Abstract
There is little information on the differential diagnosis and prognosis of hospitalized patients with hyperbilirubinemia. Here, we hypothesized that hyperbilirubinemia in hospitalized patients is associated with specific diseases and outcomes. This retrospective cohort analysis included patients admitted to the Medical University of South Carolina with a total bilirubin >3 mg/dL from January 9, 2015 to August 25, 2017. Collected clinical data included demographics, primary diagnosis, Charlson Comorbidity Index (CCI), laboratory data, and clinical outcomes. We separated and analyzed the cohort into seven primary diagnostic groups. We identified 1693 patients with a bilirubin level >3 mg/dL. The cohort was 42% female, had an average age of 54, average CCI of 4.8, and average length of stay of 13 days. The causes of hyperbilirubinemia included the following: primary liver disease (868/1693; 51%) with cirrhosis being most common (385/1693; 23%), benign biliary obstruction (252/1693; 15%), hemolytic anemia (149/1693; 9%), malignant biliary obstruction (121/1693; 7%), unknown etiology (108/1693; 6%), primary liver cancer (74/1693; 4%), and metastatic cancer to the liver (57/1693; 3%). Overall, the mortality/discharge to hospice rate in patients with a bilirubin >3 mg/dL was 30%, and was proportional to the severity of hyperbilirubinemia, including when controlling for the underlying severity of illness. Mortality was highest in patients with primary liver disease and malignancy and was lowest in patients with non-cancerous obstruction or hemolytic jaundice. Hyperbilirubinemia in hospitalized patients is most often due to primary liver disease, and identifies patients with a poor prognosis, particularly when caused by primary liver disease or cancer.
High serum bilirubin, or hyperbilirubinemia, has multiple etiologies, but it is most often caused by hepatic or biliary disease.
Causes, diagnoses, and evaluation of hyperbilirubinemia have been poorly studied in large populations.
With high levels, free unconjugated bilirubin is a known neurotoxin that can cause kernicterus in infants. However, the effects of hyperbilirubinemia is unknown in adults.
In hospitalized patients, hyperbilirubinemia is most often caused by parenchymal liver disease
Increasing levels of bilirubin, regardless of cause, are associated with increasing in-hospital mortality/discharge to hospice
Hyperbilirubinemia caused by hemolytic disease or benign obstruction are associated with better outcomes than liver disease or malignant obstruction
This study emphasizes an important clinical concept—specifically, that hyperbilirubinemia from all causes identifies patients with a poorer prognosis, particularly if caused by primary liver disease or cancer.
Introduction
Bilirubin is an end product of hemoglobin breakdown and under normal circumstances, is rapidly removed from the circulation by hepatocytes. If hepatocytes are unable to metabolize or excrete bilirubin, levels increase in the blood, leading to jaundice, a clinical condition defined by a yellowish staining of the skin, sclera, and mucous membranes. In healthy individuals, the total serum bilirubin is less than 1 mg/dL and jaundice is typically detected when serum bilirubin rises above 3 mg/dL. 1
High serum bilirubin, or hyperbilirubinemia, has multiple etiologies, but it is most often caused by hepatic or biliary disease. Although less common in adults, hyperbilirubinemia can be caused by hematologic and inherited disorders. 2 Previous studies of hyperbilirubinemia have focused on the value of serum bilirubin in assessing outcome in patients with specific hepatic diseases, most notably in acute liver failure (ALF) and in patients with cirrhosis. For example, the Model for End Stage Liver Disease (MELD) was developed and identified serum bilirubin as a key component of the MELD formula for assessing prognosis in patient with cirrhosis undergoing transjugular intrahepatic portosystemic shunt (TIPS). 3 In ALF, it has been shown that total bilirubin is predictive of outcome 4 and may be a valuable adjunct in the triage of liver transplantation. 5 Further, in idiosyncratic liver injury induced by a drug, “Hy’s Law”—the presence of bilirubin 2× the upper limit of normal, has been used as a predictor of outcome.6–8 This observation has been advocated by the U.S. Food and Drug Administration for assessing the hepatotoxicity of newly developed drugs. 9 In alcoholic hepatitis, patients admitted with a bilirubin >5 had a poorer prognosis and higher rate of mortality. 10 Further, serum bilirubin >3.0 appeared to be a useful predictor of long-term mortality (survival rate after 1 year was 28% for those with total bilirubin >3.0 mg/dL and 68% for those with <3.0 mg/dL (p < 0.0001) 11 ). In cholestatic disease, total serum bilirubin may also have prognostic value; in patients with primary biliary cholangitis and primary sclerosing cholangitis (PSC), elevated serum bilirubin has been identified as an important prognostic indicator.12–15 Finally, elevated serum bilirubin may also be of prognostic importance in patients with neoplastic biliary tract obstruction. 16
Causes, diagnoses, and evaluation of hyperbilirubinemia have been poorly studied in large populations, with little information on prognosis and outcomes of patients who have hyperbilirubinemia. In this study, we hypothesized that hyperbilirubinemia in patients admitted to the hospital is caused by specific diseases and further, that hyperbilirubinemia may be associated with poorer outcomes. Therefore, we aimed to examine not only the causes of hyperbilirubinemia, but also the association of total serum bilirubin and mortality, including as a function of the underlying etiology of hyperbilirubinemia.
