A novel clinical risk score predicted the chance of future dementia with high accuracy, but raised questions from dementia experts.
Using data from nearly 450,000 people in the UK Biobank cohort, Xi-jian Dai, PhD, of The Second Affiliated Hospital of Nanchang University in China, and colleagues developed a practical tool to predict individual risk of developing dementia.
Age, low education, sleep patterns, and comorbidities like diabetes and cerebrovascular, cardiovascular, and respiratory disease were key variables that predicted dementia risk. While men and women shared some modifiable risk and protective factors, independent variables accounted for 31.7% of men and 53.35% of women developing dementia, Dai and co-authors reported in JAMA Network Open.
Using baseline Biobank measurements, the researchers assigned points for dementia predictors in both men and women. For example, a man with a history of diabetes (1 point) had a lower dementia risk than a woman with a history of diabetes (2 points) if they had the same background.
Total points in the risk score model ranged from -18 to 30 in men and -17 to 30 in women.
For a 70-year-old (10 points) man who was underweight (3 points) with a low educational level (1 point) and a history of diabetes (1 point) and cerebrovascular disease (5 points), a sum of 20 points , the tool predicted a corresponding risk of dementia of 9% at 5 years, 31% at 9 years, and 54% at 13 years.
In people with a score of 30 points, the model had 97.59% predictive accuracy for 9-year dementia risk in men and 99.59% in women, and an almost 100% predictive accuracy for 13-year dementia risk in both sexes. It was unclear what percentage of participants had 30 points.
In the UK, some experts viewed the findings with skepticism. “The claim that ‘the risk score model yielded nearly 100% prediction accuracy of 13-year dementia risk’ is extremely misleading,” David Curtis, MD, PhD, of University College London (UCL) in England, posted on the Science Media Centre website. “The score does not accurately predict whether or not one will develop dementia in 13 years, rather it provides the probability that somebody will develop dementia,” he pointed out.
“This would be like me claiming that I can predict the risk of getting heads when I toss a coin with 100% accuracy — the risk of getting a head is 0.5,” Curtis continued. “The score is a poor predictor of whether somebody will get dementia or not; it only predicts their chances of getting dementia.”
But others, like Ivan Koychev, PhD, MRCPsych, of the University of Oxford, thought the research had merit. The paper has sound methodology and the benefit of working with one of the largest datasets currently available, Koychev observed.
“The authors have accounted for confounders within the dataset,” he noted. “The limitations are that UK Biobank participants are not fully representative of the general population and some of the measures, for example sleep, rely on participants’ self-report instead of objective measures of sleep quality that are available within the dataset.”
To build their prediction model, Dai and colleagues looked at data from 444,695 participants in the ongoing UK Biobank study who were dementia-free at baseline. The researchers grouped participants into training and testing data sets to perform internal validation.
A total of 205,187 participants were men with a mean baseline age of about 57, and 239,508 were women with a mean baseline age of about 56.
Dementia occurrence over 13 years was 0.7% for men and 0.5% for women. The dementia group included 1,473 men and 1,261 women with a mean baseline age of nearly 65.
The C statistic of the final model was 0.86 for men and 0.85 for women in the training data set, and 0.85 for men and 0.87 for women in the testing data set.
A limitation is that the researchers did not provide false positive and detection rates, noted Mika Kivimaki, PhD, also of UCL. “These are particularly important metrics when evaluating the prediction of an incurable and feared illness by individuals, such as dementia,” he wrote.
“A false positive test result in dementia risk assessment can elicit psychological distress for many of the affected individuals,” Kivimaki stated. “Receiving a false negative result, in turn, may discourage a person from taking up preventive measures.”
This study was supported by the Guangdong Basic and Applied Basic Research Foundation, Natural Science Foundation of Hunan Province, and University of Macau.
Dai and co-authors reported no conflicts of interest.
Kivimaki has an academic interest in risk prediction but no commercial interests. Koychev is a medical advisor to Five Lives, a digital technology company involved in dementia risk prediction. Curtis reported no conflicts of interest.