AI Regulations in Life Sciences
| BioBuzz is thrilled to welcome guest posts and contributing writers to the “hive.” Enjoy the read from one of our expert contributors and be sure to get in touch if you have an idea or perspective you’d like to share. We’d love to buzz about it to see if it’s a fit! |
Innovation in AI has been going through cycles of rapid development, termed AI spring, and long dormant periods known as AI winters since its inception. At this time we are once again upon an AI spring, some would even characterize it as a Cambrian explosion period for AI, where many revolutionary generative AI models are being proposed and explored every day. Throughout these explosive periods of growth, regulatory bodies have been slow to catch up, and it is no exception in life sciences.
However, many regulatory bodies across the world have been keenly aware of the AI development and have been making necessary steps to address it. As we in the DMV area are close to the seat of government and many organizations that regulate the life science sector, this piece intends to summarize some of those steps. There has also been some cross-collaboration across regulatory bodies in different countries.
Liu et al. paper indicated that regulatory submissions involving AI increased almost tenfold from 2020 to 2021. The vast majority of these submissions are Investigational New Drug Applications (IND) which signals that we can anticipate more drug approval applications to involve AI in the future. Foreseeing these changes, FDA has published the following documents Artificial Intelligence in Drug Manufacturing (Center for Drug Evaluation and Research) and Using Artificial Intelligence & Machine Learning in the Development of Drug & Biological Products (Multiple sub agencies within FDA).
The first document highlights the following examples where AI could be applied in drug manufacturing:
- Process design and scale-up
- Advanced process control
- Process monitoring and fault detection
- Trend monitoring
It also includes a feedback form to gather insights from experts in this area. The second document details the application of AI/ML in four main drug development stages, namely:
- Drug discovery
- Nonclinical research
- Clinical and research
- Postmarketing safety surveillance along with manufacturing
Along with the two articles above, the FDA has also published a document on Good Machine Learning Practices (GMLP) for medical devices in collaboration with Health Canada and Medicines & Healthcare products Regulatory Agency in the UK. GxP is a common set of guidelines used in manufacturing, lab and clinical practices and the goal is to adapt similar practices in machine learning as well. A few weeks ago, the FDA also established an advisory committee on Digital Health Technologies (DHT) to gather expertise on AI / ML applications in Life Sciences.
AI / ML regulations reaches far beyond life sciences, and the U.S. Government Accountability Office (GAO) has put forward an accountability framework for AI practices for federal agencies. Many of the practices that the FDA is discussing are based on this framework, and we could expect they would closely follow the document by the GAO.
A few common threads stand apart across all these documents, they are:
- Human-centered AI development
- Bias management / reduction
- Proper practices to ensure consistency in results
All of the above will undoubtedly accelerate the drug development process while reining in any unintended consequences.
