Prediction and prophylaxis of hepatocellular carcinoma occurrence and postoperative recurrence in chronic hepatitis B virus-infected subjects

Table 1 Hepatocellular carcinoma risk prediction scores using data from hepatitis B virus-infected cohorts

Ref.	Population and follow-up	Parameters	Validity and reliability(AUROC, sensitivity, specificity)	Advantages and disadvantages
Yuen et al[17], 2009 Hong Kong	n = 820, mean 76.8 mo	Age, sex, HBV DNA, core promoter mutations, cirrhosis	Leave-one-out cross-validation	Highlighted the importance of prevention and treatment of remediable cirrhosis with antiviral treatment
			Optimal cut-off was 101 for 5-yr (sensitivity = 87.9% and specificity = 76.2%) and 10-yr (sensitivity = 100% and specificity = 79.1%) development of HCC
				Developed from a hospital population without external validation, thus may have limited generalizability. HBV mutations may not be easily available in some clinics
Wong et al[21], 2010 Hong Kong	n = 1005 (training cohort, median 9.94 yr); n = 424 (validation cohort, median 10.53 yr)	Age, albumin, bilirubin, HBV DNA, cirrhosis	Internal validation	Included commonly measured parameters in the score, which could facilitate its routine clinical implementation
			The score ranged from 0 to 44.5, with the best discriminatory cutoff points of 5 and 20 to categorize patients into the low-, medium-, and high-risk groups. Sensitivity and NPV: 88.6% and 97.8% in the training cohort; 82.2% and 97.3% in the validation cohort
				Derived from a hospital-based cohort with internal validation, thus cannot be readily applied to populations with very different characteristics
Wong et al[22], 2014 Hong Kong,	n = 1035 (training cohort); n = 520 (validation cohort). mean 69 mo	Age, LSM, albumin, HBV DNA	Internal validation	Indicated the importance of complete viral suppression in the clinical management.
			The score ranged from 0-30, best cut-off value was 11. For 5-yr HCC risk prediction, the sensitivity was 87.9%, and NPV was 99.4% in the training cohort. In the validation cohort, sensitivity was 100% and NPV was 100% at year 3, and sensitivity was 92.3% and NPV was 99.7% at year 5
				Did not measure LSM in the follow-up, and had relatively small number of events. There was no external validation, thus the results cannot be generalized to other populations with different characteristics
			AUROC was 0.89 at 3 yr and 0.83 at 5 yr in the validation cohort
Yang et al[23], 2010 Taiwan	n = 2435 (model derivation); n = 1218 (model validation). About 12 yr (REVEL-HBV cohort)	All three models included: sex, age in 5-yr increments, family history of HCC, alcohol consumption, ALT.	Internal validation	Risk profiling allows accurate estimation of future HCC risk and appropriate recognition of patients with several seemingly marginal risk factors but may overall have high risk of HCC occurrence and thus need clinical awareness. Developed prediction models for various clinical settings with reasonably satisfactory performances
			The range of score was 0 to 17 in model 1, 0 to 20 in model 2 and model 3
			The discrimination capability was satisfactory (AUROC > 80%). The calibration capability was excellent (correlation coefficients between observed risk and estimated mean predicted risk greater than 0.9)
		Model 1: + HBeAg; Model 2: HBeAg + HBV DNA; Model 3: HBeAg + HBV DNA + HBV genotype
				Need extra caution when applied to patients with different characteristics
Yang et al[24], 2011 Taiwan	n = 3584 (development cohort, median 12 yr, REVEL-HBV cohort); n = 1505 (validation cohort, mean 7.3 yr)	Age, sex, ALT, HBeAg, HBV DNA	External validation (Hong Kong, South Korea)	This risk score can be used to guide treatment for the group of patients who do not meet existing treatment initiation recommendation by focusing on the long-term outcome. It allows the determination of risk at different time points. It provides a possibility to assess the change in risk after therapy initiation
			A 17-point risk score. AUROCs were 81.1% at 3 yr, 79.6% at 5 yr, and 76.9% at 10 yr in the validation cohort; and 90.2%, 78.3%, and 80.6%, respectively in the validation cohort excluding cirrhosis patients. The predicted risk was well calibrated with observed risk, with a correlation coefficient of 0.975 at 3 yr, 0.991 at 5 yr, and 0.999 at 10 yr in the non-cirrhotic model
				This risk score might be inappropriate in immune-tolerant patients, or those with ALT flares, and patients with evidence of cirrhosis
Lin et al[25], 2013 Taiwan	n = 1822, median 5.9 yr (REVEL-HBV cohort)	Model 1: Age, sex, ALT, HBV seromarkers	The best cutoff score: 6 for Model 1, 22 for Model 2, 19 for Model 3. Sensitivity and specificity: 0.93 and 0.57 for Model 1, 0.70 and 0.93 for Model 2, 0.80 and 0.88 for Model 3.	The findings suggest interventions to reverse fibrosis and cirrhosis are also important to reduce HCC risk. Provide an estimate of 6-yr HCC risk, allowing clinicians give clinical advice.
		Model 2: Age, sex, ALT, AAR, ALT, AFP, GGT, albumin, alpha-1 globulin
			For the 6-yr prediction, the AUROC, best Youden index, LR+, LR- were 0.83, 0.50, 2.17, and 0.11 for Model 1; 0.89, 0.63, 10.38, and 0.33 for Model 2; 0.91, 0.68, 6.52, 0.22 for Model 3	Application in patients infected in adulthood, younger chronic HBV carriers, infected with other HBV genotypes (not B/C) needs further validation
		Model 3: Model 2+ HBV seromarkers
Lee et al[26], 2013 Taiwan	n = 3340 (derivation:validation = 2:1), 53551 person-years	Age, sex, ALT, HBeAg, HBV DNA, HBsAg, HBV genotype, family history of HCC	Internal validation	Incorporating both baseline and follow-up values into the model may increase the predictability, but may not be practical at the one-shot clinical consultation. The generalizability to younger and older patients needs further evaluation.
			Risk score < 9 (low-risk), 9-12 (medium-risk), ≥ 13 (high-risk). The AUROC was 0.89, 0.85, and 0.86 for the 5-yr, 10-yr, 15-yr predicted risk in the derivation set and 0.84, 0.86, and 0.87 for the 5-yr, 10-yr, 15-yr predicted risk.
	(revel-HBV cohort)
Kim et al[27], 2013 South Korea	n = 1110, median 30.7 mo	Age, sex, liver stiffness, HBV DNA	Bootstrap to assess discrimination	Included simple, not exhaustive, non-invasive, obtainable and objective variables. Adjusted the influence of antiviral treatment.
			Average AUROC = 0.802 (95%CI: 0.791-0.812). Correlation coefficient of predicted and observed risk = 0.905
				The use of transient elastography is limited. No long-term follow-up and no external validation

AAR: AST/ALT ratio; AFP: Alpha-fetoprotein; ALT: Alanine aminotransferase; AST: Aspartate aminotransferase; AUROC: Areas under receiver operating curve; GGT: Gamma-glutamyl transferase; HBeAg: Hepatitis B e antigen; HBsAg: HBV surface antigen; HBV: Hepatitis B virus; HCC: Hepatocellular carcinoma; LSM: Liver stiffness measurements; NPV: Negative predictive value.