Mitigating AI Bias In Healthcare.

There is an opportunity to leverage AI to transform healthcare—if we build it with everyone in mind. We need a person-centred/community-centred approach to co-develop models. Otherwise, AI can become harmful. It may overlook those who often fall through the cracks because of systemic barriers. These include underrepresented groups in medicine and racialized communities.

A study by Obermeyer et al. (2019) revealed that a widely used health-risk algorithm flagged only 17.7% of Black patients with complex needs. In contrast, it flagged 46.5% of White patients with similar burdens. This disparity occurred because it used healthcare costs as a stand-in for medical need. Black patients historically incur lower costs due to unequal access. At any given risk score, Black patients were measurably sicker. They experienced more complications than their White counterparts. This suggests that the model may have missed those who are often overlooked in the system. Another recent review in NPJ Digital Medicine examined how biases are introduced into AI throughout its entire lifecycle. The review discussed how to prevent them at every step.

Bias can be introduced at every stage. This includes conception, data collection, data processing, and model deployment. Co-creating interventions and models that can address the needs of everyone means capturing those who fall through the cracks. We can do this by:

• Listening to the needs of individuals and communities being served: Bring patients, clinicians, community leaders, and data experts together. They can map out unmet needs and co-create solutions.
• Designing Together: Co-designing prototypes and testing and iterating these with them: Ask: “Does everyone understand why this recommendation came up? Does it feel fair?” “Who are we missing?” Who should be included?”
• Building Feedback Mechanisms: Embed feedback mechanisms and iterative testing that allows you to compare AI performance across groups. Modify, test, and iterate.
• Adaptation and implementation: Create processes that enable you to adapt feedback to your models. Implement and obtain real-time feedback on an ongoing basis. Ensure the models are updated as the needs of the communities change.

To leverage our expertise and better integrate health equity considerations into your models, partner with D.A.S. Innovative Hub Consulting for a hands-on co-design workshop and embed our recommended best practices into your health solutions today.

Reach out to us by clicking the Contact Us link: Contact Us. We look forward to working with you.

Referenced Studies: Dissecting racial bias in an algorithm used to manage the health of populations – PubMed and Bias recognition and mitigation strategies in artificial intelligence healthcare applications | npj Digital Medicine