Circulating Cell-free DNA-based Epigenetic Biomarker mSEPT9 for Hepatocellular Carcinoma Detection in Cirrhosis

Clinical Trial Rationale and Design Epigenetic alterations represent a fundamental hallmark of human malignancy. Although individual epigenetic biomarkers have begun to enter clinical practice, their translational validation remains limited. This limitation is particularly evident in hepatocellular carcinoma (HCC), the most common primary malignant tumor of the liver. Alpha-fetoprotein (AFP) has long been used as a diagnostic biomarker for HCC; however, as stated in international guidelines from the American Association for the Study of Liver Diseases and the European Association for the Study of the Liver, AFP lacks sufficient sensitivity and specificity for effective screening or early detection. Aberrant DNA methylation events are a frequent molecular feature of malignant transformation and can be detected in the circulation of patients with cancer using polymerase chain reaction-based assays. Among epigenetic targets, the SEPT9 gene has emerged as a key regulator of cell division and a tumor suppressor whose promoter hypermethylation is closely associated with carcinogenesis. Experimental evidence implicates SEPT9 in the early stages of hepatocarcinogenesis, and promoter hypermethylation of SEPT9 has been consistently observed in human HCC. Loss or downregulation of SEPT9 expression due to aberrant promoter methylation has also been reported in several other malignancies. Building on a proof-of-concept study conducted in France and an independent replication study in Germany, we previously demonstrated that plasma methylated SEPT9 (mSEPT9), a circulating cell-free DNA-based epigenetic biomarker, shows strong potential for the noninvasive diagnosis of HCC in patients with cirrhosis (EBioMedicine 2018:30:138-147). The SEPT9\_CROSS study is a phase III, multicenter, cross-sectional diagnostic accuracy investigation designed to validate the clinical performance of plasma methylated SEPT9 (mSEPT9) for the diagnosis of HCC in a large-scale cohort of 639 patients with cirrhosis initially enrolled across participating centers. After application of the predefined eligibility and exclusion criteria, 65 patients were excluded, resulting in 574 analyzable patients retained in the final study population. Patient Recruitment and Sample Handling Patients with cirrhosis who met the study eligibility criteria were enrolled after providing informed consent. Each participant signed a participation agreement form, and documentation of study participation was recorded in the medical record. As part of their routine follow-up, all patients underwent clinical, biochemical, and imaging assessments, including measurement of alpha-fetoprotein, in accordance with international guidelines. During routine phlebotomy performed for biochemical evaluation, an additional 20 mL of blood was collected into two 10 mL EDTA tubes dedicated to the mSEPT9 assay. Plasma was separated from these samples, aliquoted, and handled and stored under standardized conditions in the CRB Lorrain biobank at -80 °C until analysis of mSEPT9. Data Collection and Management Patient data were entered into an electronic case report form (eCRF) using the CleanWeb system, with role-based access control. Data were analyzed by the methodologist-statistician of the Methodology, Data Management, and Statistics Unit (UMDS) within the Department of Clinical Research and Innovation (DRCI) at the University Hospital of Nancy (France). All data were recorded and securely stored on the protected network of the University Hospital of Nancy (CHRU de Nancy). Sample Size and Power Calculation According to Substantial Amendment No. 2 (MS No. 2, October 5, 2021), the sample size distribution between study groups was modified based on updated recruitment data. Based on the revised power calculation, the theoretical sample size required to achieve adequate statistical power was 530 patients, including 106 HCC-positive cirrhotic patients (cases) and 424 HCC-free cirrhotic patients (controls), instead of the initially planned 220 patients per group. This adjustment was derived from a recalculated statistical power analysis designed to detect an increase of at least 0.05 in the area under the receiver operating characteristic curve (AUC) for mSEPT9 compared with alpha-fetoprotein (AFP), with a two-sided α level of 0.05, a β risk of 0.20 (power of 80%), and a case-to-control ratio of 1:4. Under these assumptions, 101 cases and 404 controls were required for analysis. Allowing for approximately 5% of unevaluable data, the final theoretical target enrollment was set at 106 cases and 424 controls, yielding a total of 530 patients. In total, 639 patients with cirrhosis were initially enrolled across participating centers in the SEPT9-CROSS study. After