e15641 Background: Colorectal cancer (CRC) remains one of the leading causes of cancer morbidity and mortality worldwide. However, early detection and timely intervention are strongly associated with improved survival and quality of life. The Dxcover Liquid Biopsy Platform is a rapid, multiomic blood-based assay that interrogates samples using infrared spectroscopy to generate a distinctive spectral signature reflective of the global biomolecular composition of the analysed sample. This technology has previously been evaluated as a standalone test for CRC detection, but it also has potential for integration with complementary data sources, including biomarkers and clinical risk factors. Methods: In this study, samples from 1377 patients were collected across sites in the USA (n = 989) and UK (n = 388). Blood was obtained from patients either prior to scheduled colonoscopy or before surgical resection and any anti-cancer therapies. Streck plasma samples were analyzed by the Dxcover Liquid Biopsy Platform to generate spectral data. Carcinoembryonic Antigen (CEA) values were determined for all samples. Fecal hemoglobin levels from fecal immunochemical testing (FIT) were also obtained for the UK samples. Machine learning algorithms were developed to compare test performance and assess combinations. Results: Initially, machine learning models were developed for the spectral dataset alone. The area under the curve (AUC) was 0.95, test sensitivity and specificity were 90% and the model reported consistent detection rates across CRC stages. There was limited diagnostic utility reported for CEA alone (37% sensitivity with 80% specificity). There was no improvement to the spectral model with the inclusion of CEA. For the UK cohort with FIT results, the FIT only model (AUC = 0.83) was enhanced by the addition of spectral data, with the combined model (spectra+FIT) reporting an AUC of 0.90. Conclusions: Integration of this liquid biopsy with orthogonal tests, including FIT or blood-based biomarkers, may enhance early-stage CRC detection. Limitations of existing screening programs underscore the need for novel technologies to support earlier diagnosis and improved outcomes.
Cameron et al. (Thu,) studied this question.