Chair: Elizabeth DeLong
NIH Representative: David Murray
Members: Chul Ahn, Bryan Comstock, Andrea Cook, Constantine Gatsonis, Dan Gillen, Roee Gutman, Patrick Heagerty, Jesse Hsu, Ken Kleinman, J. Richard Landis, Michael Leo, Qian Li, Joan Russo, Susan Shortreed, Liz Turner, William Vollmer, Jin Wang, Rui Want, Song Zhang
Project Manager: Darcy Louzao
Pragmatic clinical trials, including cluster-randomized trials, present biostatistical and study design issues in addition to those typically encountered with traditional clinical trials. The Biostatistics and Study Design Core works with the NIH and Demonstration Project teams to create guidance and technical documents regarding study design and biostatistical issues relevant to pragmatic clinical trials.
For example, when randomizing clusters rather than individuals, several issues need attention. These include the trade-off between sample size and potential contamination, the intra-class correlation at different levels, varying cluster size, and the need for stratification or matching.
Additionally, special consideration must be given to handling informative missing follow-up data when using electronic health records as the basis for follow-up data collection. Individuals who are less healthy and have more chronic conditions will have more healthcare visits per year. If an intervention is effective in improving general health, then those who received the intervention would be more likely to have missing follow-up data compared with those who did not receive the intervention. Ignoring this missing data issue could lead to biased results, including concluding that the effective intervention is not effective.
Missing data, unequal cluster sizes, pair-matching vs stratification, and the intraclass correlation coefficient are some of the many topics being addressed by the Core, and the list will continue to grow as new challenges are faced.
The Core’s objectives include the following:
Work with Demonstration Projects to address gaps and limitations in their statistical plans and study designs during the planning phase
Understand which methods can be directly applied in a distributed data setting, which can potentially be modified for the distributed data setting, and which will require individual-level data
Document new statistical and methodological issues that arise and share knowledge through these case studies
Gather input on key methodological issues from NIH collaborators, investigators, and academic institutions
Identify areas in need of methods development and work to address these methodology challenges
Generate new knowledge by studying the application of statistical techniques (e.g., constrained randomization) in pragmatic and cluster-randomized trial designs