Performance of a targeted enzymatic methylation-based early detection test by different colorectal cancer subgroups.

Abstract

3596 Background: Colorectal cancer (CRC) is the second most frequently diagnosed cancer in China and early detection could prevent over 90% of CRC-related deaths. Blood-based tests that analyze molecular features of CRC cell-free DNA (cfDNA), such as methylation and fragmentation patterns, hold great promise for early detection. However, the impact of molecular characteristics related to tumor location or mismatch repair (MMR) status on test performance has not been thoroughly investigated. In this study, we developed a blood-based CRC early detection test and analysed its performance across different CRC subgroups. Methods: A targeted enzymatic methyl sequencing panel was developed to identify tumor-specific hyper- and hypo-methylation markers and fragmentation profiles. A case-control cohort of 536 participants (268 CRC patients, 268 controls) was enrolled and startified into training and validation sets base on case/control status and cancer stage with 5-fold cross-validation. A gradient-boosted tree model was built by combining probabilities from methylomic and fragmentomic features. The optimal cutoff value for the early detection was determined by Youden's index, High specificity and High sensitivity methods, respectively. Results: The overall performance of Youden's index, High specificity and High sensitivity methods was as follows: specificity of 93.7%, 99.3%, 90.3%, and sensitivity of 96.6%, 86.2%, 97.0%, respectively. The area under the curve (AUC) value is 0.989 (95% CI: 0.981-0.996) , which is higher than those in current reports. When employing the High specificity method, the sensitivities were comparable between left and right-sided colon cancer (86.3% vs 85.7%, p = 1.0), and also similar between the dMMR (deficient mismatch repair) and pMMR (proficient mismatch repair) (87.5% vs 85.8%, p = 1.0), indicating that this model is applicable to various CRC subtypes. Additionally the TNM staging, pathological differentiation status, and the expression of Ki67, which are closely related to aggressiveness, were correlated with the sensitivity (Table). Conclusions: We have established CRC early detection model based on ctDNA methylation and fragmentation profiles, which shows excellent overall performance. Notably, this newly developed blood-based model shows no significant differences in sensitivity between distinct tumor locations or varying MMR statuses, suggesting its broader applicability across different types of CRC. Subgroup Positive/Total no. Sensitivity p_value Left-sidedRight-sided 195/22636/42 86.3%85.7% 1 dMMR pMMR 7/8200/233 87.5%85.8% 1 Stage IStage II Stage III Stage IV 34/5092/10872/7633/34 68%85.2%94.7%100% ＜0.001 Well differentiated Moderately differentiated Poorly differentiated 3/3160/19241/42 100%83.3%97.6% 0.029 Ki67_highKi67_low 138/15932/42 86.8%76.2% 0.147