PhD Scientific Days 2022

Mental Health Sciences I. (Poster discussion will take place in the Aula during the Coffee Break)

Classification with Abstention: the Case of Cognitive Fitness to Drive

Text of the abstract

INTRODUCTION
Cognitive abilities that are crucial for safe driving are often impaired after a stroke. The need for a reliable in-office method for the assessment of driving skills following a stroke is high, and in many cases it could replace the costly and time consuming on-road testing.
LITERATURE REVIEW
Assessment of cognitive fitness-to-drive, as a research field, is struggling with two main methodological issues. The first problem is the analysis of correlations between the cognitive test results and the on-road performance of the patients, instead of the definition of specific cut-off points. Treating fitness-to-drive as a dichotome variable (i.e.: the patient is either fit or unfit to drive) is the second main problem. Dichotomization - while causing a higher proportion of false decisions - is unavoidable when the clinician has to make an immediate decision, but trichotomization should be recommended in cases where on-road testing is available, providing the opportunity of abstention.
CONCLUSION
According to the literature, many of the already existing neuropsychological tools show good predictive properties which could be further enhanced by using novel data analysis approaches. Thus, future clinical studies in the field should be focusing on the application of different statistical methods to replace the common correlation analysis.
STUDY PLAN
The research setup in this planning-stage study is standard: at least one hundred stroke patients are going to take a comprehensive cognitive testing and a standardised on-road test, which is considered the “golden standard” of driving skills. In the poster we are presenting serial trichotomization and the Random Forest Classification model: two alternative ways of data analysis that could help clinical decision making.