Potential conflict of interest: Nothing to report.
This work was supported by the Kaiser Permanente Northern California Community Benefit Research Grant Program.
The natural history of hepatitis B virus (HBV) infection in a U.S. population has not been well described. We identified the causes of death in 6,689 health plan members infected with HBV who were followed between March 1, 1996 and December 31, 2005. Causes of death were grouped into HBV-related (subdivided into decompensated cirrhosis [DCC] and hepatocellular carcinoma [HCC]), cancer, cardiovascular, and other/unknown. The study cohort included 3,244 females and 3,445 males; 68.3% were of Asian-Pacific Islander (API) descent, 11.8% were white (non-Hispanic), and 19.9% were of other or unknown race. Exposure to HBV antivirals and preexisting comorbidities were uncommon. Males had higher overall 10-year death rates than females, both for total deaths (8.9% versus 4.1%) and for HBV-related deaths (4.8% versus 1.2%). The death rate rose markedly with increasing age, and approximately 40% of all deaths in subjects over the age of 40 were HBV related. The death rate from HCC was twice that of DCC. HCC deaths represented 70% of cancer deaths in males and 37% in females. On multivariable analysis, when subjects with antecedent HCC and DCC were excluded, the only significant predictor of HBV mortality in both sexes was age. Conclusion: HBV was the cause of death in over 40% of those who died during the study, and the mortality increased markedly with increasing age over 40 in males and over 50 in females. HBV-related mortality was four times more common in males than in females and was as common in non-Asians as in those of API origin. HBV-related deaths were twice as common from HCC as from DCC. (HEPATOLOGY 2013)
If you can't find a tool you're looking for, please click the link at the top of the page to "Go to old article view". Alternatively, view our Knowledge Base articles for additional help. Your feedback is important to us, so please let us know if you have comments or ideas for improvement.
In many individuals, the natural history of hepatitis B virus (HBV) infection is one of progression to cirrhosis, decompensated cirrhosis (DCC), and hepatocellular carcinoma (HCC). The most widely quoted assessment is that “approximately 15%-40% of infected patients will develop cirrhosis, liver failure, or HCC.”1 These data are largely derived from natural history studies in Asia. The study quoted most frequently is a 1981 prospective study of 22,707 male Taiwanese government employees, of whom 3,454 were positive for hepatitis B surface antigen (HBsAg+). Of those who were HBsAg+, 79% were between the ages of 40 and 59. Over an average follow-up of 3.3 years, 54% of the deaths in HBV-infected subjects were from HBV-related causes, with the death rate rising progressively with increasing age.2
Data from other parts of the world show disparate results. Two long-term studies from Italy showed minimal rates of HBV-related mortality,3, 4 whereas a third showed a much higher rate.5 Two studies from Canada and one from Alaska6-8 also reported low HBV-related mortality rates, but a report from California reported a very high rate, particularly of hepatoma.9 The different results in these studies probably reflect entry criteria that differed among the studies and that selected out for different age groups within the total HBV population.
Given the disparate conclusions about the natural history of hepatitis B, and given the lack of population-wide data in the United States about the effect of HBV infection on mortality, we undertook a study to examine this question. We performed a retrospective study of rates of mortality resulting from HBV-related causes versus all other causes within a cohort of patients with hepatitis B within a large integrated healthcare delivery system in Northern California who were followed from 1996 through the end of 2005. The study incorporated all age groups and both males and females. The dates were chosen deliberately to include a period when few patients were treated long term with antivirals and thus to reflect a largely untreated natural history. Given the widespread and increasing use of antiviral medications in recent years, this is probably the latest time that such a largely untreated cohort will be studied within the United States.
AFP, alpha-fetoprotein; ALT, alanine aminotransferase; API, Asian-Pacific Islander; APRI, aspartate aminotransferase platelet ratio index; AST, aspartate aminotransferase; CHB, chronic hepatitis B; CPT-4, Current Procedural Terminology, Fourth Edition; CVD, cardiovascular disease; DCC, decompensated cirrhosis; DM, diabetes mellitus; HBeAb, hepatitis B e antibody; HBeAg, hepatitis B e antigen; HBV, hepatitis B virus; HCC, hepatocellular carcinoma; ICD-9, International Classification of Diseases, Ninth Revision; INR, international normalized ratio; KPNC, Kaiser Permanente Northern California.
