PSI One-Day Meeting: Missing Data in Clinical Trials - Past, Present and Future

Add to:

Logo with sponsored by

Date: Tuesday 4th & Wednesday 5th May 2021
Time: [Tuesday 4th 10:30-13:30] and [Wednesday 5th 09:00-12:00]
Speakers: Khadija Rantell (MHRA), Jiawei Wei (Novartis), David Wright (AstraZeneca), Bohdana Ratitch (Bayer), Sylvia Li (IQVIA), Michael O’Kelly (IQVIA), Daniel Bratton (GSK) and James Bell (Elderbrook solutions GmbH).

Who is this event intended for? Statisticians involved in designing and analysing clinical trials.
What is the benefit of attending? Gain awareness of the increasing importance of appropriate missing data handling in clinical trials by learning about state of the art methodology and recent case studies.


Incomplete datasets due to missing data is an issue that has been, and will be, around for a long time. At this meeting we will present the evolution of missing data approaches, looking at how they have been handled in the past, the current established missing data approaches and the impact of the new ICH E9 R1 addendum on the handling of missing data, focussing in particular on the treatment policy estimand.


You can now register for this event. Registration fees are as follows:
- Members of PSI = £20+VAT
- Non-Members of PSI = £115+VAT*
*Please note: Non-Member rate includes membership for the rest of the 2021 calendar year.
To register for the session, please click here.

Speaker Details

DAY 1 – Tuesday 4th May 2021




Khadija Rantell

Khadija Rantell (nee Rerhou) is a Senior Statistical Assessor at the Medicines and Healthcare Products Regulatory Agency (MHRA) where she has worked since 2013.

Khadija is also a member of the EFSPI (European Federation for Statisticians in the Pharmaceutical Industry) Scientific Committee and a member of the EFPIA/EFSPI Estimands Implementation and Pharmaceutical Industry Biostatistics Working Groups.

History of missing data in regulatory settings.

The problem of handling missing data has evolved greatly over the last decade along with statistical methodologies for dealing with missing data. Several recent documents have been developed that lay out a set of general principles and techniques for addressing the problems raised by missing data in clinical trials. However, prospective prevention of missing data occurrence, through carefully designing and implementing of a research study, remains the single best approach. This session will focus on the evolution of regulatory framework for handling missing data in confirmatory trials and the impact of the ICH E9 (R1) addendum on the handling of missing data.

Jiawei Wei

Dr. Jiawei Wei joined Novartis in 2011, where she is currently a Director Biostatistician in the Advanced Methodology and Data Science group. She is interested in supporting the methodological development in various areas of pharmaceutical statistics, including estimand, recurrent event data, multiple testing, etc. Before joining Novartis, Jiawei got her PhD in statistics from Texas A&M University in 2010, and then one year assistant professor. Jiawei is awarded leading scientist at Novartis, she is associate editor of Statistics in Biopharmaceutical Research, and also a part time advisor at Fudan University in China.

On the role of hypothetical estimand in clinical trials and its estimation.

The ICH E9(R1) Addendum on 'Estimands and Sensitivity Analysis in Clinical Trials' introduced various strategies for addressing intercurrent events when defining the clinical question of interest. For hypothetical estimand strategies, a scenario is envisaged in which the intercurrent event would not occur. The value of the variable that reflects the clinical question of interest is the value which the variable would have taken in the hypothetical scenario. If a hypothetical strategy is proposed, it should be made clear what hypothetical scenario is envisaged. The ICH E9(R1) Addendum acknowledges that a wide variety of hypothetical scenarios can be envisaged, but it also clarifies that some scenarios are likely to be of more clinical or regulatory interest than others.

In this talk we will not only discuss the role of hypothetical estimand in clinical trials, but also the estimation of hypothetical estimand. One basic consideration is that the estimation of hypothetical estimand requires the prediction of hypothetical trajectories for all patient who have the intercurrent event. Importantly, the assumptions for the predictions need to be aligned with the hypothetical strategy, which begs the questions: From where or from whom do we borrow information to predict the hypothetical measurements of interest? Do we have sufficient data or information to borrow from? Moreover, the prediction uncertainty needs to be adequately accounted for, e.g. through multiple predictions. And finally, sensitivity analyses need to be performed to investigate the robustness of our conclusions. All the above considerations will be discussed and illustrated by a case study.

