Circulating Methylated SEPT9 for Detection of Hepatocellular Carcinoma in Cirrhosis Authors: Abderrahim Oussalah, Maël Silva Rodriguez, Guillaume Conroy, Mouni Bensenane, Hélène Rousseau, Ahmet Ayav, Jean-Louis Guéant, Houda Camara, Cédric Baumann, Valérie Laurent, Jean-Pierre Bronowicki Published: 2026 ABSTRACT Key PointsQuestionDoes circulating methylated septin 9 (SEPT9) improve detection of hepatocellular carcinoma when combined with α-fetoprotein in patients with cirrhosis undergoing surveillance?FindingsIn this diagnostic study including 574 patients with cirrhosis, the combination of at least single-positive methylated SEPT9 or α-fetoprotein greater than 20 ng/mL achieved 87.8% sensitivity compared with 44.3% for α-fetoprotein alone, a significant difference, and recovered 78% of cancers missed by α-fetoprotein, with an improvement in early-stage sensitivity.MeaningFindings from this study indicate that a blood-based combination of epigenomic and protein biomarkers may substantially improve hepatocellular carcinoma detection during surveillance, particularly for early-stage disease amenable to curative treatment. Introduction Hepatocellular carcinoma (HCC) is the most common primary liver malignant neoplasm and ranks as the third leading cause of cancer-related death worldwide.1,2,3 In 2022, approximately 866 000 new cases of liver cancer and 759 000 related deaths were recorded globally.1 Major risk factors include cirrhosis from chronic viral hepatitis, alcohol consumption, and metabolic dysfunction-associated steatohepatitis.4 Surveillance in patients with cirrhosis has been associated with earlier tumor detection and improved survival.5 Current guidelines from the American Association for the Study of Liver Diseases (AASLD) and the European Association for the Study of the Liver (EASL) recommend surveillance with abdominal ultrasonography every 6 months, with or without measurement of α-fetoprotein (AFP), to detect HCC at stages amenable to curative treatment.2,5 However, early-stage sensitivity remains limited, reaching only 63% when ultrasonography is combined with AFP.2,5,6,7 As a standalone biomarker, AFP achieves a sensitivity of 49% and a specificity of 88%.8 Hypermethylation of the septin 9 (SEPT9) promoter, encoding a tumor suppressor, is an epigenetic driver associated with hepatic carcinogenesis.9,10 In a previous retrospective study, we reported promising diagnostic accuracy of methylated SEPT9, a circulating cell-free DNA–based epigenetic biomarker, for the detection of HCC in patients with cirrhosis.11 In the current study, the primary objective was to evaluate the diagnostic accuracy of plasma methylated SEPT9 compared with AFP for detecting HCC in patients with cirrhosis and to assess whether methylated SEPT9 could enhance or complement the biomarker component (AFP) of current surveillance strategies. Methods Study Design and Participants This blinded, prospective, cross-sectional diagnostic study12,13 was conducted at the University Hospital of Nancy and the Metz-Thionville Hospital Center in France. This study is reported in accordance with the Standards for Reporting of Diagnostic Accuracy (STARD) reporting guideline.13 The study was designed to compare the diagnostic performance of circulating biomarkers, not to evaluate imaging modalities; accordingly, AFP served as the primary comparator. Patients with cirrhosis undergoing routine HCC surveillance with ultrasonography and AFP were prospectively enrolled from February 12, 2018, through October 17, 2024. During routine phlebotomy at the enrollment visit, an additional 20 mL of blood was collected into ethylenediaminetetraacetic acid tubes for methylated SEPT9 analysis. Plasma was separated, aliquoted, and stored at −80 °C until analysis (eMethods in Supplement 1). Inclusion criteria required age of 18 years or older with confirmed cirrhosis from any etiology. Key exclusion criteria included previously treated HCC, other malignant liver tumors, associated cancers within 5 years, hematological malignant neoplasms without time restriction, pregnancy, and hemodialysis. Complete inclusion