Use of Machine Learning Techniques for Serial Assessment of Systemic Inflammatory Markers in Breast Cancer Patients

Overview

Breast cancer is the most common cancer in women globally, with 2.3 million new cases diagnosed in 2020. Hormone receptor positive (HR+), human epidermal growth factor receptor 2 negative (HER2-) breast cancer is the most prevalent subtype, comprising 69% of all breast cancers in the USA. Within the tumor immune microenvironment, a higher intensity of myeloid cell infiltration and low levels of lymphocyte infiltration have been associated with worse outcomes. Markers in peripheral blood have emerged as predictive biomarkers that can be easily obtained non-invasively and at low cost. Experiments have confirmed the relative components of these tests (such as the immune cells) directly or indirectly participated in tumour occurrence, development, and immune escape, underscoring the potential use of laboratory tests as tumour biomarkers

In breast cancer, increased neutrophil levels and decreased lymphocyte levels in peripheral blood are associated with worse overall survival (OS). In HR+, HER2- metastatic breast cancers, low pretreatment NLR and high pretreatment absolute lymphocyte count (ALC) were related with better progression-free survival (PFS) and OS. The development of predictive models, based on machine learning (ML) algorithms it has been used in prognostication and assist in the diagnosis of different types of cancer. Although regular laboratory tests have potential to be breast cancer biomarkers, a single test is yet to provide adequate sensitivity or specificity. Artificial intelligence (AI) could help with integrating data from multiple tests to aid diagnosis. Technical improvements such as data storage capacity, computing power, and better algorithms mean that ML can process clinically meaningful information from laboratory test data. Models' generalisability and stability still need to be confirmed, in view of limitations such as the absence of various pathological types, small cohorts, and lack of external validation. Therefore, a competitive model is also essential to achieve more accurate stratification of patients with breast cancer. The purpose of this retrospective multicentre study is to systematically evaluate the ability of laboratory tests to predict breast cancer, and develop a robust and generalisable model to assist in identifying patients with breast cancer.