Improved Performance of Serum Alpha-Fetoprotein for Hepatocellular Carcinoma Diagnosis in HCV Cirrhosis with Normal Alanine Transaminase

Patients with hepatocellular carcinoma (HCC) have an extremely poor prognosis as they frequently present with advanced stage (1). The overall survival of patients with HCC has been improving over the past 3 decades, in part due to earlier detection of cancers that are amenable to potentially curative treatments (2, 3). As early diagnosis is a key determinant of clinical outcomes in patients with HCC, the major liver societies recommend semiannual HCC surveillance in high-risk patients, including patients with cirrhotic liver disease (4–6).

Serum alpha-fetoprotein (AFP) is the tumor marker that has been most widely used in clinical practice for HCC surveillance. However, the utility of serum AFP as a surveillance test has been the subject of active debate. The American Association for the Study of Liver Diseases (AASLD) and European Association for the Study of the Liver currently recommend liver ultrasound (US) as the primary surveillance test and the use of serum AFP for surveillance only when high-quality liver US is not available (4, 6). In contrast, guidelines from the Asia Pacific Association for the Study of the Liver recommend both liver US and serum AFP every 6 months for HCC surveillance (5).

The AFP can be falsely elevated in cirrhosis patients with active hepatitis, particularly viral hepatitis (7). However, the best evidence for the efficacy of surveillance in improving survival and decreasing mortality of patients with HCC is from studies that have used US and AFP in combination (8). A recent study from Taiwan showed that the use of AFP in addition to US significantly improves the sensitivity of surveillance compared with US alone without significant loss of specificity (9). Large population-based studies have also shown a benefit for surveillance using AFP alone (10, 11).

In this study, we aimed to identify factors associated with elevated AFP in patients with cirrhosis and early-stage HCC to define the patient subgroups for whom AFP is most effective as a surveillance test for HCC.

Data from the NCI Early Detection Research Network phase II biomarker case–control study for HCC were obtained (12). Briefly, the study included 233 consecutive early-stage HCC patients and 412 cirrhotic patients without HCC seen between February 2005 and August 2007 at seven tertiary referral centers in the United States. The study was performed in compliance with Institutional Review Board approvals from the participating centers.

Clinical information, including patient demographics and clinical characteristics, was extracted from the previous phase II biomarker study database (12). Etiology of liver disease was determined on the basis of the viral serology [positive anti–hepatitis C virus (HCV) antibody and/or HCV RNA for HCV; positive hepatitis B virus (HBV) surface antigen for HBV] and history of excessive use of alcohol (alcohol abuse, alcohol dependence, or alcoholic liver disease). Nonviral/alcohol etiology included etiologies other than HCV, HBV, or alcohol. This group was not further subdivided due to the small number of subjects in each subgroup.

HCC was defined by histopathologic examination or by the specific radiologic characteristics endorsed by AASLD (3). HCC stage was determined using the Barcelona Clinic Liver Cancer staging system, and only very early or early-stage HCC were included in the current study (13).

The presence of cirrhosis was defined by histology or clinical evidence of portal hypertension in subjects with chronic liver disease. Subjects in the control group had an US, CT, or MRI showing no evidence of a hepatic mass within 6 months prior to enrolment. Patients with an AFP ≥20 ng/mL at enrolment were also required to have a CT or MRI showing no mass suggestive of HCC within the 3 months prior to enrolment or up to 2 weeks after consent. All controls were assessed by AFP and imaging 6 months after enrolment to ensure that they did not have HCC. Serum AFP was measured by automated systems (Wako) at the time of enrolment prior to HCC-specific treatment (12).

We used a previously characterized cohort of HCV cirrhosis patients with and without HCC from Parkland Health and Hospital System as a validation set (14). All HCC cases met diagnostic criteria per AASLD guidelines; we only included the subset of very early or early-stage HCC cases for this analysis, Serum AFP was measured prior to any HCC-directed therapy as part of routine clinical care. Non-HCC cirrhosis controls had a clinical diagnosis of cirrhosis as documented by their clinic provider (based on histology, imaging with a cirrhotic appearing liver, or clinical signs of cirrhosis) and were required to have at least 6 months of follow-up after AFP assessment to confirm the absence of HCC.

