Best Practices for Implementing AI and ML Technology in a Regulated Environment

Artificial intelligence (AI) and machine learning (ML) algorithms have been significant drivers of innovation in recent years, allowing for new applications for data analytics and automation. In many industries, AI and ML have already delivered evidence of their importance in improving and optimizing processes. For this reason, their implementation in the pharmaceutical industry is essential.

However, the integration of new technologies into a regulated environment can be a challenge. New technology has the potential to open organizations up to new security risks. While it is possible with traditional software and devices to generate reproducible data and thus guarantee security, AI behaves differently. It evolves over time and learns continuously to improve accuracy. Hence, it takes time and intense oversight to develop new solutions that meet the many layers of regulatory requirements, ensure adequate data protection, and support ISPE data integrity.

In its 2nd edition of GAMP 5 Good Practice Guide: A Risk-Based Approach to Compliant GxP Computerized systems, ISPE addresses these types of systems to give guidance on how to use AI- and ML-driven solutions within a regulated environment.

This article will highlight the three-phased approach to implementation outlined in GAMP 5: the concept phase, the project phase, and the operation phase. By following this approach, organizations can ensure that their AI and ML technology implementation will benefit their organization and fully comply with all regulatory requirements.

1. Concept phase

In the concept phase, the organization identifies the business need or opportunity that the AI or ML system will address. Identifying this requirement provides critical context and a precise gauge of project progress.

For AI and ML systems, data management also begins here. An initial set of data is identified, selected, and acquired, either internally or externally. SMEs can start assessing the data for quality and possible bias, following ISPE data integrity best practices. The data must be curated, with attention to reprocessing, classification, and transformation. This allows the AI developer to start prototyping and select the most suitable model and an initial set of hyperparameters.

2. Project phase

The project phase is when the implementation and validation of the system begin. The ML sub-system is usually built iteratively and incrementally. These activities include the production and selection of model designs, engineering, model training, testing, evaluation, and hyperparameter tuning.

As data is an integral part of the algorithm, sound data management is necessary. It is critical to maintain a clean split between training and testing data. Risk management must also take into account the specific challenges of AI solutions. Changes to production or errant data, the use of external vendors/supplier data, or system performance and downtime must be considered and assessed as additional risks. In general, additional controls, periodic reviews, and monitoring of the model performance are recommended to identify potential bias, overfitting, or harmful developments.

3. Operation phase

The operation phase encompasses more than the point when your AI and/or ML solution goes live—it also includes the time when the system performance is monitored and evaluated. Over time, as new data becomes available, the AI algorithm will require enhancements. To ensure ongoing improvements and the best possible results, it is essential to have an effective change and configuration management process in place.

Conclusion

AI and ML solutions will become increasingly entwined with pharmaceutical industry best practices. Installing processes for implementing these advanced technologies will help pharma companies gain a competitive edge while maintaining compliance at every stage.

Innovation always comes with challenges, which makes it invaluable to have an experienced partner at your side when you’re ready to embrace change. GxP-CC is YOUR partner in implementing AI and ML in all lifecycle phases. GxP-CC has a broad knowledge base that includes IT, engineering, and mathematics, as well as activities in AI and ML industry groups, and extensive experience in Life Sciences. With this perfect background, GxP-CC offers you comprehensive support and advice in any circumstance. Contact us today to learn how we can move forward together.