application of the predefined eligibility and exclusion criteria, 65 patients were excluded because they did not meet inclusion criteria or were identified as duplicates. The final analyzable population included 574 patients, comprising 118 HCC-positive cirrhotic patients (cases) and 456 HCC-free cirrhotic patients (controls) retained for statistical analysis. Study Duration According to Substantial Amendment No. 4 (MS No. 4, March 1, 2023), the inclusion period was extended by an additional 12 months, from 60 to 72 months, resulting in a total study duration of 76 months instead of the initially planned 64 months. The amendment also included a change of principal investigator for the Metz-Thionville University Hospital center. Statistical and Analytical Framework The statistical analysis plan (version 3), finalized prior to database lock, prespecified all analytical procedures. The primary objective was to evaluate the diagnostic accuracy of plasma methylated SEPT9 (mSEPT9) for detecting hepatocellular carcinoma (HCC) in patients with cirrhosis and to compare its performance with that of alpha-fetoprotein (AFP). Outcome Definitions The primary outcome was the presence of hepatocellular carcinoma at enrollment. A sensitivity analysis restricted controls to patients who remained free of hepatocellular carcinoma throughout 12 months of follow-up. A prespecified composite outcome encompassed prevalent hepatocellular carcinoma at baseline and incident hepatocellular carcinoma diagnosed within 12 months after enrollment. Endpoint definitions The primary endpoint was diagnostic accuracy for detecting hepatocellular carcinoma, assessed by area under the receiver operating characteristic curve, comparing methylated SEPT9 with alpha-fetoprotein greater than 20 nanograms per milliliter. Secondary endpoints included diagnostic performance stratified by Barcelona Clinic Liver Cancer stage, with prespecified subgroup analyses for very early and early-stage disease (stages 0 and A) versus more advanced stages (stages B, C, and D). Exploratory endpoints included evaluation of combination diagnostic strategies (Tier 1 approaches maximizing sensitivity and Tier 2 approach maximizing specificity), net reclassification improvement, conditional recovery analysis, development of a risk stratification nomogram, and validation of the biological replicate design through correlation analysis between the number of positive replicates and cycle threshold values. Analytical framework Nine diagnostic algorithms were prespecified in the statistical analysis plan and evaluated across all outcome definitions: methylated SEPT9 as a continuous variable (0 to 3 positive replicates); methylated SEPT9 with three binary thresholds (single-positive, double-positive, and triple-positive); alpha-fetoprotein greater than 20 nanograms per milliliter; three Tier 1 strategies using OR logic (methylated SEPT9 at each threshold or alpha-fetoprotein greater than 20 nanograms per milliliter) to maximize sensitivity; and Tier 2 strategy using AND logic (triple-positive methylated SEPT9 and alpha-fetoprotein greater than 20 nanograms per milliliter) to maximize specificity. Performance metrics included area under the receiver operating characteristic curve, sensitivity, specificity, positive and negative predictive values, and diagnostic odds ratio, with 95 percent confidence intervals obtained through bootstrap resampling with 10,000 iterations using bias-corrected and accelerated estimation. Bayesian modeling was applied to estimate posterior distributions of area under the receiver operating characteristic curve values, with Bayes factors computed to quantify the strength of evidence supporting comparative performance across diagnostic algorithms. Prespecified subgroup analyses evaluated performance in patients with very early or early-stage hepatocellular carcinoma (Barcelona Clinic Liver Cancer stage 0 or A), representing the clinically relevant population eligible for curative interventions. Prespecified exploratory analyses included net reclassification improvement to measure the proportion of patients correctly reclassified when adding methylated SEPT9 to alpha-fetoprotein-based testing, conditional recovery analysis to assess the complementarity of diagnostic strategies, development of a multivariable risk-stratification nomogram to provide individualized probability estimates of hepatocellular carcinoma, and validation of the biological replicate design through correlation analysis between the number of positive replicates and cycle threshold values. All analyses were performed using R version 4.3.0 and Python. Custom Python scripts were implemented to automate receiver operating characteristic curve simulations, bootstrap iterations, Bayesian inference, and predictive modeling, ensuring analytical reproducibility.