The χ² test or Fisher exact test was used to compare categorical variables and the nonparametric Kruskal–Wallis Test for continuous variables. Factors associated with elevated AFP were tested using logistic regression analysis. An AFP cutoff of 10.9 mg/mL was used on the basis of the maximum sum of sensitivity and specificity for early detection of HCC from our previous study; alternatively, a cutoff of 20 ng/mL was used on the basis of other previous studies (7, 12). Backward elimination logistic regression was used to construct the best multivariate model.

The sensitivity and specificity of AFP and the corresponding 95% confidence intervals (CI) were calculated using the Youden index method in the subgroup of patients with combinations of different etiologies, races, and serum alanine transaminase (ALT). The ROC curves were plotted for each subgroup of patients. The area under the ROC curve (AUC) was calculated, and its 95% CI was determined via 1,000 bootstrap samples. DeLong test was used for the comparison of different ROC curves. Cross-validation (between 2- and 10-fold) was performed to protect against overfitting in a predictive model, considering the limited sample size in each subgroup. Statistical analysis was carried out using SAS 9.3 (SAS Institute) and R version 3.0.2 (R Foundation).

A total of 412 patients with cirrhosis without HCC and 233 patients with early-stage HCC were included in the study. Clinical characteristics of the patients are summarized in Table 1. HCV was the leading etiology of both cirrhosis (57.2%) and HCC (52.4%). Nonviral/alcohol etiology was the second most common cause of both cirrhosis (25.5%) and HCC (17.6%). Cryptogenic cirrhosis was the most common cause in nonviral/alcohol etiology (accounted for 40% of nonviral/alcohol etiology in the cirrhosis group and 42% of nonviral/alcohol etiology in the HCC group).

In the cirrhosis group, a higher proportion of HCV patients had serum ALTs >40 IU/L, AFP ≥10.9 ng/mL or ≥20 ng/mL than in the HBV, alcohol, or nonviral/alcohol groups of patients. Similarly, a higher proportion of HCV patients had serum ALT >40 IU/L or AFP ≥10.9 ng/mL than the rest of the patients in the HCC group.

The distribution of serum AFP in each of the race, etiology, and serum ALT subgroups is shown in Supplementary Fig. S1. Table 2 summarizes factors associated with elevated AFP in the cirrhosis group. For serum AFP ≥10.9 ng/mL or 20 ng/mL, non-White race, HCV etiology and serum ALT were independent predictors of elevated AFP. Factors associated with elevated AFP in the HCC group are summarized in Table 3. For serum AFP ≥10.9 ng/mL, serum ALT was independently associated with elevation of AFP. When the cutoff for serum AFP was increased to ≥20 ng/mL, serum ALT, male gender, and non-White race were independently associated with elevation of AFP.

As elevation of serum AFP was associated with etiology of liver disease, race, and serum ALT, we generated ROC curves for each subgroup of patients based on etiology, race, and serum ALT (Fig. 1A–C). The best test performance for AFP was achieved in the subgroup with HBV etiology, with an overall AUC of 0.85, compared with AUCs of 0.84 for alcohol etiology, 0.80 for HCV etiology, and 0.76 for nonviral/alcohol etiology (0.85 vs. 0.76, P = 0.17; Fig. 1A).

The AUC for AFP was marginally higher but not significantly different in the non-White group compared with the White group (P = 0.31; (Fig. 1B). There was a trend toward higher AUC for AFP in the serum ALT ≤40 U/L group compared with the group with serum ALT >40 U/L (P = 0.14; Fig. 1C).

Table 4 shows the best cutoffs for serum AFP determined by the point in the ROC curve that maximizes sensitivity and specificity. The sensitivity of serum AFP with specificity set at 90% and the specificity when sensitivity was set at 70% are reported in Supplementary Table S1.

