Can AI revolutionize health management in Burkina Faso?

By Cal Pierce, Marie-Jeanne Offosse, and Dai Hozumi

Introduction

Burkina Faso, a landlocked country in West Africa, faces numerous challenges in its health care system. District health managers are at the forefront, battling resource shortages, data management issues, and the ever-present need for improved patient care. What if these district health managers had a powerful ally in artificial intelligence (AI) to help them navigate these complexities? ThinkWell’s recent initiative aims to do just that, exploring the potential of OpenAI’s GPT-4 to transform health management in this dynamic yet resource-limited setting.

Meet Dr. Sanou

Meet Dr. Sanou*, a devoted physician and district health manager responsible for overseeing 20 health facilities in his district. Each day, he juggles a myriad of responsibilities—from managing drug supplies to coordinating staff and ensuring quality care for patients. However, Dr. Sanou often finds himself struggling to make sense of data from different health information systems. The time-consuming nature of analyzing and interpreting various datasets, combined with potential lack of confidence in data analysis, can hinder effective decision-making. This is a common scenario for many health care professionals in low- and middle-income countries (LMICs), where the need for reliable data and analytical support is paramount.

Exploring AI’s Potential

To address these challenges, ThinkWell embarked on an exciting journey to assess how readily available and popular large language models (LLMs) like GPT-4 could support district health management, without the need to create new applications or tools. Our mission was to see if these models could handle imperfect, real-world data and still deliver valuable insights that could empower health managers like Dr. Sanou. Throughout the process, ThinkWell’s team in Burkina Faso collaborated extensively with district health managers to ensure that the solutions developed were closely aligned with their specific needs and operational realities.

A Streamlined Approach

Our approach was straightforward yet strategic: we began with simpler tasks using smaller, pre-prepared data sets and then progressively tackled more complex analysis on complex, real-world data sets.

During our testing phases, GPT-4 demonstrated its ability to handle various tasks effectively, including the following:

  • Conducting simple quantitative analyses: We asked GPT-4 to tell us how many days in a month the average facility faced a stock out of essential drugs. The AI processed the data and provided clear, actionable numbers.
  • Generating insightful data visualizations: GPT-4 was tasked with creating visual representations of stock-out trends over time. The resulting charts helped district managers quickly identify problem periods and patterns.
  • Offering preliminary recommendations: We queried the AI on which facilities required urgent attention based on their stock-out rates. GPT-4 analyzed the data and recommended targeted interventions for the worst-performing facilities.

Key Learnings

Throughout this project, we uncovered three critical factors for effectively using LLMs in to support performance management at the district level:

  • Data Quality and Structure: The importance of clean, well-structured data cannot be overstated. Inconsistent naming conventions and incomplete data fields significantly hinder the accuracy of insights generated by the AI.
  • Question Framing: How you ask questions matters. Specific, detailed prompts lead to more relevant and useful outputs, making it crucial to frame queries carefully.
  • Iterative Interaction: Achieving actionable results often requires a back-and-forth dialogue with the AI. This iterative process is essential for refining and enhancing the quality of the insights provided.

Optimistic Outcomes and Future Directions

The potential for AI to enhance district health management in Burkina Faso is truly exciting. After sharing the results of our exercise, one of our district health managers captured this sentiment perfectly, saying, “The AI’s ability to process and analyze complex data sets quickly has the potential to transform our approach.”

This feedback reinforces our belief in the transformative power of AI. With the promising results from our initial tests, we are eager to push forward. Here’s how we plan to do it:

  • Expand Testing on Larger Data Sets: We will continue to refine our approach by incorporating more extensive and varied real-world data.
  • Test Across Multiple Settings: To ensure versatility of our approach, we’ll expand our testing to include a range of different environments and conditions. This will validate the AI’s effectiveness across diverse health care management settings.
  • Continue Real-World Evaluations with District Health Managers: Ongoing collaboration with district health managers will be essential. Their real-world feedback and insights will guide us in fine-tuning our approach to better meet their needs.

Moving forward, our journey with AI in Burkina Faso is just beginning. The initial successes have laid a strong foundation, but there is much more to explore and achieve. We plan to continue our close collaboration with district health managers, refining our approach and exploring new ways to leverage AI tools effectively within their unique contexts.

We’re exploring different ways to make AI more useful in day-to-day health management. Over the coming weeks, we plan to develop an easy-to-follow guide that helps managers organize their data in a way that works well with AI. We’re also looking into compiling a list of helpful questions that managers can ask the AI to get the most useful information for their work.

Conclusion

Our work in Burkina Faso highlights the exciting potential of AI in global health. By leveraging available and accessible technologies like GPT-4, we can begin to tackle some of the day-to-day (or on-the-ground) health management challenges in low- and middle-income countries. However, much more needs to be done. Testing and refining  effective use of AI tools with real-world data, enhancing data quality, and scaling solutions to diverse settings are essential steps forward. The journey is promising, but it is just the beginning. With continued innovation and collaboration, we can unlock even greater benefits for health care systems worldwide.

*Dr. Sanou’s name has been changed for the purposes of this article.


Acknowledgements

This work was funded by ThinkWell, but would not have been possible without ongoing support from the Bill & Melinda Gates Foundation in Burkina Faso through the Strategic Purchasing for Primary Care (SP4PHC) project.