Sample selection may bias the outcome of an adolescent mental health survey: results from a five-year follow-up of 4171 adolescents.