Researchers Estimate Risk for Dementia Through ML and NLP Algorithms

A team of researchers from Massachusetts General Hospital (MGH) has developed a software-based method to scan electronic health records (EHRs) in order to estimate the risk that a healthy person will receive a dementia diagnosis in a possible future. The algorithm of this software uses machine learning (ML) to first enlist the key clinical terms associated with cognitive symptoms identified by clinical experts and later the team used NLP (natural language processing) to look through the EHRs looking for those terms. At last, the results were used to estimate patients’ risk of developing dementia. According to Thomas McCoy, Jr., MD, first author of the paper, “the most exciting thing is that we are able to predict the risk of new dementia diagnosis up to eight years in advance.” This team was consisting of members of MGH’s Center for Quantitative Health, the Harvard T.H. Chan School of Public Health, and the Harvard Brain Tissue Resource Center. The paper got published recently in “Alzheimer’s & Dementia.” Here is the abstract from the study:  

Introduction	Preventing dementia, or modifying disease course, requires identification of presymptomatic or minimally symptomatic high-risk individuals.
Methods	The team used longitudinal electronic health records from two large academic medical centers and applied a validated natural language processing tool to estimate cognitive symptomatology. Researchers used survival analysis to examine the association of cognitive symptoms with incident dementia diagnosis during up to 8 years of follow-up.
Results
Discussions	A cognitive symptom measure identified in discharge notes facilitated stratification of risk for dementia up to 8 years before diagnosis.

In America, Alzheimer’s affects more than 5.5 million people at present and the number is expected to increase shortly. Therefore, early diagnosis of dementia could be an effective measure to improve care and treatments for it. However, the early detection tools that exist today require additional, potentially costly, data collection. But the tool developed at MGH is completely based on software that can better use data already generated during routine clinical care. This approach for early risk detection can potentially accelerate the research efforts for slowing progression or reverse early disease. Thomas McCoy notes that “this method was originally developed as a general ‘cognitive symptom’ assessment tool. But we were able to apply it to answer particular questions about dementia.” He further explained that “this study contributes to a growing body of work on the usefulness of calculating broad symptom burden scores across neuropsychiatric conditions.”  

Experts’ Quotes

Roy Perlis, MD, senior author of the study and director of the MGH Center for Quantitative Health said, “We need to detect dementia as early as possible to have the best opportunity to bend the curve. With this approach we are using clinical data that is already in the health record, that doesn’t require anything but a willingness to make use of the data.” Rudolph Tanzi, PhD, a member of the research team, “this approach could be duplicated around the world, giving us more data and more evidence for trials looking at potential treatments.” Rudolph is also the vice-chair of Neurology, and Co-Director of the MGH McCance Center for Brain Health at the MGH Institute for Neurodegenerative Diseases.