Incidence and Predictors of Hepatocellular Carcinoma in Patients With Cirrhosis

Clinical Gastroenterology and Hepatology Aug 2007

George N. Ioannou_, Meaghan F. Splan, Noel S. Weiss, George B. McDonald_, Laura Beretta#, Sum P. Lee_ _ Division of Gastroenterology, Department of Medicine, Veterans Affairs Puget Sound Health Care System, University of Washington, Seattle, Washington Research Enhancement Award Program, Veterans Affairs Puget Sound Health Care System, University of Washington, Seattle, Washington Hepatitis C Resource Center, Veterans Affairs Puget Sound Health Care System, University of Washington, Seattle, Washington Department of Epidemiology, University of Washington, Seattle, Washington Clinical Research Division, Fred Hutchinson Cancer Research Center, Seattle, Washington # Public Health Sciences Division, Fred Hutchinson Cancer Research Center, Seattle, Washington

In summary, we identified 7 important predictors of hepatocellular carcinoma in multivariate models (including relatively novel ones such as increased BMI, HBV core antibody, and low platelet count), which suggests a means of predicting the risk of hepatocellular carcinoma in patients with cirrhosis and optimizing surveillance strategies.

Background & Aims: Independent predictors of hepatocellular carcinoma in patients with cirrhosis are not well established. Methods: We created a cohort of 2126 patients (41% with hepatitis C virus [HCV] infection) who sought care from all Veterans Affairs health care centers in the northwest United States from 1994 to 2005 and who had a diagnosis of cirrhosis recorded in inpatient or outpatient medical records.

Results: During a mean follow-up period of 3.6 years, 100 patients were diagnosed with hepatocellular carcinoma (incidence, 1.3 per 100 patient-years). Important predictors of hepatocellular carcinoma in multivariate models included HCV infection (adjusted hazard ratio [ahr], 3.0; 95% confidence interval [CI], 1.7-5.3); hepatitis B virus (HBV) surface antigen (ahr, 3.3; 95% CI, 1.4-7.7); HBV core antibody (ahr, 1.7; 95% CI, 1.1-2.8); obesity (ahr, 2.5; 95% CI, 1.3-4.9), and overweight (ahr, 2.8; 95% CI, 1.5-5.4) relative to patients with a body mass index of < 25 kg/m2, diabetes (ahr, 1.5; 95% CI, 0.9-2.5), and low platelet count (relative to patients with a platelet count of >266 thousands/µL, the ahr was 2.1 [95% CI, 0.8-5.6] in patients with a platelet count of 180-266 thousands/µL, 3.3 [95% CI, 1.3-8.0] in patients with a platelet count of 111-179 thousands/µL, and the ahr was 4.7 [95% CI, 2.0-11.4] in patients with a platelet count of ?110 thousands/µL).

Conclusions: We identified 6 important predictors of hepatocellular carcinoma in multivariate models (including relatively novel predictors such as increased body mass index, HBV core antibody, and low platelet count), which suggest a means of predicting the risk of hepatocellular carcinoma in patients with cirrhosis and optimizing surveillance strategies.

In the United States, hepatocellular carcinoma occurs almost exclusively in the presence of cirrhosis, with the exception of patients with chronic hepatitis B virus (HBV) infection, who can develop hepatocellular carcinoma without pre-existing cirrhosis. However, only a relatively small proportion of persons with cirrhosis go on to develop hepatocellular carcinoma per year. Identifying independent predictors of hepatocellular carcinoma would allow stratification of patients with cirrhosis based on their expected incidence of hepatocellular carcinoma. In turn, in the future this would allow optimization of surveillance strategies for hepatocellular carcinoma.

