Analysis of prognostic biomarkers related to inflammation and body composition: a retrospective study using machine learning

• Grant number: 24/15490-8
• Support Opportunities: Scholarships in Brazil - Post-Doctoral
• Start date: July 01, 2025
• End date: June 30, 2027
• Field of knowledge: Health Sciences - Medicine - Medical Clinics
• Principal Investigator: Gilberto de Castro Junior
• Grantee: Willian das Neves Silva
• Host Institution: Instituto do Câncer do Estado de São Paulo Octavio Frias de Oliveira (ICESP). Coordenadoria de Serviços de Saúde (CSS). Secretaria da Saúde (São Paulo - Estado). São Paulo , SP, Brazil

Abstract

Background and aims: Lung cancer (LC) is the leading cause of cancer death worldwide, accounting for 11.4% of new global cases in 2020. The prognosis depends on several factors, including cancer stage, histological type, genetic aspects, and patient functional capacity. Several biomarkers have been investigated to estimate the prognosis; however, there is no ideal method to estimate the prognosis of patients with lung cancer. Considering different variables such as prognostic biomarkers and the increasing use of machine learning in predicting the prognosis of cancer patients, the present study aims to analyze the associations between different systemic inflammation indices with muscle depletion in patients diagnosed with LC undergoing treatment, as well as to evaluate whether these factors are related to tumor characteristics, toxicities, overall survival, and diseasefree survival, integrating all these variables through machine learning. Methodology: Patients included in the retrospective cohort will be adults with non-small cell lung cancer enrolled at the São Paulo State Cancer Institute (ICESP) between January 2009 and December 2022. Patient characteristics will be demographic, staging, histological data, metastasis sites, blood cell count, and body mass index (BMI). Inflammatory indices will be evaluated, as well as muscle and fat area obtained through tomography exams. Survival analysis will be the primary outcome and will be performed through the median overall survival, determined by Kaplan-Meier analysis. Machine learning will be adopted by a supervised model with 500 to 1000 patients, using ROC curves, correlations, and collinearity tests. After this stage, the same number of patients will be used in the model testing phase. In addition, a correlation network analysis will be performed between characteristics, and the influence of body composition and performance variables on the performance of the created model will be tested. The ethics committee will approve the study, and we will follow the guidelines for developing and reporting predictive machine learning models in biomedical research. Perspectives: We believe that the results of this study will help in planning an appropriate and specific intervention for the population studied, favoring quality, evidence-based action and enabling more efficient and personalized care. Furthermore, as an innovation proposal, we intend to generate a tool that will be incorporated into the PACS system that, in the future, may guide new conducts and treatment optimization.