Advertisement

HCC prediction post SVR: Many tools yet limited generalizability!

      Linked Article

      To the Editor:
      Despite attaining a sustained virological response (SVR), the risk of hepatocellular carcinoma (HCC) remains a significant concern in patients with chronic hepatitis C (CHC). The EASL guidelines advise HCC screening in a population with a high incidence of HCC, considering cost, expertise, treatment options, and rate of tumor growth.
      EASL Clinical Practice Guidelines
      Management of hepatocellular carcinoma.
      Accordingly, HCC screening is recommended in patients with CHC and >F3 fibrosis. Several prediction tools have been applied in various studies for HCC prediction; however, none is generalizable to the global population to date. Recently, novel HCC prediction models based on artificial intelligence have shown the superiority of deep learning models over traditional statistical models. However, these models are not restricted to patients who attain SVR and are yet to be validated
      • D’Ambrosio R.
      • Degasperi E.
      • Lampertico P.
      Predicting hepatocellular carcinoma risk in patients with chronic HCV infection and a sustained virological response to direct-acting antivirals.
      ,
      • Ahn J.C.
      • Qureshi T.A.
      • Singal A.G.
      • Li D.
      • Yang J.D.
      Deep learning in hepatocellular carcinoma: current status and future perspectives.
      (Table 1). Semmler et al., in a recent study, developed a simple bedside score incorporating post SVR variables such as age, albumin, liver stiffness measurement (LSM), alpha-fetoprotein, and alcohol consumption to stratify future risk of HCC among patients with CHC and compensated advanced chronic liver disease.
      • Semmler G.
      • Meyer E.L.
      • Kozbial K.
      • Schwabl P.
      • Hametner-Schreil S.
      • Zanetto A.
      • et al.
      HCC risk stratification after cure of hepatitis C in patients with compensated advanced chronic liver disease.
      They suggested avoiding HCC screening in low-risk (<1% person-year) individuals; however, certain issues in the study merit further discussion.
      Table 1Studies reporting predictive scores for HCC.
      • D’Ambrosio R.
      • Degasperi E.
      • Lampertico P.
      Predicting hepatocellular carcinoma risk in patients with chronic HCV infection and a sustained virological response to direct-acting antivirals.
      ,
      • Ahn J.C.
      • Qureshi T.A.
      • Singal A.G.
      • Li D.
      • Yang J.D.
      Deep learning in hepatocellular carcinoma: current status and future perspectives.
      StudyPatient population, fibrosis stageScore nameVariables includedAlgorithmRisk classesHCC rate according to risk classes
      Abe, 2020CHC, F4, Post SVRNAALBI score, platelet, diabetes status0 or 1 points for each:

      ALBI score ≤ or >2-3, platelet ≥ or < 8.2 × 104/μL, absence or presence of diabetes
      0–1: Low-score

      2–3: High-score
      Low- vs. high-score group

      0.7% vs. 12.5% at 1 year

      2.2% vs. 15.2% at 2 years

      3.1% vs. 33.9% at 3 years

      3.1% vs. 41.2% at 4 years
      Fan, 2020CHC, F4, Post SVRaMAPAge, gender bilirubin albumin, plateletMathematical formula<50: Low-risk

      50–60: Intermediate risk

      >60: High-risk
      Low vs. intermediate vs. high-risk

      0–0.8% vs. 1.5–4.8% vs. 8.1–17.8% at 3–5 year
      Shiha, 2020CHC, F3-F4, Post SVRGESAge, male gender, AFP, albumin, fibrosis0 points to 3.5 points:

      Female or Male, Age ≤ or > 54 years, Albumin ≥ or <3.8 g/dl, AFP ≤ or >20 ng/mL

      F3 or F4
      GES ≤6: Low Risk

      GES 6–7.5: Intermediate Risk

      GES >7.5: High Risk
      Low vs. intermediate vs. high-risk

      0.1% vs. 0.7% vs. 1.2% at 1 year

      1.2% vs. 3.3% vs. 7.1% at 2 years

      1.9% vs. 5.8% vs. 9.5% at 3 years
      Alonso-Lopez, 2020CHC, F3-F4, Post SVRNALSM model:

      Albumin, LSM, SVR48 ΔLSM

      FIB-4 model: Albumin, FIB-4, SVR48 FIB-4, SVR48 GGT
      LSM model (0 or 1 points): Albumin ≥ or < 4.2 g/dl, LSM ≤ or > 17.3 kPa, ΔLSM ≥ or <25.5%

      FIB-4 model (0 to 2 points): Albumin≥ or <4.2 g/dl, FIB-4 ≤ or >3.7, SVR48 FIB-4 ≤ or >3.3, SVR48 GGT ≤ or >42 U/l
      LSM model:

      Score 0-1-2-3

      FIB-4 model:

      Score 1–2 vs. 3–4 vs. 5–6
      LSM model

      Score 0 vs. 1 vs. 2 vs. 3

      0% vs. 2.1% vs. 5.8% vs. 16.3% at 3 years

      FIB-4 model

      Score 1–2 vs. 3–4 vs. 5–6

      0.4% vs. 1.7% vs. 6.5 vs. 19% at 3 years
      Watanabe, 2019CHC, Post SVRNAPre-DAA model: FIB-4, albumin, gender

      Post-DAA model: end of treatment FIB-4, AFP
      Pre-DAA model: 0 or 1 points for FIB-4 < or ≥4.0, albumin > or ≤3.8 g/dl, female or male

      Post-DAA model: 0 or 1 points for FIB-4 < or ≥4.0, AFP< or ≥6.0 ng/mL
      Pre-DAA model: 0: Low Risk, 1–2: Intermediate Risk, 3: High Risk

      Post-DAA model: 0: Low, 1: Intermediate, 2: High
      Pre-DAA model

      Low vs. intermediate vs. high risk

      0.4% vs. 2.1% vs. 9.5% at 1 year

      0.4% vs. 4.4% vs. 16.4% at 2 years

      Post-DAA model

      Low vs. intermediate vs. high risk

      0.4% vs. 1.4% vs. 6.1% at 1 year

      0.4% vs. 3.2% vs. 14.4% at 2 years
      Hiraoka, 2019CHC, Post SVRADRESGender, SVR24, FIB-4, SVR24-AFP1 point to each variable: male, FIB-4 >3.25, AFP >5 ng/mLADRES 0-1-2-3ADRES 0 vs. 1 vs. 2 vs. 3

      0% vs. 0.5% vs. 8.4% vs. 18% at 1 year

      0% vs. 1.6% vs. 13.4% vs. 32.8% at 2 years
      Iio, 2019CHC, Post SVRNASVR24-AFP, SVR24-FIB-4, TLL1 AA/TT1 point to each variable: AFP >4.6 ng/mL, FIB-4 >2.67, TLL1 AA/TT0: Low Risk, 1–2: Intermediate Risk, 3: High RiskLow vs. intermediate vs. high risk

      0% vs. 2.2% vs. 10.4% at 1 year

      0% vs. 3.0% vs. 13.6% at 2 years
      Tani, 2020CHC, Post SVR+NAEnd of treatment: Age, AFP0 to 1 point:

      Age < or ≥75 years, AFP < or ≥6 ng/mL
      Score 0-1-2Score 0 vs. 1 vs. 2

      0.3% vs. 1.05% vs. 4.92% at 1 year

      0.3% vs. 6.27% vs. 18.37% at 2 years

      1.26% vs. 10.45% vs. 18.37% at 3 years
      Ioannou, 2020
      Denotes deep learning models using artificial intelligence.
      CHC Cirrhosis, Both SVR and Non-SVRRNNClinical variablesRecurrent neural network-RNN predicted HCC development with AUC of 0.759, and AUC of 0.806 among SVR