Unfortunately, independent predictors of hepatocellular carcinoma in patients with cirrhosis are not well characterized, especially among non-Asian patients, who have a dramatically different epidemiology of cirrhosis and hepatocellular carcinoma than Asian patients. A recent cohort study from France identified increased body mass index (BMI), diabetes, age, male sex, and hepatitis C virus (HCV) infection as independent predictors of hepatocellular carcinoma.1 A Japanese cohort study of patients with hepatitis C cirrhosis2 identified and validated the following predictors of hepatocellular carcinoma: increased ƒ¿-fetoprotein level, age, male sex, and low platelet count. Chronic HBV infection is the most important risk factor for hepatocellular carcinoma worldwide because of the very high prevalence of infection in Asia and Africa.3, 4 However, it is unclear to what extent patients with cirrhosis with chronic HBV infection have a higher incidence of hepatocellular carcinoma than patients with cirrhosis of a different cause in Europe or North America. Occult HBV infection, generally defined as the detection of HBV DNA in the serum or liver tissue of patients who test negative for hepatitis B surface antigen,5 may be a risk factor for hepatocellular carcinoma.6

We aimed to document the incidence and identify the predictors of hepatocellular carcinoma in a cohort of patients with cirrhosis who sought care from any of the Veterans Affairs health care centers in the northwest United States. This information may be useful in developing surveillance strategies for hepatocellular carcinoma in patients with cirrhosis.

Discussion
In a cohort of patients with cirrhosis who sought care in Veterans Affairs health care centers, independent predictors of developing hepatocellular carcinoma included HCV infection, HBV infection, presence of HBV core antibody, diabetes mellitus, increased BMI, and reduced platelet count (Table 1). The incidence of hepatocellular carcinoma varied by more than 20-fold based on the presence or absence of combinations of these predictors (Table 2).

Previous studies using surveillance programs for hepatocellular carcinoma in patients with cirrhosis reported incidences of 1.5,7 2.5,8 or 6.71 per 100 patient-years among all patients and 2.0,9 2.5,10 or 3.911 per 100 patient-years among HCV-infected patients. In the current study, in which there was no study protocol-related surveillance program, we identified a similar overall incidence (2.4 per 100 patient-years) among HCV-positive patients with cirrhosis. However, what our results highlight best is that the incidence varies greatly depending on the presence of viral hepatitis, diabetes, obesity, and platelet count.

Surveillance of patients with cirrhosis for hepatocellular carcinoma has been advocated by the American Association for the Study of Liver Diseases using ultrasonography every 6-12 months.12 There is only one available randomized controlled trial of surveillance vs no surveillance and it observed improved survival rates in Chinese patients assigned to receive surveillance with ƒ¿ fetoprotein levels and ultrasound scans performed every 6 months.13 Most recommendations for or against surveillance are based on theoretic cost-effectiveness analyses14, 15, 16, 17 and expert opinion. If surveillance does indeed save some lives, the survival benefit16 and cost effectiveness17 of surveillance programs for hepatocellular carcinoma would be expected to decrease when the incidence of hepatocellular carcinoma decreases. Our study identifies groups of patients with cirrhosis with an increased incidence of hepatocellular carcinoma (Table 2) who stand to benefit most from surveillance and in whom surveillance is likely to be relatively more cost effective.

A recent, excellent systematic review18 cited 13 case-control and 13 cohort studies that investigated the association between diabetes and hepatocellular carcinoma, showing an overall positive association (an additional cohort study has been published since then by Lai et al19). Although these studies strongly suggest that diabetes is a risk factor for hepatocellular carcinoma, they do not address what the excess risk of hepatocellular carcinoma caused by diabetes is in patients with cirrhosis because the controls they used were noncirrhotic persons. Furthermore, the association between diabetes and hepatocellular carcinoma may be confounded by BMI and HCV. Of these 27 studies, only 3 adjusted for BMI,20, 21, 22 of which 2 also adjusted for a self-reported history of hepatitis (these were conducted before serologic testing for HCV became available20, 21). In the current study, when we adjusted for viral hepatitis, age, and platelet count, diabetes was still a predictor of the incidence of hepatocellular carcinoma (hazard ratio, 1.7; P = .05, this result is not shown in Table 1, Table 2). When we additionally adjusted for BMI the hazard ratio for the association between diabetes and hepatocellular carcinoma predictably decreased (1.4; 95% CI, 0.8-2.4). Similar to the results of a recently published cohort study from Taiwan,19 we found the association between diabetes and hepatocellular carcinoma was much weaker in the subset of HCV-infected patients with cirrhosis. Our results are nearly identical to the results of a recently published French study that reported an adjusted hazard ratio of 1.6 (95% CI, 1.2-2.1) for the association between diabetes and hepatocellular carcinoma in patients with cirrhosis,1 even though the absolute incidence of hepatocellular carcinoma in that study was much higher than the incidence of hepatocellular carcinoma in the current study.

