While artificial intelligence (AI) and machine learning have the potential to identify patients at risk for dementia, significant improvements are required before its use in clinical practice, according to study findings published in the journal Alzheimer’s & Dementia. 

With the prevalence of dementia growing worldwide due to increases in life expectancy, researchers have been seeking ways to delay its onset. Research in this area has focused on early-stage treatment and prevention as there is no current definitive cure for dementia. In addition, literature regarding the impact of treating dementia risk factors to help slow dementia progression remains weak. 

For the study, the researchers assessed how AI and machine learning could help identify risk factors and potential interventions for patients with dementia. Machine learning is a subset of AI with a focus on models and algorithms that allow computers to make decisions or predictions. Machine learning and other risk-profiling tools, such as logistic regression, support vector machines, random forest, and gradient-boosted trees, are not widely used in dementia prevention. 

These models can work to identify potential causes of risk factors and assist with identifying patient variables to assist with clinical trial recruitment, decreasing costs of trials, and accelerating the discovery of treatments. 

In the Lancet Commission on Dementia Prevention, Intervention, and Care, first published in 2017, some risk factors identified for the development of dementia included fewer years of education, smoking, depression, physical inactivity, social isolation, and type 2 diabetes. In the updated 2020 report, the Lancet Commission added alcohol, traumatic brain injury (TBI), and air pollution as risk factors.

While these factors have been identified by several systematic reviews, it is important to consider the relationship of these risk factors across an individual’s lifespan. Machine learning has the ability to use longitudinal data to understand disease trajectories and disease heterogeneity. 

One example of a machine learning model used for dementia is the Cardiovascular Risk Factors, Ageing, and Dementia (CAIDE) model, which uses patient data and mid-life risk factors to predict dementia risk in 20 years with moderate accuracy (area under the curve [AUC], 0.77; 95% CI, 0.71-0.83). 

A recent study has shown that the use of a variety of machine learning algorithms is superior to using common dementia risk models alone, such as the CAIDE model, in predicting 2-year dementia risk; however, the clinical utility of these models is in question as they are unlikely to help with prevention of dementia progression since neurodegeneration will already have occurred. 

The use of multimodal data in the CAIDE model is more reflective of clinical practice, while using a variety of machine learning models can increase sensitivity and specificity, potentially at the expense of clinical translatability. 

Some major current challenges in using machine learning to predict dementia risk are that risk factors can exert effects through a variety of different biological pathways, which can make it hard to identify mechanistic targets. 

Study limitations include the inability to establish causality using machine learning models, biased machine learning models due to lack of inclusion of minority populations, and difficulties understanding or interpreting the models. 

“Multidisciplinary collaboration is required to harness the potential of ML [machine learning], maximize utilization of available resources and data access, and enhance traditional approaches to advance dementia prevention research,” the researchers wrote.

Disclosure: Some study authors declared affiliations with biotech, pharmaceutical, and/or device companies. Please see the original reference for a full list of authors’ disclosures.

This article originally appeared on Neurology Advisor