Dementia is a collection of neurodegenerative diseases, which 55 million people around the globe are living with, in the present, as we read this article. The number of people suffering from this syndrome is increasing as the proportion of older people rises in most countries. Currently, no cure is available to treat dementia. Early diagnosis offers the best chance of managing this condition. Most people visit memory clinics when they start experiencing cognitive impairments such as forgetfulness. However, only around 10% of those who experience mild cognitive impairment (MCI) will develop dementia even after 10 years. 

Identifying patients who will go on to develop dementia and those who will not is of vital importance. Accurate tools for helping clinicians predict this can help start treatment, care and management to delay the onset of symptoms of dementia. Previously, tools such as CAIDE (Cardiovascular Risk Factors, Aging, and Incidence of Dementia) risk score and BDSI (Brief Dementia Screening Indicator) have been developed to estimate the long and midterm chances of developing dementia. Specific AI-based methods were also developed to predict Alzheimer’s disease conditions, the most common type of dementia, five years in advance. However, no tool was previously available to accurately predict if patients suffering from MCI will develop dementia or not in a clinically relevant timescale.

The new model for dementia prediction

Researchers have developed a novel machine-learning algorithm to act as a diagnostic aid for predicting dementia development in a person. It is superior to previously established models for dementia risk prediction within 2 years of the assessment. The findings have been published recently in a study in the journal JAMA Network Open. The model has a 92% accuracy of predicting whether a patient experiencing MCI will develop dementia within 2 years.

It was developed using the data collected from memory clinics in the US between 2005 and 2015. The sample consisted of 15,307 patients who did not have dementia when they first visited the clinic, but around 10% (1568 patients) developed it within 2 years. The ML algorithm was able to predict the onset of dementia with greater than 90% accuracy. It also identified more than 80% of cases where the patient was given an initial incorrect dementia diagnosis.

Even though the model was initially trained using 258 variables for dementia-related risk factors to achieve 92% accuracy, further analysis showed that only 6 key factors were required to achieve 91% accuracy. This is an important finding as one of the drawbacks of ML models for real-life application purposes are the large number of parameters they need for making correct predictions.

Future applications

The researchers hope that this ML model will serve as an excellent diagnostic aid for early detection of the diseases causing dementia. Doctors would be able to recommend necessary lifestyle changes at the initial stages of assessment to delay dementia onset. As this is a highly feared disease condition, using this ML model will also help prevent the unnecessary stress faced in the case of misdiagnosis, which occurs 8% of the time.

Neurodegenerative diseases have been an intense area of research for a long time. Several institutions, tech companies, and startups have tried to develop cures for such conditions, however, with little success till now. Diagnosing early is still the best available method for managing degenerative diseases such as dementia. Using tested AI models for diagnosis purposes will help doctors and clinicians make the correct predictions quickly and accurately.