Patients and Methods
The study was approved by the Kaiser Permanente Northern California (KPNC) Institutional Review Board. Waiver of informed consent was obtained because of the nature of the study. We identified individuals who were KPNC health plan members in March 1996 and who were diagnosed with hepatitis B either before or after that date based on the approach described below. March was selected as the starting month to allow for lag time in year-end enrollment data within KPNC. We also restricted the study cohort to those who had been KPNC members for the preceding 12 months (March 1995-February 1996). The year 1995 was the first year when computerized laboratory records became regionally available in KPNC; 2005 was the last year for which mortality data were available at the time the study was conducted. Age at the onset of the study was calculated as of March 1, 1996.
Diagnosis of hepatitis B was made according to criteria and levels of certainty established within the Northern California Viral Hepatitis Registry (see Supporting Table 1). These criteria rely on a combination of diagnostic codes and laboratory data available in health plan electronic databases and include HBsAg, hepatitis B e antigen (HBeAg), hepatitis B e antibody (HBeAb), HBV DNA, liver biopsy, and diagnosis of hepatitis. We excluded patients coinfected with human immunodeficiency virus or with hepatitis C virus because these viruses strongly influence the risk of death. The hepatitis B diagnosis could be made at any time before or during the study period. To validate our entry criteria, we ran the same analysis on a subset of the cohort meeting a stricter definition of chronic hepatitis B (CHB) (termed “Level 1”). This criterion consisted of repeated HBsAg+ results at least 6 months apart, HBV DNA or HBeAg positivity after a positive HBsAg result, liver biopsy compatible with CH, or treatment for hepatitis B (Supporting Table 1).
Table 1. Baseline Characteristics of the HBV Cohort
DCC was defined as hepatic encephalopathy, ascites, or portal hypertension (including esophageal or gastric varices). The International Classification of Diseases, Ninth Revision (ICD-9) diagnosis codes used were 197.6, 456.1, 456.2, 456.8, 567.2, 567.8, 572.2, and 789.5 at clinic visits and 348.30, 348.31, 348.39, 456.0, 456.1, 456.2, 456.8, 567.23, 572.2-4, and 789.5 during hospitalizations. We also included Current Procedural Terminology, Fourth Edition (CPT-4) or ICD-9 procedure codes for endoscopic treatment of varices, paracentesis, and shunt creations (CPT-4: 43204, 43243, 43244, 65800, and 65805; ICD-9: 39.1, 39.26, and 54.94).
HCC was strictly defined as a diagnosis of HCC, excluding hepatoblastoma and cholangiocarcinoma, whether extra- or intrahepatic. These were identified by ICD-9 diagnosis codes 155, 155.0, and 155.2 and confirmed by chart review.
Treated patients were those who had exposure to interferon, lamivudine, adefovir, tenofovir, telbivudine, or entecavir based on prescription data from ambulatory health plan pharmacy databases. Almost all members (94%-95%) had drug coverage through the health plan.10
We ascertained which members had died using our Division of Research mortality linkage system. This system consists of Statistical Analysis System (SAS Institute Inc., Cary, NC) files containing records linked between KPNC data systems and membership enrollment, the California State Department of Vital Statistics mortality files of death certificate information, and the U.S. Social Security Administration Death Master File. With these linked records, it is possible to ascertain, with a high degree of confidence, which KPNC members may have died, even if they have left the health plan.11 Causes of death were first reviewed within the KPNC Viral Hepatitis Registry, and any differences among different systems resolved. Subsequently, a board-certified transplant hepatologist (J.L.S.) reviewed the available computer records to verify the causes of death, supplemented by reviewing additional paper records. Computer records were available within the last 3 months of life for 87% of those who died and within the last year of life for 93%. Almost half of all deaths (43.7%) either occurred within a KPNC facility or had death summaries available.
The following comorbidity groups were identified at study entry: cardiovascular disease (CVD); kidney disease; cancer; and pulmonary disease. Diabetes mellitus (DM) could be identified at any time during the study period. The following ICD-9 diagnosis codes were used for each comorbidity group: (1) CVD: including 410-413, 425, 428 (cardiovascular), and 430-438 (cerebrovascular); (2) kidney: 402-403, 581-582, 585, and 587-588; (3) cancer: 140-154, 156-209, 200-203, and 235-238; and (4) pulmonary: 491-492, 494, and 496.
