The Problem
There is no uniform approach to AI implementation, evaluation, and validation across countries and there is limited information about contextual differences across settings. Furthermore, the social and technical interactions throughout the procurement, development and adaptation, clinical integration, and lifecycle management stages of AI implementation in different countries are poorly understood. Best practices developed in one national context are not efficiently adapted and adopted in different countries. And there is limited understanding of how to best diffuse or transport AI technologies across national contexts.
Our Solution
In the initial phase of the project, we focus on capacity building, best practice diffusion, and AI solution scaling between Duke Health and Aga Khan University in Kirachi, Pakistan. We first systematically reviewed AI procurement, integration, monitoring, and evaluation frameworks from across the world. Then, we conducted interviews with stakeholders across the healthcare ecosystem in Pakistan to understand barriers and enablers to AI adoption. We leveraged work from Health AI Partnership in the US context to complement our findings from Pakistan. To surface social and technical interactions throughout the stages of AI adoption, we then produced algorithm journey maps across settings. Using our nuanced understanding of stakeholders and steps in the AI adoption process, we built milestone-specific documentation templates to guide healthcare delivery organizations through the process. Lastly, we compare local manufacturing and global importation approaches to scale AI across borders between the US and Pakistan. In order to develop and validate AI across settings, we built data curation and data quality assurance capabilities in Pakistan.
Anticipated Impact
Our project has rapidly advanced the capabilities and expertise among colleagues at Duke and AKU to develop, implement, and scale AI solutions across the world. By deeply understanding differences across US and Pakistan contexts, the teams are able to navigate different levels of digital infrastructure, government regulation, data availability, and clinician readiness to develop and implement AI. This project also has built essential capabilities in Pakistan to curate data for AI development.
This project is funded by the Patrick J. McGovern Foundation