Materials and methods
Study population
Patients were identified via the Medical University of South Carolina’s Clinical Data Warehouse, which searches the institution’s electronic medical record (EMR). We obtained all adult inpatient encounters with a maximum total bilirubin >3 g/dL between September 1, 2015 and August 31, 2017. We excluded patients less than 18 years of age and those with incomplete medical records.
The following clinical data were abstracted from the EMR: demographic information including age, race, and sex, total bilirubin, additional data for each admission were collected including direct bilirubin, max AST, max ALT, primary diagnosis, growth on blood cultures, transfusion of blood products, disposition upon discharge, and the Charlson Comorbidity Index (CCI). Outcomes included mortality/discharge to Hospice, total length of admission (in days), and admission to the ICU.
The patient’s primary diagnosis was categorized into underlying diagnostic groups as follows: primary liver disease, primary liver cancer (confirmed with standard LI-RADs imaging criteria for hepatocellular cancer 17 and/or histological analysis), metastatic liver disease (requiring histological confirmation), malignant biliary obstruction (requiring histological confirmation), benign biliary obstruction (based on imaging findings, and excluding malignancy), hemolytic anemia/known metabolic disease, and unknown. The primary liver disease group was also separated into subgroups including cirrhosis of any etiology (based on histological, and/or clinical grounds), 18 acute or chronic viral hepatitis (based on serologic criteria, without clinical or histological evidence of advanced fibrosis or complications associated with cirrhosis), autoimmune hepatitis (as defined by the international autoimmune hepatitis group 19 ), vascular involvement of the liver (including ischemic hepatitis as previously defined 20 ), ALF (as previously defined—acute liver injury with encephalopathy, INR of ≥1.5, and no cirrhosis or preexisting liver disease 21 ), alcoholic hepatitis—also alcohol-related liver disease, 22 hereditary or metabolic liver disease (the latter including nonalcoholic fatty liver disease (NAFLD) and nonalcoholic steatohepatitis and cirrhosis), and miscellaneous liver diseases.
Bilirubin levels were binned (categorized) into groups that spanned 5 mg/dL intervals, beginning with ≥3.0 to <8 mg/dL up to >33 mg/dL.
Exclusion criteria
For patients in the dataset with multiple encounters, we included only their latest encounter and excluded all other previous encounters. Encounters with incorrect or missing data were excluded.
Data analysis
Demographic and clinical data were extracted and analyzed. Descriptive statistics including means, medians, frequencies, and percentages were used to summarize the data. A chi-square test (or Fisher’s exact test in the case of small sample) was used to compare groups of continuous and categorical variables, respectively. Due to non-normality of the data, non-parametric Spearman rank correlations were used to investigate the associations between bilirubin levels, CCI, and hospital length of stay, and Wilcoxon rank sum tests were used to compare bilirubin levels between patients who lived and those who died. Multiple logistic regression was used to assess the association between mortality and bilirubin levels while adjusting for CCI. All p-values reported are two-sided; a level of 0.05 was considered statistically significant. All data were collected and analyzed within Microsoft Excel (Microsoft Corporation, Redmond, WA, USA) and SAS v9.4 (SAS Institute, Cary, NC, USA).
Results
During the study period, we identified n = 2407 patients with a bilirubin level greater than 3 mg/dL; of this group, n = 714 were excluded, leaving a total of n = 1693 patient for analysis (Figure 1). Women represented 42% of the cohort, and the average age was 54 years. Caucasians made up 60% of the cohort, 36% were African American, and 4% were of other ethnicities (Table 1). The mean of maximum total bilirubin was 9.6 g/dL, direct bilirubin of 5.9 g/dL, AST of 873 U/L, ALT of 481 U/L, and alkaline phosphatase of 245 IU/L.