Bohdana Ratitch

Bohdana is a Principal Statistician at Bayer. She has over 15 years of experience in biopharmaceutical industry including statistical methodology research and consulting, with numerous publications in missing data, estimands, subgroup identification, and machine learning. She is a co-author of two books, “Estimands, Estimators and Sensitivity Analysis in Clinical Trials” (2020) and “Clinical Trials with Missing Data: a Guide for Practitioners” (2014). Bohdana is a member of the DIA SWG on Missing data and EFSPI Subgroup Special Interest Group.

Statistical Issues and Recommendations for Clinical Trials Conducted During the COVID-19 Pandemic.

The COVID-19 pandemic continues to impact planned and ongoing clinical trials. Any study that is conducted in part or in whole during the pandemic must anticipate, assess, and mitigate these impacts to assure safety of participants and address operational issues. From a statistical perspective, the study teams need to ensure the ability to conduct the trial and collect data in alignment with the study objectives. Statistical analysis plans should incorporate strategies to deal with potential increases or distinct patterns of treatment discontinuations, interruptions, protocol deviations, and missing data. In this presentation, we will discuss potential impacts of COVID-19 pandemic disruptions on clinical trials with a focus on statistical aspects and mitigation strategies. It is beneficial to do so in a structured and systematic way through the estimand framework, regardless of whether the estimand is formally defined in the trial protocol. We will also discuss issues of missing data handling and supplemental analyses that may be needed for the trial.

Day 2 – Wednesday 5th May 2021

David Wright (AstraZeneca)

David became the Head of Statistical Innovation at AstraZeneca in September 2016. David leads a team of expert statistical methodologists who advise colleagues within AstraZeneca on novel trial design and analysis issues. Between 1999 and 2016 David worked for the Medicines and Healthcare products Regulatory Agency (MHRA) as a Statistical Assessor and was Chair of the Biostatistics Working Party from 2011-2016. David was heavily involved in the revision to the CHMP missing data guideline. He is currently involved in how the ICH E9 Addendum on Estimands impacts the design, analysis and reporting of clinical trials within AstraZeneca and is also a member of the EFPIA/EFSPI Estimand Implementation Working group.

Aligning how subjects with missing data due to study discontinuation are handled in the primary analysis with the primary estimands.

ICH E9 R1 explains the distinction between treatment discontinuation and study withdrawal. The former is an intercurrent event whilst the latter gives rise to missing data to be addressed in the statistical analysis. However, E9 R1 also states that “methods to address the problem presented by missing data can be selected to align with the estimand”. In this presentation I will stress that in fact the chosen methods should be selected to align with the estimand and failure to do this would lead to an incoherent analysis strategy. The treatment policy strategy will be considered and it will be highlighted that there are a number of methods of estimation available to address missing data. Which methods are fully aligned with the treatment policy estimand will be discussed.

James Bell
(Elderbrook solutions GmbH)

James originally trained as a chemist, receiving his PhD from the University of Cambridge in 2009. Following a short spell in industry in computational drug design, he completed his MSc in statistics at UCL in 2013. Starting at Boehringer Ingelheim, he spent three years as trial statistician before joining its newly-formed statistical methodology group. James now works as a consultant methodology statistician for the CRO Elderbrook Solutions GmbH, providing services to a major pharmaceutical company. James is an active member of cross-industry WGs in his two main areas of research; estimands and missing data handling. Other topics of particular interest include event prediction and the process of how clinical trials are designed.

The Practicalities of Treatment Policy Estimation.

Treatment policy estimands were introduced by ICH E9(R1) as an approach to include the effects of intercurrent events within the treatment effect of interest. They have been commonly thought of as both equivalent to ITT analysis and robust to estimate. However, it has become increasingly clear that these properties start to break down in the presence of missing data – almost an inevitability in real clinical trials. Indeed, due to missing data typically being associated with discontinuation of randomised treatment, treatment policy estimation will usually deviate from traditional ITT approaches and in doing so, situations can easily arise where no reliable estimation is possible.

