Multiple imputation 40 years on, where are we now?

Share Download as iCal file
Click the icons above to Share, Tweet or add this event to your calendar (iCal)
RSS sections/special interest groups, Medical Section

Tuesday 04 December 2018, 03:00pm - 06:00pm

Location 15 Hatfields, Chadwick Court, London, SE1 8DJ


About the Event

It has been exactly 40 years since the seminal paper introducing multiple imputation to handle the problem of missing data was published. Since then, this approach has led the way in developing principled statistical methodology to address this problem. The approach has also proved to be widely appealing, with uptake popular in many disciplines. In this session we aim to bring together state of the art developments in multiple imputation research in the area of medical statistics. In particular, we will look at the use of Multiple Imputation beyond in RCTs. The session speakers comprise some of the world leaders and experts in the field.

15:00 – 15:40

Speaker: Ian White, University College London

Title: Multiple imputation: the universal panacea, and its limitations

Abstract: Ian will review some of the developments in theory and software that led to multiple imputation being seen by some people as a universal solution to missing data problems. He will then explain why it isn’t, discussing alternatives to multiple imputation, difficulties of imputing multilevel data, and avoiding the untestable missing at random assumption.


15:40 – 16:20

Speaker: Tra Pham, University College London

Title: Population-calibrated multiple imputation for a binary/categorical covariate in categorical regression models

Abstract: Multiple imputation (MI) has become popular for analyses with missing data in medical research. The standard implementation of MI is based on the assumption of data being missing at random (MAR). However, for missing data generated by missing not at random (MNAR) mechanisms, MI performed assuming MAR might not be satisfactory. For an incomplete variable in a given data set, its corresponding population marginal distribution might also be available in an external data source. We show how this information can be utilised in the imputation model to calibrate inference to the population by incorporating an appropriately calculated offset termed the "calibrated-? adjustment". We describe the derivation of this offset from the population distribution of the incomplete variable and show how, in applications, it can be used to closely (and often exactly) match the post-imputation distribution to the population level. Through analytic and simulation studies of a binary/categorical covariate in categorical regression models, we show that our proposed calibrated-? adjustment MI method can give the same inference as standard MI when data are MAR, and can produce more accurate inference under two general MNAR mechanisms. The method is used to impute missing ethnicity data in a type 2 diabetes prevalence case study using UK primary care electronic health records. Calibrated-? adjustment MI represents a pragmatic approach for utilising available population-level information in a sensitivity analysis to explore potential departures from the MAR assumption.


16:20 – 16:50: Break for refreshments


16:50 – 17:30

Speaker: James Carpenter, London School of Hygiene and Tropical Medicine

16:50 – 17:30

Speaker: James Carpenter, London School of Hygiene and Tropical Medicine
Title: Sensitivity analysis for missing trial outcomes: what can it do for you?

Abstract: James Carpenter (1,2) and Suzie Cro (3)James Carpenter (1,2) and Suzie Cro (3)

1. Department of Medical Statistics, London School of Hygiene & Tropical Medicine
2. MRC Clinical Trials Unit at UCL
3. Imperial Clinical Trials Unit, Imperial College London

Missing outcome data are almost inevitable in clinical trials, for example due to inter-current events such as treatment withdrawal, treatment switching or loss to follow-up. In such settings, the analysis can only proceed on the basis of an untestable assumption about the missing outcome data. In applications, it is therefore important to understand the robustness of conclusions to a range of plausible assumptions about the distribution of the missing outcomes.
In this talk, we outline two approaches for this: (a) reference based imputation, where missing outcomes are imputed by reference to other patient groups [1], and (b) eliciting expert opinion on the distribution of missing values and incorporating this into the analysis [2]. For each approach, we discuss the assumptions made about the missing data, implementation using multiple imputation, and give an illustrative application. We conclude with a discussion of the pros and cons of each approach, and how they may be used to address some of the challenges raised by the ICH-E9 addendum on estimands.

[1] Cro, S., Carpenter, J. R. and Kenward, M. G. (2018) Information-anchored sensitivity analysis: theory and application. Journal of the Royal Statistical Society, Series A.
[2] Mason, A. J., Gomes M., Grieve, M. Ulug, P., Powell, J. T. and Carpenter J. R. (2017). Development of a practical approach to expert elicitation for trials with missing health outcomes: application to the IMPROVE trial. Clinical Trials, 14, 357-367.

17:30 – 18:00, Panel discussion

Tea and coffee will be available from 2.30pm

Registration is required - the event is free to RSS Fellows with a £25 registration fee for non-Fellows


The meeting will be preceded by the Annual General Meeting of the Medical Section

Organiser Name Robin Mitra

Email Address This email address is being protected from spambots. You need JavaScript enabled to view it.

Organising Group(s) RSS Medical Section





Join the RSS

Join the RSS

Become part of an organisation which works to advance statistics and support statisticians

Copyright 2019 Royal Statistical Society. All Rights Reserved.
12 Errol Street, London, EC1Y 8LX. UK registered charity in England and Wales. No.306096

Twitter Facebook YouTube RSS feed RSS feed RSS newsletter

We use cookies to understand how you use our site and to improve your experience. By continuing to use our site, you accept our use of cookies and Terms of Use.