Comparing Methods to Predict Dementia
Author Information
Author(s): João Maroco, Dina Silva, Ana Rodrigues, Manuela Guerreiro, Isabel Santana, Alexandre de Mendonça
Primary Institution: Unidade de Investigação em Psicologia e Saúde & Departamento de Estatística, ISPA - Instituto Universitário
Hypothesis
Newer statistical classification methods derived from data mining and machine learning can improve the accuracy, sensitivity, and specificity of predictions obtained from neuropsychological testing for dementia.
Conclusion
Random Forests and Linear Discriminant Analysis are the most effective methods for predicting dementia from neuropsychological tests.
Supporting Evidence
- All classifiers performed better than chance alone (p < 0.05).
- Support Vector Machines showed the largest overall classification accuracy (Median = 0.76).
- Random Forest ranked second in overall accuracy (Median = 0.73).
- Linear Discriminant Analysis showed acceptable overall accuracy (Median = 0.66).
- Sensitivity was low for Support Vector Machines (Median = 0.3).
- Random Forests and Linear Discriminant Analysis ranked first in sensitivity and specificity.
Takeaway
This study looked at different ways to predict if older people with memory problems will develop dementia, finding that some methods work better than others.
Methodology
The study compared seven data mining classifiers and three traditional classifiers using neuropsychological tests on a sample of elderly patients with Mild Cognitive Impairment.
Potential Biases
The performance of classifiers may depend on the tuning parameters chosen, which could introduce bias.
Limitations
The sample size may limit the performance of some data mining methods, and the results are based on a specific dataset.
Participant Demographics
The sample consisted of 400 elderly patients, with 275 diagnosed with Mild Cognitive Impairment and 125 with Dementia.
Statistical Information
P-Value
p<0.05
Statistical Significance
p<0.05
Digital Object Identifier (DOI)
Want to read the original?
Access the complete publication on the publisher's website