Clinical trials with missing data

Features

  • Author: Michael O'Kelly and Statistics Views
  • Date: 25 Mar 2014
  • Copyright: Photograph appears courtey of Quintiles.

This month, Wiley are proud to publish Clinical Trials with Missing Data: A practitioner’s guide, by Michael O’Kelly and Bohdana Ratitch.

This book provides practical guidance for statisticians, clinicians, and researchers involved in clinical trials in the biopharmaceutical industry, medical and public health organisations. Academics and students needing an introduction to handling missing data will also find this book invaluable.

The authors describe how missing data can affect the outcome and credibility of a clinical trial, show by examples how a clinical team can work to prevent missing data, and present the reader with approaches to address missing data effectively.

The book is illustrated throughout with realistic case studies and worked examples, and presents clear and concise guidelines to enable good planning for missing data. The authors show how to handle missing data in a way that is transparent and easy to understand for clinicians, regulators and patients. New developments are presented to improve the choice and implementation of primary and sensitivity analyses for missing data. Many SAS code examples are included – the reader is given a toolbox for implementing analyses under a variety of assumptions.

Here Michael O'Kelly describes what we can expect from this exciting new work.

thumbnail image: Clinical trials with missing data

Data are missing from almost all clinical trials, because some subjects withdraw from the trial before the crucial final measurements. Without those final measurements, our picture of the new treatment is imperfect, and it could actually be wrong. What if most of the withdrawals from the new treatment arm dropped out because they felt they were getting worse, or because of strong side effects? Believe it or not, it took the pharmaceutical industry a long time to realise that missing data was not something that statisticians could just “take care of”. In the last ten years, people in the pharmaceutical industry – and regulators like the U.S. Food and Drug Administration (FDA) and the European Medicines Agency (EMA) – began to see that missing data is missing, and we can’t just go with some assumption about what the missing data might have been and leave it at that. The result from the planned main analysis of a clinical trial, no matter what its assumptions about withdrawals, could not be taken as the one and only answer. People saw that one should test the results of the trial for robustness to missing data by presenting, alongside the main result, the results from a number of “what-if” analyses that made different assumptions about withdrawals – preferably assumptions somewhat “hostile” to the new treatment.

There was clearly an opening for a book that working statisticians could use as a practical guide to the variety of ways one can handle missing data and, as important, the statistician would need also to know how to test the results of the trial for robustness to missing data. The key question any clinician, regulator or patient would like answered, in this respect, is “If we are sceptical about what happened to withdrawals, is the new treatment still on the whole better than the control treatment?”

Vice President and Managing Director of Quintiles’ Dublin office John Kiernan (left) with Michael O’Kelly at the launch of “Clinical trials with missing data: a guide for practitioners”. Michael presented a copy of the book to Michelle McCrory, Statistical Scientist, a specialist in missing data at the office.

So when Wiley asked me for ideas for a book some years ago, missing data was a natural choice of subject. There were many good books that covered the theory and methods for missing data, but there seemed to be an opening for a book that could directly help the statistician working on a clinical trial. I thought it would be nice to produce a book that I myself would have liked to have had, when I was statistical team lead for a clinical trial that was likely to have missing data. Such a book could also be useful, I felt, as practical introduction to the subject for the non-statistician, and in addition could be useful for undergraduates who wanted a grounding in approaches to missing data in the area of medical science.

Co-author Bohdana Ratitch and I began to think about approaches that seemed to work well and hold up scientifically. We realised that each clinical trial should plan a holistic approach to the issue of missing data. Clinical, project management, data monitoring and even clinical trial recruitment specialists could all work with the statisticians to improve the credibility of clinical trials with respect to missing data. A scientific approach to missing data did not just involve dealing with it after the fact via clever statistics. The first priority for a study team should be to prevent missing data in the first place. I was lucky enough to have heard Sara Hughes, head of Clinical Statistics at GlaxoSmithKline, talk about her work in preventing missing data. She and her team have pioneered practical ways that statisticians could work with others to improve the retention of subjects in a clinical trial and in this way to minimise the problem of missing data. Sara has written a great chapter on this for the book. Those and similar ideas are now being used more and more in clinical trials. For example, one simple improvement that has worked well in my own company is to allow for a flexible consent form. With this a subject is, of course, free to withdraw from the trial at any time, but can opt to return for a final assessment at the time that was originally planned; or can choose to allow follow-up by telephone. We have found that this strategy has reduced missing data dramatically.

