Last month I had a great pleasure of participating in an AAIH sponsored panel at BIO 2024 in San Diego joined by an esteemed panel of experts in drug development, SaaS, and law. The panel focused on AI and regulatory innovation in healthcare. With great interest and thoughtful questions from the audience, we explored how recent AI regulation has evolved and examined through the lens of a SWOT analysis (i.e. strength, weakness, opportunity, and threats) how these changes can impact the future.
Feel free to review the key points and a brief SWOT analysis and watch the panel.
I decided to add more detail around an area I find pressing to work out.
Balancing Innovation and Safety
The potential for artificial intelligence (AI) to revolutionize healthcare delivery and improve countless lives is immense. However, striking the right balance between innovation and oversight is a delicate task, especially in a highly risk-averse field like healthcare. In a previous article, "Implementation: Where AI Programs Go to Die," I explored some of the reasons why this occurs and provided recommendations on how to resolve the complexity and challenges in integrating AI into healthcare workflows. But before implementation, we must experiment and innovate, which requires getting comfortable with risk and failure. Given the high stakes in healthcare, it is crucial to have a clear approach from design to implementation.
Practical Recommendations for Balancing Innovation & Safety
Start with the optimal results and work backwards: Assess what is needed and what the solution would look like. Keep in mind that you might not need a full-blown overhaul of your current workflow; small changes introduced by automation and AI could be a good starting point. Always aim for a simple and elegant solution.
Assemble an interdisciplinary team of experts: Bring together professionals from various fields to guide the design and map out risks and hurdles. This is key to assessing potential safety and compliance challenges.
Focus on non-clinical administrative areas initially: Begin with areas that pose a lower risk and can serve as a success case study before moving on to clinical decision-making workflows. While clinical workflows have major upside, they are not the ideal place to learn and experiment. If considering clinical use cases, ensure comprehensive testing alongside the standard workflow before transitioning, and run both workflows side by side, feeding the results back into the model for improvement.
Collaborate with clinicians throughout the process: When appropriate, involve clinicians early and throughout the design and testing phases to improve the performance, accuracy, and sensitivity of your algorithm.
Roll out AI to an enthusiastic small group first: This will help identify and resolve any bugs before making your larger team lose patience and interest.
Embrace failures as a necessary part of innovation: While failures are understandable and inevitable, they are not without repercussions. However, they are a fundamental part of the innovation process. Considering the previous steps will help reduce risk and potential blindsides, but it's essential to become comfortable with temporary setbacks.
Case Study: Implementing AI in Radiology
One successful example of AI implementation in healthcare is the use of AI algorithms to assist radiologists in detecting and diagnosing diseases from medical images. A leading healthcare organization assembled a diverse team of radiologists, data scientists, and IT professionals to develop an AI system that could analyze chest X-rays and identify potential cases of pneumonia.The team began by focusing on the desired outcome—improving the accuracy and efficiency of pneumonia diagnosis—and worked backwards to determine the necessary steps and resources. They collaborated closely with radiologists to ensure that the AI algorithm was trained on high-quality, diverse datasets and that its performance was rigorously tested against the standard diagnostic workflow. Before deploying the AI system across the organization, the team piloted it with a small group of enthusiastic radiologists who provided valuable feedback and helped refine the algorithm. The pilot also allowed the team to identify and resolve any technical issues or workflow challenges before the full-scale implementation.Throughout the process, the team remained open to learning from failures and iterating on their approach. They continuously monitored the AI system's performance and made adjustments as needed to ensure its accuracy and reliability. As a result of this careful, collaborative approach, the organization successfully integrated the AI algorithm into its radiology workflow, leading to improved diagnostic accuracy and efficiency.
Ongoing Monitoring and Evaluation
Once AI systems are implemented in healthcare settings, it is crucial to continuously monitor and evaluate their performance to ensure they continue to operate as intended and to identify any unintended consequences or biases that may emerge over time. Regular audits, performance assessments, and updates to the AI models and workflows should be conducted to maintain their accuracy and reliability. This is another area where careful planning ahead of time is crucial to understand the level of long-term investment and resources are needed before committing to the program.
Looking Ahead
Balancing innovation and safety is a critical challenge in implementing AI in healthcare. By following practical recommendations, such as starting with optimal results, assembling interdisciplinary teams, focusing on low-risk areas initially, collaborating with clinicians, and embracing failures as learning opportunities, healthcare organizations can successfully integrate AI into their workflows while mitigating potential risks. Ongoing monitoring and evaluation, addressing ethical considerations, and fostering collaboration and harmonization efforts will be essential to realizing the full potential of AI in healthcare. As we continue to navigate this exciting but complex landscape, ongoing dialogue and adaptation will be key to ensuring that AI is used responsibly and effectively to improve patient outcomes and transform healthcare delivery.
As always, feel free to reach out to me via email to discuss how we can support your business.
Cheers,
Mida