Patient cohort. A Consolidated Standards of Reporting Trials (CONSORT) participant flow diagram of the cohort is shown.
Patient cohort demographics (n = 1693).
The most common underlying cause of hyperbilirubinemia in the cohort was primary liver disease (Table 2), found in (868/1693; 51%) of patients. Among the primary liver diseases, cirrhosis was most common (422/1693; 25%). The major causes of cirrhosis included hepatitis C virus (n = 42/422; 10%), alcohol related liver disease (n = 203/422; 48%), and nonalcoholic steatohepatitis (NASH) (n=106/422; 25%). However, other causes of liver disease were prominent, including hepatitis and vascular disorders of the liver. “Benign” causes of hyperbilirubinemia, including hemolysis and non-malignant obstruction made up nearly 25% of causes of hyperbilirubinemia. Hemolysis rarely led to marked elevations in bilirubin levels; in the two patients (of 149) with very high elevations—both had sickle cell anemia with other concomitant conditions (such as G6PD deficiency and drug-induced liver injury). In approximately 7% (108/1693) of patients, the etiology of hyperbilirubinemia was unknown, and the elevation in bilirubin was typically modest, uncommonly above 8 mg/dL. Primary liver cancer, sepsis, ALF, and miscellaneous liver diseases were uncommon causes of hyperbilirubinemia.
Diagnostic frequencies of causes of hyperbilirubinemia.
Other includes the following: Acute fatty liver of pregnancy, HELLP syndrome, Liver laceration, Cystic fibrosis-related liver disease, Ruptured hepatic adenoma, Alagille syndrome, Hepatic artery stenosis, Granulomatous hepatitis, Caroli disease.
We also examined the degree of hyperbilirubinemia as a function of diagnosis. Several notable patterns emerged. First, there was a significant (p < 0.0001) association between bilirubin levels and underlying diagnosis (Table 2). Of patients with profound hyperbilirubinemia (especially bilirubin levels greater than 28 mg/dL), cirrhosis was by far the most common cause. In fact, less than 10% of the cohort with a bilirubin level greater than 28 mg/dL did not have primary parenchymal liver disease. In patients who did not have underlying liver disease and a strikingly elevated bilirubin, combinations of different contributing conditions were often present—such as sepsis with cholangitis and/or hemolysis. In patients with hemolysis alone, it was uncommon for bilirubin levels to be elevated to greater than 8 mg/dL; in fact, 135 of 151 (89%) patents with hemolysis had a bilirubin ≤7.9 mg/dL, and only four patients had a bilirubin greater than 12.9 mg/dL—an overwhelming majority of these patients had sickle cell anemia, with a smaller number of patients having hematologic malignancies such as leukemia, lymphoma, or multiple myeloma. Five patients had hemolytic anemia, and one had G6PD deficiency. Patients with non-malignant obstruction also had comparatively lower elevations in bilirubin; 219 of 252 (87%) had a bilirubin less than 12.9 mg/dL. In this group, there was a relatively even distribution of diagnoses—including pancreatitis, cholecystitis, choledocholithiasis, cholangitis, and bile duct injury/strictures. Some more uncommon causes of hyperbilirubinemia in this group included vanishing bile duct syndrome or AIDS cholangiopathy. We also noted that sepsis and metastatic disease to the liver were characterized by lower elevations in bilirubin, typically with bilirubin levels less than 8 mg/dL. The most frequent primary source of cancer for metastatic disease was lung cancer, but other primary cancers such as breast and colon cancer were also present.
The average CCI was 4.8, with an average length of stay of 13 days, and 52% of patients were admitted to the ICU (Table 3). The overall mortality/discharge to hospice rate was 30%. The rate of death/discharge to hospice increased in proportion to increasing bilirubin levels (p < 0.0001). The mortality rate increased progressively—from 23% among patients with bilirubin levels ranging from 3.0 to 7.9 mg/dL to 61% among patients with bilirubin levels >33 mg/dL. There was a strong correlation (rho = 0.81, p < 0.0001) between bilirubin levels and mortality (Figure 2); this finding remained highly statistically significant (p < 0.0001) even when controlling for CCI in the multiple logistic regression model. When comparing mortality rates for all those who had liver disease vs those who did not, the mortality rates were substantially higher for those with primary liver disease (35% vs 26%, respectively,p < 0.0001). Notably, there were no significant associations between the CCI and total bilirubin (Spearman correlation = 0.01, p = 0.83) (Table 3). There was a statistically significant positive association between total bilirubin and hospital length of stay; however, the magnitude of this association was not strong (Spearman correlation = 0.05, p = 0.02).
