Routine blood markers panel based on artificial intelligence for breast cancer early detection

Abstract

In Brazil, women over 50 years of age have access to a biannual mammogram through the SUS, but 77% of Brazilian municipalities do not have mammography equipment. Furthermore, the screening program does not accommodate younger women or men - who are the group at most significant risk for more severe or late-diagnosed diseases. In this scenario, HUNA proposes the creation of a breast cancer screening test, applicable to the entire population and at an affordable cost, employing Machine Learning to blood test markers and clinical-demographic data. In consonance with the medical-scientific community, HUNA and collaborators point to developing "liquid biopsies." It has been carried out from blood analysis where it is possible to identify, using cutting-edge Artificial Intelligence tools, tumor genetic markers, and specific blood metabolites to detect breast cancer, possibly before the first clinical symptoms appear. C-Reactive Protein (CRP), HDL and LDL cholesterol, iron, neutrophil/lymphocyte index, platelets, RDW, CA153, MPV, glucose, CEA, CA19-9, CA125, CA153, TPS, D-dimer, progesterone, prolactin, among others are the non-genetic blood analytes identified to enable a product for the early identification of breast cancer. To this end, Grupo Fleury will provide, upon approval by the Research Ethics Committee and guaranteeing anonymity protection to patients, approximately 30,000 medical records containing: laboratory tests, mammographic (ultrasound or breast MRI) reports, excisional biopsies of female patients older than 18 years admitted to the institution for screening, diagnosis or monitoring of breast CA. In possession of this data, we intend to develop, test, and validate Machine Learning models, using data previously highlighted from medical records and blood tests to determine the best predictive models. Then, formulate, test, and validate a "maximum diagnostic prediction window," temporally establishing the appearance of the first indicators of breast alteration in the best predictive models when compared to the diagnosis (BI-RADS mammographic result > 3 with subsequent confirmation by biopsy).Considering the current screening rate in the SUS (<15% of the target population), we calculated that we could generate savings of R$ 50 million/year in a first approximation. As we believe in the acceptance by physicians, in the ease of incorporating this innovation by clinical analysis laboratories, and in the savings potentially generated for the health system, we understand that the investment costs for the implementation/maintenance of this new product would be justified. In a disruptive way, in the future, we intend to serve the segment of the population currently unassisted by the screening program - young women, men, and municipalities without mammograms - in addition to having the potential to increase the adherence to the current target population to complying with their screening routine in the face of the possibility of conducting an accessible and convenient test. (AU)