AI casts wide net to predict functional decline in the elderly

Psychology researchers have used machine learning to wring useful two-year dementia trajectory predictions from more than 500 potentially contributing risk factors.

Grouping the factors into five categories—demographics, neurocognitive performance, structural brain MRI, nuclear neuroimaging and lab results—the team pursued its broad sweep of variables to correct for a common concern in the field:

Selecting some risk factors while ignoring others can lead to imprecise prognoses and suboptimal treatment recommendations.

Such shortcomings could generate missed opportunities to curtail functional decline in the elderly, hastening the day dementia sufferers must rely on others to feed, dress and bathe themselves.

The research was carried out at The Ohio State University and is described in Brain Communications.

Lead author Kate Valerio, PhD, senior author Jasmeet Hayes, PhD, and colleagues analyzed data on just under 400 participants in a national dataset from the NIH-backed Alzheimer’s Disease Neuroimaging Initiative.

They used classification algorithms to gauge each of the five categories of variables on predictive power.

Neurocognitive test batteries proved the single best performer at predicting functional decline, achieving almost 75% accuracy.

Nuclear neuroimaging was a close second at 71%.

Meanwhile three variables taken together explained almost a third of the differences seen in degree of functional decline.

These were a daily checkup called Everyday Cognition, an Alzheimer’s disease assessment scale and a particular type of brain activity uncoverable with nuclear neuroimaging.

Valerio et al. comment:  

The machine learning approach applied here improves upon previous work examining the ability of neuroimaging, neurocognitive, demographic and genetic/fluid-based biomarkers to predict functional decline two years after baseline assessment. Neurocognitive measures showed the highest accuracy and best discriminative ability, suggesting that a set of inexpensive and non-invasive cognitive assessments can be used to predict independent functioning in lieu of expensive and more invasive measures.”

Around the web

The Palo Alto giant used exams from nearly 250,000 patients to upgrade its already robust algorithm.

Exams performed using the deep learning-based reconstruction tool also maintained high image quality, experts reported Wednesday.

Stratifying exams according to risk can reduce unnecessary imaging and downstream costs of care, Hawaiian researchers reported in Radiology.

Trimed Popup
Trimed Popup