We previously reported that obesity,23 particularly central obesity,24 is associated with cirrhosis-related death or hospitalization in the US population. The current study suggests that in addition to being a risk factor for cirrhosis, increased BMI is a strong predictor of hepatocellular carcinoma in patients with cirrhosis, over and above any effect that may be mediated through diabetes. A population-based study of 900,000 US adults reported that men with a BMI of 35 kg/m2 or greater were 4.5 times more likely to die from hepatocellular carcinoma than men of normal weight (although as in the diabetes cohort studies described earlier, this study could not distinguish whether this excess risk was owing to an excess risk of development of cirrhosis, rather than hepatocellular carcinoma per se).25 Our results are nearly identical, again, to the earlier-mentioned French study that reported adjusted hazard ratios of 2.0 (95% CI, 1.4-2.7) and 2.8 (95% CI, 2.0-4.0) for overweight and obesity, respectively.1

Patients with cirrhosis with active HBV (adjusted hazard ratio, 4.3; 95% CI, 1.6-11.9) or HCV (adjusted hazard ratio, 2.6; 95% CI, 1.4-5.0) infection had a much higher incidence of hepatocellular carcinoma than patients with alcoholic cirrhosis without viral hepatitis, even after adjusting for obesity, age, diabetes, and platelet count. Patients without viral hepatitis who did not have a diagnosis of alcoholic cirrhosis had a lower incidence of hepatocellular carcinoma than patients with alcoholic cirrhosis (adjusted hazard ratio, 0.5; 95% CI, 0.2-1.5), but this difference was not statistically significant.

European5, 6, 26, 27, 28 and Chinese29 studies have suggested that occult HBV infection (defined as the presence of low levels of HBV DNA in blood or liver tissue by sensitive polymerase chain reaction in the absence of HBV surface antigen) is a risk factor for cirrhosis and hepatocellular carcinoma. We identified a significant independent association between HBV core antibody and hepatocellular carcinoma. Thus ours is a cohort study which suggests that HBV core antibody is an independent predictor of hepatocellular carcinoma in patients with cirrhosis after adjusting for other potential confounders. It is possible that prior infection with HBV, as evidenced by HBV core antibody, may promote hepatocellular carcinogenesis even in the absence of circulating HBV surface antigen by persistence of hepatic HBV DNA.

Low platelet count, a marker of more advanced cirrhosis and portal hypertension, was one of the strongest predictors of hepatocellular carcinoma in patients with cirrhosis. This suggests that as cirrhosis becomes more advanced and liver function deteriorates, the incidence of hepatocellular carcinoma increases. If this were true, then other markers of liver function, such as serum total bilirubin level, also would be expected to predict the incidence of hepatocellular carcinoma. Serum total bilirubin level was associated with hepatocellular carcinoma in univariate but not in multivariate models. This may be because platelet count was a better predictor and explained most of the variability related to liver function, or because there were fewer patients with available serum bilirubin levels (n = 1212), limiting the power of our study. Alternatively, there may be something unique to platelet count as a predictor of hepatocellular carcinoma not directly related to its association with liver function.

The incidence of hepatocellular carcinoma in the general US population is 2 times higher in African Americans than in whites.30 This generally is believed to be caused by the 2- to 3-fold higher prevalence of hepatitis C, hepatitis B, and alcoholic cirrhosis in African Americans compared with whites.31 Sarbah et al32 reported that patients with hepatocellular carcinoma were 3 times more likely to be African American than patients with cirrhosis on a transplant database. However, this is almost certainly a result of the underrepresentation of African Americans on transplant lists. In this study we found no difference between African American and white patients with cirrhosis in the incidence of hepatocellular carcinoma. These results should be interpreted with caution because a substantial percentage of veterans (297/2126 [14%]) did not report their race.

