A PREDICTIVE prognostic model (PPM) has been developed to predict the rate at which individuals will progress to Alzheimer’s Disease (AD). Whilst most machine learning models in AD research have focused on binary disease diagnosis (whether they will develop AD or not), researchers have now developed a PPM-derived prognostic index that predicts how fast an individual may progress from mild cognitive impairment (MCI) to AD, offering a solution for patients that may be misdiagnosed with binary classification models due to the subtle differences in their disease trajectory that these models can fail to capture.
The research team devised a trajectory modelling approach based on Generalized Metric Learning Vector Quantization (GMLVQ) that leverages multimodal data to make predictions about future cognitive decline at early dementia stages by iteratively adjusting class-specific prototypes and learning class boundaries. The model was trained on routinely collected, non-invasive, and low-cost patient data (cognitive tests and medial temporal lobe grey matter density with structural MRI), to discriminate between stable MCI and progressive MCI (individuals who progressed to AD within 3 years).
To extend the model beyond solely predicting the likelihood of AD progression, the research team used a scalar projection method to generate a PPM-derived prognostic index of cognitive decline over time. This index represents the distance of an individual from the stable MCI prototype, with a higher index suggesting a higher risk of cognitive decline, and progression to AD, in the future. The model was trained on research cohort data from ADNI, and tested on independent, real-world patient data from memory clinics across the UK and Singapore to validate its generalisability.
The main findings of the study were that the model can predict whether patients at early stages of the disease (MCI) will remain stable or progress to AD, with 86% accuracy, 82.38% sensitivity, 80.94% specificity, and an area under the curve (AUC) of 0.84.
This model provides a new strategy for early identification of at-risk patients, which may enable the identification of the optimal timeframe for intervention of preventative measures to slow disease progression in AD.
Katrina Thornber, EMJ
Reference
LY Lee et al. Robust and interpretable AI-guided marker for early dementia prediction in real-world clinical settings. EClinicalMedicine. 2024;DOI:10.1016/j.eclinm.2024.102725.
