Genetic Algorithm-Based Individual Treatment Effect Model for Optimizing Induction Chemotherapy in Nasopharyngeal Carcinoma

Motivation: Improve decision-making by focusing on treatment effects rather than risk. Goal(s): Develop an individual treatment effect (ITE) model, based on structured MR reporting variables, for predicting induction chemotherapy (IC) efficacy in nasopharyngeal carcinoma (NPC). Approach: In this retrospective 2-center study (N=1207), an ITE model was developed based on real-number coding genetic algorithm. Results: The ITE model successfully classify patients into three categories. In IC-beneficial subgroup, IC treatment decreased the mortality risk by 48% (adjusted P=0·029) in testing sets. In IC-detrimental group, mortality risks increased by 2·11 (adjusted P=0·023, testing set) after IC treatment. Overall survival was not significantly different in IC-ambiguous group. Impact: By precisely stratifying the outcomes of IC treatment, ITE model has the potential to enhance individualized treatment and optimize the management of patients with NPC. Additionally, the ITE model offers a viable strategy for personalized treatment decision-making in other tumors.