Rapid cognitive testing predicts real-world driving risk in commercial and medically at-risk drivers

Road safety is a major public and occupational health issue. Safe driving requires numerous cognitive and sensorimotor skills, and past literature suggests that cognitive testing can predict safe or unsafe driving in both healthy and medically at-risk drivers. However, such testing is often time-consuming and inaccessible. In this study, we designed a modified version of the Trail Making Test (TMT) which can be completed on a smartphone in approximately 5 minutes. We recruited 4405 commercially-licensed drivers and 314 medically at-risk drivers to complete the TMT, plus an on-road test of their driving abilities. We then trained and tested a logistic regression model using 50-50 splits on each dataset. The results of the model showed that the longer it took drivers in both groups to complete the TMT, the more likely they were to fail the on-road driving test. Accuracy for the commercial group was 83.8%, with a positive predictive value (PPV) of 34.4% and a negative predictive value (NPV) of 85.3%. Accuracy for the medically at-risk group was 63.1%, with a PPV of 55.8% and an NPV of 65.8%. Overall accuracy was 82.5%, with a PPV of 43.0% and an NPV of 84.3%. Log-transformed reaction time to targets was significantly associated with on-road failure in both driving groups. The results of this study suggest that a rapid and accessible version of the TMT can predict unsafe driving with comparable accuracy to more time-consuming and administratively burdensome means of testing.