Figure 2 shows AUCs in patients with serum ALT ≤40 compared with patients with serum ALT >40 in each etiology subgroup. The AUC was significantly higher in HCV patients with serum ALT ≤40 compared with patients with serum ALT >40 (P < 0.01; Fig. 2C). The AUC increased from 0.76 to 0.91 in Whites with HCV and from 0.79 to 0.89 in non-Whites with HCV (Table 4). At 90% specificity, the sensitivity increased from 44% to 74% in Whites with HCV and from 58% to 85% in non-Whites with HCV (Supplementary Table S1). At a fixed sensitivity of 70%, the specificity increased from 69% to 96% in Whites with HCV and from 74% to 93% in non-Whites with HCV (Supplementary Table S1). In the alcohol etiology subgroup, there was a trend toward higher AUC in patients with serum ALT >40 compared with patients with serum ALT ≤40 (0.94 vs. 0.75, P = 0.06; Fig. 2A). A test for cross-validation showed similar AUCs of AFP for HCC detection in each subgroup (Supplementary Table S2).

Baseline characteristics of the validation cohort are summarized in Supplementary Table S3. Among the 110 HCV patients with HCC, 31 (29%) had ALT ≤40. Among 362 HCV patients with cirrhosis, 140 (39%) had ALT ≤40. Overall, an AFP cutoff of 9.9 ng/mL yielded a sensitivity of 65% and specificity of 66%, with an AUC of 0.73. There was a trend toward higher AUC in HCV patients with serum ALT ≤40 U/L than those with serum ALT >40 U/L (0.79 vs. 0.69, P = 0.10; Supplementary Fig. S2). When the analysis was repeated among HCV patients with normal ALT, an AFP cutoff of 8 ng/mL (cutoff that maximized sensitivity and specificity) yielded a sensitivity of 71% and specificity of 75%. The same AFP cutoff of 8 ng/mL yielded a sensitivity of 78% and specificity of 84% in the discovery set.

In the current study, we identified factors associated with elevated AFP in patients with cirrhosis and early-stage HCC, with the goal of defining the specific subgroups in which AFP shows the greatest utility for HCC screening. HCV etiology, non-White race and serum ALT were independently associated with elevated AFP in patients with cirrhosis, whereas non-White race and ALT were independent predictors of elevated AFP in patients with early-stage HCC. The performance of serum AFP was heavily influenced by serum ALT in HCV patients. After excluding HCV patients with ALT >40 U/L, the AUC of AFP for HCC increased from 0.76 to 0.91 in Whites with HCV and from 0.79 to 0.89 in non-Whites with HCV (Table 4). At 90% specificity, the sensitivity of AFP for HCC increased from 44% to 74% in Whites with HCV and from 58% to 85% in non-Whites with HCV (Supplementary Table S1). At a fixed sensitivity of 70%, the specificity of AFP for HCC increased from 69% to 96% in Whites with HCV and from 74% to 93% in non-Whites with HCV (Supplementary Table S1). The cross-validation test confirmed stable AUCs of AFP for HCC detection in each subgroup, and limited external validation again showed a trend toward improved performance of the AFP for HCC detection in HCV patients with normal ALT (Supplementary Fig. S2).