and exclusion criteria are detailed in eMethods in Supplement 1. This study was approved by the institutional review board (Comité de Protection des Personnes Ile-de-France VI, Groupe Hospitalier Pitié-Salpêtrière). All participants provided written informed consent. This study was registered at ClinicalTrials.gov (NCT03311152). HCC Ascertainment HCC was diagnosed according to AASLD and EASL criteria by adjudicating hepatologists who were unaware of methylated SEPT9 results.14,15 All patients with HCC underwent contrast-enhanced cross-sectional imaging with computed tomography or magnetic resonance imaging within 3 months preceding or following the enrollment visit. Lesions were classified using the Liver Imaging Reporting and Data System (LI-RADS).16 All imaging studies were reviewed centrally by an expert abdominal radiologist (V.L.) blinded to biomarker results. Diagnosis was established on the basis of characteristic imaging findings (LI-RADS category 5) or histopathologic confirmation when biopsy was clinically indicated.17 All confirmed HCC diagnoses were discussed at multidisciplinary tumor board meetings. Patients with cirrhosis undergoing standard surveillance with no evidence of HCC at enrollment were classified as participants without HCC. Among participants without HCC at enrollment, absence of HCC at 12 months was ascertained through medical record review to minimize misclassification from subclinical disease undetected at enrollment and to enable prespecified composite outcome and sensitivity analyses. Methylated SEPT9 Assay Plasma samples were analyzed for methylated SEPT9 using the Epi proColon 2.0 CE test (Epigenomics AG) or its rebranded equivalent, the ColoHealth test (New Day Diagnostics).11,18 The assay uses bisulfite conversion followed by real-time polymerase chain reaction. Each patient sample was divided into 3 independent aliquots analyzed separately, with each scored as positive or negative. Laboratory personnel who performed and interpreted the methylated SEPT9 assay were blinded to clinical information and to reference standard results. These biological replicates capture stochastic sampling of circulating tumor DNA molecules, with the number of positive replicates (0 to 3) serving as a biological dose indicator reflecting circulating tumor DNA burden.19,20 Samples were classified by number of positive replicates: negative, single-positive, double-positive, or triple-positive. The biological validity of this replicate design was assessed by testing the correlation between replicate positivity and mean cycle threshold values from the methylated SEPT9 assay (eMethods in Supplement 1). Diagnostic Algorithms Nine diagnostic algorithms were evaluated, comprising individual biomarker assessments and combination strategies. For individual assessments, methylated SEPT9 was evaluated as a continuous variable based on the number of positive replicates (range, 0 to 3)11 and at 3 binary positivity thresholds: single-positive (at least 1 positive replicate), double-positive (at least 2 positive replicates), and triple-positive (all 3 replicates positive). AFP was evaluated at a fixed threshold greater than 20 ng/mL. Combination strategies integrated both biomarkers. Tier 1 strategies used disjunction logic, classifying patients as having positive test results if either biomarker exceeded its threshold, thereby maximizing sensitivity. Three tier 1 algorithms were generated by varying the methylated SEPT9 threshold: tier 1a required at least single-positive methylated SEPT9 or AFP greater than 20 ng/mL; tier 1b required at least double-positive methylated SEPT9 or AFP greater than 20 ng/mL; and tier 1c required triple-positive methylated SEPT9 or AFP greater than 20 ng/mL. The tier 2 strategy used conjunction logic, requiring both triple-positive methylated SEPT9 and AFP greater than 20 ng/mL, thereby maximizing specificity. Outcomes and End Points The primary outcome was the presence of HCC at enrollment. In a sensitivity analysis, participants without HCC at enrollment were limited to those with documented imaging follow-up during the 12 