AI-DRIVEN PREDICTIVE MODELING FOR GERIATRIC ONCOLOGY OUTCOMES

Abstract

Implementing artificial intelligence (AI) in geriatric oncology is a game-changing approach to predicting the outcomes of treatment and customizing the care of geriatric cancer patients.  A mixed-methods research was conducted to develop Machine Learning models capable of predicting the possibility of hospitalization, treatment toxicity, and survival 12 months after the onset of cancer in persons above the age of 65.  In the dataset consisting of 412 patients with clinical, geriatric, and genomic data, we trained and tested such models as logistic regression, random forest, support vector machines, and XGBoost.  XGBoost fared the best with an AUC score of 0.87, F1-score of 0.79 and a balanced accuracy of 0.81.  SHAP analysis revealed the best predictors of survival were age, hemoglobin level, mutation burden and ECOG status.  Together with quantitative modeling, qualitative interviews with 30 patients identified significant themes such as anxieties related to therapy and functional independence and emotional resilience. Such themes were employed in enhancing the interpretation and application of the model to real world.  The model was also tested outwardly by an independent group of individuals (n = 98), and it has been identified that it was strong (AUC = 0.83).  The findings indicate the significance of AI-driven tools to segregate risks, assist physicians in decision-making, and inculcate the values of patient-centredness into cancer treatment of the aged.  This research is based on a paradigm that can be applied repetitively and comprehended which is the rigour in computer science and compassion in health science.