Missing Data Caused by the COVID-19 Pandemic on Clinical Trials – Overall Approach

In our final blog of the series, we examine the impact of missing data caused by the COVID-19 pandemic on clinical trials. It is generally accepted that there will be an increased amount of missing data in most clinical trials due to the COVID-19 pandemic. Missing data should be anticipated and the appropriate methods of dealing with it need to be in place. Missing data and premature termination are not new to statisticians and well-established methods exist on how to deal with these issues, but the disruptions of the COVID-19 pandemic to clinical trials are expected to be so diverse that no single solution will be appropriate for all trials.

As per regulatory guidance, all missed visits and assessments will need to be captured and summarised in the tables, figures and listings (TFLs) and CSR. This should include an assessment of the reasons for the missing data, the impact of missed visits on a case-by-case basis and for each endpoint in the trial. In general, missed visits will be classified as missing or censored (for time-to-event endpoints) in the statistical analyses.

The extent of missing data and its impact on the trial will need to be assessed as part of the risk assessment process. It is likely that the mechanism of missingness caused by the pandemic will be at random (either missing completely at random or missing at random) but it will still be necessary to further investigate and understand the source of the missing data. An imbalance of missingness between treatment arms can be of concern and it may be appropriate to utilise an IDMC to investigate the imbalance. Since trial integrity is of primary concern, any data monitoring should be on baseline and safety data rather than efficacy data in order to avoid inflating the Type I error.

Blinded reviews of the data should be carried out to explore the missing data further and to decide on the most appropriate methods of handling it in the statistical analyses. If there is a significant amount of missing data, it may be necessary to consider replacing participants and re-estimating the sample size. Methods for handling missing data should be proposed and fully documented in the SAP along with proposed plans for sensitivity and supplementary analyses. Any changes that are implemented need to be applied to all treatment arms equally and any amendments to the SAP need to be completed prior to any database locks (interim or final). Strategies for dealing with the missing data need to be considered with the estimands strategy.

Once the trial has been unblinded (and for open-label studies), summaries should be produced to determine patterns in the missing data. These summaries should be overall as well as by treatment group. It is expected that for most pandemic-related missing data, there would be no differences between the treatment groups.

With more missing data, the assumptions used can have greater leverage over the results, so it will be important to pay extra special attention to the assumptions that the method is making. It will be necessary to do a range of sensitivity analyses and supportive analyses, including methods to evaluate the missing data mechanism, the impact of alternative endpoint measurements, and the impact of intercurrent events. For example, one scenario might be the situation of imputing extremes such as in responder analyses where the sensitivity analysis planned is to impute missing or discontinued participants as non-responders. A less conservative approach may also be considered or even a range of methods. It will be important to assess the impact of the COVID-19 pandemic proactively and multiple strategies may be needed to adequately assess the impact on a trial.

Throughout this blog series we have tried to summarise just some of the information and guidance available to address the key issues raised by the COVID-19 pandemic on the conduct, analysis, and interpretation of clinical trials. Regulatory agencies are consistent with each other in the guidelines that they are publishing, but the situation is constantly changing and evolving as more becomes known about the virus and the effects of the pandemic on clinical trials become more apparent.

There is no exhaustive list of the possible effects and impacts on efficacy data in clinical trials and there are no fixed rules on how each effect should be handled. It will vary greatly from one trial to another. However, in terms of the statistical impact, our standard processes are well-equipped for us to deal with almost anything. The important thing is to carefully assess the effect of the COVID-19 pandemic on each aspect of the trial and then to decide on the appropriate methods of handling the issues, whilst maintaining the quality and integrity of both the clinical trial and the trial data, in order to ensure that the main aim of the trial can still be met.

The main strategy is to conduct an initial risk assessment, implement an appropriate mitigation strategy with ongoing further risk assessments throughout the conduct of the trial, adhering to the estimands framework to assess the impact of the pandemic on the key objectives of the trial, minimise missing data as far as possible and utilise appropriate methods to handle missing data. Discussing all proposals with regulatory agencies will be critical.

References

FDA Guidance on Conduct of Clinical Trials of Medical Products during COVID-19 Public Health Emergency, Guidance for Industry, Investigators, and Institutional Review Boards (March 2020, Updated on June 03, 2020)

FDA Guidance on Statistical Considerations for Clinical Trials During the COVID-19 Public Health Emergency Guidance for Industry (June 2020)

EMA Guidance on the Management of Clinical Trials during the COVID 19 (Coronavirus) pandemic. Version 3 (28/04/2020)

EMA Points to consider on implications of Coronavirus disease (COVID-19) on methodological aspects of ongoing clinical trials (draft)

ICH Addendum on estimands and sensitivity analysis in clinical trials to the guideline on statistical principles for clinical trials E9(R1)