Related Experiment Videos
Selecting pre-screening items for early intervention trials of dementia--a case study
Lang Li1, Jeffrey Huang, Sharon Sun
1Division of Biostatistics, Indiana University School of Medicine, 1050 Wishard Boulevard, RG4, Indianapolis, IN 46202, USA.
Statistics in Medicine
|January 13, 2004
Summary
Statistical methods, including neural networks and logistic regression, show promise for identifying cognitive impairment. These models generally outperform expert opinion, offering better predictive power for dementia screening.
Area of Science:
- Neurology
- Biostatistics
- Machine Learning
Background:
- Accurate discrimination between normal cognition and cognitive impairment or dementia is crucial for timely intervention.
- Existing statistical methods require evaluation for their predictive power and efficiency in clinical settings.
Observation:
- Six statistical methods (logistic regression, neural networks, decision trees, with and without LASSO variable selection and boosting) were compared against expert opinion.
- Models were trained on baseline data and validated on longitudinal data from a dementia study cohort.
Findings:
- Statistical methods generally outperformed expert opinion in predicting cognitive status.
- Neural networks demonstrated superior performance compared to logistic regression and decision trees.
- Least Absolute Shrinkage and Selection Operator (LASSO) enhanced logistic and neural network models, though variable reduction was minimal for neural networks.
- A single decision tree model proved effective for pre-screening, matching logistic model performance with fewer variables.
Implications:
- The findings support the use of advanced statistical models for cognitive impairment screening.
- Comparing various classification methods is recommended to optimize diagnostic tools for specific clinical needs.
- Decision trees show potential as efficient pre-screening tools for dementia and cognitive impairment.