Being able to predict who will suffer from dementia with accuracy thanks to artificial intelligence. That's the research topic of a team of researchers from the University of Exeter in the UK. The results of their study were published in JAMA Network Open and show how theydeveloped machine learning algorithms that can predict the incidence of dementia over 2 years. Data from over 15,000 US patients was integrated which showed that the AI systems were very efficient as 92% of the results were correct. Among patients in clinics treating memory disorders such as Alzheimer's disease, for example, those suffering from dementia at the start of treatment are in the minority. The challenge for doctors is to identify patients who are at risk of developing dementia. Until now, they have relied on mild cognitive impairment (MCI) in the initial assessment to decide on a suitable follow-up for upcoming dementia. Machine learning makes it possible to exploit complex masses of data. These algorithms can integrate information not usually used by clinics such as advanced neuroimaging, genetic testing and cerebrospinal fluid bio-markers, etc. On the other hand, researchers and specialists benefit from this clinical application.

Algorithms with 92% accuracy

The Exeter researchers' study focused on the possibility of Machine Learning algorithms predicting dementia at 2 years and comparing these predictions to existing models. For this purpose, they used data from 15,307 patients without dementia at baseline, treated in 30 memory clinics of the National Alzheimer Coordinating Center in the USA between 2005 and 2015. They then implemented four ML algorithms: logistic regression (LR), support vector machine (SVM), random forest (RF), and gradient-boosted tree (XGB) to classify patients according to the likelihood or not of developing dementia. 1568 of these patients were diagnosed with dementia 2 years after their first visit during these ten years, i.e. about 10%, which the machine learning algorithms requiring only 6 variables predicted with an accuracy of up to 92%. These models even revealed 80% of the diagnostic errors, which the researchers found to be around 8% (130 cases corrected after the error was identified). Professor David Llewellyn, Alan Turing Fellow based at the University of Exeter, said:
"We are now able to teach computers to accurately predict who will develop dementia within two years. We are also delighted to learn that our machine learning approach has been able to identify patients who may have been misdiagnosed. This offers an opportunity to reduce guesswork in clinical practice and significantly improve the diagnostic pathway, helping families access the support they need as quickly and accurately as possible."
The conclusion of this study is that ML models are more reliable in their prediction of incident dementia at 2 years than other existing models. On the other hand, six key factors for dementia risk identified in this study have the potential to improve clinical practice and decision making. Dr Janice Ranson, a researcher at the University of Exeter Medical School, said:
"We know that dementia is a very feared disease [...] Integrating machine learning into memory clinics could help ensure that diagnosis is much more accurate, reducing the unnecessary distress that a misdiagnosis could cause. "
References: Performance of Machine Learning Algorithms for Predicting Progression to Dementia in Memory Clinic Patients, Charlotte James, Janice M. Ranson, Richard Everson, David J. Llewellyn - JAMA Netw Open. 2021;4(12):e2136553. doi:10.1001/jamanetworkopen.2021.36553 Translated from Les performances de l'IA pour la prédiction de la démence au coeur d’une étude de l’Université d’Exeter

Our articles will then appear first in Google Top Stories.