months after enrollment showing no HCC, no suspected HCC, and no indeterminate hepatic nodule (eTable 2 in Supplement 1). A prespecified composite outcome encompassed prevalent HCC at baseline and incident HCC diagnosed within 12 months after enrollment. These additional outcome definitions were prespecified to address potential misclassification from subclinical disease undetected at enrollment. The primary end point was the comparison of diagnostic accuracy between methylated SEPT9, evaluated as a continuous variable (range, 0 to 3 positive replicates), and AFP greater than 20 ng/mL, assessed by area under the receiver operating characteristic curve (AUROC) for detecting HCC. Secondary end points included diagnostic performance stratified by Barcelona Clinic Liver Cancer (BCLC) stage, with prespecified subgroup analyses for very early-stage disease (BCLC stage 0) and early-stage disease (BCLC stage A) vs more advanced stages (BCLC stages B, C, and D).2,4 Exploratory end points included evaluation of combination diagnostic strategies, bayesian estimation of posttest probabilities after positive and negative test results using likelihood ratios across pretest prevalences ranging from 1% to 99% with Fagan nomogram construction, net reclassification improvement, conditional recovery analysis, development of a risk stratification nomogram combining methylated SEPT9 replicate count and AFP concentration to provide individualized HCC probability estimates, and validation of the biological replicate design through correlation analysis between the number of positive replicates and cycle threshold values (eMethods in Supplement 1). Sample Size Calculation The study was powered to detect a minimum difference of 0.05 in the AUROC between methylated SEPT9 and AFP, assuming a 2-sided α of .05, 80% power, and a ratio of participants with HCC to those without HCC of 1:4, requiring 101 participants with HCC and 404 without HCC (eMethods in Supplement 1). Statistical Analysis AUROC values with 95% CIs were estimated using bootstrap resampling with 10 000 iterations. For the primary end point, pairwise comparison between continuous methylated SEPT9 and AFP greater than 20 ng/mL was performed using the DeLong test.21 P values for individual algorithm performance against the null hypothesis were adjusted using the Benjamini-Hochberg false discovery rate correction. For the continuous methylated SEPT9 variable, the optimal threshold was determined using the Youden index. Sensitivity, specificity, positive predictive value (PPV), negative predictive value (NPV), and likelihood ratios were calculated with bootstrap-derived confidence intervals. Bayesian analyses were performed as sensitivity analyses using Jeffreys prior Beta(0.5, 0.5), with Bayes factors quantifying evidence strength. Missing AFP values (1.9%) were not imputed, and affected patients were excluded from corresponding analyses. Missing laboratory parameters (less than 2%) were addressed using multiple imputation by chained equations with random forest regression; exact missing data rates per variable and imputation methods are detailed in the eMethods in Supplement 1. All hypothesis tests were 2-sided, and statistical significance was defined as P < .05. Continuous variables were compared using the Mann-Whitney U test, and categorical variables using the Pearson χ2 test or the Fisher exact test when any expected cell frequency was less than 5. Full statistical procedures, including the tests applied to the primary, secondary, and exploratory analyses, are detailed in the eMethods in Supplement 1. All analyses were performed using Python version 3.8 (Python Software Foundation) or later with a consistent random seed ensuring reproducibility (eMethods in Supplement 1). Data were analyzed from October 2025 to January 2026. Results Study Population and Baseline Characteristics Of 639 patients with cirrhosis enrolled across participating centers, 65 were excluded because they did not meet inclusion criteria or represented duplicate entries, resulting in 574 patients in the final analysis (eFigure 1 in Supplement 1). Among 