Abstract
Objective. We set up this study to determine the predictive value of approaches for which a statistical association with driving performance has been documented.
Methods. We determined the statistical association (magnitude of association and probability of occurrence by chance alone) between four different predictors (the Mini-Mental State Examination, Trails A test, Useful Field of View [UFOV], and a composite measure of past driving incidents) and driving performance. We then explored the predictive value of these measures with receiver operating characteristic (ROC) curves and various cutoff values.
Results. We identified associations between the predictors and driving performance well beyond the play of chance (p < .01). Nonetheless, the predictors had limited predictive value with areas under the curve ranging from .51 to .82.
Conclusions. Statistical associations are not sufficient to infer adequate predictive value, especially when crucial decisions such as whether one can continue driving are at stake. The predictors we examined have limited predictive value if used as stand-alone screening tests.
ACKNOWLEDGMENTS
Funding for this research was provided through research grants from AUTO21, Network of Centres of Excellence, the Ontario Neurotrauma Foundation, the Manitoba Health Research Council, and the Thunder Bay Foundation. Michel Bédard and Michelle M. Porter are members of CanDRIVE, a New Emerging Team funded by the Canadian Institutes of Health Research, Institute of Aging. Michel Bédard is also a Canada Research Chair in Aging and Health (www.chairs.gc.ca); he acknowledges the support of both programs. Portions of this work were presented to the 60th Annual Scientific Meeting of the Gerontological Society of America (November 2007) and at the International Conference on Aging, Disability and Independence (ICADI; February 2008).
Notes
∗For Trails A, the ORs are for a 10-s increment. For the UFOV, the OR is for a 50-ms increment.
∗For the MMSE higher scores are better; for the Trails A, the UFOV, and the driving record it is the reverse. Given that Trails A data are in seconds and UFOV data are in milliseconds not all cutoffs are presented to simplify the presentation.
1We suggest that the proportion would be at least 38% because with prevalence of failure at a more realistic level within the general population, PPV would be lower, and 1 – PPV would be higher.