The utility of serum AFP as a screening test for use in HCC surveillance has been controversial, due to its reported low sensitivity for detection of early stage HCC (8, 10, 11, 15, 16). Cost effectiveness analysis showed that semiannual AFP and US is the most effective strategy in reducing HCC mortality with 46% reduction in HCC mortality (17). A single-center prospective cohort study of 446 cirrhosis patients showed that US and AFP had sensitivities of 44% and 66% and specificities of 92% and 91%, respectively, for the detection of HCC. The sensitivity significantly improved to 90%, with a minimal decrease in specificity to 83% when AFP and US were used in combination, suggesting a benefit of the combination of US and AFP as HCC surveillance tests (18). A more recent study from Taiwan clearly demonstrated the benefit of AFP in HCC surveillance (9). In this large retrospective cohort study of 1,597 Taiwanese cirrhosis patients, US had a sensitivity and specificity of 92.0% and 74.2%, respectively. Using an AFP cutoff of 20 ng/mL or an AFP level increase of ≥2-fold from its nadir during the previous 1 year as a trigger for cross-sectional imaging, the combination of US and AFP improved the overall sensitivity of surveillance from 92% to 99.2%, with a minimal decrease in specificity from 74.2% to 71.5%, supporting the use of AFP as an adjunct to US for HCC surveillance. When stratified by etiology, the area under the ROC curve was highest in patients with HBV-induced HCC, suggesting that AFP performs best in patients with HBV. A large population-based study from Alaska also showed that serial AFP testing in patients with chronic HBV infection was associated with earlier diagnosis of HCC, a higher likelihood of resection, and improved overall survival (11).

Prior studies have also shown worse performance of AFP as a screening test for HCC in patients with HCV. A multicenter retrospective study showed that approximately 20% of individuals with HCV cirrhosis had an AFP ≥20 ng/mL in the absence of HCC (19). In the prospective Hepatitis C Antiviral Long-Term Treatment against Cirrhosis (HALT-C) study of HCV patients with advanced fibrosis or cirrhosis, a significant proportion of the participants had falsely elevated AFP in the absence of HCC; 27% of subjects with HCV cirrhosis had a serum AFP ≥20 ng/dL at enrolment (20). Interestingly, the HALT-C study results showed that as the HCV RNA level declined with antiviral treatment, the serum AFP concentration also declined, suggesting that active replication of HCV virus and inflammation of the liver contribute to the false-positive elevation of AFP levels (20). The satisfactory performance of serum AFP in the subset of HCV patients with normal ALT in the current study is in line with the evidence reported in the literature (21, 22). As more and more patients with HCV cirrhosis achieve sustained virologic response in the era of highly potent directly acting antiviral (DAA) therapy, serum AFP may prove to be an excellent surveillance test in this increasing subgroup of patients who remained at high risk of developing HCC (23, 24).

The association between race and the performance of AFP in surveillance has not been evaluated rigorously in the literature. Several studies have reported higher frequencies of falsely elevated AFP levels in African American patients with cirrhosis, primarily patients with HCV cirrhosis (14, 19, 20). Whether there is an independent association between African American race and false-positive elevation of AFP levels or the apparent association was confounded by other factors, such as virologic features or activity of hepatitis, remains to be determined. The small number of African American subjects (7%) in our study precluded us from evaluating the performance of AFP in African Americans.

The association between elevation of AFP and severity of hepatic inflammation has been well reported, but the underlying mechanism of such association is poorly understood (25). The association between elevated ALT and AFP in the absence of HCC suggests that AFP production increases in the presence of enhanced hepatocyte destruction and regeneration of liver progenitor cells with a less differentiated phenotype (26). However, this association seems to be etiology and race specific given that both race and etiology are important determinants of elevated AFP even after controlling for serum ALT. Underlying mechanisms connecting severity of hepatic inflammation, non-White race, viral etiology of liver disease, and elevated AFP levels should be further investigated in future studies.

Unique features of our study are the focus on identifying factors associated with elevated AFP in patients with cirrhosis versus early-stage HCC. We also specifically evaluated the performance of AFP as a screening test in different etiologic and racial subgroups, allowing comparison between the subgroups. Our results strongly suggest that AFP is a good test for early detection of HCC in patients with HCV cirrhosis with minimal hepatic inflammation.

