Comparative evaluation of multimarker algorithms for early-stage HCC detection in multicenter prospective studies

Abstract

Background & aims: We compared the clinical performance of the novel GAAD (gender [biological sex], age, alpha-fetoprotein [AFP], des-gamma carboxyprothrombin [DCP]) and GALAD (gender [biological sex], age, AFP, Lens culinaris agglutinin-reactive AFP [AFP-L3], DCP) algorithms to deduce the clinical utility of AFP-L3 for detecting early-stage hepatocellular carcinoma (HCC) from chronic liver disease (CLD).

Methods: An algorithm development study (STOP-HCC-ARP) and clinical validation study (STOP-HCC-MCE) were conducted, recruiting adult participants with HCC (confirmed by radiology or pathology) or CLD in an international, multicenter, case-control design. Serum biomarkers were measured using Elecsys assays (GAAD and GALAD [Cobas]) or μTASWAKO assays (GALAD [μTASWAKO]) while blinded to case/control status.

Results: In STOP-HCC-ARP (algorithm development cohort), 1,006 patients {297 HCC (41.4% early-stage [Barcelona Clinic Liver Cancer {BCLC} 0/A) and 709 CLD} were included. Area under the curve (AUCs) for discriminating between early-stage HCC vs. CLD were 91.4%, 91.4%, and 90.8% for GAAD (Cobas), GALAD (Cobas), and GALAD (μTASWAKO), respectively. The clinical validation cohort of STOP-HCC-MCE comprised 1,142 patients, (366 HCC cases [48% early-stage], 468 specificity samples and 302 CLD); AUCs for GAAD (Cobas), GALAD (Cobas), and GALAD (μTASWAKO) for discriminating between early-stage HCC vs. CLD were 91.4%, 91.5%, and 91.0%, respectively; AUCs were 94.7-95.0% for all-stage HCC. The GAAD and GALAD algorithms demonstrated similar good performance regardless of disease etiology, presence of cirrhosis, geographical region, and within pan-tumor specificity panels (p <0.001).

Conclusions: GAAD (Cobas) demonstrated good clinical performance, similar to GALAD (Cobas and μTASWAKO) algorithms, in differentiating HCC and CLD controls, across all disease stages, etiologies, and regions; therefore, AFP-L3 may have a negligible role in GALAD for HCC surveillance.

Impact and implications: To improve the detection of early-stage hepatocellular carcinoma (HCC) from benign chronic liver disease (CLD), algorithms combining demographic characteristics and serum biomarkers, such as GAAD and GALAD, have been developed. GAAD combines gender (biological sex), age, alpha-fetoprotein (AFP), des-gamma carboxy-prothrombin (DCP); GALAD combines the same characteristics and biomarkers as GAAD with the addition of Lens culinaris agglutinin-reactive AFP (AFP-L3). Changing disease etiologies and treatment paradigms have raised questions regarding the utility of AFP-L3 in HCC surveillance. Our work demonstrates that the GAAD (Cobas) algorithm demonstrated good clinical performance and was as sensitive and specific as the GALAD (Cobas) and GALAD (μTASWAKO) algorithms in differentiating HCC and CLD controls, across all disease stages, etiologies, and geographical regions; therefore, AFP-L3 may have a negligible role in HCC detection. Our study provides supporting evidence that in participants with CLD undergoing guideline-directed HCC surveillance, the GAAD (Cobas) algorithm may be used as an effective method for the detection of HCC, potentially resulting in improved patient outcomes.

Keywords: Algorithm; GAAD; GALAD; Hepatocellular carcinoma; Surveillance.

Graphical abstract

Fig. 1

Development, clinical evaluation, and performance of GAAD (Cobas), GALAD (Cobas), and GALAD (μTASWAKO) algorithmic scores in STOP-HCC-ARP. Algorithm development strategy (A) and study disposition (B); clinical performance in differentiating early-stage (C); late-stage (D) and all-stage HCC (E) from CLD controls. Sensitivities and specificities are shown in table (F). ∗Cut-off value corresponds to matching GAAD (Cobas) specificity of 90%. AUC, area under the curve; BCLC, Barcelona Clinic Liver Cancer; CCA, cholangiocarcinoma; CLD, chronic liver disease; GAAD, gender (biological sex), age, AFP, DCP (PIVKA-II); GALAD, gender (biological sex), age, AFP-L3, AFP, DCP (PIVKA-II); HCC, hepatocellular carcinoma.

