Biomarker & Subgroup Analysis & Identification Subteam of QSPI Multiplicity Working Group Goals and tasks Ilya Lipkovich, Quintiles Copyright © 2012 Quintiles General objectives • Formed under multiplicity team of QSPI • http://biopharmnet.com/wiki/QSPI_Subgroup_Analysis_Working_Group > The subteam is planning to create a unifying and transparent framework for the current state of subgroup analysis/identification methods and their application to clinical and observational studies: > Describing/developing common principles for confirmatory subgroup analysis versus subgroup search/identification in pharmaceutical applications. > Creating a systematic review of key published methodologies and available algorithms for subgroup analysis/identification methods for data from randomized clinical trials. > Evaluating operating characteristics and relative advantages/disadvantages of existing methods via simulation studies. > Developing better validation/confirmation strategies for subgroup identification. > Connecting subgroup identification methodologies used in randomized clinical trials with those utilized in observational studies. > Developing educational tools to facilitate application of subgroup analysis/subgroup identification methods 2 Group members & affiliations Current members: Arunava Chakravartty (Novartis), Josh Chen (Merck), Ying Ding (University of Pittsburgh), Alex Dmitrienko (Quintiles), Birol Emir (Pfizer), Adarsh Goshi (Gilead), Zhou Kefei (Amgen), Yumi Kubo (Amgen), Ilya Lipkovich (Quintiles), Lei Liu (Northwestern University),Yufan Liu (Rutgers University), Cristiana Mayer (Johnson & Johnson), Sandeep Menon (Pfizer), Padraic Neville (SAS Institute), John O’Gorman (Biogen Idec), Gautier Paux (Servier), Tom Parke (Tessella), Jane Qian (Abbvie), Lei Shen (Eli Lilly), Li Wang (Abbvie), Lin Wang (Sanofi), Lei Xu (Biogen Idec), Stan Young (National Institute of Statistical Sciences),Donghui Zhang (Sanofi). 3 Goals Overview # Goal Task Responsible group members/Lead Deliverable Timeline G1 Common/existing practices for Subgroup analysis (SA) in pharmaceutical industry (1) Quantitative review of SA/SI features from SAP & protocols (2) Qualitative review of main tasks involving SA/SI Cristiana (J &J), Ilya & Alex (Quintiles) , Qian (Abbvie), Yumi (Amgen) Q4 G2 Creating systematic review of the state of art Key published methodologies and available algorithms. Arunava (Novartis), Josh (Merck), Sandeep (Pfizer), Ilya & Alex (Quintiles), Lin Wang (Abbvie), Ying (Pitt) Industry paper Review paper on existing subgroup analysis practices/method ologies and presentation of survey results G3 Evaluating operating characteristics of existing methods via simulation and case studies. (1) List of available software implementations (2) Common simulation scenarios (3) Common criteria for comparison Lei Shen (LEAD, Eli Lilly), Birol (Pfizer),Ilya (Quintiles), Lin Wang (Sanofi), Adarsh Joshi (Gilead), Tom Parke (Tessella Ltd), Donghui Zhang (Sanofi); Padraic Neville (SAS), Zhou Kefei (Amgen), Yufan (Rutgers) Improved tools for data generation/perfor mance evaluation, Library of test data sets for developers Paper Q3 G4 Developing better validation/confirmation strategies for SA/SA Based on G1-G3 Recommendation s/ guidelines G5 Developing educational tools to facilitate application of SA/SI Based on G1-G3 Slide deck G7 SA/SI in observational studies Literature review and simulation study comparing 2-3 methods in several scenarios of confounding Q4 2014 Ilya/Stan 4 Common/existing practices for Subgroup analysis in pharmaceutical industry (G1) • Idea – collect information on both general SA practices and specific SA instances using > Data Collection Tool finalized: Google forms > https://docs.google.com/forms/d/1hFcxljo9nF7rY8S__aJJUs6s87ArBjKEXjJF1q1Sd8/viewform > Suggested 4 types of SA – did not mean to impose, respondents may have their own views, however tried to keep it reasonably structured > 8 basic questions, some of them conditioned depending on what type of SA the respondent will be answering • Next steps > run data collection tool among group members > Additional questions/instructions > Can we add glossary? • Suggestions on should we and how to disseminate it among broader audience (ASA Biopharmaceutical section?) 5 Creating systematic review of the state of art (G2) • Divided subgroup analyses (SA) into 4 domains Types of SA Strategy (preplanned vs. .post-hoc) Multiplicity control (Y/N) Data (biomarker) scope Pre-specified subgroups (Y/N) Confirmatory SA (SA1) Preplanned Yes Narrow Yes Exploratory SA (SA2) Preplanned Sometimes Narrow Yes Post-hoc SA (SA3) Post-hoc No Narrow Yes Biomarker & subgroup discovery (SA4) Preplanned Sometimes Broad No (data-driven) • Plan is to draft reviews 5-6 pages per each domain with review of different approaches, challenges, etc and case studies by end of November > > > > Confirmatory SA (SA1) - Sandeep (Pfizer), Arunava (Novartis) , Ying (University of Pittsburg) Exploratory SA (SA2) - Li Wang (Abbvie) Post-hoc SA (SA3) - Josh Chen (Merck) Biomarker & subgroup discovery (SA4) - Ilya and Alex (Quintiles) (see review as part of book chapter http://multxpert.com/wiki/Lipkovich_Dmitrienko_2014) 6 Evaluating operating characteristics of existing methods via simulation and case studies (G3) • A three stage process: Data Simulation & Analysis & Performance Evaluation: > Simulator generates data set with known subgroups and stores data and “key”, > Analyst analyzes the data and provides “solution” that scores subjects and biomarkers > Evaluator runs the solution against the truth. • Lei Shen’s group at Eli Lilly developed a flexible R -based data generation & performance evaluation tools that are now available at > http://biopharmnet.com/wiki/Data_Simulation_and_Evaluation_Subgroup_Identificati on_Results • Several subgroup identification methods are available now > http://biopharmnet.com/wiki/Software_for_Subgroup_Identification_and_Analysis 7 Building a library of tools for subgroup analysis • See “Software for Subgroup Identification and Analysis http://biopharmnet.com/wiki/Software_for_Subgroup_Identification_and_Anal ysis > Ilya and Alex (Quintiles) published a beta version of SIDESxl excel macro (along with illustrative example and the user guide) implementing SIDES and SIDEScreen procedures. We will update the code regularly, “testers” are welcome > R functions for SIDES procedures provided by Jue Hou > R functions for Interaction Trees provided by the author (Xiaogang Su) > Virtual Twins method (provided by one of the authors Jared Foster) > The GUIDE package for classification and regression trees by prof. Wei-Yin Loh now includes features for subgroup identification. • Please let us know what other tools/packages can be used and provide links to the software or your own code 8