574 participants, 160 (27.9%) were female, 414 (72.1%) were male, and the median (IQR) age was 63 (57-70) years. The study population comprised 118 patients with HCC and 456 without HCC at enrollment (Table 1 and eTables 1 to 3 in Supplement 1). The most common causes of cirrhosis were alcohol-related liver disease (315 [54.9%]) and metabolic dysfunction-associated steatohepatitis (190 [33.1%]). The median (IQR) Child-Pugh score was 5 (5-6) points, consistent with well-compensated cirrhosis. Among patients with HCC, 51 (43.2%) had very early or early-stage disease (BCLC 0-A). Excluded and included patients did not differ significantly in sex distribution or liver disease etiology (eTable 4 in Supplement 1). During the 12 months after enrollment, 11 of 456 participants (2.4%) without HCC at enrollment had incident HCC, yielding 129 HCC diagnoses for the composite outcome analysis (eResults in Supplement 1). Table 1. Baseline Characteristics of Study ParticipantsCharacteristicTotal, No. (%)Total, No.574Age, median (IQR), y63.0 (57.0-70.0)Sex Female160 (27.9) Male414 (72.1)Comorbidities History of cardiac disease106 (18.5) History of chronic obstructive pulmonary disease49 (8.5) History of kidney insufficiency28 (4.9)Oncological history Colorectal cancer6 (1.0) Cancer other than colorectal cancer24 (4.2)Liver disease etiology Alcohol-related liver disease315 (54.9) MASH190 (33.1) Hepatitis C virus infection122 (21.3) Hepatitis B virus infection29 (5.1) Autoimmune hepatitis12 (2.1) Primary biliary cholangitis12 (2.1) Primary sclerosing cholangitis3 (0.5) Hemochromatosis8 (1.4) Other etiology36 (6.3)Hepatic function and prognosis scores MELD-3.0 score, median (IQR)9.0 (8.0-12.0) Child-Pugh classa A457 (79.6) B96 (16.7) C21 (3.7)Last hepatic imaging diagnosis No HCC374 (65.2) HCC present118 (20.6) Indeterminate hepatic nodule74 (12.9) Suspected HCC requiring close follow-upb8 (1.4)HCC occurrence during the 12 mo following inclusionc Yes11 (2.4) No404 (88.6) Not available21 (4.6)Abbreviations: HCC, hepatocellular carcinoma; MASH, metabolic dysfunction-associated steatohepatitis; MELD, Model for End-Stage Liver Disease. a Child-Pugh classification included class A (compensated cirrhosis), class B (moderately impaired hepatic function), and class C (decompensated cirrhosis). b Lesions suspicious for hepatocellular carcinoma, classified as Liver Imaging Reporting and Data System category LR-4. c Includes 436 participants. The 20 remaining participants (4.4%) had a follow-up imaging diagnosis of indeterminate hepatic nodule (18 [3.9%]) or suspected HCC requiring close follow-up (2 [0.4%]) during the 12-month period following inclusion. These participants were classified as without HCC in the primary and composite outcomes and were excluded from the corresponding sensitivity analyses. Baseline Characteristics of Study Participants Abbreviations: HCC, hepatocellular carcinoma; MASH, metabolic dysfunction-associated steatohepatitis; MELD, Model for End-Stage Liver Disease. Child-Pugh classification included class A (compensated cirrhosis), class B (moderately impaired hepatic function), and class C (decompensated cirrhosis). Lesions suspicious for hepatocellular carcinoma, classified as Liver Imaging Reporting and Data System category LR-4. Includes 436 participants. The 20 remaining participants (4.4%) had a follow-up imaging diagnosis of indeterminate hepatic nodule (18 [3.9%]) or suspected HCC requiring close follow-up (2 [0.4%]) during the 12-month period following inclusion. These participants were classified as without HCC in the primary and composite outcomes and were excluded from the corresponding sensitivity analyses. Diagnostic Performance of Individual Biomarkers The median (IQR) methylated SEPT9 count was 3 (1-3) in patients with HCC vs 0 (0-1) in those without HCC (P < .001) (eTable 5 in Supplement 1). For the primary outcome, the AUROC was 0.79 (95% CI, 0.74-0.84) for methylated SEPT9 assessed as a continuous variable compared with 0.71 (95% CI, 0.66-0.76) for AFP greater than 20 ng/mL (difference, 0.09; 95% CI, 0.03-0.15; P = .002) (Table 2; eTables 6 and 