MATCH-Net for Risk Prognosis

Dynamic Prediction with Time-Dependent Covariates in Clinical Survival Analysis of Alzheimer's Disease using Convolutional Neural Networks

Try the Demo

Personalized Risk Prognosis

We present the first interactive atlas of Alzheimer's disease trajectories, powered by MATCH-Net. Select sample patients below and explore their risk trajectories using the timeline navigator. The top panel shows the historical risk surface for each case, as well as the dynamically computed forward risk estimates at each time step. The bottom panel shows the evolution of covariate measurements. Features can be toggled using the selector panel, or even edited on the fly with the data table to explore their effects. The dynamic influence of each feature and timestep can be observed in the right-hand panel.

Hint: Some interesting examples to look at include patients 61, 361, 867, and 2087, among others. Trajectories that eventually end in diagnosis of Alzheimer's disease are shaded in red to indicate failure, and censored trajectories are shaded in gray.

How It Works

In Alzheimer's disease—the annual cost of which exceeds $800 billion globally—the effectiveness of therapeutic treatments is often limited by the challenge of identifying patients at early enough stages of disease progression for treatments to be of potential use. As a result, early detection through accurate and personalized prognosis during earlier stages of cognitive decline is critical for effective intervention and subject selection in clinical trials. MATCH-Net: a Missingness-Aware Temporal Convolutional Hitting-time Network, is designed to capture temporal dependencies and heterogeneous interactions in covariate trajectories and patterns of missingness, and issues dynamic predictions for personalized risk prognosis.

MATCH-Net accepts as input a sliding window of observed covariates, as well as a corresponding binary mask of missing-value indicators. The convolutional dual-stream architecture learns representations of longitudinal covariate trajectories and missingness by extracting local features from temporal patterns in the data. Employing a multi-task approach, each prediction in the output block is associated with a single softmax layer, producing the array of failure estimates for pre-specified intervals. Using real-world data from the Alzheimer's Disease Neuroimaging Initiative (ADNI), we demonstrate state-of-the-art performance without making any assumptions regarding underlying longitudinal or time-to-event processes.

Jarrett, D., Yoon, J., and van der Schaar, M. (2018). MATCH-Net: Dynamic Prediction in Survival Analysis using Convolutional Neural Networks. NIPS 2018 ML4H Workshop (arXiv:1811.10746).

This website does not provide medical or professional services advice. The content on this website is intended to be a general information resource in regard to the subject matter covered, is provided solely on an "as is" and "as available" basis without warranty of any kind, neither express nor implied, and is not intended or implied to be a substitute for medical advice offered by a physician. If you desire or need such services or advice, regarding the information contained on this website or otherwise, you should consult a professional healthcare provider. Your use of the website or any content on the website is at your own risk. We specifically disclaim any liability, whether based in contract, tort, strict liability or otherwise, for any direct, indirect, incidental, consequential, or special damages arising out of or in any way connected with access to or use of the website or the content on the website including, but not limited to, the risk calculator.

Data used in preparation of this article were obtained from the Alzheimer’s Disease Neuroimaging Initiative (ADNI) database ( As such, the investigators within the ADNI contributed to the design and implementation of ADNI and/or provided data but did not participate in analysis or writing of this report. A complete listing of ADNI investigators can be found here. The ADNI was launched in 2003 as a public-private partnership, led by Principal Investigator Michael W. Weiner, MD. The primary goal of ADNI has been to test whether serial magnetic resonance imaging (MRI), positron emission tomography (PET), other biological markers, and clinical and neuropsychological assessment can be combined to measure the progression of mild cognitive impairment (MCI) and early Alzheimer’s disease (AD). For up-to-date information, see Further details of the ADNI data sharing and publication policy can be found here.

Copyright © 2018 ● Dan Jarrett
All Rights Reserved