NEWS

Comment on new draft guidance on Bayesian methods by Han Chang Chiam

2026.01.13

The FDA has just released a significant new draft guidance titled “Use of Bayesian Methodology in Clinical Trials of Drug and Biological Products” (January 2026, link). The draft guidance focuses on Bayesian methods for primary inference in clinical trials evaluating drug effectiveness and safety.

Clinical trials do not exist in isolation. They are built on accumulating evidence, ranging from preclinical knowledge to previous trials and biological plausibility. Bayesian methods make this explicit by translating previous evidence into a prior, then updating it with trial data using Bayes´ theorem. This approach is particularly valuable in settings such as pediatrics or rare diseases, where standard large-scale RCTs are challenging. This is also the context in which borrowing external information becomes attractive.

The draft guidance highlights specific use cases such as borrowing from previous clinical trials and augmenting a randomized concurrent control using an external control. These topics are closely related to my PhD work at the Medical University of Vienna within the SHARE-CTD network (https://www.sharectd.eu/).

However, these opportunities come with real risks. The draft guidance stresses prespecification of how priors are constructed and how external sources are selected, to avoid the “preferential selection of favorable studies”. It also notes that borrowing often benefits from patient-level data, since this supports a thorough evaluation of relevance and covariate adjustment when populations are not fully aligned.

This connects directly to what I am working on, prespecification and selection bias in trials that utilize historical data, and to a practical challenge many of us face: data accessibility. Access to patient-level data is often difficult, whether for borrowing in new trials or for evidence synthesis more generally, such as meta-analyses and systematic reviews, especially when relevant studies are spread across different sponsors. My recent preprint examines how outcome-dependent choices of external sources can introduce bias, and why prespecification matters when there is such flexibility (https://doi.org/10.48550/arXiv.2510.04829).

Finally, the draft guidance puts strong emphasis on success criteria and operating characteristics, including simulation-based evaluation, sensitivity analyses, and clear reporting of prior influence, such as effective sample size. This shows the need for comprehensive assessment when designing complex innovative trials that borrow external information.

Overall, I see this guidance as both encouragement and a call to action. We are in an era where data generation and data sharing are more feasible than ever, and using shared historical data to improve future trials can be one of the highest impact ways to honor the contribution of trial participants. But to do it well, we need good policies, better access to patient-level data, and a culture of prespecified, reproducible methodology. Through SHARE-CTD, I have learned a lot about both the methodological and practical aspects of doing this responsibly, and I hope this guidance encourages more researchers, sponsors, and regulators to engage with this field and help maximize the impact of clinical data.

Reply to Trinquart and Stockler’s comments on clinical trial data sharing

2026.01.09

Trinquart and Stockler (https://lnkd.in/eJGhQcyR) propose in a viewpoint, published during summer 2025 in JAMA Oncology, the routine release of original stripped-down datasets as pragmatic entry points toward greater readiness for data sharing. However, the degree to which these approaches advance genuine data-sharing readiness warrants careful consideration.

A stripped-down dataset is a deliberately simplified version of a full research dataset. They are inherently assumption-dependent and support only a narrow range of inferential questions. Different underlying data settings can yield identical stripped-down summaries, limiting their usefulness for assessing bias, missing data mechanisms, or alternative analytic strategies (identical stripped survival data ≠ identical estimands).

We suggest positioning stripped-down datasets as a stopgap, exploratory approach, while prioritizing efforts to improve the efficiency and proportionality of data access mechanisms. Streamlined data-use agreements, tiered access models, and clearer journal expectations for IPD availability are more likely to foster sustainable and trustworthy data sharing in oncology and elsewhere.

The full comment can be found at Zenodo: https://lnkd.in/eqiWGVku.

Third Schooling in June 2026 online

2026.01.07

Led by Prof. Dr. Florian Naudet (University of Rennes, UNIVREN) the third SHARE-CTD schooling will take place online from June 1 to June 12, 2026.

Third Datathon in September 2026 in Göttingen

2025.12.19

Led by Prof. Dr. Ulrich Sax (University Medical Center Göttingen, UMG) the third SHARE-CTD datathon will take place at Rechenzentrum Göttingen from 7th to 11th of September 2026.

"Creativity"

2025.12.17

On March 4, 2026, another online workshop will take place for our 11 doctoral students. The topic of the workshop is “Creativity” and will be held by C. Rémi.

Second Datathon in February 2026 in Utrecht

2025.09.09

Under the lead of PhD Assistant Professor Valentijn de Jong (University Medical Center Utrecht) the second SHARE-CTD Datathon will take place in Zeist from February 9 to 13, 2026.