This article is more than 5 years old. Information may no longer be current.

Disease State Index may predict late-life dementia

Recent findings suggest that the Disease State Index, a validated supervised machine learning method, may be useful for prediction of late-life dementia.

“The results of our study are very promising, as it is the first time this machine learning approach was used for estimating dementia risk in a cognitively normal general population,” study researcher Alina Solomon, MD, PhD, of the University of Eastern Finland, said in a press release. “The risk index was designed to support clinical decision making, and we are very keen on exploring its potential practical use. However, we still need to validate this risk index in other populations outside Finland. We also need to investigate if it works in people older than 80 years, and if it can monitor changes in dementia risk over time, for example as a response to lifestyle interventions. These are some of the next steps we are planning now.”

To assess the Disease State Index (DSI) as a late-life dementia prediction model, researchers evaluated study participants from the CAIDE study, a longitudinal population-based study that assessed participants at multiple times in midlife and late-life. Mean age was 50.6 years at midlife, 71.3 years at first late-life examination and 78.6 years at second re-examination. The main study cohort included 709 cognitively normal individuals, of whom 39 had incident dementia at second re-examination. An extended population (n = 1,009) included non-participants and non-survivors, of whom 151 had incident dementia.

Composite DSI had an area under the curve of 0.79 (95% CI, 0.79-0.8) in the main study cohort and 0.75 (95% CI, 0.74-0.75) in the extended cohort.

Areas under the curve for individual factors and factor groups were similarly lower for the extended and main cohort.

“Large health information databases contain a lot of valuable information which is still partly hidden and under-exploited. Modern machine learning methods can be used to extract patterns of data that may be difficult to observe just by looking at the data by eye,” study researcher Jyrki Lötjönen, PhD, chief scientific officer at Combinostics Ltd. Tampere, Finland, said in the release. “Our objective has been to detect patterns that predict whether a person is more likely to get dementia in the future. Another area of interest has been how to present all these complex data in a simple form to make these modern technologies useful for clinicians and general public interested in dementia prevention.” – by Amanda Oldt

Disclosure: Please see the study for a full list of relevant financial disclosures.