Article type
Abstract
Background: Maternal mortality, a significant global health concern, was approximately 287,000 deaths in 2020, affecting low- and middle-income countries.(1,2) This statistic underscores the need for enhanced obstetric care. Addressing this, our study introduces an innovative artificial intelligence (AI)–powered chatbot that integrates large language models (LLMs) like GPT-3.5 and GPT-4 with clinical practice guidelines (CPG) to optimize decision-making in obstetric emergencies, with a particular focus on obstetric hemorrhage, a leading cause of maternal mortality.
Objectives: The primary goal is to develop and evaluate a specialized obstetrics chatbot. This AI tool aims to facilitate rapid, evidence-based decision-making in emergency scenarios, thereby improving the quality of care.
Methods: We developed the chatbot using advanced LLMs, GPT-3.5 and GPT-4, integrated with a database of high-quality obstetric CPG. The chatbot's core is powered by the retrieval-augmented generation (RAG) process and the hypothetical document embeddings (HyDE) methodology (3), which ensure that the medical advice provided is accurate and aligned with the latest clinical practices and evidence.
Results: In its initial testing phase, the chatbot demonstrated a capacity for accurately retrieving and providing guideline-specific information. Its performance, evaluated on accuracy, relevance, and completeness parameters, indicated a promising potential in enhancing emergency obstetric care. This represents a significant step toward integrating AI into critical medical decision-making processes.
Conclusions: This chatbot exemplifies a pilot novel, potentially transformative approach in maternal health care. It stands as a testament to the possibilities of AI in augmenting medical decision-making, particularly in resource-strained environments. However, further research and extensive real-world application are essential to establish its effectiveness and reliability in diverse clinical settings.
Limitations: Our study acknowledges several limitations, including the chatbot's reliance on the currency and comprehensiveness of GPC, potential inaccuracies in AI interpretations, and the preliminary nature of current evaluations, which might not encapsulate the chatbot's utility in varied and complex real-world scenarios.
Relevance to Patients: This chatbot could provide health care providers with immediate, evidence-based decision support during obstetric emergencies, potentially improving health outcomes for mothers and offspring. Its development could mark a significant step toward influencing technology to bridge gaps in maternal health care, particularly in underserved regions.
Objectives: The primary goal is to develop and evaluate a specialized obstetrics chatbot. This AI tool aims to facilitate rapid, evidence-based decision-making in emergency scenarios, thereby improving the quality of care.
Methods: We developed the chatbot using advanced LLMs, GPT-3.5 and GPT-4, integrated with a database of high-quality obstetric CPG. The chatbot's core is powered by the retrieval-augmented generation (RAG) process and the hypothetical document embeddings (HyDE) methodology (3), which ensure that the medical advice provided is accurate and aligned with the latest clinical practices and evidence.
Results: In its initial testing phase, the chatbot demonstrated a capacity for accurately retrieving and providing guideline-specific information. Its performance, evaluated on accuracy, relevance, and completeness parameters, indicated a promising potential in enhancing emergency obstetric care. This represents a significant step toward integrating AI into critical medical decision-making processes.
Conclusions: This chatbot exemplifies a pilot novel, potentially transformative approach in maternal health care. It stands as a testament to the possibilities of AI in augmenting medical decision-making, particularly in resource-strained environments. However, further research and extensive real-world application are essential to establish its effectiveness and reliability in diverse clinical settings.
Limitations: Our study acknowledges several limitations, including the chatbot's reliance on the currency and comprehensiveness of GPC, potential inaccuracies in AI interpretations, and the preliminary nature of current evaluations, which might not encapsulate the chatbot's utility in varied and complex real-world scenarios.
Relevance to Patients: This chatbot could provide health care providers with immediate, evidence-based decision support during obstetric emergencies, potentially improving health outcomes for mothers and offspring. Its development could mark a significant step toward influencing technology to bridge gaps in maternal health care, particularly in underserved regions.