A small study of adolescents with attention-deficit/hyperactivity disorder (ADHD) found clear differences in school experiences, family circumstances, peer relationships and treatment history between those with and without conduct disorder.
The study included 138 young people treated at a child and adolescent psychiatry clinic at Erciyes University in Kayseri, Turkey, between August 2024 and February 2025. Of the participants, 100 had ADHD alone and 38 had both ADHD and conduct disorder.
Researchers used machine-learning models to examine which of 99 collected variables best distinguished the two groups. They described the results as preliminary and stressed that the model was not a diagnostic or forecasting tool.
More discipline and absences in the combined-diagnosis group
All participants were between 11 and 18 years old, and a child psychiatrist confirmed diagnoses using DSM-5 criteria. The groups were similar in age. Boys made up 86.8% of the group with both diagnoses and 72% of the ADHD-only group, a difference that did not reach statistical significance.
School-related differences were among the clearest findings. Disciplinary punishment was reported for 60.5% of teens with ADHD and conduct disorder, compared with 17% of those with ADHD alone. Frequent absences occurred in 47.3% of the combined-diagnosis group and 12% of the ADHD-only group.
Conduct disorder refers to a lasting pattern of serious behavior problems. These can include aggression toward people or animals, destruction of property, repeated lying or stealing, and major rule-breaking such as skipping school or running away.
Family, peer and treatment factors ranked highly
Family criminal involvement was reported for 15.7% of teens in the conduct-disorder group, compared with 1% in the ADHD-only group. No mother in the conduct-disorder group had a university education, while 34% of mothers in the ADHD-only group did.
Lower maternal education, family criminal history, physical abuse, school punishment, frequent absences, and friendships involving smoking or troublemaking ranked highly in the researchers’ analyses. Breaks from ADHD medication and receiving an ADHD diagnosis at age 12 or later also ranked highly.
Infant sleep problems and a history of speech delay ranked highly in two of the three variable-ranking methods. The researchers collected information from family backgrounds, school histories, ADHD treatment records, peer behavior and questionnaires completed by teens and parents.
What the machine-learning models showed
The research team tested six types of machine-learning models using different sets of the top-ranked variables: the top 10, 20, 30 or 50 variables, as well as all 99. This produced 30 model combinations, which were evaluated with five-fold cross-validation.
A support vector machine using the top 20 variables had the strongest combination of accuracy and F1-score among the 30 combinations. It classified 87% of participants correctly, with a 95% confidence interval of 81% to 92%, and had an AUC of 0.90.
That model had 98% specificity, correctly leaving 98 of 100 teens with ADHD alone unflagged. Its sensitivity was 58%: it identified 22 of the 38 teens with conduct disorder and missed 16.
Results varied depending on the model. A top-30-variable model had the highest AUC, at 0.92, but identified fewer than 40% of cases. A model using all 99 variables identified only three of the 38 conduct-disorder cases, corresponding to a sensitivity of 7.9%.
Why the findings do not establish causes
The study captured participants at one point in time. As a result, it cannot show whether school punishment, absences or peer problems contributed to conduct disorder, or whether they were already part of the behavior pattern associated with the diagnosis.
The same applies to treatment gaps. Breaks from ADHD medication and later diagnosis were more common among teens with conduct disorder, but the study cannot show that steadier treatment would have prevented the condition. Those patterns may instead reflect a more difficult clinical course from the beginning.
The researchers suggest that school difficulties, missed classes, treatment gaps, later ADHD diagnosis and peer relationships could help guide closer follow-up and future programs involving families and schools. The current model, however, classified teens who already had a diagnosis; it cannot identify which children with ADHD will later develop conduct disorder.
Important limits on applying the results
The study’s conduct-disorder group included only 38 teens. The model was not tested in a new group or at another clinic. In addition, the researchers selected highly ranked variables using all participants before cross-validation, which can make performance appear better than it is.
The team also compared 30 model setups using the same results without a rule set in advance for choosing the best model. Teens with anxiety, depression or other diagnoses were excluded, even though such conditions often occur alongside ADHD.
A 2025 meta-analysis cited by the researchers estimated that 10.7% of 39,894 young people with ADHD across 121 studies also had conduct disorder, with higher rates in boys. That broader estimate provides context, but it does not remove the limitations of this single-clinic study.
AI tools were used to assist with the preparation of this article under our editorial standards.
Leave a Reply