In the New England Journal of Medicine article on Missing Data - Table 1 provides 8 ideas for limiting missing data in the design of clinical trials.
Please choose the idea you think is the "best" among those proposed for limiting missing data and discuss why you believe that approach may be the best for limiting missing data.
Next, choose the idea that seems the easiest and the idea that seems the hardest to implement and discuss why one method is more challenging (harder) to implement than the other.
The Prevention and Treatment of Missing Data in Clinical Trials n engl j med 367;14 nejm.org october 4, 2012 1355 s p e c i a l r e p o r t T h e n e w e ngl a nd j o u r na l o f m e dic i n e background Missing data have seriously compromised infer- ences from clinical trials, yet the topic has re- ceived little attention in the clinical-trial commu- nity.1 Existing regulatory guidances2-4 on the design, conduct, and analysis of clinical trials have little specific advice on how to address the problem of missing data. A recent National Re- search Council (NRC) report5 on the topic seeks to address this gap, and this article summarizes some of the main findings and recommenda- tions of that report. The authors of this article served on the panel that prepared the report. Missing data have seriously compromised infer- ences from clinical trials.1 For example, editorials in the Journal have noted how missing data have limited the ability to draw definitive conclusions from weight-loss trials6 or could lead to incor- rect inferences about drug safety.7 High rates of missing data that can affect conclusions occur in trials of treatments for many diseases.8-13 Since existing regulatory guidances2-4 lack specificity, in 2008 the Food and Drug Administration (FDA) requested that the NRC convene an expert panel to prepare “a report with recommendations that would be useful for FDA’s development of guid- ance for clinical trials on appropriate study de- signs and follow-up methods to reduce missing data and on appropriate statistical methods to address missing data for analysis of results.” This article summarizes some of the main find- ings and recommendations of the report5 of that panel. More details are provided elsewhere.14 The report focused primarily on phase 3 con- firmatory clinical trials for assessing the safety and efficacy of drugs, biologic products, and some medical devices, for which the bar of sci- entific rigor is set high. The use of randomized study-group assignments predominates in such studies, since this design feature ensures com- parability of study groups and allows assessment of causation. However, many of the recommen- dations are applicable to early-phase random- ized trials and epidemiologic studies in general. Missing data are defined as values that are not available and that would be meaningful for analysis if they were observed. For example, measures of quality of life are usually not mean- ingful for patients who have died and hence would not be considered as missing data under this definition. We focus on missing outcome data here, though analysis methods have also been developed to handle missing covariates and auxiliary data. key findings Substantial instances of missing data are a seri- ous problem that undermines the scientific cred- ibility of causal conclusions from clinical trials. The assumption that analysis methods can com- pensate for such missing data are not justified, so aspects of trial design that limit the likelihood of missing data should be an important objective. In addition to specific aspects of trial design, many components of clinical-trial conduct can limit the extent of missing data. Finally, in stud- ies with missing data, analysis methods that are based on plausible scientific assumptions should be used. For example, this consideration often rules out simple fixes, such as imputation by the last observation carried forward.10 Although there are better analysis alternatives to that method, they all require unverifiable assumptions. Thus, The Prevention and Treatment of Missing Data in Clinical Trials Roderick J. Little, Ph.D., Ralph D’Agostino, Ph.D., Michael L. Cohen, Ph.D., Kay Dickersin, Ph.D., Scott S. Emerson, M.D., Ph.D., John T. Farrar, M.D., Ph.D., Constantine Frangakis, Ph.D., Joseph W. Hogan, Sc.D., Geert Molenberghs, Ph.D., Susan A. Murphy, Ph.D., James D. Neaton, Ph.D., Andrea Rotnitzky, Ph.D., Daniel Scharfstein, Sc.D., Weichung J. Shih, Ph.D., Jay P. Siegel, M.D., and Hal Stern, Ph.D. The New England Journal of Medicine Downloaded from nejm.org on November 15, 2020. For personal use only. No other uses without permission. Copyright © 2012 Massachusetts Medical Society. All rights reserved. T h e n e w e ngl a nd j o u r na l o f m e dic i n e n engl j med 367;14 nejm.org october 4, 20121356 sensitivity analyses should be conducted to assess the robustness of findings to plausible alterna- tive assumptions about the missing data. We now consider a number of specific missing- data issues that are intended to be representative and informative rather than comprehensive. Follow-up af ter Treatment Discontinuation A major source of missing data in clinical trials is participants who discontinue the assigned treatment because of adverse events, lack of toler- ability, lack of efficacy, or simple inconvenience. Too many investigators incorrectly equate treat- ment discontinuation with study dropout; that