7 and eFigures 2 and 3 in Supplement 1). The optimal threshold for methylated SEPT9 determined by the Youden index was triple-positive (all 3 replicates positive), which achieved 63.6% (95% CI, 54.5-72.4) sensitivity and 88.2% (95% CI, 85.1-91.0) specificity, whereas AFP greater than 20 ng/mL achieved 44.3% (95% CI, 35.3-53.5) sensitivity and 97.5% (95% CI, 96.0-98.9) specificity (Table 2; Figure 1; Figure 2). Specificity increased progressively with higher positivity thresholds: 60.7% (95% CI, 56.3-65.2) for single-positive results, 78.9% (95% CI, 75.1-82.6) for double-positive results, and 88.2% (95% CI, 85.1-91.0) for triple-positive results (Table 2; Figure 1; Figure 2). Negative and positive likelihood ratios, predictive values, and diagnostic odds ratios are reported in Table 2 and eTable 8 and eFigures 4 to 8 in Supplement 1. Bayesian analysis confirmed methylated SEPT9 superiority over AFP for the primary outcome (posterior probability of superiority >0.998; posterior median difference in AUROC, 0.09; 95% credible interval, 0.03-0.15; Bayes factor, 713), with consistent superiority across the other outcome definitions (Bayes factors, 1249 to 4999) and strong concordance between bayesian and frequentist estimates (eTables 7 and 9 and eFigures 9 to 15 in Supplement 1). Subgroup analyses by patient characteristics are provided in eTable 10 in Supplement 1. Table 2. Performance Characteristics of Biomarkers and Diagnostic Algorithms for Hepatocellular Carcinoma Detection in the Primary Outcome AnalysisAlgorithmaAUROC (95% CI)% (95% CI)PLR (95% CI)NLR (95% CI)P valuebSensitivitySpecificityPPVNPVMethylated SEPT9 Continuous0.79 (0.74-0.84)63.6 (54.5-72.4)88.2 (85.1-91.0)58.1 (49.6-66.4)90.3 (87.5-93.1)5.4 (4.1-7.2)0.4 (0.3-0.5)<.001 Single positive0.71 (0.67-0.76)82.2 (75.0-88.9)60.7 (56.3-65.2)35.1 (29.5-40.9)93.0 (89.9-95.7)2.1 (1.8-2.4)0.3 (0.2-0.4)<.001 Double positive0.75 (0.71-0.79)71.2 (62.8-79.3)78.9 (75.1-82.6)46.7 (39.5-53.9)91.4 (88.5-94.1)3.4 (2.8-4.2)0.4 (0.3-0.5)<.001 Triple positive0.76 (0.71-0.81)63.6 (54.5-72.4)88.2 (85.1-91.0)58.1 (49.6-66.4)90.3 (87.5-93.1)5.4 (4.1-7.2)0.4 (0.3-0.5)<.001AFP >20 ng/mL0.71 (0.66-0.76)44.3 (35.3-53.5)97.5 (96.0-98.9)82.3 (72.1-91.2)87.2 (84.2-90.1)18.1 (10.5-39.7)0.6 (0.5-0.7)<.001Tier 1a (methylated SEPT9 single positive or AFP)0.74 (0.70-0.77)87.8 (81.6-93.5)59.8 (55.3-64.4)35.9 (30.3-41.6)95.0 (92.3-97.4)2.2 (1.9-2.5)0.2 (0.1-0.3)<.001Tier 1b (methylated SEPT9 double positive or AFP)0.77 (0.73-0.82)77.4 (69.6-85.0)77.5 (73.5-81.3)46.8 (39.7-54.0)93.0 (90.4-95.5)3.4 (2.8-4.2)0.3 (0.2-0.4)<.001Tier 1c (methylated SEPT9 triple positive or AFP)0.80 (0.76-0.85)74.8 (66.7-82.5)85.9 (82.6-89.1)57.7 (49.7-65.6)93.0 (90.5-95.3)5.3 (4.2-7.0)0.3 (0.2-0.4)<.001Tier 2 (methylated SEPT9 triple positive and AFP)0.67 (0.62-0.71)33.9 (25.4-42.6)99.6 (98.9-100.0)95.1 (87.5-100.0)85.4 (82.3-88.4)76.0 (26.9-181.0)0.7 (0.6-0.7)<.001Abbreviations: AFP, α-fetoprotein; AUROC, area under the receiver operating characteristic curve; HCC, hepatocellular carcinoma; NLR, negative likelihood ratio; NPV, negative predictive value; PLR, positive likelihood ratio; PPV, positive predictive value; SEPT9, septin 9. a Single positive, double positive, and triple positive indicate at least 1 positive replicate, at least 2 positive replicates, or all 3 replicates positive, respectively, of 3 methylated SEPT9 measurements. Tier 1 strategies (disjunction logic) maximize sensitivity for rule-out testing, with variants using single-positive, double-positive, and triple-positive methylated SEPT9 results combined with AFP greater than 20 ng/mL; tier 2 strategy (conjunction logic using triple-positive methylated SEPT9 and AFP greater than 20 ng/mL) maximizes specificity for rule-in testing. Performance metrics are presented with 95% CIs calculated using bootstrap methods (10 000 iterations). PLR indicates how much the odds of disease increase when the test is positive; NLR indicates how much the odds of disease decrease when the test is negative. b P values are for comparison of AUROC against the null hypothesis (AUROC of 0.50) and were adjusted for