Fig. 2

Study design and sample disposition in STOP-HCC-MCE. ^aExcluded because of exclusion criteria (67 owing to renal failure, 17 owing to ICF issues, and 124 owing to lab parameters, sample processing, other cancer/missing diagnosis). ^bExcluded because of interferences with assays. ^cOne non-HCC control subject was included for GAAD evaluation but excluded for GALAD evaluation due to bilirubin interference with AFP-L3 assay. AFP-L3, Lens culinaris agglutinin-reactive alpha-fetoprotein; E/I, exclusion/inclusion; GAAD, gender (biological sex), age, AFP, DCP (PIVKA-II); GALAD, gender (biological sex), age, AFP-L3, AFP, DCP (PIVKA-II); HCC, hepatocellular carcinoma; ICF, International Classification of Functioning, Disability and Health.

Fig. 3

Weighted Deming regression fit and specificty panel data in STOP-HCC-MCE. Weighted Deming regression fit of agreement between GAAD (Cobas) and GALAD (μTASWAKO), GALAD (Cobas) and GALAD (μTASWAKO) in STOP-HCC-MCE (A), specificity panel data across disease groups for GAAD (Cobas) (B), GALAD (Cobas) (C), and GALAD (μTASWAKO) (D) in STOP-HCC-MCE. Levels of significance: ∗∗p <0.01. (p values were calculated using binomial tests, Bonferroni correction for every marker, and specificity values with the corresponding cut-offs for the markers: GAAD [2.57, 90%], GALAD [2.47, 90%], WAKO GALAD [−1.89, 90%].) ALD, alcoholic-related liver disease; GAAD, gender (biological sex), age, AFP, DCP (PIVKA-II); GALAD, gender (biological sex), age, AFP-L3, AFP, DCP (PIVKA-II).

Fig. 4

Distribution of GAAD (Cobas) and GALAD (Cobas) scores. Distribution by HCC and CLD controls (A), BCLC (B), etiology (C), and geographical region (D) in STOP-HCC-MCE. Levels of significance: ∗∗∗p <0.001 compared with control (Welch’s t test). ALD, alcoholic-related liver disease; BCLC, Barcelona Clinic Liver Cancer; CLD, chronic liver disease; CLD, chronic liver disease; GAAD, gender (biological sex), age, AFP, DCP (PIVKA-II); GALAD, gender (biological sex), age, AFP-L3, AFP, DCP (PIVKA-II); HBV, hepatitis B virus; HCC, hepatocellular carcinoma; HCV, hepatitis C virus; NASH, non-alcoholic steatohepatitis.

Fig. 5

Clinical performance of individual Elecsys assays, AFP, PIVKA-II, and AFP-L3 and algorithmic scores, GAAD (Cobas), GALAD (Cobas), and GALAD (μTASWAKO) in STOP-HCC-MCE. Clinical performance in differentiating early-stage (A); late-stage (B); and all-stage HCC (C) from CLD controls. Levels of significance: early-stage HCC: GAAD (Cobas) vs. GALAD (Cobas) p <0.001; GAAD (Cobas) vs. GALAD (μTASWAKO) p = 0.0202; GALAD (Cobas) vs. GALAD (μTASWAKO) p = 0.0166; late- and all-stage HCC: p <0.001 (p values comparing non-inferiority of AUCs were calculated using the H0 hypothesis AUC (Test 1) < AUC (Test 2) - 0.01 with a studentized bootstrap approach using N = 1,000 bootstrap replicates.) Sensitivities and specificities are shown in table (D). ^aCut-off value corresponds to matching GAAD (Cobas) specificity of 90%. ∗AUC is not significantly worse by 1%. AFP, alpha-fetoprotein; AFP-L3, Lens culinaris agglutinin-reactive AFP; AUC, area under the curve; BCLC, Barcelona Clinic Liver Cancer; CLD, chronic liver disease; GAAD, gender (biological sex), age, AFP, DCP (PIVKA-II); GALAD, gender (biological sex), age, AFP-L3, AFP, DCP (PIVKA-II); HCC, hepatocellular carcinoma; PIVKA-II, protein induced by vitamin K absence or antagonist-II.

Fig. 6

ROC curves of GAAD (Cobas), GALAD (Cobas) scores for discriminating between HCC and CLD controls in STOP-HCC-MCE. Overall and by etiology group in early-stage, late-stage and all-stage HCC (A); and by geographical region in early-stage, late-stage and all-stage HCC (B). AUC, area under the curve; CLD, chronic liver disease; GAAD, gender (biological sex), age, AFP, DCP (PIVKA-II); GALAD, gender (biological sex), age, AFP-L3, AFP, DCP (PIVKA-II); HCC, hepatocellular carcinoma.