Graded dosages of alcohol (.01-.12 BAC) were regressed against a ten test computerised battery administered to 40 subjects in order to examine sensitivity and specificity. Multiple regression analyses comparing performance decrements to alcohol dosage were statistically predictive (R's > 0.60; p's <.001). Multiple cut-off analyses, using proportion of tests passed and size of the cut-off (in percent), permitted specificity (percentage correctly identified as being fit out of the total that are fit) above 97%; sensitivity (percentage identified as being unfit, out of the total who are unfit) was >80% for high dosages of alcohol; 60+% for low. We believe that these data support the feasibility of a performance-based test battery that is a simple and economic measure of alcohol impairment. Like fitness-for-duty metrics, automobile interlock devices should carefully weigh sensitivity and specificity as a source of error. Although both criteria should be as high as possible, in practical applications we believe specificity is the more important consideration. In the workplace, if multiple testings are possible, we believe 50-60% sensitivity with 97%+ specificity should be a design goal for fitness-for-duty applications and close to that (70-80% sensitivity, 90%+ specificity) for an automobile interlock device if used for habitual offenders.
Abstract