$1.1M grant supports learning more about early Alzheimer’s with machine learning

A newly funded project aims to develop machine learning tools that can improve our understanding of patient risk for Alzheimer’s disease through the study of longitudinal clinical data. The project, called “Tackling Progressive Disease – Learning from Longitudinal Observational Clinical Data in the Presence of Noise and Confounding,” earned a $1.1M grant from the NSF’s Smart Health and Biomedical Research in the Era of Artificial Intelligence and Advanced Data Science program.

Data from patient records could provide a valuable historical perspective on which factors increase Alzheimer’s risk. The data in these records is challenging to work with, however, suffering from two major shortcomings: it lacks ground truth diagnoses, offering instead only proxies like billing codes and prescribed treatments; and the dataset covers irregular time intervals, only being updated when the patient seeks care.

But this noisy, confounded data can still be valuable in better understanding the sixth leading cause of death in the US, according to the project leads.

“There is a pressing need to improve our ability to stratify and treat patients at risk of developing Alzheimer’s,” write Profs. Jenna Wiens, PI of the project and Co-Director of Michigan Precision Health, Bruno Giordani, Chief Psychologist at the U-M Department of Psychiatry, and Raymond Migrino, Cardiologist at the Phoenix VA Health Care System. “While curated research datasets have been crucial to advancing our understanding of Alzheimer’s, the disease progresses slowly over the course of decades. It will be some time before such curated datasets will contain enough data to provide insights about early disease progression.”

It’s this early progression and the period leading up to an Alzheimer’s diagnosis that they seek to learn more about from patient data. 

The researchers aim to overcome the technical limitations through several key advances in machine learning techniques. One key outcome proposed by the team is a set of new approaches for multi-event survival analysis that model the probability of events hierarchically at different time scales while accounting for noisy labels and novel machine learning techniques that advance our ability to estimate causal effects using observational data.

“This project will lead to key advancements in the fields of machine learning for patient risk stratification and individual treatment effect estimation,” the investigators write.

While the project is inspired by the important challenge of predicting patient risk of Alzheimer’s, the use of these tools to estimate patient risk from observational data like patient records has the potential to generalize from Alzheimer’s to a number of other conditions.

“We expect the proposed work to lay the groundwork for clinical systems that directly impact society by identifying patients most likely to benefit from early intervention.”