Our study has several limitations. This is a retrospective phase II biomarker case–control study that could have been affected by unmeasured potential biases. Evaluating the performance of AFP for HCC in the presence of known HCC may not be the same as evaluating the performance of AFP for HCC detection in a prospective cohort of patients with cirrhosis. A larger phase III prospective multicenter cohort biomarker study is now underway to validate the current findings (27). This ongoing biomarker study will be able to address whether baseline serum AFP can predict subsequent development of HCC on imaging. Because of sample size limitations, subgroup analysis results might not be as reliable, particularly in non-HCV subgroups. Sample sizes became smaller when further broken down by race or ALT, which has limited the power of statistical analysis. Similarly, it was hard to interpret data on the lack of improvement of ROC in non-HCV subgroup patients with normal ALT and this should be further evaluated in future studies. It is important to note that the HCV cohort in the current study was established before highly potent DAA treatments became available. Therefore, the proportion of patients with normal ALT was lowest in the HCV subgroup. More recently, a large proportion of HCV cirrhosis patients are recipients of DAA treatment. Thus, it is likely that the proportion of HCV cirrhosis patients with normal ALT will further increase with time, and the good performance of AFP for HCC detection in HCV patients with normal ALT will be more generalizable.

Although the normal serum ALT cutoff is somewhat controversial, the normal ALT cutoff was set at 40 U/L in the current study due to the small sample size, as setting up a normal ALT cutoff at 20 or 30 U/L did not provide large enough sample sizes for subgroup analyses. Our study was also not able to assess the performance of longitudinal trends in the serum AFP. As shown in previous studies and applied in clinical practice, we anticipate that assessing longitudinal trends in AFP levels would further improve the sensitivity and specificity of AFP compared with the levels reported in the current study (9, 28–30). Finally, our study does not have information on the performance of US in HCC screening. Thus, the results of the current study cannot address whether AFP is of additional benefit to US in HCC surveillance. However, our study shows that AFP may perform better than the historically reported performance of US for the diagnosis of HCC, particularly in subjects with HCV cirrhosis and normal serum ALT.

In conclusion, our study suggests that race, etiology of HCC, and severity of hepatic inflammation are associated with the elevation of AFP in patients with cirrhosis with/without HCC. The good performance of serum AFP for early-stage HCC diagnosis in HCV patients with normal ALT is clinically relevant in the era of highly potent DAA therapy. Given the relatively small sample size of the HCV patients with normal ALT and the retrospective study design, the performance of AFP in HCV with normal ALT should be further validated in large prospective cohort studies.

A.G. Singal has received speakers bureau honoraria from Bayer and is a consultant/advisory board member for Bayer and Wako Diagnostics. No potential conflicts of interest were disclosed by the other authors.

The contents of this article are solely the responsibility of the authors and do not necessarily represent the official views of the NIH.

Conception and design: J.D. Yang, B. Addissie, M.H. Nguyen, A.S. Befeler, M. Schwartz, Z. Feng, J.A. Marrero, L.R. Roberts

Development of methodology: J.D. Yang, J. Dai, B. Addissie, M.H. Nguyen, M. Schwartz, H. Yamada, J.A. Marrero, L.R. Roberts

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): A.G. Singal, B. Addissie, M.H. Nguyen, A.S. Befeler, K.R. Reddy, M. Schwartz, D.M. Harnois, H. Yamada, L.R. Roberts

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): J.D. Yang, J. Dai, A.G. Singal, B. Addissie, M.H. Nguyen, K.R. Reddy, Z. Feng, J.A. Marrero, L.R. Roberts

Writing, review, and/or revision of the manuscript: J.D. Yang, A.G. Singal, P. Gopal, B. Addissie, M.H. Nguyen, A.S. Befeler, K.R. Reddy, M. Schwartz, D.M. Harnois, H. Yamada, J.A. Marrero, L.R. Roberts

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): Z. Feng, L.R. Roberts

Study supervision: G.J. Gores, Z. Feng, L.R. Roberts

Other (review and revision of the manuscript): J. Dai

Other [measured all of the study samples and provided the biomarker's (serum alpha-fetoprotein) values]: H. Yamada

We appreciate generous support from American Liver Foundation: 2016 Hans Popper Memorial Postdoctoral Research Fellowship award (to J. Yang).

This publication was supported by grant number T32 DK07198 from the National Institute of Diabetes and Digestive and Kidney Diseases (NIDDK; to J. Yang) and CA165076 from the NCI (to L.R. Roberts).

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.