| Dementia and cognitive disorders  

Estimating Dementia Risk Using Multifactorial Prediction Models

A challenge in preventing dementia lies in the ability to predict who may develop a neurodegenerative disease in advance of symptom presentation.

Several risk assessment tools exist to predict future incidence of dementia, however their clinical validity is not clear. Kivimaki and colleagues investigated the predictive accuracy of four assessment scores in a prospective study of 465,929 people enrolled in the UK biobank cohort.

Scores used were:

  1. Cardiovascular risk factors, aging, and dementia (CAIDE)
  2. The CAIDE-APOE-supplement
  3. Brief dementia screening indicatory (BDSI)
  4. Australian national university Alzheimer disease risk index (ANU-ADRI)

The primary endpoint of interest was incident dementia at ten years from baseline assessment. Concordance, or ‘C’, statistics were used to assess predictive accuracy. These measure the probability that a randomly selected person with dementia at ten years was more likely to have a high risk assessment score at baseline. Authors also calculated per score detection and false positive rates. Age was used as a comparator variable across analyses.

3,421 People were diagnosed with dementia at 10 year follow up (0.7%). C-statistics were between 0.59-0.73 for the four questionnaires, and 0.79 for the model only containing age as a predictive variable (highly predictive C-scores usually > 0.8). When the cut-offs were calibrated for each questionnaire to achieve a 5% false positive rate, they could only detect 9-16% of incident dementia. On the other hand, when they were calibrated to detect >50% of incident dementia, the ratio of true: false positive exceeded 1:66, indicating a high error rate.   

Key points

  1. Dementia prevention is contingent on accurate and robust predictive tools
  2. Kivimaki showed in 465,929 UK biobank participants that four existing dementia risk assessment tools demonstrate lower predictive accuracy, as indexed by c-statistics, when predicting 10 year incident dementia. Age alone was a superior predictive measure than risk assessment tools.
  3. These data support the clear need for more accurate tools to improve disease prediction. These may include digital, neuroimaging, and AI guided clinical data extraction methods.

References
Kivimäki M, Livingston G, Singh-Manoux A, et al. Estimating Dementia Risk Using Multifactorial Prediction Models. JAMA Netw Open. 2023;6(6):e2318132. doi:10.1001/jamanetworkopen.2023.18132

Publish on behalf of the Scientific Panel Dementia and Cognitive Disorders

Back