multiple comparisons using the Benjamini-Hochberg false discovery rate method; 5 algorithmic approaches were evaluated, generating 9 diagnostic algorithms based on threshold variations of methylated SEPT9 and AFP. Sensitivity analyses used stricter outcome definitions to test the robustness of findings. Performance Characteristics of Biomarkers and Diagnostic Algorithms for Hepatocellular Carcinoma Detection in the Primary Outcome Analysis Abbreviations: AFP, α-fetoprotein; AUROC, area under the receiver operating characteristic curve; HCC, hepatocellular carcinoma; NLR, negative likelihood ratio; NPV, negative predictive value; PLR, positive likelihood ratio; PPV, positive predictive value; SEPT9, septin 9. Single positive, double positive, and triple positive indicate at least 1 positive replicate, at least 2 positive replicates, or all 3 replicates positive, respectively, of 3 methylated SEPT9 measurements. Tier 1 strategies (disjunction logic) maximize sensitivity for rule-out testing, with variants using single-positive, double-positive, and triple-positive methylated SEPT9 results combined with AFP greater than 20 ng/mL; tier 2 strategy (conjunction logic using triple-positive methylated SEPT9 and AFP greater than 20 ng/mL) maximizes specificity for rule-in testing. Performance metrics are presented with 95% CIs calculated using bootstrap methods (10 000 iterations). PLR indicates how much the odds of disease increase when the test is positive; NLR indicates how much the odds of disease decrease when the test is negative. P values are for comparison of AUROC against the null hypothesis (AUROC of 0.50) and were adjusted for multiple comparisons using the Benjamini-Hochberg false discovery rate method; 5 algorithmic approaches were evaluated, generating 9 diagnostic algorithms based on threshold variations of methylated SEPT9 and AFP. Sensitivity analyses used stricter outcome definitions to test the robustness of findings. Dot Plot of Sensitivity of Methylated Septin 9 (mSEPT9), α-Fetoprotein (AFP), and Combination Algorithms for Hepatocellular Carcinoma Detection Across Outcome Definitions Dot plot showing sensitivity estimates with 95% CIs for 9 diagnostic algorithms across 4 outcome definitions. Tier 1 strategies used disjunction logic; tier 2 used conjunction logic. Dot Plot of Specificity of Methylated Septin 9 (mSEPT9), α-Fetoprotein (AFP), and Combination Algorithms for Hepatocellular Carcinoma Detection Across Outcome Definitions Dot plot showing specificity estimates with 95% CIs for 9 diagnostic algorithms across 4 outcome definitions. Tier 1 strategies used disjunction logic; tier 2 used conjunction logic. Exploratory Analysis of Combination Algorithms and Bayesian Posttest Probability Analysis In the exploratory analysis of combination strategies, the tier 1a algorithm achieved 87.8% (95% CI, 81.6-93.5) sensitivity and 95.0% (95% CI, 92.3-97.4) NPV, representing a 98% relative improvement in sensitivity compared with AFP greater than 20 ng/mL alone (from 44.3% [95% CI, 35.3-53.5]) (Table 2; Figure 1; Figure 2; eFigures 6 and 7 and eResults in Supplement 1). The tier 1c algorithm achieved the highest AUROC of 0.80 (95% CI, 0.76-0.85), with 74.8% (95% CI, 66.7-82.5) sensitivity and 85.9% (95% CI, 82.6-89.1) specificity. The tier 2 algorithm achieved 99.6% (95% CI, 98.9-100) specificity and 95.1% (95% CI, 87.5-100) PPV. Diagnostic odds ratios for this algorithm ranged from 112.1 (95% CI, 37.8-294.7) for the primary outcome to 199.6 (95% CI, 48.7-273.6) for the composite outcome (eTable 8 in Supplement 1). Bayesian posttest probabilities were evaluated across pretest probabilities encountered in surveillance populations (Figure 3). Following a positive tier 2 result, posttest probability exceeded 70% across all pretest probabilities above 3% (positive likelihood ratio range, 76.0 [95% CI, 26.9-181.0] to 138.7 [95% CI, 35.7-169.9]). Following a negative tier 1a result, posttest probability remained below 5% for pretest probabilities up to 20% (negative likelihood ratio ranging from 0.2 [95% CI, 0.1-0.3] for the primary outcome to 0.2 [95% CI, 