This webinar is free to both Members of PSI and Non-Members.
\nTo register for this event\, please click here
\nThe rapid adoption of AI in pharmaceutical de
velopment has created both challenges and opportunities for CMC statistici
ans. While initial AI enthusiasm sometimes prioritized speed over statisti
cal rigor\, leading voices now recognize that sustainable AI implementatio
n requires balancing agility with robust data governance and statistical r
igour.
\nThis webinar presents two complementary perspectives on succ
essfully integrating statistical expertise into AI initiatives. Statistici
ans bring unique value through asking the right questions\, applying criti
cal thinking\, and understanding both the nature of the objective and the
data. They articulate and communicate model limitations while advocating f
or data integrity\, interoperability\, and contextualization. These core c
ompetencies become particularly critical in assessing and validating agent
ic models\, where statistical frameworks ensure AI systems meet the string
ent quality standards required in pharmaceutical manufacturing while maint
aining the agility needed for innovation.
\nTina Lang (Bayer) present
s "The Importance of Being Human\," sharing lessons on meaningful statisti
cal inclusion in AI programs. Christian Schmid (Roche) follows with "Stati
stical Frameworks for Computerized Validation of Generative AI".
\nTo
gether\, these presentations illustrate that as AI matures in pharma\, the
statistician's role evolves from gatekeeper to strategic partner: ensurin
g innovations are both rapid and solid.
| Speaker \; | \nBiography \; | \n\;Abstract | \ n
| Tina Lang\, PhD \n Bayer Pharmace uticals AG \; | \n Tina Lang studied statistics at TU Dortmund where she also earned her PhD. She joined Bayer in 2009 as a preclinical statisticia n\, supporting in vitro and in vivo experiments across the indication port folio. Additionally\, she is highly invested in elevating statistical lite racy within the company and leads several training initiatives. In 2024 sh e transferred to the department of Data Science &\; Artificial Intellig ence\, where she has more insights into AI tools and their use within the drug development process. | \nQuality in Preclinical Research - The Impor
tance of Being Human \; \n Preclinical experi ments form the empirical foundation of translational medicine by assessing the feasibility\, safety\, and efficacy of new therapeutic approaches. Ye t\, unlike the highly regulated standards of clinical trials\, preclinical research often exhibits substantial methodological heterogeneity\, leadin g to concerns about reproducibility\, bias\, and the robustness of conclus ions. These challenges are further compounded by the emerging use of artif icial intelligence (AI). While AI has the potential to increase efficiency and support data analysis\, uncontrolled or poorly understood application s can amplify existing weaknesses in study quality. In this presentation\, we discuss the critical importance of human judgment\, statistical rigor\ , and transparent study design for ensuring reliable preclinical evidence. We examine key dimensions of analytical quality across study design\, dat a generation\, and evaluation\, and consider how AI - particularly large l anguage models (LLMs) and foundation models - can support\, rather than un dermine\, methodological integrity. From a biometrical perspective\, safeg uards for uncertainty quantification\, error control\, and interpretabilit y are essential when integrating AI into preclinical research. Ultimately\ , methodological progress depends on a careful balance between human exper tise\, statistical inference\, and computational tools. Illustrative examp les highlight both opportunities and pitfalls at the interface of biostati stics\, data science\, and translational medicine\, emphasizing that maint aining human oversight is crucial for generating reproducible and interpre table evidence. \; | \n
| Christian Schmid\, Ph
D \n F. Hoffmann &ndash\; La Roche AG | \n
Christian Schmid i s a Senior Statistical Scientist at Roche and a founding member of the CMC Statistical Network Europe\, where he has served on the steering committe e since its inception. In his role\, he primarily supports Roche&rsquo\;s commercial manufacturing and has recently expanded his work to include sev eral AI validation initiatives. He earned his PhD in Statistics from The P ennsylvania State University\, with a specialization in computational meth ods. | \nValidating GenAI (Generative Artificial IntelligenceI) in GxP (G
ood x Practice) environments requires moving beyond traditional "exact mat
ch" testing to a framework built on statistical confidence and risk manage
ment \n This presentation proposes a validation f ramework for GenAI applications\, where rigorous testing strategies are us ed to quantify the reliability of non-deterministic outputs. We will explo re technical controls for risk mitigation\, specifically the use of Retrie val-Augmented Generation (RAG) architectures and parameter adjustments\, s uch as temperature and top-p\, to control output variability. The session will cover specific methods for combining automated scoring with structure d Subject Matter Expert (SME) review to evaluate correctness and detect ha llucinations\, ensuring that probabilistic systems meet the safety standar ds required for regulated healthcare applications. | \n