As the population of the US ages, the number of older drivers is increasing. In 2020, there were 54 million adults aged 65 years and older, and this number is projected to rise to 84 million in 2050, accounting for 20% of the total population.
Driving is essential for many older adults as it allows them to maintain their independence, self-control, social connections, and life satisfaction. However, age-related declines in cognitive functions and health, medical conditions, and side effects of medication can increase the risk of crashes, with older drivers having higher mileage-based crash rates than most other age groups. Stopping driving can also have adverse health outcomes for older adults. Therefore, it is essential to detect early indications of mild cognitive impairment (MCI) and dementia in older drivers.
Multiple recent studies have suggested that certain unusual changes in driving behaviour may serve as early indications of MCI and dementia. However, these studies have been subject to limitations due to their small sample sizes and short follow-up periods.
To address these limitations, the Longitudinal Research on Aging Drivers (LongROAD) project collected naturalistic driving data from 2,977 cognitively healthy participants over a period of up to 44 months. The driving trajectories were processed and aggregated into 31 time-series driving variables. A classification method using a statistic called Influence Score (I-score) was employed to predict MCI and dementia based on the naturalistic driving data. The findings were published in the journal Artificial Intelligence in Medicine.
The I-score is a statistical measure that can be used to evaluate a variable’s predictive ability, particularly in large datasets. It has been shown to effectively differentiate between predictive and “noisy” variables, as well as to identify influential variable modules or groups that capture compound interactions among explanatory variables.
Studies have demonstrated that I-score can improve the performance of classifiers over imbalanced datasets by virtue of its association with the F1 score, which is defined as the harmonic mean of sensitivity and specificity.
Using the predictive variables selected by I-score, researchers constructed an interaction-based residual block classifier and employed ensemble learning to aggregate these predictors in order to boost the overall classifier’s prediction ability.
Experiments were conducted using naturalistic driving data to evaluate the proposed classification method. Results showed that the method achieved an accuracy of 96% in predicting MCI and dementia, which was the highest accuracy among the tested methods.
The study also conducted a feature importance analysis and found that the right-to-left turn ratio and the number of hard-braking events are the most important driving variables in predicting MCI and dementia. Furthermore, region and age are the two most important demographic features, demonstrating how geographic and age variance can affect driving behaviour.
Early detection of MCI and dementia is crucial for older drivers, and this study has clinical and practical implications. It could lead to timely evaluation, diagnosis, and interventions, and promote earlier implementation of supportive or assistive services. However, the study has limitations, such as combining participants with different disease levels into one class and insufficient data for training a robust classifier. The study could be improved by including more driving data from different types of drivers to enhance the predictive power. The work will be extended to capture the longitudinal information of each participant and develop a personalised time-dependent classifier to predict the risk of MCI/dementia for each individual as time progresses.
The proposed classification method using I-score and interaction-based residual block classifiers has shown promising results in predicting MCI and dementia in older drivers. The findings of this study provide valuable insights into the relationship between demographics, driving behaviour, and the risk of MCI and dementia. Future research can build on these findings to develop personalised and effective interventions that can improve the quality of life for older drivers.