0.1-0.3] for the composite outcome sensitivity analysis) (eFigures 16 to 18 in Supplement 1). Bayesian Posttest Probability Curves for Hepatocellular Carcinoma Detection With Tier 1a, Tier 1c, and Tier 2 Combination Algorithms Probability curves showing conversion of pretest to posttest probability. A, Posttest probability after a positive test result. B, Posttest probability after a negative test result. Diagonal dashed line indicates a likelihood ratio (LR) of 1; shaded bands represent 95% credible intervals. AFP indicates α-fetoprotein. Subgroup Analysis Among Patients With Very Early or Early-Stage HCC AFP greater than 20 ng/mL demonstrated markedly limited sensitivity in BCLC 0-A disease, detecting only 23.5% (95% CI, 12.5-35.8) of cases. In contrast, the tier 1a algorithm achieved 74.5% (95% CI, 62.2-86.5) sensitivity, representing a 217% relative improvement compared with AFP alone (absolute gain, 51.0 percentage points; 3.2-fold increase). This improvement translated into an NPV of 95.4% (95% CI, 92.9-97.8). The tier 2 algorithm maintained robust rule-in performance in BCLC 0-A disease, achieving 99.6% (95% CI, 98.9-100) specificity. Performance of all 9 diagnostic algorithms in BCLC stage subgroups is detailed in eTable 11 in Supplement 1. Net Reclassification Improvement and Conditional Recovery Analysis Compared with AFP greater than 20 ng/mL alone, the tier 1c algorithm correctly reclassified a net 19.1% (95% CI, 10.5-28.2; P < .001) of patients, with similar improvements across all outcome definitions (eTable 12 in Supplement 1). Among 64 cases missed by AFP greater than 20 ng/mL, tier 1a recovered 50 patients (78.1%; 95% CI, 66.3-86.8), representing 8.9% (95% CI, 6.7-11.8) of the total cohort. In contrast, among 19 cases missed by methylated SEPT9, AFP recovered only 5 patients (26.3%; 95% CI, 10.0-50.6), corresponding to 0.9% of the total cohort. Detailed recovery rates are provided in eTable 13, eFigures 19 to 22, and eResults in Supplement 1. Risk Stratification and Fagan Nomograms A multivariable nomogram integrating methylated SEPT9 count (0 to 3 positive replicates) and AFP concentration was developed to provide individualized probability estimates for HCC (eTable 14, eFigure 23, and eResults in Supplement 1). The nomogram achieved an AUROC of 0.87 (95% CI, 0.86-0.87) after bootstrap validation with excellent calibration (Brier score, 0.09; calibration slope, 0.99). Among patients with AFP less than 20 ng/mL, the nomogram-estimated probability increased progressively with methylated SEPT9 positivity: 15% to 27% for single-positive, 27% to 45% for double-positive, and 44% to 63% for triple-positive (eFigure 23 in Supplement 1). A Fagan nomogram illustrating how likelihood ratios translate into patient-level probability estimates is provided in eFigure 24 in Supplement 1. An interactive version is available online.22 Validation of the Methylated SEPT9 Biological Replicate Design The biological validity of the replicate approach was confirmed by a strong inverse correlation between replicate positivity and cycle threshold values (Spearman ρ = −0.63; 95% CI, −0.70 to −0.55; P < .001), demonstrating that replicate count quantitatively reflects circulating tumor DNA burden (eFigure 25 and eResults in Supplement 1). Discussion In this prospective, cross-sectional, diagnostic study involving 574 patients with cirrhosis undergoing surveillance with ultrasonography and AFP, circulating methylated SEPT9 demonstrated superior diagnostic accuracy compared with AFP for HCC detection (posterior probability of superiority >99.8%). Combination algorithms substantially improved sensitivity, particularly for early-stage disease, while maintaining high specificity for disease confirmation. A 3-tier algorithmic framework combining both biomarkers enables tailored surveillance strategies across the sensitivity-specificity spectrum. Ultrasonography with or without AFP achieves only 47% sensitivity for early-stage HCC, with substantial operator-dependent variation.6,23,24 Surveillance adherence remains suboptimal at approximately 25% in community practice, partly due