Generative AI–Powered clinical decision support in germline predisposition to myeloid neoplasms

Abstract

Abstract Introduction The exploration of germline predispositions to myeloid neoplasms (MNs) has gained significant traction, particularly with the recognition of these conditions in the current World Health Organization's classification. This abstract outlines the development and implementation of a specialized chatbot designed to assist healthcare professionals in managing patients with germline predispositions to MNs. The chatbot aims to enhance clinical decision-making by providing evidence-based recommendations and facilitating genetic counseling. The introduction of this tool is rooted in the increasing need for precise and timely information regarding genetic predispositions to MNs, which include acute myeloid leukemia (AML) and myelodysplastic syndromes (MDS). These conditions require careful consideration in clinical management, particularly concerning therapeutic strategies and donor selection for allogeneic hematopoietic stem cell transplantation (allo-HSCT). Our goal is to bridge the gap in knowledge and streamline the decision-making process for clinicians. Objective As germline predisposition to MNs is a field requiring a high level of expertise, the primary objective of this initiative is to support healthcare providers by offering a comprehensive resource that integrates current guidelines and cutting-edge research findings. The chatbot is designed to deliver tailored information based on specific clinical scenarios, thereby improving patient management by optimizing treatment strategies. Methods The chatbot utilizes advanced Natural Language Processing (NLP) techniques in combination with GPT-4 large language model (LLM) to interpret and respond to clinical queries. It also incorporates a Retrieval-Augmented Generation (RAG) approach, enabling it to dynamically draw on a curated database of peer-reviewed literature and clinical guidelines. Moreover, a multimodal feature has been added through the OpenAI API, allowing the chatbot to describe medical images—such as radiological or histopathological findings—and provide relevant insights. Collaboration with experts in hematology-oncology and genetics ensured alignment with the latest scientific evidence and best clinical practices. Results Preliminary findings indicate that the chatbot significantly enhances consultation efficiency, offering quick, point-of-care access to critical information. Users report increased confidence in their clinical decision-making and a deeper understanding of germline mutations in myeloid neoplasms. The chatbot's ability to assist with genetic counseling and family risk assessment has also been highlighted as a major advantage. By adding medical image description, the chatbot now provides a multimodal solution that further optimizes decision-making. It can be accessed via Google Chrome at: https://chatbot.codigorojo.tech/chatbots/26/. Conclusion Implementing this advanced chatbot is a significant step forward in managing germline predispositions to myeloid neoplasms. By integrating LLMs and multimodal medical image interpretation, it offers an accessible, evidence-based resource that enhances clinical decision-making, supports genetic counseling and family risk assessment, and ultimately improves patient outcomes. Future efforts will focus on refining its capabilities—particularly through deeper data integration and seamless interoperability with electronic health records—to streamline workflows and provide more personalized care.