Abstract 5394: Detecting pulmonary malignancy against benign nodules using non-invasive cfDNA fragmentomics assay

Abstract Background: Early screening using Low-Dose Computed Tomography (LDCT) can reduce mortality by non-small-cell-lung cancer (NSCLC), which contributed the most cancer-related death worldwide. Yet ~25% of the “suspicious” nodules identified by LDCT are confirmed to be benign through resection surgery, which adds to patients’ discomfort and the burden of the healthcare system. In this prospective study, we set to develop a non-invasive liquid biopsy assay for distinguishing pulmonary malignancy from benign lung nodules using cfDNA fragmentomic profiling. Methods: An independent training cohort, which consisted of 193 patients with malignant nodule and 44 patients with benign nodule, was used to construct a machine learning model. Base models employing 4 different fragmentomics profiles were optimized using an automated machine learning (autoML) approach before being ensemble stacked to create the final predictive model. An independent validation cohort included 96 malignant nodules and 22 benign nodules, as well as an external test cohort containing 58 malignant nodules and 41 benign nodules, were used to assess the performance of the ensemble stacked model. Results: Our machine learning models showed excellent performances in detecting patients with malignant nodules. The AUCs reached 0.857 (0.782 - 0.932) and 0.860 (95% CI: 0.788 - 0.933) in the independent validation cohort and the external test cohort, respectively. The validation cohort achieved an excellent 68.2% specificity (95% CI: 45.1-86.1%) at the targeted 90% sensitivity (89.6%, 95% CI: 81.7-94.9%). An equivalently good performance was observed while applying the cutoff to the external cohort, which reached a specificity of 63.4% (95% CI: 46.9-77.9%) at 89.7% sensitivity (95% CI: 78.8-96.1%). Subgroup analysis for the independent validation cohort showed that the sensitivities for detecting various subgroups of nodule size (< 1cm: 91.7%; 1 - 3cm: 88.1%; > 3cm: 100%; Unknown: 100%) and smoking history (Yes: 88.2%; No: 89.9%) all remained high among the lung cancer group. Additionally, the specificities for successfully identifying the benign nodules among different subgroups decrease as the size increase (< 1cm: 80.0%, 1 - 3cm: 66.7%, > 3cm: 33.3%). Conclusions: Our cfDNA fragmentomics assay can provide a non-invasive approach to distinguish malignant nodules from the radiographically suspicious but pathologically benign nodules. Citation Format: Shun Xu, Ji Luo, Wanxiangfu Tang, Hua Bao, Jiajun Wang, Shuang Chang, Haimeng Tang, Zifang Zou, Xiaoxi Fan, Yang Liu, Changrui Jiang, Xue Wu, Yang Shao. Detecting pulmonary malignancy against benign nodules using non-invasive cfDNA fragmentomics assay. [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2023; Part 1 (Regular and Invited Abstracts); 2023 Apr 14-19; Orlando, FL. Philadelphia (PA): AACR; Cancer Res 2023;83(7_Suppl):Abstract nr 5394.