The following short courses will be held on Sunday, 3rd of September 2023 (the day before the start of the main conference). Note that main conference registration is a prerequisite to be able to book short courses.
|Virtual attendance possible?
|Advanced group-sequential and adaptive confirmatory clinical trial designs, with R practicals using rpact
|Marcel Wolbers, Kaspar Rufibach, Gernot Wassmer and Marc Vandemeulebroecke
|Bayesian methods for missing covariates in longitudinal studies
|Nicole Erler and Emmanuel Lesaffre
|Implementing the estimand framework in global drug development: Application of causal inference approaches
|Mouna Akacha, Björn Bornkamp, Alex Ocampo and Jiawei Wei
|Go fastR: High Performance Computing with R
|Michael Mayer and Lukas Widmer
|Target Trial Emulation for Causal Inference from Real-World Data
|Vanessa Didelez and Maria Geers
|Improving Precision and Power in Randomized Trials by Leveraging Baseline Variables
|Kelly Van Lancker, Michael Rosenblum and Josh Betz
|Model and Algorithm Evaluation in Supervised Machine Learning
|Max Westphal and Rieke Alpers
Keywords: interim analyses; sample size re-calculation; multi-arm multi-stage designs
This course is intended for biostatisticians from pharma and academia who are interested in learning more about advanced topics in group-sequential and adaptive clinical trial designs. Topics covered are the efficient use of interim analyses in group-sequential trials, an introduction to adaptive trials and sample size recalculation, the use of closed testing procedures for adaptive trials with multiple objectives, and multi-arm multi-stage designs. Examples from real clinical trials will be used throughout the presentations. We also aim to discuss operational aspects of implementing such designs in practice.
The course will be a mix of presentations and practicals using the R
package rpact, a free and fully validated package for the design and
analysis of group-sequential and adaptive trials. We will assume basic
familiarity with group-sequential designs and R. Participants are asked
to bring a laptop with R and rpact installed. It is the ambition of the
instructors to make this course very interactive.
Keywords: Bayesian Methods; Missing Data; Longitudinal Data; Imputation
Missing values commonly complicate the analysis of observational
data. Multiple imputation (MI) is considered the “gold standard” for
handling incomplete covariates. MI, developed at the beginning of the
Computer Age, is based on Bayesian ideas. In complex settings,
e.g. involving non-linear associations or multi-level data, the
assumptions of the commonly used MI algorithms are, however, often
violated, leading to possibly biased results. Thanks to the current
computational power, a fully Bayesian approach, allowing us to
simultaneously estimate parameters of interest and impute missing
values, is now feasible. This approach is theoretically valid and
superior to MI in complex settings. Highly complex non-standard missing
data models can relatively easily be implemented with the help of freely
available software such as the R package JointAI. In this course, we
briefly review the essentials of multi-level data, Bayesian concepts and
(multiple) imputation. The main focus is on the Bayesian approach to
missing values in covariates in multi-level and longitudinal studies,
which is motivated and illustrated using examples from clinical and
epidemiological studies. Practical sessions will be organized to show
the capabilities of the R package JointAI, starting with simpler
standard settings and extending to highly complex joint models for
longitudinal and survival data and imputation in non-standard settings.
Keywords: causal inference; conditional estimand; marginal estimand; standardization; ICH E9(R1); hypothetical estimand; principal stratum estimand
This half-day short course introduces how causal inference approaches are relevant and used in the implementation of estimands framework in drug development. It includes 4 lectures:
Lecture 1 - Introduction to Estimands and Causal Inference:
Lecture 2 - Estimation Methods of Causal Effects Targeting at Hypothetical Estimands:
Lecture 3 - Principal Stratum:
Lecture 4 - Conditional and Marginal Treatment Effects:
Keywords: statistical computing, R; HPC; clustermq; batchtools; Stan; Bayesian; clinical trial simulations
This course will help participants to optimize their R code as well as parallelizing and debugging it on their own machines as well as high-performance computing environments. Example use cases include commonly performed activities for trial design, bootstrapping, cross-validation and related workloads. The following topics will be covered:
Part I: Identifying bottlenecks in your R code, debugging, and optimizing
Part II: R parallelization on high performance computing environments (HPCE)
Part III: Case studies and code examples
Keywords: observational data, avoding self-inflicted biases, comparing treatment strategies
Target trial emulation (TTE) is a general principle to organize and structure the analysis of observational data, such as electronic health records, claims or registry data, so as to minimize common but avoidable sources of bias, e.g. immortal-time bias. Moreover, formulating a target trial is helpful to elicit practically meaningful causal research questions (aka “estimands”) with a clear interpretation. The workshop will explain the principle of TTE using examples from cancer screening, drug safety as well as nutritional epidemiology. For instance, we will illustrate how to emulate a target trial on screening colonoscopy, how this avoids design-related and other biases, while showing how results are badly affected if a naive study design is chosen that suffers from these biases. A brief overview of some relevant statistical methods will be given, such as the clone-censor-weight approach or the parametric g-formula. However, as will become clear, TTE is a fundamental principle that can be combined with various causal inference methods.
Organization of the workshop:
There will be theoretical parts as well as worked examples, with hands-on tasks for the participants.
Participants will (i) be able to recognise avoidable sources of bias in naïve studies using observational data; (ii) become aware of basic techniques to avoid these issues; (iii) acquire a basic understanding of TTE that will facilitate studying the more advanced literature.
Keywords: covariate adjustment; causal inference; standardization; treatment policy; robustness; group sequential designs
In May 2021, the U.S. Food and Drug Administration (FDA) released a revised draft guidance for industry on “Adjustment for Covariates in Randomized Clinical Trials for Drugs and Biological Products”. Covariate adjustment is a statistical analysis method for improving precision in clinical trials by adjusting for pre-specified, prognostic baseline variables (e.g., age, BMI, comorbidities). The resulting sample size reductions can lead to substantial cost savings, and also more ethical trials since they avoid exposing more participants than necessary to experimental treatments. Though covariate adjustment is recommended by the FDA and the European Medicines Agency, many trials do not fully exploit the available information in baseline variables.
In Part 1, we explain what covariate adjustment is, how it works, when it may be useful, and how to implement it (in a preplanned way that is robust to model misspecification) for a variety of scenarios.
In Part 2, we present a new method that enables us to easily combine covariate adjustment with group sequential designs. This approach can lead to faster trials, without sacrificing validity or power, even when the experimental treatment is ineffective.
In Part 3, we show the impact of covariate adjustment using completed
trial datasets in multiple disease areas. We provide step-by-step, clear
documentation of how to apply the software in each setting. Participants
will have the time to apply software tools on the different datasets.
Keywords: prediction; performance; validation; comparison; benchmark
The statistical evaluation of developed models and algorithms is an essential part of applied machine learning and predictive modelling. This half-day course is suitable as a concise introduction or refresher for this important topic. It is divided into three parts with sufficient time for participant questions and breaks in between.
Initially, we will repeat essential machine learning basics and cover core concepts of model evaluation. We will mainly consider classification tasks and the most relevant assessment criteria (discrimination, calibration) but also summarize adaptations for regression and survival problems. In the main part, we discuss common pitfalls (leakage, multiplicity, ….) in model evaluation and appropriate best practices to avoid and/or rectify them. Finally, we touch upon some advanced topics and cover important practical aspects (software, reporting, reproducibility) that are required for a successful evaluation study.
The course contents are illustrated by means of real-world data examples, including R code to showcase how the numerical results were obtained. There are no explicit coding sessions in this short course, so a laptop is not necessarily required. The course materials will be made available so that participants have the opportunity to individually reproduce the numerical examples after the course.