An interpretable decision tree model for prioritizing functional and cognitive assessments in early diagnosis of Alzheimer’s disease

The extensive clinical implementation of machine learning models for diagnosing Alzheimer’s disease (AD) is often hindered by the fact that they are “black box” technologies. This lack of transparency undermines clinician trust. This study addressed this issue by developing and validating a high-performing, interpretable decision tree classifier. This model prioritizes features that are both statistically significant and clinically relevant for the early diagnosis of AD. The study used a publicly available dataset containing demographic, clinical, and cognitive data from 2,149 patients. A strategic feature selection process was implemented. The final model was meticulously trained and thoroughly evaluated using a separate test set. The evaluation revealed the model’s exceptional diagnostic capability, demonstrating an overall accuracy of 91.8% and an area under the curve of 0.94. The model also demonstrated high sensitivity (92.1%) and specificity (95.7%). Feature importance analysis revealed that predictions were primarily influenced by direct measures of cognitive and functional status. Specifically, the most significant factors were the mini-mental state examination score, the activities of daily living score, and the Functional Assessment score. These core features overshadowed the predictive influence of established risk factors such as family history. Despite their strong statistical association with the disease, the model assigned negligible importance to these risk factors in its final classification. This work demonstrates that an interpretable decision tree can accurately diagnose AD by transparently prioritizing functional and cognitive assessments. The model’s clarity and reliance on routine clinical data highlight its potential as a decision-support tool. It is ready to improve diagnostic pathways and promote further research in this area.