Big data, AI used in bid to predict dementia early

Victoria Bela

Chinese scientists say they have made a major step towards predicting a patient’s risk of dementia up to a decade and a half before symptoms start by analysing a sample of their blood.

The scientists used a database of more than 50,000 people to identify proteins linked to a risk of developing different types of dementia, and with the help of AI the team created a predictive model to assess disease risk.

The use of artificial intelligence was “one of the key factors for the success of this research”, said Yu Jintai, study author and neurology professor at the Fudan University-affiliated Huashan Hospital.

Using the data-driven strategy, the scientists “innovatively identified important plasma biomarkers for future dementia prediction” the team wrote in a paper published in the journal Nature Ageing last week.

Blood tests for diagnosing types of dementia such as Alzheimer’s have been growing in popularity. From a single drop of blood they can determine whether a patient who has started showing symptoms has the disease.

But scientists have bigger goals in mind for blood biomarker tools, such as using them to predict whether a patient could develop the disease in future, even before exhibiting clinical symptoms.

There is no cure for dementia but being able to understand if a person could develop it might allow for early diagnosis and intervention, according to the authors of the paper.

The study of proteins – also called proteomics – can be used to find potential drug or diagnostic interventions for diseases. However, systematically studying proteins in the blood has proved difficult because of technical constraints and a lack of comparison methods, the team wrote.

To overcome this barrier, the team employed the help of the massive UK Biobank cohort, which enrolled more than 50,000 people aged 40 to 69 and had a median follow-up period of 14 years starting in the mid-2000s.

Some 1,400 of the subjects in the cohort – who all provided biological samples and demographic information – developed dementia within 10 years of the initial data collection.

The biobank recently released a new data set of more than 1,400 blood proteins found in the participants’ samples.

This data release gave the team an “unprecedented opportunity” to conduct a proteomics study on blood proteins associated with the development of dementia, the paper said.

It allowed them to “trace the trajectories of plasma proteins back from the time of dementia diagnosis and assess when each protein begins to deviate from normal control values”.

The scientists focused their study on a handful of “important proteins” found to have begun changing in expression up to at least a decade before the clinical onset of dementia.

These proteins were evaluated using an AI algorithm called a light gradient boosting machine, which used machine learning to screen out proteins and combinations that were most closely related to dementia risk, Yu said.

The algorithm was used to determine which proteins created a better prediction model and checked it against the biobank data that showed which subjects had developed dementia.

The algorithm had “powerful pattern recognition and prediction capabilities”, allowing for more efficient screening of the large-scale data set, Yu said

According to the team, including protein data by itself into a prediction model was “unlikely to attain the highest predictive accuracy”.

To develop an “optimal predictive algorithm that is non-invasive, cost-effective and easily accessible”, they combined data on a protein called GFAP – which they found was associated with more than double the risk of dementia – with demographic information such as age and gender.

Their final, combined predictive model showed promise for being able to provide an accurate prediction of future dementia, even more than 10 years before the diagnosis, according to the paper.

And compared with imaging scans or spinal taps used to screen people for disease risk, their method could also offer considerable cost benefits, the scientists added.