Blood-Based Machine Learning Assay Noninvasively Detects Ovarian Cancer

The disease is particularly lethal because it often doesn't cause symptoms in its early stages. The absence of effective screening tools and the disease's asymptomatic nature contribute to its diagnoses during the later stages when treatment options are less effective. A cost-effective, accessible detection method could revolutionize the clinical approach to ovarian cancer screening and potentially save lives. Although liquid biopsy technologies, which analyze blood for tumor-derived DNA, have been explored for noninvasive cancer detection, their utility in ovarian cancer has been limited. Now, a retrospective study presented at AACR 2024 has demonstrated that a blood-based machine learning assay, which combines cell-free DNA (cfDNA) fragment patterns with levels of the proteins CA125 and HE4, can effectively distinguish patients with ovarian cancer from healthy controls or patients with benign ovarian masses.

The DELFI (DNA Evaluation of Fragments for early Interception) method employs a novel liquid biopsy approach called fragmentomics. This technique improves the accuracy of tests by detecting circulation changes in the size and distribution of cfDNA fragments across the genome. Researchers at the Johns Hopkins Kimmel Cancer Center (Baltimore, MD, USA) applied DELFI to analyze the fragmentomes of individuals with and without ovarian cancer. The study included plasma samples from 134 women with ovarian cancer, 204 women without cancer, and 203 women with benign adnexal masses. They trained a machine learning algorithm to integrate this fragmentome data with plasma levels of CA125 and HE4, two established biomarkers for ovarian cancer.

The researchers developed two models: one for screening ovarian cancer in an asymptomatic population and another for noninvasively differentiating benign from cancerous masses. At a specificity of over 99% (virtually eliminating false positives), the screening model detected 69%, 76%, 85%, and 100% of ovarian cancer cases from stages I to IV, respectively; the area under the curve (a measure of test accuracy) was 0.97 across all stages, significantly outperforming current biomarkers. For comparison, using CA125 levels alone identified 40%, 66%, 62%, and 100% of cases staged I-IV, respectively. The diagnostic model distinguished ovarian cancer from benign masses with an area under the curve of 0.87. The researchers plan to validate their models in larger cohorts to confirm these findings, but the initial results are promising.

“This study contributes to a large body of work from our group demonstrating the power of genome-wide cell-free DNA fragmentation and machine learning to detect cancers with high performance,” said Victor Velculescu, MD, PhD, FAACR, senior author of the study. “Our findings indicate that this combined approach resulted in improved performance for screening compared to existing biomarkers.”

Related Links:
Johns Hopkins Medicine