The current COVID-19 pandemic has the potential to impact the conduct of clinical trials of medical products. Trial participants may not be able to come to investigational sites for protocol-specified visits, which will likely cause rates of missing and incomplete data that are higher than anticipated when the study was designed. This whitepaper outlines statistical methods and strategies to mitigate the effects of missing and incomplete data on the interpretability of clinical trials data.
Minimize the impact of missing data
- Alternative methods of data collection, such as questionnaires administered via telephone or using mobile or wearable devices, that do not require in-person visits
- Outcome measures such as AUEC (area under the effect curve) that use all data collected, can accommodate missing observations and data collection times that vary from subject to subject
- Statistical models that include participants with incomplete observations and do not require imputation of missing data, such as mixed models for repeated measures (MMRM)
- Methods that can minimize bias due to sparse data, such as penalized estimation
Assess the impact of missing data
- Sensitivity analysis can be conducted to determine the impact of missing data on trial results. Missing data values for primary endpoint(s) can be imputed based on “worst case” (consistent with the null hypothesis) and/or “best-case” (consistent with the alternative hypothesis) assumptions.
- For multi-site studies, COVID-19 may result in site differences, since the pandemic may affect sites’ ability to follow the study protocol and/or collect needed outcome data. A poolability analysis to determine if study site influences the primary outcome(s) is advisable.