Advancing Education, Research, and Quality of Care for the Head and Neck oncology patient.
Background: Oropharyngeal cancer (OPC) presents a unique clinical challenge due to its multidisciplinary management and the occurrence of clinical equipoise, where different treatment options—such as Transoral Robotic Surgery (TORS), definitive chemoradiation (chemoRT)—offer comparable outcomes. This makes shared decision-making (SDM) a crucial part of OPC care, as patients must carefully weigh the benefits and risks of each option. However, newly diagnosed cancer patients often experience cognitive overload, struggling to process complex medical information, which can hinder their ability to engage effectively in SDM. To address this, we developed NavigaTORS, a fine-tuned language model designed to provide OPC patients with personalized, accessible guidance by leveraging after-visit summaries (AVS) and educational resources. NavigaTORS supports patients throughout diagnosis, treatment planning, and postoperative care, aiming to improve their capacity for SDM and overall treatment experience.
Methods: Two different large language models, a basic model using retriever augmented generation (RAG), and a fine-tuned+RAG, prompt-engineered language model, (NavigaTORS), were compared across five different evaluation criteria: accuracy, clarity, tone, relevance, and actionability. Five experts from a single institution evaluated the responses generated for six thematic prompts (e.g., NPV, treatment options, side-effects, decision-making, and preoperative guidance). Statistical analysis was conducted using the Wilcoxon Signed-Rank Test to determine significant differences between the models.
Results: NavigaTORS demonstrated significantly improved performance over the RAG-only model across all evaluated criteria. In an expert evaluation of prompts focused on diagnosis, staging, treatment options, TORS post-operative care, and treatment side-effects related to oropharyngeal cancer (OPC), NavigaTORS achieved statistically higher scores in accuracy, clarity, tone, relevance, and actionability. Wilcoxon Signed-Rank Test results indicated notable improvements in clarity (p = 0.0002) and tone (p = 0.00004), highlighting the model’s superior capacity for delivering clear, empathetic responses. The fine-tuned model also showed significant improvements in accuracy (p = 0.0036), relevance (p = 0.0456), and actionability (p = 0.0277), suggesting a greater alignment with patient queries and a stronger ability to provide actionable guidance. These findings highlight OPC’s potential to enhance patient support, facilitate shared decision-making, and reduce the informational burden following an oropharynx cancer diagnosis.
Conclusion: NavigaTORS represents an invaluable tool in OPC patient care, enhancing SDM by empowering patients with accessible, relevant information tailored to their needs. By addressing common challenges in information processing, NavigaTORS has the potential to improve adherence to postoperative care instructions and enhance the overall cancer treatment experience. Future studies will refine NavigaTORS based on patient feedback, while personalizing responses according to each patient’s treatment phase to enhance support and streamline care coordination.