TL;DR:
- UNC Health focuses on responsible AI framework development in healthcare.
- Partnership with external vendors and in-house AI development.
- Emphasis on addressing bias and discrimination in AI algorithms.
- Structured framework to ensure effectiveness and safety in AI applications.
- Transparent communication with end-users for trust-building.
- Inclusive decision-making involving various stakeholders.
- Commitment to data security, ethics, and patient trust.
- Formation of a multidisciplinary group for responsible AI decisions.
- Collaboration with vendors on responsible AI discussions.
- Enterprise data warehouse with robust data protection measures.
Main AI News:
UNC Health, a distinguished healthcare institution based in North Carolina, is spearheading a remarkable journey toward responsible AI adoption. In a recent discussion during a WEDI webinar on the best practices in artificial intelligence within the healthcare sector, Rachini Ahmadi-Moosavi, the Chief Analytics Officer, and Ram Rimal, the Manager of Data Science Engineering, both of UNC Health, shared insights into their visionary approach.
Their ambitious endeavor dates back to 2016 when UNC Health embarked on a mission to harness the potential of AI to advance healthcare delivery. While partnering with external companies, such as Optum, to integrate cutting-edge AI technology like computer-assisted coding, UNC Health was concurrently developing its proprietary AI solutions. Use cases ranged from enhancing case duration accuracy to augmenting sepsis detection capabilities.
With this technological influx, the imperative for responsible AI practices became paramount. Ahmadi-Moosavi emphasized the need to ensure responsible development, whether through in-house creation or third-party procurement, to provide the most optimal solutions to the healthcare system.
Rimal delved into the rationale behind UNC Health’s conscientious approach to AI governance. He elucidated how biases and discrimination can become embedded in algorithms, emphasizing the importance of robust checks and balances. A comprehensive responsible AI system, he noted, scrutinizes every facet to mitigate bias and discrimination effectively.
Having a structured framework serves as a cornerstone for ensuring healthcare’s due diligence towards responsible AI. As Rimal eloquently put it, “The lives of everyone coming to the healthcare system are valuable.” Thus, it is paramount to ensure that every AI application is not only effective but also safe.
Communication about AI models emerges as a critical challenge. Rimal, drawing from his background as a data scientist, stressed the importance of transparent communication with end-users. Establishing trust hinges on effectively conveying what the model does and how it can be relied upon. This necessitates a rigorous and transparent process facilitated by the right framework.
Rimal further underscored the significance of inclusivity in decision-making. He advocated for involving all stakeholders from the outset, whether patients, clinicians, or customers. This inclusive approach, he argued, is facilitated by the framework of responsible AI.
Data security and ethical considerations also loom large. UNC Health recognizes the importance of ensuring patient trust through privacy, security, and ethics. The responsible AI framework plays a pivotal role in reaffirming these commitments.
A telling example from the past was the development of a custom sepsis model within the Epic system. This project expanded its scope by engaging multiple technical teams and clinical experts, allowing for an iterative process that addressed not only modeling but also workflow issues.
Fast forward to 2023, UNC Health has evolved significantly. The challenges surrounding AI utilization prompted the formation of a systemwide multidisciplinary group. Comprising experts from various domains, including IT, finance, ethics, law, human resources, and clinical fields, this group ensures diverse voices converge to make responsible AI decisions. Furthermore, UNC Health has extended its commitment to responsible AI by engaging with vendors in meaningful discussions, despite the complexities of intellectual property protection.
Ahmadi-Moosavi, addressing concerns about data privacy and security related to AI, highlighted their proactive approach. They have established a robust enterprise data warehouse with multiple layers of data protection and security. This ensures that information flows securely from source systems to various consumption layers, while also guaranteeing appropriate access rights for the data science team.
In the grand scheme of data management, security discussions are encompassed by UNC Health’s data governance council. Collaborations with data security, privacy, and legal teams ensure a holistic approach when considering data usage for AI and other transformative outcomes.
Conclusion:
UNC Health’s proactive approach to responsible AI governance sets a significant precedent in the healthcare market. Their commitment to transparency, inclusivity, and ethical considerations not only ensures patient trust but also positions them as a leader in responsible AI adoption, potentially influencing the broader healthcare industry’s approach to AI integration.
