Trialists should describe how they deal with missing data


A review of randomized controlled trials (RCTs) in four leading medical journals has revealed that the majority of studies did not use imputation to deal with missing quality of life (QoL) outcome data. The results of this study, conducted at the University of Aberdeen, were published this week in Trials, a BioMed Central journal.

Quality of life outcomes are becoming an increasingly important factor for decision-making in clinical trials of new treatments. Often these outcomes are collected via postal questionnaires and consequently subject to a substantial proportion of missing information. Imputation, whereby a reasonable alternative value is substituted for one that is missing, is one of a range of methods used by researchers to account for missing QoL data and thus try to eliminate potential bias in the results.

Fielding et al. conducted a PubMed search to identify a random selection of 285 RCTs published during 2005 and 2006 in the BMJ, The Lancet, The New England Journal of Medicine and the Journal of the American Medical Association. A QoL outcome was reported in 61 papers, which formed the basis of the review. Nineteen (31%) of these trials employed an imputation method to address missing data. Last value carried forward (LVCF) was a popular imputation strategy, although the rationale behind the choice of method was not discussed by any of the articles. A complete-case analysis, whereby the missing data is completely ignored, was undertaken in the majority of studies that did not adopt some form of imputation.


A review of RCTs in four medical journals to assess the use of imputation to overcome missing data in quality of life outcomes
Shona Fielding, Graeme Maclennan, Jonathan A Cook, Craig R Ramsay
Trials 2008, 9:51 (11 August 2008)
[Abstract] [Provisional PDF]

The authors highlight the need for avoiding missing data from the outset, clearer reporting and explanation of the methods used to overcome missing data, and discussion of the potential impact of this absent data on results. They conclude that missing data should be more reliably accounted for when analysing QoL outcomes in RCTs.

Trials encompasses all aspects of the design, performance and findings of RCTs in any discipline related to health care. This broad scope includes secondary analyses, information relating trial design and discussion of the challenges face or ‘lessons learned’ in conducting trials. The journal also considers articles about the methodology of trials in general and critical commentaries of trial results published elsewhere. Traditional trial reports are also welcome, regardless of the outcome or significance of the findings. For more information please contact the editorial office.

Abigail Jones
Assistant Editor – Trials 

View the latest posts on the On Medicine homepage

One Comment

By commenting, you’re agreeing to follow our community guidelines.

Your email address will not be published. Required fields are marked *

adamson s. muula

Missing data must be reported: but will reviewers and readers care?

The article by Shona Fielding and others [1] should be a ‘wake up’ call to all researchers, reviewers and journal editors on responsible reporting of clinical and other human research endeavors. Most of us conduct complete case analyses, and fail to disclose the proportion of missingness in any of the key variables. We may even not have taken care to learn what the mechanisms of missingness were; how else would we know when we were not looking for it. The standard statistical and epidemiology text books that categorize missingness into essentially three groups i.e. Missing at Random (MAR), Missing Completely at Random (MCAR) and Missing Not At Random (MNAR) or non-ignorable, are often understood to suggest that these categories are entirely distinct rather than viewing the groups as a continuum [2]. What we could all consider is that missing data is valuable information, that should, at the least be reported that it is missing. It is also worthwhile to remember that not all missingness is important; but at least if the author reported, the reader can then be the judge.
Finally, manuscript reviewers and readers come in all shapes and sizes; will they care if missing data are fully disclosed and the authors detail how they dealt with the problem?

1.Fielding S, Maclennan G, Cook JA, Ramsay CR. A review of RCTs in four medical journals to assess the use of imputation to overcome missing data in Trials 2008, 9:51
2. Graham GW. Missing Data Analysis: Making It Work in the Real World.
Annu Rev Psychol. 2008 Jul 24. [Epub ahead of print]