Limitations
Data were derived from a research/administrative database (CHIPS) of the Veterans Affairs Northwest network, instead of by identification, examination, interview, and medical record abstraction of individual patients. The ICD-9 codes used in this study from CHIPS are derived from the same source (Veterans Health Information Systems and Technology Architecture) as the better-known national Veterans Affairs databases, such as the patient treatment file for inpatient diagnoses and the outpatient care file for outpatient diagnoses. These ICD-9 codes are entered into the local computer systems at each facility by the physicians or by professional coders, who assign these codes based on recorded findings in progress reports and discharge summaries in patients' records. By using chart review to verify the patient treatment file and outpatient care data, the reliability was found to be adequate for demographic data (? ~ 0.92) and for selected diagnoses (eg, schizophrenia, ? = 1; cardiac dysrhythmias, ? = 0.888; acute myocardial infarction, ? = 0.869; alcohol dependence syndrome, ? = 0.866; and diabetes, ? = 0.823).33 The use of ICD-9 code 155.0 to examine the temporal trends of hospitalization with hepatocellular carcinoma in the VA system has produced identical results to the incidence rates of histologically confirmed hepatocellular carcinoma in national cancer registries,34, 35 thus indicating the validity of using this code as an indicator of hepatocellular carcinoma. Both cirrhosis and liver cancer are often the principal diagnoses of many inpatient and outpatient visits and we therefore would expect them to be coded reliably and completely. Numerous studies have used ICD-9 codes for the diagnosis of cirrhosis and liver cancer from national VA databases.34, 35, 36 All the laboratory results that are recorded in the electronic medical records of the 8 participating facilities are transferred electronically to CHIPS.37 Therefore, CHIPS should be almost 100% complete and valid with regards to laboratory test results.

The definitions of cirrhosis and hepatocellular carcinoma were based on diagnoses recorded by the patients' physicians in their medical records rather than on specific criteria determined a priori for the purposes of this study. However, a clinical diagnosis of cirrhosis made while the patient is alive has been shown to have almost 100% specificity when compared with autopsy findings.38 Hepatocellular carcinoma might have been underdiagnosed because there was no universal screening program. However, the incidence of hepatocellular carcinoma that we report among HCV-positive patients is very similar to that reported in smaller studies that routinely screened for hepatocellular carcinoma.9, 10 If physicians were more likely to screen for hepatocellular carcinoma in patients who had certain perceived risk factors (such as viral hepatitis), this would inflate the hazard ratios that we reported for these risk factors. However existing guidelines do not recommend using different screening protocols for hepatocellular carcinoma in patients with cirrhosis depending on the presence of risk factors. Alcohol consumption was not ascertained in our study during the at-risk time period; instead we had available data on a diagnosis of alcohol abuse or dependence, which could have been current or past. Hence, our data did not allow us to determine whether alcohol consumption in patients with cirrhosis influences the development of hepatocellular carcinoma. The same limitation applied to tobacco and marijuana use. We were not able to estimate the duration of hepatitis B or C infection. BMI may have been influenced by the development of ascites in patients with cirrhosis. However, we found near-identical results when we used the earliest weight recorded for each patient, which was frequently many years before the diagnosis of cirrhosis, rather than the weight recorded within 90 days of the diagnosis of cirrhosis. Our results were based primarily on male VA users with available viral hepatitis B and C serology results and therefore should be extrapolated to other populations with caution. Future studies are necessary to validate our models in different patient populations.

In summary, we identified 7 important predictors of hepatocellular carcinoma in multivariate models (including relatively novel ones such as increased BMI, HBV core antibody, and low platelet count), which suggests a means of predicting the risk of hepatocellular carcinoma in patients with cirrhosis and optimizing surveillance strategies.

Results
Forty-one percent of the 2126 patients with cirrhosis had evidence of chronic HCV infection, 3.5% were positive for HBV surface antigen, and 35% were positive for hepatitis B core antibody (Table 1). In addition, 21% had type 2 diabetes mellitus, 25% had a serum total bilirubin level of 2.0 mg/dL or greater, 25% had a platelet count of 110 thousands/µL or less, 36% were overweight, 33% were obese, at least 72% were of non-Hispanic white ethnicity, and only 3% were women, consistent with the study population derived from American Veterans (Table 1). Among these patients, 100 were diagnosed with hepatocellular carcinoma during 7691 patient-years of follow-up evaluation (mean follow-up period, 3.6 y; incidence of hepatocellular carcinoma, 1.3 per 100 patient-years). By the end of the follow-up period (July 12, 2005), 851 of the initial 2126 patients with cirrhosis had died.

