A simple language test developed using artificial intelligence (AI) is able to predict, with a high level of accuracy, which cognitively normal individuals will go on to develop Alzheimer's disease (AD), new research suggests.
The findings show that the test has a 70% accuracy rate in predicting AD onset years before cognitive decline begins and is more accurate than traditional predictive methods, such as neuropsychological testing.
"The bottom line is that we're finding tiny markers that when you put them together are providing a significant amount of information to predict dementia," study author Guillermo Cecchi, PhD, Computational Biology Center, IBM Research, Yorktown Heights, New York, told Medscape Medical News.
The researchers are hopeful the new results will eventually lead to the use of simple, inexpensive speech probes that detect early dementia and that monitor its progression.
"The value of this [type of test] is that it can be done quickly; it's not intrusive and can done at any time," said Cecci.
The study was published online October 22 in eClinicalMedicine.
The Cookie Thief
A key priority in AD research is developing early interventions to decrease risk, delay onset, and/or slow disease progression. This requires identifying patients who are likely to benefit from such interventions, and this is where language comes in.
Research shows that various aspects of language are an important component of age-related cognitive decline. Even mundane linguistic abilities, such as object naming, engage extensive brain networks.
For the study, the investigators used data from the Framingham Heart Study (FHS), a longitudinal investigation of more than 5000 individuals spanning several decades. As part of the FHS, participants complete a neuropsychological test battery that includes the cookie-theft picture description task (CTT) from the Boston Aphasia Diagnostic Examination.
In this test, study participants are asked to describe in writing the cookie-theft picture. The picture depicts three characters in a kitchen ― a woman at an overflowing sink; a boy reaching into a cookie jar in the cupboard; and a girl expecting to get a cookie from the boy. Investigators extracted linguistic variables from responses to the CTT. In total, 87 linguistic variables were computed.
Using these variables, researchers developed computer models to predict whether a participant would develop mild cognitive impairment (MCI) leading to AD.
A review panel with at least one neurologist and one neuropsychologist reviewed possible cognitive decline and dementia cases. They based a diagnosis of AD on criteria from the National Institute of Neurological and Communicative Disorders and Stroke–Alzheimer's Disease and Related Disorders Association.
Case patients included those who developed MCI because of AD at or before age 85. Age-, sex-, and education-matched control persons included those who remained dementia-free until at least age 85.
From more than 3000 responses to the CTT, the researchers created models with which to test 80 participants – 40 case patients, and 40 control persons.
Analyses showed that future onset of AD was associated with repetitiveness, misspellings, and telegraphic speech, which is defined as speech that lacks fluid grammatical structure and continuity.
Another important variable, said Cecchi, is the "granularity" of "reference" or lack of such reference, for example, referring to the older woman in the picture as a "mother" or "wife" instead of the using the more general term, "woman."
Using linguistic variables yielded significant predictive power, with an area under the curve of 0.74 and an accuracy of 0.70. The mean time to diagnosis of AD was 7.59 years.
The researchers noted that the model is based on data that were collected when study participants were cognitively normal. The study showed that predictions were harder regarding participants who had a college degree than regarding less-educated participants.
Linguistic competence is a behavioral marker of educational and occupational attainment, both of which have been suggested to increase "cognitive reserve," the investigators note.
The study showed that it is much easier to predict conversion to AD in women than men. The authors note that the prevalence of AD is significantly higher in women than men and that the rate of progression after onset of cognitive impairment is faster in women.
Performance was better with predictive models that used linguistic variables than predictive models that incorporated more traditional variables associated with AD risk, such as neuropsychological test scores, demographic information, and APOE status. For disease predictions based on a combination of these traditional variables, the accuracy was 59%, said Cecchi.
The authors note that neuropsychological tests and other biomarkers, including cerebrospinal fluid assessments and brain imaging, have been used to predict progression of MCI to AD. In addition, there have been very promising results using neurofilament light chain for disease progression for patients at early presymptomatic stages of familial AD.
"However, these are still technologically or logistically demanding, and require significant specialist involvement," they note.
Cecchi and his colleagues hope this new research will lead to the development of a simple, accessible tool to accurately assess AD risk.
Cecchi envisions a simple test that can be used regularly by clinicians and family members alike to track a patient's disease progression, thereby dispensing with the need for long appointments at the doctor's office.
Patients who were identified by the test as being at risk could potentially make lifestyle changes to help delay cognitive decline. These could include following a healthy diet, becoming physically active, and enhancing social and cognitive engagement, said Cecchi.
In addition, physicians could use this tool to identify patients who might benefit from enrollment in clinical trials of potential preventive therapies, he said.
Exciting but Early Research
Commenting on the research for Medscape Medical News, Heather Snyder, PhD, vice president of medical and scientific relations, the Alzheimer's Association, described the research as "exciting" and "intriguing" but cautioned that it's still "early work."
Research into identifying speech or language biomarkers of cognitive change is "definitely" one of the "emerging new areas," said Snyder. The Alzheimer's Association has funded studies in this area.
"We are increasingly seeing studies that are looking at, and trying to understand, speech and speech patterns and how that might be an indicator of what might be happening in the brain," she said.
However, Snyder pointed out that the current study used a rather small and "pretty defined population of individuals."
She also noted that language is tied into hearing and other senses. "So this is not as straightforward as it might seem."
She emphasized that there is definitely some work still to be done before this type of test will be useful as a diagnostic tool
Pfizer, Inc, provided funding to obtain the data from the Framingham Heart Study Consortium and to support the involvement of IBM Research in the initial phase of the study. Cecchi worked as a salaried employee of IBM Corp for the full duration of the project. IBM holds a patent for the extraction of one of the features used in the linguistic model.
EClinicalMedicine. Published online October 22, 2020. Full text
Medscape Medical News © 2020
Send news tips to firstname.lastname@example.org.
Cite this: Simple Language Test May Predict Alzheimer's Years Before Symptoms - Medscape - Oct 22, 2020.