When and how to use GenAI in late-stage clinical development

GenAI and related tools can help improve the design of clinical trials and speed data gathering and regulatory reporting.

In brief

• Using GenAI and other technologies like agentic AI in late-stage clinical trials can be the next step for biopharmaceutical companies to improve drug discovery.
• Considerations such as data security, costs and talent requirements need to be addressed for companies to get more comfortable using these tools.
• EY-Parthenon teams work with companies in using GenAI in late-stage clinical development has shown benefits in several parts of the trial process.

Biopharmaceutical companies are increasingly considering or using generative artificial intelligence (GenAI) across the drug discovery and early-development processes, harnessing the technology’s potential to interpret large data sets and suggest novel molecules and/or structures tailored to the treatment of a diverse set of diseases.

Now may be the time for CEOs, chief strategy officers and chief technology officers to move to the next step: leveraging the power of GenAI, as well as related technologies such as machine learning and agentic AI, in late-stage clinical development. These technologies offer potential cost reduction through automation of data analysis and other administrative tasks; enhanced clinical trial design along with more efficient and effective trials; and streamlined preparation of regulatory documents and improved accuracy of submissions.

However, pharma and biotechnology executives need to address several areas to get more comfortable with using GenAI in late-stage clinical development. EY-Parthenon teams’ work with pharmaceutical companies shows several key considerations about when GenAI can help in clinical trials and steps to effectively utilize the technology in those cases.

These considerations and steps include:

• Cost and benefit implications: A thorough cost-benefit analysis is critical to understanding the financial implications of adopting GenAI, including evaluating the potential for cost reduction, efficiency gains in clinical trials, and return on investment (ROI).
• Potential integration with existing processes: Determine whether GenAI will integrate with current workflows and systems. This includes identifying the specific goals the AI model will address and establishing alignment with clinical and product-quality objectives.
• Regulatory compliance: Confirm that GenAI applications meet regulatory requirements and guidelines, such as those outlined by the FDA.
• Data governance policies: Establish robust data governance practices to achieve data integrity, quality and security.
• Risk management: Assess the risks associated with AI models, including validation, reproducibility and explainability. Decision-makers need to evaluate the potential impact of incorrect AI outputs and implement risk mitigation strategies.
• Continuous learning and adaptation: Implement feedback loops to continuously update AI models with new data from ongoing trials and studies, improving model accuracy and relevance over time. Encourage a culture of experimentation, where teams are open to testing and refining AI applications based on real-world outcomes.

Practicing good data hygiene

GenAI is only as good as the data that goes into the models, but the use of patient data can be especially sensitive. In fact, data quality and security are key concerns we regularly hear about from pharma executives about using GenAI. Pharma companies can help keep data safe by developing and implementing data encryption solutions.

In terms of data quality and quantity, companies can generate new high-quality data like RNA-seq and multiomics (which combines data from disciplines such as genomics and metabolomics). They can also unlock value from decades of existing data assets held by biopharma. Pharma service providers and contract research organizations (CROs) also have copious amounts of data, which can be used for portfolio management, target identification and candidate enhancement.

Validation of results and continuous monitoring of the results the GenAI model serves up are also essential, as is metadata management. This in turn leads to the importance of metadata management to help ensure that data is properly categorized, searchable, and traceable throughout its lifecycle, thereby enhancing the overall effectiveness of the governance model.

Reasons pharma companies hesitate to use GenAI

Given the potential benefits, what keeps companies from utilizing GenAI? Some of the concerns we most hear from clients include organizational barriers such as securing leadership support to eliminate organizational push back; poor quality of existing sources of data; finding people with the technical expertise to understand and manage AI; fear or disinterest across the organization, which can make it difficult to scale AI solutions; lack of consistent metrics to measure success and general resistance to change.

Other barriers include concerns about protecting patient information, the ever-changing regulatory environment, and unintended consequences that GenAI may include unconscious biases in the output of its models. In late-stage clinical development, where decisions have significant real-world implications, these biases could perpetuate health disparities or lead to inequitable access to treatments. Addressing these concerns requires diverse data sets, inclusive development practices, and the use of responsible AI to achieve transparency and fairness.

Next steps for implementing GenAI in late-stage clinical trials

Given these concerns, what steps can biopharma leaders take to move forward with GenAI?

• Establish a governance steering committee of executive leadership from across the organization to champion this initiative and define the specific use cases. To limit any ethical- or bias-related issues associated with the GenAI model, develop appropriate algorithm auditing and explainability guidelines as well as robustly validate outputs in real-world settings.
• Assess gaps in talent and technology infrastructure between the current state of the organization and its future GenAI needs while encouraging a culture of experimentation in the process. Specifically, when building a GenAI-enabled tool organizations need to have data analysts or architects on the team who have built not only these tools but have experience in using health care data, given the stringent rules around how health care data can be used.
• Develop and pilot an agile test of GenAI to see how it can benefit a clinical trial compared to typical trials in a given therapeutic area or indication.
• Define the value you are trying to achieve and assess, prioritize and track AI initiatives based on it such as trial efficiency (time to first person in (FPI), trial duration), cost savings (cost per patient, overall trial cost), data quality and integrity, regulatory compliance (regulatory approval time, compliance rate), and patient outcomes (adverse event detection, patient retention rate)

Conclusion

GenAI and related technologies are already useful for improving productivity in basic drug development operations. As the data used in the models becomes richer, the value to pharma companies will increase.

Now is the time to take steps to realize that value in cost savings, identification of new markets and uses for treatments and getting those treatments to patients more quickly.

Adam Berman, EY-Parthenon, and Sanjay Jaiswal also contributed to this article.