Myeloma and Lymphoma Medical Helpers: AI-Powered Decision Support Tools for Hematologic Malignancies

Abstract

Introduction: Recent advancements in therapeutic options for multiple myeloma (MM) and lymphoma have significantly improved patient prognosis and survival. However, the variety of treatments complicates decision-making for optimal therapy selection, with new drugs often presenting challenging adverse effects, interactions and dose adjustments. To address these challenges, we developed Myeloma Medical Helper and Lymphoma Medical Helper, AI-driven tools to assist hematologists in clinical decision-making based on the latest medical guidelines and scientific evidence. Objectives: Develop AI-based decision support tools for MM and lymphoma that provide easy and rapid access to updated medical guidelines and recommendations on optimal therapy sequencing and adverse effect management. Material and Methods: For Myeloma Medical Helper, we conducted a review of recent literature on treatment recommendations and toxicity management for MM, incorporating various clinical guidelines and studies. These included first-line therapies, relapse management, and the prevention and treatment of specific toxicities such as cardiotoxicity and infections. For Lymphoma Medical Helper, we utilized the latest clinical guidelines from national cooperative groups, including guidelines for the treatment of Non-Hodgkin B and T-cell lymphomas, Hodgkin lymphoma and central nervous system prophylaxis recommendations. We developed APIs using OpenAI's GPT-4 (OpenAI et al., 2023), creating specialized chatbots through a Retrieval Assisted Generation (RAG) framework. This approach combines the pre-trained model's natural language processing abilities with a curated database of medical literature and guidelines, ensuring accurate, contextually appropriate and referenceable responses. Text was extracted from sources, divided into sentences, and merged into chunks using KMeans clustering based on embedding similarities using the model embedding-3-small. These chunks form the corpus of information available for the RAG framework. The APIs, developed in Python, handle queries and return relevant responses. A webpage accessing these APIs was created using JavaScript and PHP. Responses were manually validated for medical accuracy and relevance. Results: Myeloma Medical Helper and Lymphoma Medical Helper are AI tools capable of addressing specific clinical scenarios in MM and lymphoma patients. These chatbots provide clinicians with rapid, referenced responses regarding therapeutic strategies based on current scientific knowledge. They offer suggestions to improve patient management, focusing on preventing and managing significant complications such as cardiovascular toxicity, infections, and cytokine release syndrome. The systems evaluate management options, recommending strategies to optimize efficacy and safety for each patient. Both tools deliver evidence-based responses, facilitating the adoption and management of new targeted therapies, including bispecific antibodies and CAR-T cell therapies. By providing updated, referenced information on toxicity management, these tools contribute to improving clinical outcomes and patient quality of life. Conclusion: Myeloma Medical Helper and Lymphoma Medical Helper enhance MM and lymphoma treatment by facilitating the adoption of novel therapies and managing toxicities. These chatbots optimize clinical outcomes through personalized treatment strategies. Periodic updates with the latest information are crucial for maintaining their relevance and efficacy. Both decision support systems are freely available upon registration at https://codigorojo.tech.

Ähnliche Arbeiten