is, outcomes are not recorded for participants who discontinue treatment. However, enrollees committed to participating in the study, not just to receiving the assigned treatment. When a study treatment is discontinued, efforts should be made to obtain the participant’s consent for the collection of data on treatments and outcomes. When such efforts are successful, gathering these data after treatment discontinuation preserves the ability to analyze end points for all partici- pants who underwent randomization and thus to make possible intention-to-treat inferences, which are grounded in randomization. It also allows exploration of whether the assigned ther- apy affected the use and efficacy of subsequent therapies and provides the ability to monitor side effects that might occur or persist after the discontinuation of treatment.7 The consensus of the panel was that in many studies, the benefits of collecting outcomes after participants have discontinued treatment outweigh the costs.5 Trial Design Since there is no foolproof way to analyze data in the face of substantial amounts of missing data, we emphasize the role of design and trial conduct to limit the effect of missing data on regulatory decisions. Good clinical-trial design should clearly define the target population, along with efficacy and safety outcomes, and the likely effect of missing data should factor into deci- sions about reasonable alternative choices. The report5 states that “investigators, sponsors, and regulators should design clinical trials consistent with the goal of maximizing the number of par- ticipants who are maintained on the protocol- specified intervention until the outcome data are collected.” Design elements for clinical trials can help to prevent missing data by reducing the number of participants for whom primary end-point data will be missing. A variety of design ideas are discussed in the report, and eight of them are shown in Table 1. Their relevance varies greatly according to setting, and they may have limita- tions or drawbacks that need to be considered. An important and relatively neglected design issue is how to account for the loss of power from missing data in statistical inferences such as hypothesis tests or confidence intervals. The most common approach simply inflates the re- Table 1. Eight Ideas for Limiting Missing Data in the Design of Clinical Trials. Target a population that is not adequately served by current treatments and hence has an incentive to remain in the study. Include a run-in period in which all patients are assigned to the active treatment, after which only those who tolerated and adhered to the therapy undergo randomization. Allow a flexible treatment regimen that accommodates individual differences in efficacy and side effects in order to reduce the dropout rate because of a lack of efficacy or tolerability. Consider add-on designs, in which a study treatment is added to an existing treatment, typically with a different mechanism of action known to be effective in previous studies. Shorten the follow-up period for the primary outcome. Allow the use of rescue medications that are designated as components of a treatment regimen in the study protocol. For assessment of long-term efficacy (which is associated with an increased dropout rate), consider a randomized withdrawal design, in which only participants who have already received a study treatment without dropping out undergo randomization to continue to receive the treatment or switch to placebo. Avoid outcome measures that are likely to lead to substantial missing data. In some cases, it may be appropriate to consider the time until the use of a rescue treatment as an outcome measure or the discontinuation of a study treatment as a form of treatment failure. The New England Journal of Medicine Downloaded from nejm.org on November 15, 2020. For personal use only. No other uses without permission. Copyright © 2012 Massachusetts Medical Society. All rights reserved. n engl j med 367;14 nejm.org october 4, 2012 1357 quired sample size in the absence of missing data to achieve the same sample size under the anticipated dropout rate, estimated from similar trials. This approach is generally flawed, since inflating the sample size accounts for a reduc- tion in precision of the study from missing data but does not account for bias that results when the missing data differ in substantive ways from the observed data. In the extreme case in which the amount of bias from missing data is similar to or greater than the anticipated size of the treatment effect, detection of the true treat- ment effect is unlikely, regardless of the sample size, and the study is noninformative. When per- forming power calculations, one should consider sample-size computations for an intention-to- treat analysis that uses a hypothesized popula- tion treatment effect that is attenuated because of the inability of some study participants to adhere to the treatment. Alternatively, one could develop power analyses for statistical procedures that explicitly account for missing data and its associated uncertainty, as discussed below. Trial Pl anning and Conduc t The incidence of missing data varies greatly across clinical trials. Some of this variation is context-specific, but in many cases more careful attention to limiting missing data in trial planning and conduct can substantially reduce the problem. Eight practical