We extracted entry laboratory data on our cohort, but they were available in only a minority of our patients, in part because the start date of our cohort was so soon after the implementation of electronic records. No chart review was conducted for paper copies of the lab tests of interest. The laboratory data were thus not analyzed as risk factors related to the outcomes of interest, but they are included in the supplemental records (Supporting Table 2). These laboratory data at entry included alanine aminotransferase (ALT), bilirubin, international normalized ratio (INR; excluded if the patient was taking warfarin), alpha-fetoprotein (AFP), platelet count, and a calculated aspartate aminotransferase platelet ratio index score (APRI score), with a value of 2.0 set as a marker of cirrhosis.12 Viral loads were not available within our computerized database at the onset of this time period.
Table 2. All-Cause and HBV-Related Death Rates per 1000 Person-Years
n (Row %)
n (Row %)
n (Row %)
n (Row %)
n (Row %)
n (Row %)
Age at entry, years
65 and over
Risk factors that were analyzed to assess the risk of HBV-related death were age, sex, race, and comorbid conditions. Subjects with preexisting HCC, DCC, and liver transplant were included only in the initial analysis and were omitted in subsequent analyses. Race/ethnicity was determined by KPNC out- and inpatient administrative databases primarily using self-reported race, language of choice, and surname. We created three groupings for analysis: Asian-Pacific Islander (API); White non-Hispanic; and other/unknown (grouped because of small numbers in each subcategory). The other/unknown group included 1,330 members, with 321 African American (24.1%), 253 Hispanic (19.0%), 95 mixed (7.1%), 9 Native American (0.7%), and 652 unknown (49.0%). Risk factors were also analyzed separately for males and females. A number of important risk factors, such as alcohol use,13 cigarette use,14 and coffee intake,15 were unable to be analyzed because of a lack of computerized data on those factors.
We conducted a bivariate analysis to determine risk factors of all-cause and HBV-related deaths. Then, we performed a survival analysis to estimate the survival rates during the study period. Survival time was defined as time from the study start date (March 1, 1996) to death with censoring at the end of 2005. Survival analysis was performed using Kaplan-Meier's method to estimate the survival function. The log-rank test was used to compare survival functions stratified by subgroups, such as API, White, and other/unknown. Cox's proportional hazards models were used to identify factors predictive of death from any cause and from HBV-related causes, with the threshold of significance set at P < 0.05. For HBV-related deaths, deaths from other causes were censored. We created a multivariable model that adjusted for sex and one that was stratified by sex. We also calculated the lifetime risk of dying from an HBV-related cause, compared to the risk of dying from any cause. SAS version 9.12 (SAS Institute) was used for all statistical analyses, except the lifetime risk of dying from hepatitis B, which was calculated manually.
Description of Cohort.
We identified 6,689 members who met our inclusion criteria. The baseline characteristics of the study cohort are shown in Table 1. The cohort was almost equally divided between males and females. A majority of the cohort (>75%) was between the ages of 30 and 64, and two thirds were API. Overall, females were slightly younger than males. Preexisting comorbid conditions were rarely identified in the cohort. Diabetes was diagnosed in 953 (14.2%) of the cohort members during the course of the study. Only 12 subjects had preexisting HCC, 21 had DCC, 3 had both HCC and DCC, and 4 had had liver transplants. Laboratory data at study entry were available for only a small minority of members and therefore were not used for further analyses of risk factors for the endpoints of interest. When laboratory data were available, they were often abnormal: INR results were greater than 1.2 in 15.8%, bilirubin was greater than 1.3 mg/dL in 9%, platelet counts were low in 8%, and APRI score was higher than 2 in 6% (Supporting Table 2). During the course of the study period, 43 (0.6%) patients received a liver transplant, of whom 9 died, 5 of HBV-related causes.
Very few members of our cohort (n = 19; 0.3%) were exposed to HBV medications before the onset of our study, but the number increased over time. Slightly more than 3% were treated by the end of year 5 of our study and 8.5% by the end of the study. Thus, two thirds of the exposure occurred in the latter half of the study period (Supporting Table 3).
