'Machine Learning' Tool May Predict Dementia Development Up to 10 Years Later

Deborah Brauser

December 21, 2016

A novel late-life dementia-prediction model, known as the Disease State Index (DSI), may help identify risk for the disorder up to a decade later, new research suggests.

The DSI is a validated and supervised risk assessment tool that uses "machine learning" methods to analyze large amounts of health information.

"We were very happy to see how well this tool performed in predicting dementia," senior author Alina Solomon, MD, PhD, Department of Neurology at the University of Eastern Finland in Kuopio, told Medscape Medical News.

Dr Alina Solomon

"There are other available tools. But what we saw with the one we created was in the upper range of their performance results. So it did very well."

She stressed, however, that the DSI shouldn't be used to diagnose dementia but to help in making decisions and developing strategies for dementia prevention.

The findings were published online December 6 in the Journal of Alzheimer's Disease.

Big-Data Artificial Intelligence

The investigators explained in a release that machine learning is the process of using several methods to teach computers to make improved predictions. This type of big-data artificial intelligence is "just starting to be used in the context of dementia prevention," they add.

"Compared to previously used methods for developing dementia risk scores, the main strengths of DSI are its ability to deal with larger amounts of heterogeneous data, to handle missing data well, and to use unprocessed data," write the researchers, adding that the DSI hasn't been investigated before in a cognitively unimpaired general population.

Interestingly, DSI is partnered with the Disease State Fingerprint method to present findings "in an easily and quickly interpretable visual form."

"We have worked before with more simple predictor tools, such as a pen-and-paper test. But we wanted a tool that could handle more complex data," said Dr Solomon.

The original Cardiovascular Risk Factors, Aging, and Dementia (CAIDE) study included 1449 participants aged 65 to 79 years and was conducted in Eastern Finland. A wide assortment of measures were conducted at both follow-up visits, including MRI, blood tests, and neuropsychological assessments.

The investigators included 709 participants from the CAIDE study who were classified as cognitively normal at first study visit and who completed a second re-examination up to 10 years later.

Validated Results

For the current analysis, the first late-life follow-up was considered to be "baseline." Among the participants who had no dementia or mild cognitive impairment at this baseline, 39 had incident dementia at the second re-examination.

The composite DSI, which included cognitive and vascular factors, subjective memory status, and APOE genotype, showed a significant area under the curve (AUC) of 0.79 — with a very narrow 95% confidence interval (CI; 0.79 - 0.80).

The investigators also examined general population records for 1009 individuals from the Finnish Hospital Discharge Register, the Drug Reimbursement Register, and the Causes of Death Register.

Among these participants, 151 had dementia after a mean follow-up of 9.0 years. The composite DSI AUC for this extended population was also significant (0.75; 95% CI, 0.74 - 0.75).

"Results were validated by comparison with a [support vector machine] classification, trained with a linear kernel using the same set of factors and cross-validation procedure," report the researchers.

Table. Other Significant Factors in the DSI

Variables CAIDE Study Population: AUC (95% CI) Extended Study Population: AUC (95% CI)
Basic model    
  Cognition 0.73 (0.73 - 0.74) 0.69 (0.69 - 0.70)
  Sociodemographics 0.67 (0.65 - 0.68) 0.66 (0.66 - 0.67)
  Vascular factors 0.65 (0.64 - 0.66) 0.53 (0.52 - 0.53)
  Subjective memory 0.64 (0.63 - 0.66) 0.58 (0.57 - 0.58)
  APOE genotype 0.59 (0.58 - 0.60) 0.60 (0.59 - 0.61)
Basic model + vascular factor changes    
  Composite DSI 0.80 (0.79 - 0.81) 0.78 (0.77 - 0.79)
  Vascular factors (such as body mass index change) 0.68 (0.66 - 0.69) 0.65 (0.64 - 0.66)


"The detailed and visually easy to interpret individual risk profiles" could possibly help in the design of future dementia prevention interventions, write the investigators.

In the release, Dr Solomon noted that the DSI was created to support clinical decision making, "and we are very keen on exploring its potential practical use."

"However, we still need to validate it in other non-Finnish populations and see if it works in other age groups or if it can monitor risk level over time," she added to Medscape Medical News.

"This is just one tool in the prediction toolbox. I think we need a range of tools — from the simple to the more complex ones," said Dr Solomon. "And I think we need to figure out how to link the prediction with the prevention so we can more likely identify people who are most likely to benefit from an intervention."

The study was funded by the European Union (EU) 7th Framework Program, the MIND-AD Academy of Finland and Swedish Research Council through the EU Joint Program-Neurodegenerative Disease Research, the Center for Innovative Medicine in Sweden, Alzheimerfonden Sweden, the AXA Research Fund, and grants from the Academy of Finland. Dr Solomon has disclosed no relevant financial relationships. Disclosures for the coauthors are posted in supplemental material on the journal's website.

J Alzheimers Dis. Published online December 6, 2016. Full article

Follow Deborah Brauser on Twitter: @MedscapeDeb. For more Medscape Neurology news, join us on Facebook and Twitter


Comments on Medscape are moderated and should be professional in tone and on topic. You must declare any conflicts of interest related to your comments and responses. Please see our Commenting Guide for further information. We reserve the right to remove posts at our sole discretion.