      achieved patients
      Phan, 2020
      Denotes deep learning models using artificial intelligence.
      CHB and CHCCNNDisease historyConvolutional neural network;-CNN achieved an accuracy of 0.980 and AUC of 0.886 for predicting the development of HCC among viral hepatitis patients
      Nam, 2020
      Denotes deep learning models using artificial intelligence.
      CHB and CHC cirrhosisDLClinical variablesDeep Learning-DL model achieved accuracy of 0.763 and AUC of 0.782 in validation cohort and outperformed the previous models
      AFP, alpha-fetoprotein; CHB, chronic hepatitis B; CHC, chronic hepatitis C; DAA, direct-acting antiviral; HCC, hepatocellular carcinoma; LSM, liver stiffness measurement; SVR, sustained virological response.
      Denotes deep learning models using artificial intelligence.
      Firstly, the selection of cohorts – the derivation cohort included 527 patients from 3 institutions in Europe in whom hepatic venous pressure gradient, LSM, and histopathology were assessed. These patients were derived from subgroups of patients from 6 different studies where HCC detection was not a target outcome. Such a retrospective inclusion may not accurately represent the real-world population of patients with CHC. Further, the validation cohort included 1,500 patients across multiple European centers. However, it is worth noting that the largest share of patients with CHC reside in Asia and Africa (about 78%).
      • Petruzziello A.
      • Marigliano S.
      • Loquercio G.
      • Cozzolino A.
      • Cacciapuoti C.
      Global epidemiology of hepatitis C virus infection: an up-date of the distribution and circulation of hepatitis C virus genotypes.
      Whether the findings generalize to such populations or if the score needs further fine-tuning is still questionable.
      Secondly, patients with CHC are known to have a progressive reduction in inflammation and fibrosis, including LSM, till 24-96 weeks post-SVR.
      • Pietsch V.
      • Deterding K.
      • Attia D.
      • Ringe K.I.
      • Heidrich B.
      • Cornberg M.
      • et al.
      Long-term changes in liver elasticity in hepatitis C virus-infected patients with sustained virologic response after treatment with direct-acting antivirals.
      In the present study, the authors included LSM at 12 weeks post-treatment in the derivation cohort and 48 weeks post-treatment in the validation cohort. Indeed, they represent 2 different time frames with varying values of LSM and relate to dynamic changes in liver inflammation, fibrosis, and normalization of liver functions.
      Thirdly, the authors provided a simple score-based approach to stratify the risk of HCC as low or high. However, to generalize this score to a large population, the performance characteristics of the final score in terms of discrimination (c-index), sensitivity, specificity, and accuracy are highly desired for policy decisions. It is also unclear whether the score is biased toward better categorizing patients without HCC due to class imbalance problems in the derivation cohort.
      • Ling C.X.
      • Sheng V.S.
      Class imbalance problem.
      Fourthly, the authors based their research discussion on the proposition of <50,000 USD per life-year saved as a cost-effective approach.
      EASL Clinical Practice Guidelines
      Management of hepatocellular carcinoma.
      However, we feel that the cost of testing and surveillance varies significantly across the world. The cost-effectiveness thresholds derived from developed countries for a disease that is much more prevalent in developing and under-developed countries with different economic standards, remain biased and merit further research.
      Fifthly, the points given for variables in the final score were derived from sub-distribution hazard ratios (SHRs) of a competing risk model with death and liver transplant as competing events and the development of HCC as a target event. These individual points were added to obtain the final risk prediction score. However, adding such SHRs has been identified as mathematically incorrect and misleading in prediction models. Literature suggests a mathematical addition of the coefficients of variables rather than exponential of coefficients (i.e., SHR) in the model as an appropriate approach.
      • Moons K.G.M.
      • Harrell F.E.
      • Steyerberg E.W.
      Should scoring rules be based on odds ratios or regression coefficients?.
      Finally, the study provides an easy single-time prediction score for risk stratification of HCC in a limited setting. Further multi-ethnic studies are warranted for HCC prediction with better generalizability.

      Financial support

      The authors received no financial support to produce this manuscript.

      Authors’ contributions

      NB – Writing – original draft, NV – Conceptualization, Writing – review and editing, VS – Conceptualization, Writing – review and editing.

      Conflict of interest

      The authors declare no conflicts of interest.
      Please refer to the accompanying ICMJE disclosure forms for further details.

      Supplementary data

      The following are the supplementary data to this article:

      References

        • EASL Clinical Practice Guidelines
        Management of hepatocellular carcinoma.
        J Hepatol. 2018; 69: 182-236
        • D’Ambrosio R.
        • Degasperi E.
        • Lampertico P.
        Predicting hepatocellular carcinoma risk in patients with chronic HCV infection and a sustained virological response to direct-acting antivirals.
        J Hepatocell Carcinoma. 2021; 8: 713-739
        • Ahn J.C.
        • Qureshi T.A.
        • Singal A.G.
        • Li D.
        • Yang J.D.
        Deep learning in hepatocellular carcinoma: current status and future perspectives.
        World J Hepatol. 2021 Dec 27; 13: 2039-2051
        • Semmler G.
        • Meyer E.L.
        • Kozbial K.
        • Schwabl P.
        • Hametner-Schreil S.
        • Zanetto A.
        • et al.
        HCC risk stratification after cure of hepatitis C in patients with compensated advanced chronic liver disease.
        J Hepatol. 2022; 76: 812-821
        • Petruzziello A.
        • Marigliano S.
        • Loquercio G.
        • Cozzolino A.
        • Cacciapuoti C.
        Global epidemiology of hepatitis C virus infection: an up-date of the distribution and circulation of hepatitis C virus genotypes.
        World J Gastroenterol. 2016; 22: 7824-7840
        • Pietsch V.
        • Deterding K.
        • Attia D.
        • Ringe K.I.
        • Heidrich B.
        • Cornberg M.
        • et al.
        Long-term changes in liver elasticity in hepatitis C virus-infected patients with sustained virologic response after treatment with direct-acting antivirals.
        United European Gastroenterol J. 2018; 6: 1188-1198
        • Ling C.X.
        • Sheng V.S.
        Class imbalance problem.
        in: Sammut C. Webb G.I. Encyclopedia of machine learning. Springer US, Boston, MA2010 (171-171)
        • Moons K.G.M.
        • Harrell F.E.
        • Steyerberg E.W.
        Should scoring rules be based on odds ratios or regression coefficients?.
        J Clin Epidemiol. 2002; 55: 1054-1055