【斯坦福大学】2025生成式AI在中低收入国家健康领域的应用白皮书_52页_4mb

Summary of "Generative AI for Health in Low & Middle Income Countries"

Core Content

This white paper explores the current state and potential of Generative AI (GenAI) in improving health and healthcare in Low- and Middle-Income Countries (LMICs). It provides an overview of use cases, challenges, and opportunities for leveraging GenAI in health-related behavioral change (HBC) and broader healthcare applications. The report includes case studies, survey results, and insights from experts, stakeholders, and implementers in the field.

Main Viewpoints

• GenAI is a type of artificial intelligence that can generate new, meaningful content such as text, images, or audio from training data.
• Large Language Models (LLMs) are a subset of GenAI that are trained on vast amounts of text and can understand and generate human-like language.
• LMICs are defined by the World Bank using gross national income (GNI) per capita.
• Health behaviors are actions taken by individuals that impact their health or mortality, such as smoking, diet, and adherence to medical treatments.
• Human-in-the-loop (HITL) is a critical component in many GenAI applications, ensuring human oversight and intervention to control outcomes.
• Retrieval-Augmented Generation (RAG) is a technique used to improve the accuracy of LLM outputs by integrating relevant external data.

Key Use Cases

GenAI applications in LMICs are typically categorized into three main types:

1. Direct-to-Consumer

   • Provide personalized counseling and recommendations for sensitive health topics.
   • Use conversational agents and voice-based systems to improve engagement, especially in low-literacy settings.

2. Direct-to-Provider

   • Support healthcare workers with communication, triaging, and translation tasks.
   • Enhance efficiency in handling patient queries and improve clinical decision-making.

3. System-Level

   • Enable early warning systems for potential pandemics.
   • Optimize healthcare workflows and support medical research.

Key Findings from Scoping Analysis

• 285 grants from 14 GenAI accelerator programs, funded by 10 organizations, were analyzed, supporting 279 projects.
• Health systems strengthening (HSS) was the most commonly funded health area, accounting for 31.9% of projects.
• Communicable diseases (16.1%) and Maternal, Newborn, and Child Health (MNCH) (14.4%) were the next most funded health areas.
• System-level interventions represented 41% of GenAI use cases, followed by direct-to-consumer (37%) and direct-to-provider (22%).

Key Recommendations

1. Share Learnings

   • Develop and maintain a regular update process for practical guidance on identifying and validating LLM applications.
   • Use consistent outcome metrics to enable meaningful comparisons and benchmarking.

2. Focus on Actionable Measurement

   • Establish clear standards for measuring benefits, costs, and risks.
   • Encourage partnerships between implementers and academics to ensure rigorous and timely evaluation.

3. Improve Language & Localization

   • Create standardized measures to evaluate model performance across languages and health contexts.
   • Curate high-quality, region-specific datasets, including voice data for low-literacy populations.

4. Improve Technical Capacity & Shared Infrastructure

   • Identify and centralize technical infrastructure elements.
   • Offer technical consulting and capacity-building support to health system leaders, funders, and implementers.

5. Improve Digital & Basic Health Infrastructure

   • Prioritize investment in both digital and physical health infrastructure.
   • Evaluate an organization's digital readiness before deploying AI tools.
   • Ensure AI tools align with existing infrastructure to avoid costly failures.

Survey Results

• 145 respondents (25% response rate) participated in the survey, representing various roles: health implementers (37%), tech facilitators (19%), academics (18%), health funders (12%), and health system experts (5%).
• 51% of respondents were actively involved in GenAI projects for health in LMICs.
• Health education and awareness (50%) was the top priority use case, followed by clinical decision support (44%) and health-related behavior change (43%).

Case Studies

• Five case studies were highlighted, focusing on health-related behavior change in LMICs.
• These include:
  • Viamo: Ask Viamo Anything (AVA)
  • Girl Effect
  • Audere: Self-Care From Anywhere
  • Noora Health
  • Other HBC-focused applications

Conclusion

The paper emphasizes the importance of collaboration, localization, and evidence-based implementation in realizing the full potential of GenAI in LMIC healthcare. It calls for sustained focus on improving technical capacity, digital infrastructure, and health systems to support scalable and impactful applications of AI in health behavior change and broader healthcare services.

Acknowledgements & Research Team

• The report acknowledges the contributions of various stakeholders and organizations.
• It was authored by a team including experts in Generative AI and digital health, with input from case study teams, interviewees, and participants in roundtables and surveys.

Appendix & References

• The report includes an appendix with the list of included GenAI accelerator programs.
• References to supporting literature and data sources are provided for further reading.