In univariate analyses, significant predictors of hepatocellular carcinoma included chronic HCV infection, HBV surface antigen, HBV core antibody, cause of cirrhosis, type 2 diabetes mellitus, reduced platelet count, and increased BMI, serum bilirubin level, alanine aminotransferase level, and aspartate aminotransferase level. The majority of hepatocellular carcinomas (72%) occurred in the 41% of patients with cirrhosis who had chronic HCV (Figure 1). Female sex and non-Hispanic black race were associated weakly with a lower incidence of hepatocellular carcinoma.

The predictors of hepatocellular carcinoma that were retained in the final multivariate model were as follows: HCV infection (adjusted hazard ratio [ahr], 3.0; 95% confidence interval [CI], 1.7-5.3); HBV surface antigen (ahr, 3.3; 95% CI, 1.4-7.7); HBV core antibody (ahr, 1.7; 95% CI, 1.1-2.8); obesity (ahr, 2.5; 95% CI, 1.3-4.9), overweight (ahr, 2.8; 95% CI, 1.5-5.4) relative to patients with a BMI of less than 25 kg/m2; diabetes (ahr, 1.5; 95% CI, 0.9-2.5); and a low platelet count (relative to patients with a platelet count of >266 thousand/µL, the ahr was 2.1 [95% CI, 0.8-5.6] in patients with a platelet count of 180-266 thousands/µL, the ahr was 3.3 [95% CI, 1.3-8.0] in patients with a platelet count of 111-179 thousands/µL, and the ahr was 4.7 [95% CI, 2.0-11.4] in patients with a platelet count of ?110 thousands/µL). Supplementary Table 1 shows the important predictors only; (see supplementary material online at www.cghjournal.org). All of these predictors were statistically significant in the multivariate model, except age and diabetes, which were associated weakly with hepatocellular carcinoma but were retained because their inclusion affected the coefficients of other predictors (Table 1). Because some patients did not have a platelet count measured within 60 days of the first diagnosis of cirrhosis, the multivariate model was based on 1770 patients, of whom 83 developed hepatocellular carcinoma. Cause of cirrhosis was also a significant predictor when used in the multivariate model instead of separate variables for HCV infection and HBV surface antigen (Table 1). We did not find evidence of statistically significant interactions between important predictors of hepatocellular carcinoma included in our multivariate model.

The observed incidence of hepatocellular carcinoma varied dramatically according to the presence and level of the 7 predictors of hepatocellular carcinoma retained in the final multivariate model (Table 2). For example, patients with cirrhosis with no evidence of viral hepatitis, obesity, or diabetes had a hepatocellular carcinoma incidence of only 0.2%. On the other hand, overweight patients with HCV infection and a platelet count of less than 110 thousand/µL had an incidence of 5.8%.

Because the majority of hepatocellular carcinomas (72%) occurred in patients with chronic HCV infection, additional analyses were performed limited to patients with chronic HCV infection (Supplementary Table 2; see supplementary material online at www.cghjournal.org). In multivariate models, important predictors of hepatocellular carcinoma in HCV-positive patients with cirrhosis included the following: overweight (ahr, 3.4; 95% CI, 1.5-7.5); obesity (ahr, 2.2; 95% CI, 0.9-5.0); low platelet count (relative to patients with a platelet count of >266 thousand/µL, the ahr was 2.3 [95% CI, 0.6-8.8] in patients with a platelet count of 180-266, the ahr was 3.1 [95% CI, 0.9-10.9] in patients with a platelet count of 111-179, and the ahr was 5.3 [95% CI, 1.6-17.5] in patients with a platelet count of ?110 thousands/µL).

Methods
Study Setting and Data Collection
The Veterans Health Administration Northwest Network serves the health care needs of veterans at 8 facilities in 4 states (Alaska, Idaho, Oregon, and Washington). Selected patient data from these 8 health care facilities are stored in the Consumer Health Information and Performance Sets (CHIPS) database, which is designed for research, decision support, quality assurance, performance measurement, and testing guidelines. CHIPS contains data on all inpatient and outpatient diagnostic codes, appointments, laboratory results, medication prescriptions, imaging tests, surgical procedures, and vital signs, recorded for all patients who attended these health care facilities.

