Can data-driven patient grouping predict dementia better than standard clinical criteria?

What is it about?

Mild cognitive impairment (MCI) is a clinically recognised condition that increases risk of conversion to dementia, but standard clinical subtyping based on Petersen and Morris criteria is a categorical, criteria-based approach that may not capture the full complexity of cognitive profiles. This study compared that standard clinical approach against latent profile analysis (LPA), an empirical statistical method that derives groupings directly from continuous neuropsychological test data without imposing predetermined categories. Memory clinic participants with MCI (n=139) and age-matched controls (n=98) from Queen's University Belfast were recruited and given a full cognitive assessment. The study examined how closely standard clinical subtypes corresponded to the empirically derived LPA clusters, and compared the two approaches in their ability to predict which patients converted to dementia over follow-up. Kaplan-Meier survival analyses assessed conversion risk in groups defined by each method.

Featured Image

Why is it important?

Improving the accuracy of dementia prediction in memory clinics has direct consequences for patients, families and health systems, enabling earlier intervention, better-informed prognosis and more efficient allocation of specialist resources. Standard MCI subtyping, though widely used, is a blunt tool whose categories were developed on theoretical grounds rather than derived empirically from patient data. This study is among the first to formally test whether a data-driven statistical approach like LPA can outperform standard clinical subtyping in predicting conversion to dementia in a real memory clinic population.