Outcomes by groups of increasing levels of bilirubin.
p < 0.0001 by chi-square test comparing outcomes across total bilirubin categories.
p < 0.05 when testing the null hypothesis that the Spearman rank correlation is 0 between the outcome and total bilirubin.
In-hospital mortality rate by total bilirubin.
Mortality/hospice rates differed significantly (p < 0.001) as a function of underlying diagnosis (Table 4). The mortality rate was highest in patients with ALF (63%), and lowest in those with hemolysis (11%). The mortality was also remarkably high in patients with metastatic cancer to the liver (60%), sepsis (56%), vascular involvement of the liver (45%), primary liver cancer (39%), and cirrhosis (34%).
Outcomes based on diagnosis.
Other includes the following: Acute fatty liver of pregnancy, hemolysis, elevated liver enzymes, and low platelets (HELLP) syndrome, liver trauma, cystic fibrosis related liver disease, hepatic adenoma, Alagille syndrome, granulomatous hepatitis, Caroli disease.
Differences in bilirubin levels between patients who survived and those who did not were remarkable (Figure 3), represented by large effect sizes (δ), and are shown separately for each of the top 5 most prevalent underlying diagnoses. Effect sizes were largest (and highly statistically significantly different (p < 0.0001)) for cirrhosis (δ = 0.67), obstruction (benign) (δ = 0.85), and hepatitis (δ = 0.83). Effect sizes for vascular disease (δ = 0.41, p = 0.06) and hemolysis (δ = 0.32, p = 0.02) were slightly smaller, perhaps indicative of the fact that mortality in these groups was due to non-liver related conditions.
Bilirubin levels and mortality for top 5 conditions. Individual dots represent individual patients, and horizontal black bars reflect group-specific medians. Effect sizes (δ) are the mean bilirubin differences between the patients who lived and those who died, divided by the pooled standard deviation; theirp-values were obtained by two-sample t-tests.
Discussion
In this study, we have demonstrated that hyperbilirubinemia in hospitalized patients can be caused by a variety of disorders, but is most often caused by parenchymal liver disease. Further, we found that pronounced hyperbilirubinemia was most characteristic of primary liver disease, while lower degrees of hyperbilirubinemia were associated with hemolytic disorders and benign biliary obstruction. Finally, hyperbilirubinemia was independently associated with increased mortality. The mortality rate was over 30% for all diagnoses combined, and interestingly, was greater than 10% even in patients with benign disorders such as hemolysis and benign biliary obstruction.
Our findings that increasing bilirubin levels are associated with increasing mortality is consistent with findings from other studies examining various liver diseases such as end stage liver disease, 3 ALF, 5 alcoholic hepatitis,10,11 as well as obstructive disease such as PSC, 12 and neoplastic obstruction. 16 In an older study, it was found that cirrhotic patients with a bilirubin >3 mg/dL had a 62% one-year mortality rate compared to those with bilirubin <3 mg/dL, who had a 25% mortality rate. 11 Our study showed that cirrhotic patients with a bilirubin >3 mg/dL had a more modest increase in mortality rate of 34%. However, here, we studied in-hospital mortality as opposed to 1-year follow up mortality, and it is possible that 1-year mortality may have also been similarly elevated in our patients.
Although we were not surprised by the high mortality rate associated with marked elevation in bilirubin (patients with a bilirubin >28 mg/dL had a 55% mortality rate, and those with a bilirubin >33 mg/dL had an even higher mortality rate of 61%), we were surprised that patients with a bilirubin level between 3 and 7 mg/dL had a remarkably high mortality rate (of 23.3%), particularly since nearly two-thirds of the cohort (1065/1693) fell into this category. The explanation for this remarkably high mortality rate is unclear but did not appear to be due to severe underlying disease in these patients given that the CCI was similar throughout all groups.
Outcome analysis of the different diagnostic groups showed that, overall, neoplastic causes of hyperbilirubinemia had an extremely high mortality rate at 39%. Metastatic liver disease was the highest at 59%, primary liver cancer at 39%, and obstructive cancer 29%. Primary liver disease was also associated with a high mortality (34% overall). As might be expected, ALF was associated with the highest mortality at 63%, and vascular involvement had a 45% mortality. Also, as might be predicted, the best outcomes were associated with hemolysis and non-malignant biliary obstruction (12% and 11% mortality rates respectively).