The death rate per 1,000 person-years from all causes and from HBV-related causes for the entire study cohort is shown in Table 2. The crude all-cause and HBV-related mortality rates for the various age groups overall, by cause of death, and by sex are shown in Table 3 and Supporting Table 4. The crude death rate for females was half of that for males, and the HBV-related death rate was less than one quarter of that for males. The mortality rate for females within each age group over 30 was less than one third that of males. HCC deaths were twice as common in males overall as were deaths from DCC, and in males, HCC deaths were twice as common as all other forms of cancer put together.
Table 3. Mortality Rates by Age Group at Entry and Cause of Death, Stratified by Sex
n (Col %)
Cause of Death, n (Row %)
n (Col %)
Cause of Death, n (Row %)
Abbreviation: Col, column.
65 and over
To validate our inclusion criteria for HBV diagnosis, we determined the prevalence of HBV-related causes of mortality in the subgroup of patients with the highest level of certainty of HBV diagnosis, Level 1 (Supporting Tables 1 and 5-7). The prevalence was no higher in this subgroup than it was in the entire cohort.
Death rates increased markedly with increasing age (Figs. 1A-C and 2A-C). Survival was notably lower in males than in females, both for all causes (Fig. 1B,C and Supporting Fig. 1) and for HBV-related causes (Fig. 2B,C and Supporting Fig. 2) of death.
The overall survival was similar in the API and other/unknown groups, but was significantly lower in the White group (Fig. 1D). This was true for both males and females (Fig. 1E,F). In the analysis of HBV-related mortality, survival of API subjects was intermediate between that of White and other/unknown subjects, both for the overall cohort and for males (Fig. 2D,F), but not for females (Fig. 2E).
Predictors of Mortality.
On initial bi- and multivariate analysis, antecedent HCC and DCC were significant predictors of all-cause and of HBV-related mortality in both sexes. Subsequent analysis was done excluding subjects with antecedent HCC and DCC. Bivariate analysis then showed that the significant predictors of increased all-cause mortality in both sexes were age, White race, and all comorbid conditions, except pulmonary disease (Supporting Table 8). Predictors of decreased mortality were API race in females only and other/unknown race in males only. For HBV-related mortality, bivariate predictors of increased mortality in both females and males included age and diagnosis of diabetes during the study period, and in females cancer was an additional predictor. In males only, being of White race predicted increased mortality, whereas being of other/unknown mortality predicted decreased mortality.
Cox's proportional hazards model for all-cause mortality adjusted for sex showed that older age, White race, and all comorbid conditions listed were predictors of increased mortality. For HBV-related deaths, only age was a significant predictor. When stratified by sex, the model for all-cause mortality showed that older age and cardiovascular and kidney disease were predictors of increased mortality in both sexes, with cancer a predictor in women only and diabetes in men only. There was decreased mortality for API race in both sexes and for other-unknown race in females only (Table 4). For HBV-related deaths, only age was a significant predictor in both men and women.
Table 4. Multivariable Predictors of All-Cause Death and HBV-Related Death
All (N = 6,657)
Females (n = 3,237)
Males (n = 3,420)
HR (95% CI)
HR (95% CI)
HR (95% CI)
HR (95% CI)
HR (95% CI)
HR (95% CI)
Patients with preexisting decompensated cirrhosis, HCC, and/or liver transplant were excluded from the analysis.
Abbreviations: HR, hazard ratio; CI, confidence interval; N/A, not applicable.
Values have been rounded up. The actual values reflect statistical significance.
Diabetes was comorbid, but not preexisting.
CVD and kidney disease were not significant in the bivariate analysis of HBV-related deaths and therefore were not included in the multivariable analysis of HBV-related deaths.
The lifetime risk of dying from HBV-related causes was 42.2%, with a 27.6% risk for females and a 48.7% risk for males.
Finally, early deaths were extremely rare. No males died below the age of 30, and 2 died of HBV-related causes below the age of 40, 1 of HCC and 1 of DCC. For females all early HBV deaths but 1 (HCC at age 28) occurred between the ages of 45 and 49, with 3 dying from HCC and 2 from DCC (Supporting Table 9).
