Implications of Missing Data Mechanisms in Randomized Controlled Trials: Current Practices and Suggestions Utilizing the
ACTIVE
Study
Jody S. Nicholson, Pascal R. Deboeck, Waylon Howard, Brenna Gomer ABSTRACT
Background
Missingness in randomized controlled trials (RCTs) might mask or inflate effects with consequences for translating results confidently. The current study presents three steps to establishing confidence in results in light of missing data using an RCT examining the preventive potential of computerized cognitive training for the onset of dementia.
Methods
Participants ( n = 2832) consented to complete a computerized cognitive training study and ranged in age from 65 to 94 years ( M = 73.33; SD = 5.74). Variables related to baseline demographic, physical, cognitive, and psychological functioning were examined as predictors of missingness to identify auxiliary variables. Secondary data analysis was conducted using data from the Advanced Cognitive Training for Independent and Vital Elderly study (ACTIVE). Confidence in the sustained impact of the intervention group was examined using two types of MNAR models (i.e., selection and pattern mixture modeling frameworks).
Results
Using auxiliary variables, there were slight changes in the pattern of results of the cognitive training intervention, suggesting that missing data was contributing to slight increases and decreases in model estimates, but not creating major changes in the interpretation of the trial. The analysis under MNAR models did not suggest any systemic bias in the study outcomes due to missing data but highlighted potential differences in those with complete data compared to those with incomplete data.
Conclusions
The current study demonstrates how to expand examination of missingness and triangulate multiple methods for investigating bias under MAR and MNAR. It is recommended that the checklist for reporting for randomized controlled trials (i.e., CONSORT‐ROUTINE) be amended to reflect specificity for requirements necessary for missingness.