This talk will describe the practical difficulties of treatment policy estimation that arise due to missing data and discuss potential solutions. Topics covered will include trial conduct, approaches to estimation, assumptions, how much data is needed and sample size calculations.

Michael O’Kelly & Sylvia Li

Michael O’Kelly has worked as a statistician in the pharmaceutical industry for 27 years. He has been involved in many areas of biostatistics, and has a special interest in statistical modelling and simulation. He co-authored a Best Practice proposal for projects involving Modelling and Simulation, which was adopted by the PSI board in 2017. With colleagues Michael O’Kelly has developed new methods for missing data that are now widely used in clinical trials. His book authored with Bohdana Ratitch, “Clinical trials with missing data: a guide for practitioners”, was published in 2014 by Wiley. He has given courses on missing data for PSI and at many conferences and for many pharmaceutical companies. For his work in best practice and in missing data, Michael received the RSS/PSI award for Excellence in Pharmaceutical Statistics in 2017; he is Senior Director with IQVIA’s Centre for Statistics in Drug Development.

Even a “treatment policy” estimand may have missing data: how can we take account of this?

Under the “treatment policy” estimand, outcomes contribute to the estimate of effect irrespective of treatment actually taken after randomisation, and all subjects are to be followed up for the full scheduled follow-up period. It is hoped that under this estimand missing data would be limited. However, in practice, subjects may avail of their right to discontinue completely from a trial. Even under a “treatment policy” estimand, this leaves the analyst with missing data. How is the analyst to take account of this residual missing data under the “treatment policy” estimand? The new ICH E9 R1 Addendum suggests “for subjects who discontinue treatment without further data being collected, a model may use data from other subjects who discontinued treatment but for whom data collection has continued”. However, numbers of appropriate subjects with data available may be small, and compromises or assumptions may be required to do the kind of modelling suggested by ICH E9 R1. This presentation evaluates some of the options that use statistical modelling to take account of this residual missing data.

Daniel Bratton

Dan worked as a statistician at GSK since 2016. During this time he has worked on RCTs investigating the effect of various treatments in severe asthma, COPD and nasal polyps, and is now working in trials assessing therapies for COVID-19. Dan started his career as a medical statistician at the Medical Research Council Clinical Trials Unit (MRC CTU) working on RCTs in respiratory diseases and dermatology. He then completed a PhD at the same unit through UCL investigating statistical issues in the design of multi-arm multi-stage clinical trials which are increasingly being used in practice to accelerate the evaluation of new therapies. Prior to starting at GSK, Dan worked as a statistician in the Department of Pulmonology at University Hospital Zurich for one year, focusing mainly on network meta-analyses of RCTs investigating treatments for sleep apnoea.

Treatment policy estimands for recurrent event data using data collected after cessation of randomised treatment.

Trials now increasingly collect data from subjects following premature discontinuation of study treatment, as this event is irrelevant for the purposes of a treatment policy estimand. However, despite efforts to keep subjects in a trial, some will still choose to withdraw. Publications for sensitivity analyses of recurrent event data have often focused on reference‐based imputation methods, more commonly applied to continuous outcomes, where imputation for the missing data for one treatment arm is based on the observed outcomes in another arm. The existence of data following premature discontinuation of treatment now raises the opportunity to impute missing data for subjects who withdraw from study using this observed ‘off-treatment’ data, potentially allowing more plausible assumptions for the missing post‐study‐withdrawal data than other reference‐based approaches. In this poster, we describe a recent imputation method (1) for recurrent event data in which the missing post‐study‐withdrawal event rate for a subject is assumed to reflect the rate observed from subjects during the off‐treatment period. The method is illustrated in a trial in chronic obstructive pulmonary disease (COPD; GSK funded NCT02105961) where the primary endpoint was the rate of exacerbations, analysed using a negative binomial model.

Upcoming Events

Latest Jobs