Keynote 1: TBD

May 16, 2025

Speaker: Professor Shannon McWeeney, PhD

OHSU

TBD

Banquet: Statistical Bait and Switch: Inferring What is Likely to Be True

May 16, 2025

Speaker: Dr. Stephen Ruberg, PhD

Analytix Thinking, LLC and Purdue Department of Statistics.

Dr. Stephen Ruberg received a bachelor’s degree in mathematics from Thomas More College, an MS in Statistics from Miami of Ohio, and a PhD in Biostatistics from the University of Cincinnati. Dr. Ruberg was in the pharma industry for 38 years where he worked in all phases of drug development and commercialization – from R&D to Business Analytics. Throughout his career, Steve had senior leadership roles in the companies for which he worked, including VP of Statistics and Data Management at several companies. In his last 10 years at Lilly, he formed the Advanced Analytics Hub for which he was the Scientific Leader and ultimately named a Distinguished Research Fellow in Lilly R&D. He retired from Lilly at the end of 2017. Dr. Ruberg also served in many leadership roles related to the pharmaceutical industry and the statistical profession. He was on the Expert Working Group for ICH-E9 Statistical Principles for Clinical Trials and a co-author of that Guidance. Most notably, Steve served on a select Advisory Committee to the Secretary of Health and Human Services during the Bush administration for advancing the use of electronic medical records. Since 2018, Dr. Ruberg has founded his own consulting firm, Analytix Thinking, LLC, which focuses on consulting and teaching pharma companies big and small, as well as lecturing and publishing on important statistical topics. Dr. Ruberg’s current research interests include estimands, subgroup identification, Bayesian methods for clinical drug development, and digital medicine. He has been a Fellow of the American Statistical Association (ASA) since 1994, was given the Career Achievement Award by Quantitative Scientists in the Pharmaceutical Industry and was elected a Fellow of International Statistics Institute.

Abstract

In 1925, Sir Ronald Fisher published his now VERY famous book Statistical Methods for Research Workers in which he posited, “The [test statistic] value for which P =.05, or 1 in 20, is 1.96 or nearly 2; it is convenient to take this point as a limit in judging whether a deviation is to be considered significant or not.” With the later work by Jerzy Neyman and Egon Pearson in 1933 on what is now known as the Neyman-Pearson lemma, the foundations for statistical hypothesis testing (frequentist statistics) were laid and have predominated to the present day. In 1961, in the 7th edition of his renowned book Principles of Medical Statistics, Sir Austin Bradford Hill considered the difficulty of analyzing clinical trials in which patients do not follow the study protocol or adhere to their randomized study medication. He states, “we may inevitably have to keep such patients in the comparison [of treatments] and thus measure the intention to treat in a given way rather than the [effect of the] actual treatment.” The concept of intention to treat gained firm footing in the statistical community, and in the pharmaceutical drug development world was established as the default approach in ICH E9 Statistical Principles for Clinical Trials (1998). As so it is. Design a randomized controlled trial that results in a p-value<0.05 from an intent-to-treat analysis and your study is a success and the treatment under study works! Anything different is a failure. But is this a suitable way to understand the truth? How should we answer the fundamental scientific question, “Does this treatment cause that outcome?” Part 1 of this presentation will present concepts, arguments, and examples (some of which are novel) for why a Bayesian probability provides a more direct and informative answer to that scientific question than a p-value. Part 2 will discuss the topic of estimands and why the controlled direct effect of treatment is often preferrable to the effect estimated by the intention-to-treat approach.

Keynote 2: LLM-Lasso: A Robust Framework for Domain-Informed Feature Selection and Regulation with an Application to Lymphoma Diagnosis

May 17, 2025

Speaker: Professor Robert Tibshirani, PhD

Stanford University

Robert Tibshirani is a Professor of Biomedical Data Science, and of Statistics, at Stanford University. He has made important contributions to the statistical analysis of complex datasets. Some of his most well-known contributions are the Lasso, which uses L1 penalization in regression and related problems, generalized additive models and Significance Analysis of Microarrays (SAM). He also co-authored five widely used books ‘Generalized Additive Models’, ‘An Introduction to the Bootstrap’, ‘The Elements of Statistical Learning’, "An Introduction to Statistical learning", and ‘Sparsity in Statistics: the Lasso and its generalizations’. He is an active collaborator with many scientists at Stanford Medical school.Tibshirani received the COPSS Presidents' Award in 1996. Given jointly by the world's leading statistical societies, the award recognizes outstanding contributions to statistics by a statistician under the age of 40. He was elected a Fellow of the Royal Society of Canada in 2001, the National Academy of Sciences in 2012, and the Royal Society of Britain in 2019. In 2021 he received the ISI Founders of Statistics Prize for his 1996 paper Regression Shrinkage and Selection via the Lasso. In 2024 he received the COPSS Distinguished Achievement Award and WNAR/IBS Outstanding Impact Award.

Abstract

We introduce LLM-Lasso, a novel framework that leverages large language models (LLMs) to guide feature selection in Lasso ℓ1 regression. Unlike traditional methods that rely solely on numerical data, LLM-Lasso incorporate domain-specific knowledge extracted from natural language, enhanced through a retrieval-augmented generation (RAG) pipeline, to seamlessly integrate data-driven modeling with contextual insights., LLM-Lasso outperforms standard Lasso and existing feature selection baselines, all while ensuring the LLM operates without prior access to the datasets. To our knowledge, this is the first approach to effectively integrate conventional feature selection techniques directly with LLM-based domain-specific reasoning.

Joint work with:  Erica Zhang, Ryunosuke Goto, Naomi Sagan, Jurik Mutter, Nick Phillips, Ash Alizadeh, Kangwook Lee, Jose Blanchet, and Mert Pilanci