But, inevitably, a trial will end up having some missing data. Here again, we have found that the statistician will come up with better solutions by working with other experts, especially with clinicians. In the book, we tie all sections to illustrative clinical data. We show how, in a real study of a disease, it will be important to look to earlier, similar studies so as to create a strategy for missing data that will make sense to clinicians. If a strategy for missing data makes sense to clinicians, it will make sense to patients, and to regulators. A bonus is that, in a discussion at a regulatory review, a strategy based on historic data is straightforward to justify.

One reason we feel free to advocate collaboration and openness to clinical ideas, is that the statistician now has at his or her fingertips a wide variety of approaches that are likely to fit with what clinicians need. Co-author Ratitch contributed a beautifully simple way of implementing pattern-mixture models to fit a variety of clinically plausible assumptions for missing data. Ratitch’s method is available as downloadable packages at the web page of the Industry Scientific Working Group (SWG) on Missing Data (www.missingdata.org.uk). Industry will usually need independently-programmed checks of any macro used, so the book provides template SAS (c) code to allow the reader to put together his or her own version of the missing data strategy, and thus facilitate formal, auditable verification of the macro.

...inevitably, a trial will end up having some missing data. Here again, we have found that the statistician will come up with better solutions by working with other experts, especially with clinicians. In the book, we tie all sections to illustrative clinical data. We show how, in a real study of a disease, it will be important to look to earlier, similar studies so as to create a strategy for missing data that will make sense to clinicians. If a strategy for missing data makes sense to clinicians, it will make sense to patients, and to regulators. A bonus is that, in a discussion at a regulatory review, a strategy based on historic data is straightforward to justify.

Other industry and academic researchers have also made software available, providing an extensive toolbox for missing data. The chapter of the book devoted to sensitivity analyses delineates the choices of method that are available, explains each carefully, and shows how they can be applied to real-life indications. To do this, the book uses illustrative data from a number of clinical studies, including studies of Parkinson’s disease and of insomnia. The book also describes assumptions that can be made for missing binary (yes/no) outcomes, and provides suggestions and code for implementing these for a study in mania – this turns out to be a particularly tricky indication, since the symptoms of withdrawals are likely to fluctuate due to the nature of the illness.

The book provides clear explanations of all concepts related to missing data in simple language, with illustrative data and, where appropriate, example SAS code. Two chapters provide comprehensive guides to two of the “mainstream” statistical methodologies for missing data, direct likelihood and multiple imputation. Sonia Davis, erstwhile statistical lead of many clinical trials and now Professor of the Practice at the University of North Carolina, provided the chapter on direct likelihood approaches (often referred to as mixed models for repeated measures, or MMRM). Co-author Bohdana Ratitch wrote the chapter on multiple imputation. The book also includes a chapter by Belinda Hernández and Ilya Lipkovich on an up-and-coming approach called doubly robust estimation – again, a macro and template code are provided so that the reader can try out the method for herself.

What next? The SWG of which I am a member is the Drug Information Association Scientific Working Group on Missing Data. I co-ordinate its “New Tools” subgroup, and we are planning to make available some new software for clinical trials over the next year. Software is needed for missing binary and time-to-event outcomes. The SWG hopes to make software available for time-to-event outcomes soon – Quintiles colleague Ilya Lipkovich and I are working on that with Bohdana Ratitch at the moment, and there will be more. The SWG also plans to publish papers on this area soon.

My friends in Dublin who know I’m researching missing data will probably continue to ask me “Have you found it yet?” and I am pretty sure that for the next few years I will be saying “No, still looking”.

Related Topics

Related Publications

Related Content

Site Footer

Address:

This website is provided by John Wiley & Sons Limited, The Atrium, Southern Gate, Chichester, West Sussex PO19 8SQ (Company No: 00641132, VAT No: 376766987)

Published features on StatisticsViews.com are checked for statistical accuracy by a panel from the European Network for Business and Industrial Statistics (ENBIS)   to whom Wiley and StatisticsViews.com express their gratitude. This panel are: Ron Kenett, David Steinberg, Shirley Coleman, Irena Ograjenšek, Fabrizio Ruggeri, Rainer Göb, Philippe Castagliola, Xavier Tort-Martorell, Bart De Ketelaere, Antonio Pievatolo, Martina Vandebroek, Lance Mitchell, Gilbert Saporta, Helmut Waldl and Stelios Psarakis.