Using machine learning to identify parenting features prospectively related to callous-unemotional traits from infancy to early adolescence

Comprehensive Summary

This study by Paz et al. uses machine learning to identify which specific parenting features assessed across infancy and early childhood (ages 6-90 months) are most predictive of callous-unemotional (CU) traits and conduct problems in early adolescence. To do so, the researchers collected data from the Family Life Project (N = 1,292; 49% female, 41% Black, 28% below poverty line) with 74 parenting predictors assessed at eight time points between ages 6-90 months using parent-reported questionnaires, observer ratings of videotaped interactions, and home visit assessments. CU traits and conduct problems were measured via parent questionnaires in preadolescence. The researchers used recursive feature elimination with random forest machine learning algorithms, splitting the data into training and testing samples, with five-fold cross-validation repeated to identify which parenting features were most important in predicting outcomes. The parenting features explained 8.2% of CU traits variability in preadolescence, with the top predictors including early sensitive parenting and later behavior management and scaffolding practices. Most top predictors for CU traits (71%) were classified as parental emotional sensitivity, with 79% derived from behavioral coding and 64% assessed in the first 3 years of life. Parents valuing teaching about emotions at age 2 years was found to be an influential predictor. For conduct problems, prediction was weaker, with parenting explaining only 4.5% of variability. The conduct problems model showed greater variety, with top predictors more evenly distributed across emotional sensitivity (36%), behavior management (36%), scaffolding (21%), and parental mastery (7%), and these predictors were measured closer in time to preadolescence rather than in early infancy. The results support targeting parental sensitivity and behavior management, with analytic work on parenting interventions and recently adapted treatments for CU traits that emphasize good relational interactions.

Outcomes and Implications

Conduct problems can increase lifetime risk for psychiatric conditions including antisocial personality disorder, major depression, and substance use disorders, while CU traits predict even greater risk for severe conduct problems, violence, and adult psychopathy. Recent evidence demonstrates parenting represents a true environmental risk factor even when accounting for heritable factors, making it a critical target for interventions. However, questions remain about which specific aspects of parenting assessed across different developmental periods are most relevant to CU traits, knowledge that can identify the most effective parenting targets to reduce CU traits. This research is clinically relevant as it provides objective, data-driven evidence for which specific parenting practices should be targeted in interventions for CU traits and conduct problems across different developmental stages. The findings support incorporating both early parental emotional sensitivity (warmth, attunement, responsivity) and behavior management strategies into preventative interventions, which aligns with gold-standard treatments for conduct problems and recently adapted treatments emphasizing positive relational interactions for children with CU traits. The results underscore the need for universal screening methods to identify and support parents experiencing specific difficulties with early parental emotional sensitivity and behavior management. The authors do acknowledge several limitations requiring future research such as larger sample sizes for machine learning analyses, fully balanced longitudinal designs with repeated assessments of the same constructs to determine optimal timing of interventions, and prospective validation studies to confirm the identified parenting targets translate to improved intervention outcomes.