to logistic challenges.25,26 A blood-based strategy integrated with routine laboratory monitoring could improve both performance and surveillance adherence.2 The tiered algorithmic framework is intended to complement ultrasonography-based surveillance by guiding patient selection for cross-sectional imaging, not to replace current protocols. The 3 algorithmic tiers provide decision gates for distinct clinical scenarios. The tier 1a algorithm optimizes HCC exclusion, with sensitivity of 87.8% (95% CI, 81.6-93.5) and NPV of 95.0% (95% CI, 92.3-97.4), enabling disease rule out in high-risk populations, including transplant candidates and patients with decompensated cirrhosis.4,27,28 When both biomarker tests are negative, the residual risk of HCC remains below 5%, permitting confident deferral of cross-sectional imaging. Conversely, the tier 2 algorithm optimizes disease confirmation with specificity of 99.6% (95% CI, 98.9-100) and PPV of 95.1% (95% CI, 87.5-100), warranting expedited diagnostic evaluation when triple-positive methylated SEPT9 and AFP greater than 20 ng/mL occur concurrently. The diagnostic odds ratios for this algorithm ranged from 112.1 (95% CI, 37.8-294.7) to 199.6 (95% CI, 48.7-273.6) across outcome definitions, indicating strong rule-in performance. Because the diagnostic odds ratio is independent of pretest probability, these estimates allow translation into individualized posttest probabilities,29 as illustrated by the Fagan nomogram.22 Tier 1c balances sensitivity (74.8%; 95% CI, 66.7-82.5) and specificity (85.9%; 95% CI, 82.6-89.1) for routine surveillance, requiring imaging in only 14.1% of patients without cancer. Although differences in study design preclude direct comparisons, recent biomarker studies provide context. The GALAD score demonstrated a 1.3-fold improvement in early-stage sensitivity over AFP alone.30 In the ALTUS study, a multitarget blood test incorporating methylated DNA markers, AFP, and patient sex achieved a 1.3-fold improvement in early-stage sensitivity over AFP alone (from 52.6% to 66.7%) among 40 HCC cases identified from 2467 participants.31 With 118 HCC cases among 574 participants (ratio of 1:4 vs 1:61 in ALTUS), the present study afforded greater precision for sensitivity estimates and showed a 3.2-fold improvement in early-stage sensitivity (from 23.5% [95% CI, 12.5-35.8] to 74.5% [95% CI, 62.2-86.5]). The superior performance of methylated SEPT9 reflects distinct biological processes. AFP demonstrates variable expression in HCC, with particularly poor sensitivity in well-differentiated early-stage tumors.32,33 In contrast, SEPT9 promoter hypermethylation occurs consistently across diverse etiologies and tumor stages.11,34 The combination strategy integrates a protein marker reflecting tumor burden (AFP) with an epigenetic marker detecting early alterations in liver carcinogenesis (methylated SEPT9).10,35 The progressive increase in specificity with replicate positivity confirms a biological dose-response relationship between replicate count and circulating methylated DNA burden.19 Limitations Several limitations warrant consideration. The study was conducted at 2 French centers, and external validation in geographically diverse populations is needed to confirm generalizability. The predominance of alcohol-related and metabolic dysfunction-associated steatohepatitis, though viral hepatitis was represented, may limit applicability to populations with different cirrhosis etiologies. This diagnostic study does not directly measure survival outcomes; however, improved early-stage detection has clear clinical significance, as BCLC 0-A disease achieves 5-year survival exceeding 70% with curative treatment.4,5 Conclusions In this prospective, cross-sectional, diagnostic study, a blood-based combination of methylated SEPT9 and AFP was associated with improved HCC detection in patients with cirrhosis, with substantial gains in early-stage detection critical for curative treatment eligibility. Prospective outcome studies are needed to determine whether improved diagnostic performance translates into survival benefit.