Predicting on-road assessment pass and fail outcomes in older drivers with cognitive impairment using a battery of computerized sensory-motor and cognitive tests.

OBJECTIVES To generate a robust model of computerized sensory-motor and cognitive test performance to predict on-road driving assessment outcomes in older persons with diagnosed or suspected cognitive impairment. DESIGN A logistic regression model classified pass–fail outcomes of a blinded on-road driving assessment. Generalizability of the model was tested using leave-one-out cross-validation. SETTING Three specialist clinics in New Zealand. PARTICIPANTS Drivers (n=279; mean age 78.4, 65% male) with diagnosed or suspected dementia, mild cognitive impairment, unspecified cognitive impairment, or memory problems referred for a medical driving assessment. MEASUREMENTS A computerized battery of sensory-motor and cognitive tests and an on-road medical driving assessment. RESULTS One hundred fifty-five participants (55.5%) received an on-road fail score. Binary logistic regression correctly classified 75.6% of the sample into on-road pass and fail groups. The cross-validation indicated accuracy of the model of 72.0% with sensitivity for detecting on-road fails of 73.5%, specificity of 70.2%, positive predictive value of 75.5%, and negative predictive value of 68%. CONCLUSION The off-road assessment prediction model resulted in a substantial number of people who were assessed as likely to fail despite passing an on-road assessment and vice versa. Thus, despite a large multicenter sample, the use of off-road tests previously found to be useful in other older populations, and a carefully constructed and tested prediction model, off-road measures have yet to be found that are sufficiently accurate to allow acceptable determination of on-road driving safety of cognitively impaired older drivers.