A mouthwash, a few milliliters of saliva and a battery of data: it is on this unexpected combination that Chinese researchers are banking on to better monitor brain aging. Their idea: to identify early on signals which sometimes precede memory or orientation problems by several years.
This work, carried out at Chongqing Medical University and published in the journal Translational Psychiatryrely on theartificial intelligence to analyze the content of saliva: stress hormones, inflammation molecules, composition of the oral microbiota. The objective is to predict the risk of
neuropsychiatric symptoms in seniors who are still independent, considered as an alarm signal before neurodegenerative diseases.
Mood changes, first signs of future Alzheimer’s disease?
Before the “memory lapses”, many elderly people go through a more vague phase: loss of momentum, persistent sadness, anxiety, irritability, sleep problems. Specialists group these manifestations under the term neuropsychiatric symptoms, which are found in 30 to 80% of people on the cognitive continuum, well before a diagnosis of
Alzheimer’s disease or other dementia.
For the Chinese team, these psychic signals are worth tracking down early. “Neuropsychiatric symptoms are early indicators of cognitive decline linked to neurodegenerative diseases, and their rapid detection is of paramount importance“, explain Ping Liu and Zeng Yang from Chongqing Medical University. Problem: current tools, such as the Neuropsychiatric Inventory questionnaire, remain long to administer and poorly suited to systematic screening in community medicine.
From saliva to risk score: what AI really does
To overcome these limitations, the researchers recruited 338 residents aged over 60 attending community health centers in Chongqing. Each completed a questionnaire (age, sex, level of education, alcohol consumption, history), took cognitive tests, then provided a saliva sample obtained by mouthwash. In this liquid, the teams measured cortisol, several inflammatory cytokines, the Cath-B protein, and sequenced their oral microbiota.
Half of this data was used to train several types of machine learning models, before testing it on the other half of the data. The model called XGBoost proved to be the most efficient. They then used one of the models they had developed to create a platform that was easily accessible to healthcare professionals and usable for screening groups of older people.
Research has revealed interactions between cortisol, certain oral bacteria and specific metabolic pathways. These findings could offer new perspectives for exploring the connections between inflammation, microbiota and mental health.
Towards a future AI saliva test for the prevention of dementia?
Researchers have already transformed a simplified version of the model into a nomogram, a graphical tool that assigns a risk score based on a few clinical and biological parameters. The idea is that a community doctor could, after a simple saliva test, identify patients most exposed to neuropsychiatric symptoms and refer them more quickly to a memory consultation, without replacing imaging exams or specialized blood tests.
The initial results of this research highlight the potential of machine learning models to analyze biological data and identify neuropsychiatric symptoms early.
For the moment, however, this approach remains limited to this Chinese study and expensive laboratory technologies.