Predicting Alzheimer's disease progression: Results from the TADPOLE Challenge


Accurate prediction of progression in subjects at risk of Alzheimer’s disease is crucial for enrolling the right subjects in clinical trials, at the right time. However, an unbiased comparison of state-of-the-art algorithms for predicting disease onset and progression is currently lacking. Method: We present the findings of The Alzheimer’s Disease Prediction Of Longitudinal Evolution (TADPOLE) Challenge, which compared the performance of 92 algorithms (Fig. 1) from 33 international teams at predicting the future trajectory of 219 individuals at risk of Alzheimer’s disease (demographics in Fig. 2). Challenge participants were required to make predictions, for each month of a 5-year future time period, of three key outcomes: clinical diagnosis, ADAS-Cog13, and total volume of the ventricles. Results: No single submission was best at predicting all three outcomes. For clinical diagnosis and ventricle volume prediction, the best algorithms strongly outperformed simple baselines (Fig. 3). However, for ADAS-Cog13 no single submitted prediction method was significantly better than random guessing. On a limited cross-sectional dataset emulating clinical trials, performance of the best algorithms at predicting clinical diagnosis decreased only slightly (+3% error) compared to the full longitudinal dataset (Fig. 4). Two consensus methods — mean and median over all predictions — obtained top scores on almost all tasks (Fig. 3-4). Better-than-average performance at predicting clinical progression was associated with the additional inclusion of features from cerebrospinal fluid (CSF) and diffusion tensor imaging (DTI) (Fig. 5). On the other hand, better performance at ventricle volume prediction was associated with inclusion of summary statistics, such as patient-specific biomarker trends (Fig. 5). Full results can be found on the website, while code for submissions is being collated by TADPOLE SHARE: Conclusion: Our work suggests that current prediction algorithms are accurate for biomarkers related to clinical diagnosis and ventricle volume, opening up the possibility of cohort refinement in clinical trials for Alzheimer’s disease.

Alzheimer’s Association International Conference 2020
Razvan Marinescu
Assistant Professor

My research interests are in Machine Learning, and it’s applications in Healthcare and Molecular Biology. I am doing research in generative models, bayesian modelling, causal ML, compositional ML and multimodal modelling.