Combining Multiple Imputation and Bootstrapping for Prognostic Variable Selection
Author Information
Author(s): Martijn W. Heymans, Stef van Buuren, Dirk L. Knol, Willem van Mechelen, Henrica C.W. de Vet
Primary Institution: Vrije Universiteit, Institute for Health Sciences, Department of Methodology and Applied Biostatistics, Amsterdam, The Netherlands
Hypothesis
Can combining multiple imputation with bootstrapping improve prognostic variable selection in studies with missing data?
Conclusion
Combining multiple imputation with bootstrapping results in effective prognostic models for datasets with missing values.
Supporting Evidence
- The study combined data from three trials to address the issue of missing data in prognostic modeling.
- Multiple imputation was used to handle missing values, while bootstrapping was applied for model selection.
- The results indicated that the combined method improved the selection of prognostic variables.
Takeaway
This study shows that when researchers have missing data, they can use a special method that combines two techniques to make better predictions about patients' health.
Methodology
A prospective cohort study using data from three randomized controlled trials, applying multiple imputation and bootstrapping techniques for variable selection.
Potential Biases
Potential bias due to missing data and the assumptions made in the imputation process.
Limitations
The study's sample size may be too small for reliable modeling, and the methods may not account for non-linear effects or interactions.
Participant Demographics
Patients with low back pain, with a mean age of 40.6 years and 71% male.
Digital Object Identifier (DOI)
Want to read the original?
Access the complete publication on the publisher's website