From the 37 randomized participants, 17 had an ODD diagnosis (SCL-BF = 10; TAU = 7), 9 ODD/CD (SCL-BF = 4; TAU = 5), and 3 CD (SCL-BF = 2; TAU = 1) alone and 8 (SCL-BF = 2; TAU = 6), presented a T-score > 70 on the aggressive behavior and/or rule-breaking behavior subscale. Furthermore, 6 participants also had comorbid ADHD (SCL-BF = 3; TAU = 3). Baseline characteristics did not differ between groups, except for higher RPQ scores in the BF group. Details are depicted in Table 1.

Two participants were excluded due to missing baseline data. In total, 35 mITT participants were analyzed. RM-ANOVA of the MOAS questionnaire showed a significant effect of time [F(1,33) = 6.57, p = 0.015] with a small-effect size (ES = 0.27, [CI 95% = 0.41–0.498]), and irrespective of group (p = 0.208). This result did not change when only completers were analyzed (ES = 0.38, [CI 95% = 0.046–0.715], p = 0.024). Sensitivity analyses, including IQ and age as covariates, yielded the same results as the main analyses but revealed also that participants with lower IQ (p = 0.008) and younger age (p = 0.049) improved more after treatment, irrespective of group. Additionally, correlation analysis revealed that IQ (r = − 0.428, p = 0.011), but not age (r =− 0.149, p = 0.25) correlated significantly with clinical change. Exploratory within-group analysis between pre- and post-assessments were significant for the SCL-BF group only (ES = 0.36, [CI 95% = 0.036–0.689], p = 0.020). For details, Table 2.

For the CBCL, we found lower externalizing symptoms after treatment in both groups [F(1,22) = 11.699, p = 0.002] with a large-effect size (ES = 0.83, [CI 95% = 0.251–1.41]). Regarding the subdomains of the CBCL, which reflect the core domains of aggression-related symptoms, medium-to-large improvements were obtained for the ODD subscale [F(1,22) = 7.822, p = 0.011, ES = 0.81, CI 95% = 0.168–1.444] and for the CD subscale [F(1,22) = 8.151, p = 0.009, ES = 0.63, CI 95% = 0.118–1.138]. However, again, no significant group differences were found. In an exploratory within-group analysis, pre–post-differences showed medium-to-large ES and were only significant in the BF group, and not in the TAU group.

No significant changes were found in CU traits and RPQ total score and its subscales (all p > 0.152). All treatment and time effects are depicted in Table 2.

Mean effect of session was not significant (p = 0.199). However, a significant session x condition interaction emerged (p = 0.046), which revealed an increase in performance over time for the up-regulation condition across runs. Exploratory post hoc between-session comparisons revealed significant improvement between the first session and the eleventh (p = 0.0107), thirteenth (p = 0.0168), fourteenth (p = 0.0225), and sixteenth (p = 0.0442) session for the up-regulation condition. In addition, self-regulation for the up-regulation condition had lower mean percent of correct regulation, indicating that it was more difficult to carry out (up- vs down-regulation, p < 0.001). With regard to the different runs, the transfer run proved to be most difficult (p = 0.003).

Additionally, an interaction between session and run emerged, revealing that improvement over time was higher in the video run (p = 0.046). For details, see Fig. 3 and Table 3. For individual performance over time and a secondary analysis using the offline preprocessed skin-conductance data, which revealed a session x condition interaction at a trend level only (p = 0.051). Furthermore, we assessed if medication affected the SCL-BF learning but found no impact on the main model. See supplementary material for more details.

SCL-BF Performance. A SCL-BF performance across sessions and runs. B Mean performance for each run. ***p < 0.001

As expected, the learning of self-regulation during the video condition and for the mean across all conditions was related to clinical improvement. Lower externalizing symptoms (mean self-regulation: rs = − 0.621, p = 0.041, video condition: rs = − 0.726, p = 0.011), ODD (mean self-regulation: rs = − 0.761, p = 0.007, video condition: rs = − 0.852, p = 0.001), ICU (mean self-regulation: rs = − 0.621, p = 0.041, video condition: rs = − 0.697, p = 0.017), and CD (mean self-regulation: rs = − 0.696, p = 0.017, video condition: rs = − 0.682, p = 0.021), but were unrelated to the primary outcome. For details, Fig. 4. We additionally correlated clinical outcome with the number of attended sessions for the completers, which however, were not significant (all p > 0.160) Fig. 5.

Clinical outcome and SCL self-regulation improvement. Negative signs indicates more clinical improvement (Post–Pre), and positive slope better SCL performance. A ICU: Inventory of callous-unemotional traits. B CD: Conduct disorder. C ODD: Oppositional defiant disorder. D Externalizing symptoms (CBCL)

Correlation matrix for SCL Self-regulation and clinical outcome. Correlation Matrix for SCL Self-regulation (slope of the SCL-BF sessions over time) and clinical improvement. Line: Slope of line feedback; Video: Slope of video feedback, Average slope = Mean slope of all runs. Questionnaire data based on Post–Pre differences. RPQ: Reactive–proactive questionnaire; CD: CBCL conduct disorder scale; ODD: CBCL oppositional defiant disorder scale; Externalizing: CBCL Externalizing scale; ICU: Inventory of callous-unemotional traits