A person’s online interactions leave behind a “digital footprint” that often contains vast amounts of information about their feelings, behavior, and overall well-being. In a new study, researchers at the University of Minnesota (UMN) School of Public Health (SPH) will analyze these digital footprints to examine how their linguistic, syntactic, or behavioral markers could provide early evidence of cognitive impairment and dementia.
The research project, funded by a UMII Seed Grant fund through the UMN Office of the Vice President for Research, will use a novel combination of qualitative and computational methods to evaluate social media and blog data.
First, “we’ll pull a sample of Twitter users who self-disclose as having clinically diagnosed cognitive impairment,” explains Jude Mikal, a research scientist at SPH. “Using a sample of age-matched controls, we use natural language processing (NLP) tools to identify the linguistic and syntactic features of writing that occur more frequently in individuals with self-reported cognitive impairment.”
After identifying language patterns among those self-reporting impairment, researchers will use this data to seek clues in other users’ language and syntax. “We take more text-rich blog data to look for qualitative evidence of known behaviors or blog content that may signal that an individual has advancing cognitive impairment,” says Mikal.
The research has the potential to benefit millions of people, as well as providers, researchers, and policymakers. As the U.S. population continues to age, there will be an exponential increase in the number of individuals diagnosed with cognitive impairment and dementia. While biomarker-based assessment methods for prodromal neurodegenerative disease have been developed (e.g., brain imaging and cerebrospinal fluid markers of amyloid or tau), they are relatively expensive and currently difficult to access, limiting the feasibility of more widespread screening.
“With this proposal,” Mikal noted, “we’re hoping to develop a readily-accessible, inexpensive initial screening aid for mild cognitive impairment using users’ social media data.”
There is a clear need for this study. While diagnosis of prodromal dementia remains frustratingly elusive, research clearly supports the notion that early indicators of cognitive impairment can be found in written communication. Despite the fact that older adults represent the fastest-growing user base for social media platforms — producing troves of longitudinal written communication — researchers have not previously applied natural language processing tools to evaluate these writings for signs of cognitive impairment.
Mikal also emphasized that the researchers are keenly aware of the ethical implications of looking at social media data for evidence of health conditions that may be costly and do not have any actionable cure. In response, he noted, “we use only publicly available Twitter and blog data — data designed to be broadcast to public audiences through a public platform. Before we expand this project, our plan is to conduct a thorough ethical investigation into how older adults feel about having their social media data assessed for research of this nature and how they would like to see any tools developed from this research deployed in a clinical setting.”
The two-year study is a collaborative project that includes UMN Department of Psychiatry and Behavioral Sciences researchers Kelvin Lim and Laura Hemmy who specialize in later life neurodegenerative disorders. It will also leverage a long-time collaboration with Michael Conway at the University of Melbourne, who specializes in bioinformatics and natural language processing to identify health behaviors. The research team also includes Michael Beckstrand, a methodological specialist from the UMN College of Liberal Arts (CLA) Liberal Arts Technologies and Innovation Services (LATIS) program who provides high-level statistical and qualitative analysis skills.
Results and findings will be published in academic journals.