One of the most notable findings of this study was that the higher the level of bilirubin elevation, the poorer the outcome. This may be related to the fact that the highest bilirubin elevations were found in patients with underlying liver disease. Notwithstanding, an important issue raised by our study is whether hyperbilirubinemia itself affects patient outcomes, or whether an elevated bilirubin level is simply a reflection of underlying disease. With high levels, free unconjugated bilirubin is a known neurotoxin that can cause kernicterus. In kernicterus, bilirubin deposits in the brain’s grey matter, particularly the basal ganglia and hippocampus, causing long-term neurological damage. However, this is mainly seen in infants, as their blood-brain barriers are underdeveloped and levels of albumin are low—which can lead to higher levels of free unconjugated bilirubin crossing into the brain. 23 Rather, it appears that it is most likely that higher levels of bilirubin indicate more severe underlying disease. For example, in a study that examined gallbladder cancer, patients who presented with jaundice were not only more likely to have more advanced stage disease (Stage III and IV), but also significantly shorter survival (6 months vs 16 months (p = 0.0001)). 24
NAFLD is a common cause of liver disease with an increasing prevalence worldwide. While steatosis and NASH alone may cause elevated liver transaminases (ALT and AST), it is uncommon for these to cause hyperbilirubinemia in the absence of cirrhosis. In fact, there is some evidence in the literature that serum bilirubin levels are inversely associated with simple NAFLD. 25 In our cohort, NASH was responsible for cirrhosis in a considerable number of patients (106/422), consistent with the increasing prevalence of NASH. There were no patients with NAFLD alone who had elevated bilirubin >3 g/dL. Thus, we conclude that patients with a history of NAFLD presenting with high hyperbilirubinemia should prompt clinicians to search for NASH cirrhosis (or alternative or coexisting disorders as highlighted here).
We recognize limitations of this study. First, the retrospective nature of the study may have led to ascertainment bias, though we utilized rigorous definitions of diagnostic disease categories. Improper documentation could have potentially contributed to unrecognized bias, though this would not be expected to affect our findings or conclusions. Another concern was that we were unable to identify a cause of hyperbilirubinemia in a small number of patients. We suspect that many of these cases were likely due to unrecognized passive congestion (i.e., cardiac disease) or Gilbert’s syndrome. The latter is consistent with the generally lower bilirubin levels found in this group (Table 2). It is notable that since the latter disease is associated with no clear adverse outcome itself, this could have biased the study toward better outcomes than if these patients had been excluded. However, because the number of patients is small given our very large sample size, we do not believe that including them in the outcome analyses had an effect on our conclusions. Finally, our cohort consisted of 36% Black and 60% White patients, but only 4% patients of other ethnicities. Thus, our results may not be generalizable to people in other ethnic groups such as Asian and Hispanic; and the data may not be representative of patients living in a different region of the country or world.
In summary, we have shown that elevated levels of total bilirubin, regardless of cause, are associated with increased mortality. Prior studies have examined outcomes related to hyperbilirubinemia in specific disease processes (typically liver disease). The novelty of this study includes the fact that it is the first study to examine all causes of jaundice in a large cohort of hospitalized patients and correlate the related outcomes. Our study emphasizes an important clinical concept—specifically, that hyperbilirubinemia from all causes identifies patients with a poorer prognosis, particularly if caused by primary liver disease or cancer.
Footnotes
Acknowledgements
The authors would like to thank Jean Craig for her assistance in helping curate the dataset for this study.
Authors contributions
Joshua Y Kwon—study concept and design; acquisition of data; analysis and interpretation of data; drafting of the manuscript; critical revision of the manuscript for important intellectual content; statistical analysis.
Paul Nietert—analysis and interpretation of data; drafting of the manuscript; critical revision of the manuscript for important intellectual content; statistical analysis.
Don Rockey—study concept and design; acquisition of data; analysis and interpretation of data; drafting of the manuscript; critical revision of the manuscript for important intellectual content; study oversight.
All authors have reviewed and approved this final draft of the manuscript.
Declaration of conflicting interests
The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.
Funding
The author(s) disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: Drs. Nietert and Rockey were supported by a grant from the National Institute of Digestive and Kidney Diseases (P30DK123704), which supports the Clinical Component Core of the MUSC Digestive Disease Research Core Center.
Ethics approval
This study was approved by the Medical University of South Carolina Institutional Review Board.
Consent to participate
N/A.
Consent for publication
N/A.