Our study showed that 46.7% of deaths among HBV-infected adult members of an integrated healthcare delivery system were attributed to causes related to HBV infection. The risk of dying from HBV was four times higher in males than it was in females, and only 30.3% of the deaths in females were related to HBV. Deaths from all causes, including HBV-related causes, increased with age and were rare in the young. No males in the cohort that were below the age of 30 at the study onset died of HBV-related causes during the nearly 10-year follow-up, and only 2 females in the cohort below the age of 40 at the onset died of HBV-related causes. We are currently planning a study to examine the actual age at diagnosis of HCC.
Our findings are similar to those found by Beasley et al. in their all-male study of Taiwanese subjects, wherein 54% of deaths were from HBV-related causes.2 When we looked at deaths in males only, our rate of 53.8% was nearly identical to the risk reported by Beasley et al.
The difference in outcome data between our cohort and that of other cohorts can largely be explained by differing entry criteria, particularly age. Our cohort had a mean age of 41 and included eligible subjects of all age groups, including those younger than 30 and older than 64, both male and female, and both symptomatic patients and those with mild disease. By contrast, the two Italian studies3, 4 and the Montreal study8 that showed low HBV mortality tracked younger cohorts that almost all had normal ALTs and benign liver biopsies. The studies that showed mortality similar to ours or higher included one where all patients at onset had chronic active hepatitis and were followed until they reached the age when HCC became more common5 and a U.S. study from a referral center that included an older patient cohort (mean age: 48.4), one third of whom had cirrhosis.9
We also found a lower proportion of HBV-related deaths than did the study of Chen et al. in Haimen City, China.16 That study found that 73% of the deaths were from HBV-related causes, including 51.7% from HCC. A partial explanation is that there was a higher percentage of males (60%) in that study, but other factors remain to be determined. It may be that the Haimen City population was more homogeneous than was ours, which might imply more homogeneity for an unfavorable genotype of HBV.17
In our cohort, we did not find that subjects of API origin were at higher risk for death from HBV-related complications than were those of White or other/unknown origin. The similar risk of dying from HBV-related causes across racial groups was unexpected, in that it is often stated that API persons are at higher risk for bad outcomes from hepatitis B, presumably because they are often infected in childhood and therefore have disease of longer duration.18, 19 We were unable to demonstrate such a worse prognosis in our overall cohort. It is unclear whether this was because all our infected patients also acquired disease at an earlier age, because they died at a later age and thus had a similar duration of disease, or because the duration of disease might not be a dominant predictor of HBV death. Data on use of alcohol, cigarettes, and coffee were not available in our computerized records, so we were unable to address the effect of those risk factors. Electronic data on HCV status were available for almost 80% of those who died from HBV, and where HCV status was unavailable, this was largely (78%) in members of API origin. Therefore, unmeasured HCV status did not appear to be a factor.
Our study by intent encompasses a period starting 16 years ago when treatment options for HBV were limited. The American Association for the Study of Liver Diseases guidelines came out at the very end of 200120 and cautioned about “limited long term efficacy” of treatment. Many of the landmark lamivudine studies came out during our study period. Adefovir was approved in 2002, two thirds of the way through our study period, and entecavir was approved in 2005, the last year of our study. Nonetheless, the increasing use of HBV medications pointed to the dissemination of these findings among the thousands of physicians in the plan. Consequently, a natural history study of untreated members would no longer be feasible.
A potential weakness of our study is the inclusion of members identified as having HBV during the course of the study, but subsequent to the study start date. Thus, the cohort might include cases of acute hepatitis B. However, within Northern California, the odds are overwhelmingly high that a patient with HBsAg has CHB, given that the incidence of acute hepatitis B in California is extremely low and is decreasing: Between 1990 and 2002, it dropped from 1 per 10,000 population to 1 per 50,000.21 In addition, immunization has decreased its incidence across the United States from a rate of 4 per 100,000 in 1996 to 1.8 per 100,000 in 2005.22 Within California, the rate of acute HBV is lowest in the counties covered by KPNC.23 Furthermore, it is commonly believed that most API members, who formed two thirds of our cohort, acquire the disease at birth or in childhood. To validate our inclusion criteria, we examined a subset of the cohort meeting stricter criteria for CHB. This subset showed no major differences from the entire cohort. In fact, limiting our study to the highest level of certainty would have disproportionately undercounted the number dying from HCC and DCC, given that a smaller percentage died from HBV-related causes among the members meeting the stricter criteria than among the entire cohort. To the extent that our cohort included acute hepatitis B cases, that would bias the study toward underestimating the actual rate of mortality related to CHB, given the generally good prognosis of acute hepatitis B.24
Had we restricted entry to only those members who had been identified before study commencement as carrying HBV, the study cohort would have been skewed toward sicker patients, given that screening was not widespread in 1996 and that patients were therefore tested for other reasons. We would expect that patients with abnormal liver enzymes or liver disease would more likely have been tested for HBV than would those without either. The existence of this testing bias is suggested by the high proportion of abnormal laboratory values in the minority who had test results available before entry into the study. Specifically, nearly 16% of those had an INR greater than 1.2, 9% had a bilirubin greater than 1.3, more than 8% had low platelets, and more than 6% had an APRI score above 2. By including in our cohort all those who were diagnosed during the period of the study, we are thereby more likely to have included those who were asymptomatic at the entry date.
