Oncogeriatric Screening and Evaluation Program (PROTEGER)

Cancer incidence increases significantly with age, and up to 70% of cancer mortality occurs in patients aged 65 years or older. Despite this, older patients are frequently undertreated due to the high risk of secondary toxicity, which is associated with quality of life deterioration, increased hospitalizations, and higher mortality. Comprehensive Geriatric Assessment (CGA) has proven to be an effective tool to identify vulnerability, reduce chemotherapy-related toxicity, and tailor interventions. However, the lack of geriatricians, especially in Latin American public health systems, creates significant barriers to accessing CGA-guided oncological care.

To overcome these barriers, the PROTEGER program proposes an innovative digital health solution by developing and validating a machine learning-based clinical decision support system (CDSS) for oncogeriatric care. The study is an observational, multicenter, bidirectional cohort study conducted in two phases:

Phase 1: Retrospective Training Phase This phase uses anonymized clinical data (2021-2023) from the Oncogeriatric Tele-Committee of the Chilean Ministry of Health's Digital Hospital. Data from older patients with solid tumors who underwent a CGA will be used to train and test predictive models using machine learning techniques (e.g., Gradient Boosting Trees and Random Forest) following the CRISP-DM methodology. The predictive model aims to learn the Committee's treatment recommendation patterns based on patient functionality, comorbidities, and geriatric syndromes.

Phase 2: Prospective Validation Phase A prospective, multicenter cohort will be enrolled across healthcare centers in Chile, Peru, and Brazil. Eligible patients (aged 65+ with a solid tumor diagnosis) who undergo routine CGA and oncological care will be followed for 6 months. Data regarding baseline characteristics, treatment decisions (made by local oncology teams blinded to the AI model's recommendation), dose reductions, treatment discontinuation, disease progression, quality of life (EORTC QLQ-C30 and ELD14), and survival will be collected.

Study Objectives:

The primary objective is to develop, train, and clinically validate the PROTEGER machine learning predictive model to provide an accurate treatment recommendation (e.g., standard treatment, dose-adjusted treatment, or supportive care only) capable of assisting clinical decision-making by oncology teams. A secondary objective involves the design and development of an intuitive graphical user interface capable of being used by healthcare providers and patients for data management and result interpretation.

All predictive models will be evaluated using standard metrics, such as the Area Under the ROC Curve (AUC) and the C-statistic, to determine their discriminatory capacity in a real-world clinical setting.