Imputation Strategies for Missing Data in Cluster Randomized Trials
Author Information
Author(s): Ma Jinhui, Akhtar-Danesh Noori, Dolovich Lisa, Thabane Lehana
Primary Institution: McMaster University
Hypothesis
How do different imputation strategies for missing binary outcomes in cluster randomized trials compare in terms of performance?
Conclusion
Different imputation strategies yield similar results when missing data is low, but standard methods underestimate variance when missing data is high.
Supporting Evidence
- When missing data is low, different imputation strategies yield similar treatment effect estimates.
- Standard imputation methods underestimate variance when a high percentage of data is missing.
- Within-cluster and across-cluster imputation strategies are more appropriate for handling missing outcomes in CRTs.
Takeaway
When some data is missing, different ways to fill in the gaps can give similar answers if not too much is missing, but if a lot is missing, some methods can give wrong results.
Methodology
The study compared six multiple imputation strategies for handling missing binary outcomes in cluster randomized trials using simulation studies.
Potential Biases
Standard imputation strategies may underestimate variance and lead to spurious conclusions when intra-cluster correlation is not accounted for.
Limitations
The study does not address scenarios where entire clusters are missing or the mechanisms of missing data beyond MCAR and CD.
Participant Demographics
Patients aged 65 and older from 28 family practices, with a focus on blood pressure management.
Statistical Information
Confidence Interval
1.14 (0.76, 1.70)
Digital Object Identifier (DOI)
Want to read the original?
Access the complete publication on the publisher's website