A potential bias would also exist in the other direction if our cohort did not include asymptomatic HBV-infected patients in the KPNC population whose HBV had not yet been detected and who subsequently died from non-HBV causes. This would increase the denominator of all deaths and thus make HBV-related deaths a smaller percentage of all deaths. It is unlikely that HBV-related deaths would be undercounted, because testing for hepatitis B is part of the algorithm for evaluating HCC and cirrhosis; in contrast, non-HBV-related deaths in unidentified HBV carriers could have been undercounted because testing for HBV is not part of the algorithm for evaluating other causes of death. We can roughly assess whether significant undercounting is taking place by employing rates of HBV from later years, when screening was more widespread, and applying those rates of HBV backward to our cohort. Taking the 2005 adult prevalence of identified HBV in our health plan of 0.5% and applying it to the 1995 adult membership, we would have expected approximately 9,467 adult members to have HBV. This is close to the number in our study (6,689), which included the more-restrictive criterion of a 1-year membership in the plan. Further accounting for part of the difference is that the percentage of health plan members of API descent in 1996 was only 80% as high as in 2005, which would lead to a lower number of HBV-infected members given that those of Asian-Pacific descent have a higher rate of HBV infection than the general population.22, 25 Overall, our cohort appeared to be a close approximation of the ideal cohort that would be the universe of health plan members with HBV infection. However, it remains true that our measure of the rates of HBV-related deaths is more definitive than is the measure of overall deaths, so this bias would lead to an overestimation of the relative risk of dying from HBV-related causes.
Previous studies have shown the limitations of death certificates in ascribing the causes of death.26, 27 For that reason, detailed electronic records were reviewed to validate the cause of death. These records were available for the overwhelming majority of our deaths: 92.9% in the last year of life, 87.2% in the last 3 months, and 78.1% in the last month of life; 42.7% had death records available. Only 31 of those who died had the last available contact within KPNC more than 1 year before the time of death. By comparing our record of deaths to the mortality records of the Social Security System and of the state of California, we were confident that we detected almost all deaths and that HBV deaths were not undercounted. However, we could not exclude that some members moved away from the United States and died there unbeknownst to us. That would have had the effect of undercounting the number of deaths in our cohort.
Of note for screening purposes is the finding that those with hepatitis B die more often from HCC than from all other cancers put together, with HCC representing 59.6% of all cancer deaths. In males, 69.8% of cancer deaths were from HCC, and in females, 36.6% of cancer deaths were from HCC. This has implications for the priority that should be given to age-appropriate HCC screening in this population. Even more important are the appropriate screening for HBV in high risk populations, appropriate use of anti-HBV medications, and ultimate eradication of this disease through universal HBV vaccination.
In conclusion, our study of a largely untreated cohort of patients in the United States showed that hepatitis B was a major cause of mortality in those infected with HBV and that this mortality increased dramatically with increasing age. HBV-related mortality was much more common in males than in females and was as common in non-Asians as in those of API origin. HBV-related deaths were much more common from hepatoma than from DCC, and incidence increased markedly in males over the age of 40 and in females over the age of 50.
The authors thank the Kaiser Permanente Viral Hepatitis Registry at the Division of Research (Valentina Shvachko and Dr. Michele Manos) for providing data used in these analyses. The authors also thank Mary Anne Armstrong, MA, for the advice she provided throughout the development of this study.