, glossary, and checklist for clinicians. BMJ 362, k601 (2018). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • (49).Kunutsor SK Serum total bilirubin levels and coronary heart disease--Causal association or epiphenomenon? Exp Gerontol 72, 63–6 (2015). [DOI] [PubMed] [Google Scholar]
  • (50).Kunutsor SK, Kieneker LM, Burgess S, Bakker SJL & Dullaart RPF Circulating Total Bilirubin and Future Risk of Hypertension in the General Population: The Prevention of Renal and Vascular End-Stage Disease (PREVEND) Prospective Study and a Mendelian Randomization Approach. J Am Heart Assoc 6, (2017). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • (51).Yuan S & Larsson SC An atlas on risk factors for type 2 diabetes: a wide-angled Mendelian randomisation study. Diabetologia 63, 2359–71 (2020). [DOI] [PMC free article] [PubMed] [Google Scholar]

Associated Data

This section collects any data citations, data availability statements, or supplementary materials included in this article.

Supplementary Materials

tS1-6
NIHMS1746934-supplement-tS1-6.xlsx (511.6KB, xlsx)

RESOURCES