Screening for conduct problems: Does the predictive accuracy of conduct disorder symptoms improve with age?

Objectives: To investigate the influence of age on the predictive accuracy of conduct disorder (CD) symptoms for future CD and to evaluate the usefulness of CD symptoms as a screening tool in normal populations. Method: Two cohorts were derived from the Canadian National Longitudinal Survey of Children and Youth based on age at baseline (cohort 1: ages 5-6, n = 881; cohort 2: ages 8-9, n = 859). CD symptoms were assessed with behavior symptom checklists at baseline and the 2-year follow-up assessment. Predictive accuracy (PA) was evaluated with logistic regression and receiver operating characteristic curves. Results: The PA of teacher-identified CD symptoms, alone or in combination with other risk factors (gender, socioeconomic status, familial factors), was higher in the older cohort. PA increased when CD symptoms were combined with other risk factors. However, none of the predictors evaluated achieved minimum criteria for sensitivity and positive predictive value of at least 50%. Conclusions: The PA of CD symptoms improves with age. However, the results show that the PA of CD symptoms assessed at a single point in time in children in normal populations is insufficient. Screening initiatives that use this approach to detect risk for CD should be abandoned.