September 23, 2026
Special seminar: Igor Shuryak on validating causal models when there is no ground truth
PAC-AID is sponsoring an in-person special seminar on Thursday, October 8, 2026, 11:30 AM to 12:30 PM HST in the Sullivan Conference Center at the University of Hawaiʻi Cancer Center. Igor Shuryak, MD, PhD, Associate Professor of Radiation Oncology at Columbia University Irving Medical Center, presents “How do you know a causal model is right? Validating treatment-effect estimates when there is no ground truth.”
This seminar is separate from the monthly PAC-AID Talks. Please join us in person if you can. If you can’t make it, you can join on Zoom.
Meeting ID: 845 5230 5635 · Passcode: 502787
About the talk
You can check a predictive machine learning model against a held-out test set, because the quantity it predicts is eventually observed. A causal estimate has no such answer key. We never see the outcome a patient would have had under the treatment they didn’t receive, and no cohort size or better learner can supply it. This talk is about what you can do instead.
Dr. Shuryak uses an analysis of radiation dose escalation in stage III non-small cell lung cancer as a running example. He works through a practical sequence for building confidence in a treatment-effect estimate. First, the cohort is designed and the adjustment set is locked before anyone chooses an estimator. Then the result is challenged from several independent directions:
- Is it stable across estimators that rest on different assumptions?
- Does the same pipeline recover a known effect in simulated and semi-synthetic data, and return a null when there is none?
- How strong would an unmeasured confounder have to be to explain the result away?
The talk also covers what none of these checks can settle, and why a good predictive score can point the wrong way about a treatment effect. A causal estimate earns confidence by holding up under these challenges. No single test can prove it is correct.
About the speaker
Igor Shuryak, MD, PhD, is Associate Professor of Radiation Oncology at the Center for Radiological Research, Columbia University Irving Medical Center. He is also an Associate Member of the Precision Oncology and Systems Biology Program at the Herbert Irving Comprehensive Cancer Center. He holds an MD from SUNY Downstate College of Medicine and a PhD in Environmental Health Sciences from Columbia University. His PhD, awarded with distinction, developed new approaches to quantitative mechanistic modeling of radiation-induced carcinogenesis.
His current work centers on causal machine learning for radiation oncology. He estimates treatment effects from registry-scale and clinical data with doubly robust and forest-based estimators, and asks how far such an estimate can be trusted when the counterfactual outcome is never observed. He also works on mechanistic models of dose, fractionation and tumor repopulation, on radiation biodosimetry, and on cancer risk after radiation exposure. He has authored more than 135 peer-reviewed publications. His awards include the Michael Fry Award and the Jack Fowler Award of the Radiation Research Society.
Dr. Shuryak spoke at PAC-AID Talks in April 2026 on CAST, a method for modeling time-varying treatment effects in head and neck cancer. You can find that talk in the past talks archive.
See the events page for upcoming PAC-AID Talks and other events.