Abstract 914P
Background
For many patients, early detection of cancer strongly correlates with better outcomes. Unfortunately, the cancers with the poorest 5-year survival statistics are those that are most often found at late stages. Correlation between outcomes and detection may justify broad cancer screening if innovative approaches provide compelling results with acceptable performance for general screening or early detection purposes. Several diagnostic companies test blood products to identify markers for disease, e.g. liquid biopsy. Many of these sequence circulating tumor DNA, which may not be detectable in early-stage patients. To overcome this and other limitations of competitive approaches, we have developed a novel platform for disease detection using extracellular vesicle (EV) proteins and machine learning (ML) algorithms.
Methods
Using minimal patient or control volume (240 μL), we isolate EVs from the plasma and probe them for surface and intravesical protein cargoes. The platform used for EV isolation enriches particles (size 50 to 250 nm) with minimal contamination in contrast with other EV isolation technologies. Retrospective samples from patients with stage I or II cancer across sixteen different cancer types, and non-cancer controls (including benign conditions) were evaluated in a pilot early detection study and a ML classifier was developed using EV biomarkers.
Results
The most informative biomarkers were found by analyzing each cancer type individually in comparison to the controls. Using this information, we then employ a suite of machine learning algorithms to create a multi-cancer early-detection (MCED) classifier from the most informative biomarkers and phenotypic/clinical information obtaining an AUC greater than 0.95.
Conclusions
EVs isolated by our method and analyzed by multiomics may be a powerful tool for early cancer detection. Multiple challenges remain for the successful implementation of platforms like ours in screening or early detection of disease. These include partnering with academic clinics to address patient heterogeneity; identify elevated risks for specific diseases (germline mutations, precursor lesions), establish performance metrics and explore if our testing may improve pre- or post-surgical management with curative intent.
Clinical trial identification
Editorial acknowledgement
Legal entity responsible for the study
The authors.
Funding
Biological Dynamics.
Disclosure
P. Billings: Financial Interests, Institutional, Member of the Board of Directors: Biological Dynamics. H. Balcer, J.P. Hinestrosa: Financial Interests, Institutional, Full or part-time Employment: Biological Dynamics.