Study Population
Out of 29,003 persons who were tested for HCV antibody (including HCV RNA or recombinant immunoblot assay if the HCV antibody was positive), HBV surface antigen, and anti-HBV core antibody from October 1, 1994 (the earliest date that these laboratory results were entered into CHIPS) to December 31, 2003, there were 2394 who had a diagnosis of cirrhosis recorded in their inpatient or outpatient records before July 12, 2005, the date that CHIPS was last searched. The study population was limited to patients with available viral serologies because these were thought to be crucial predictors of outcomes in cirrhosis. We excluded 41 patients who had a diagnosis of hepatocellular carcinoma before the first recorded diagnosis of cirrhosis, 226 without a BMI measurement within 90 days of the diagnosis of cirrhosis, and 1 patient of unknown sex, leaving 2126 participants for most of the analyses. Analyses of the association of laboratory test results such as serum bilirubin level or platelet count with hepatocellular carcinoma were based on fewer patients because these test results were not available for all patients within 60 days of the diagnosis of cirrhosis.

Definition of Cirrhosis
Cirrhosis was defined based on the presence of one or more of the following inpatient or outpatient International Classification of Diseases, ninth revision (ICD-9) diagnoses: 571.2 (cirrhosis with alcoholism), 571.5 (cirrhosis no mention of alcohol), 571.6 (biliary cirrhosis), 456.2 (esophageal varices with bleeding), 456.21 (esophageal varices no mention of bleeding), 572.2 (hepatic coma), 572.3 (portal hypertension), 572.4 (hepatorenal syndrome), and 789.5 (nonmalignant ascites). The presence of esophageal varices, hepatic coma, portal hypertension, hepatorenal syndrome, or nonmalignant ascites was considered sufficient for the diagnosis of cirrhosis because the overwhelming majority of these conditions among Americans are caused by cirrhosis.

Ascertainment of Hepatocellular Carcinoma
The date of incidence of hepatocellular carcinoma was the earliest date on which a diagnosis of hepatocellular carcinoma (ICD-9 code 155.0) was recorded on inpatient or outpatient medical records.

Predictors of Hepatocellular Carcinoma in Patients With Cirrhosis
We evaluated the degree to which the following predictors were associated with the incidence of hepatocellular carcinoma during follow-up evaluation: age at diagnosis of cirrhosis (categorized into quartiles); sex; race-ethnicity (non-Hispanic white, non-Hispanic black, other, and unknown race-ethnicity); HCV infection defined by positive HCV RNA or recombinant immunoblot assay in a patient positive for anti-HCV antibody; HBV surface antigen; HBV core antibody; type 2 diabetes defined by ICD-9 codes 250.00-250.92 reported in inpatient or outpatient records before the diagnosis of cirrhosis; human immunodeficiency virus 1 infection defined by a positive human immunodeficiency virus 1 RNA test or by ICD-9 code 042; BMI calculated as the measured weight in kilograms divided by the square of the measured height in meters recorded within 90 days of the diagnosis of cirrhosis and categorized into normal (<25 kg/m2), overweight (25 to <30 kg/m2), and obese (=30 kg/m2); alcohol abuse or dependence defined by ICD-9 codes 305.00-305.03 (abuse) and 303.90-303.93 (dependence) recorded in inpatient or outpatient records; cannabis abuse or dependence defined by ICD-9 codes 305.20-305.23 (abuse) and 304.30-304.33 (dependence) recorded in inpatient or outpatient records; and serum bilirubin level, creatinine level, albumin level, alanine aminotransferase level, aspartate aminotransferase level, ferritin level, prothrombin time international normalized ratio, hemoglobin concentration, and platelet count recorded within 60 days of the diagnosis of cirrhosis and analyzed both as continuous variables after logarithmic transformation and as categoric variables into quartiles. If more than one measurement of any of these laboratory tests was recorded within 60 days of the diagnosis of cirrhosis, we used the least abnormal value. These laboratory test results were not available for all patients within 60 days of cirrhosis (Table 1). Finally, in addition to analyzing potential causes of cirrhosis separately, we also created a single variable in which we categorized patients based on the most likely cause of cirrhosis as follows: alcoholic cirrhosis if they had a diagnosis of cirrhosis with alcoholism (ICD-9 code 571.2), alcohol abuse, or alcohol dependence without evidence of active HBV or HCV (defined earlier); viral hepatitis C if they had hepatitis C infection without hepatitis B infection; viral hepatitis B if they had hepatitis B infection; and other for all other causes of cirrhosis.