Bundled assessment to replace on-road test on driving function in stroke patients: a binary classification model via...

This study proposes to construct a model to replace the on-road test and provide a bundled assessment on the driving function of stroke patients.

Clinical data were collected from 38 stroke patients who specified meeting criteria. Bundled assessment including the Oxford Cognitive Screen (OCS) scale ratings, eye tracking data obtained under the same eight simulated driving tasks as in subject 3, Fugl-Meyer Assessment-lower extremity (FMA-LE) scores, lower limb ankle muscle strength and active range of motion (AROM), and performance on the simulated driving machine. All patients were transported to a driving school and underwent the on-road test. The subject was classified as either Success or Unsuccess group according to whether they had completed the on-road test. A random forest algorithm was then applied to construct a binary classification model based on the data obtained from the two groups.

Compared to the Unsuccess group, the Success group had higher scores on the OCS scale for "crossing out the intact heart" (p = 0.015) and lower scores for "executive function" (p = 0.009). The analysis of eye tracking recordings revealed that the Success group exhibited a reduced pupil change rate, a higher proportion of eye movement types that were fixations, a longer mean fixation duration, and a significantly faster mean average velocity of saccade in the U-turn (p = 0.032), Left-turn (p = 0.015), and Free-driving tasks (p = 0.027). Compared to the Unsuccess group, the Success group had higher FMA-LE scores (p = 0.018), higher manual muscle strength for ankle dorsiflexion (p = 0.024) and plantarflexion (p = 0.040), and greater AROM in dorsiflexion (p = 0.020) and plantarflexion (p = 0.034). The success group demonstrated fewer collisions (p < 0.001), lane violations (p < 0.001), and incorrect maneuvers (p < 0.001) when completing the simulated driving task. The random forest model for bundled assessment demonstrated an accuracy of > 83% based on 56 statistically distinct input data sets.

The bundled assessment, which includes cognitive, eye tracking, motor, and simulated driver performance, offers a potential indicator of whether stroke patients may be able to pass the on-road test. Furthermore, the established random forest classification model has demonstrated efficacy in predicting on-road test outcomes, which is worthy of further clinical application.