Does the psychopathology of the parents predict the developmental-emotional problems of the child in early childhood? A machine-learning study

Introduction Parental psychopathology has been defined in respect of psychopathological development in early childhood. This study aimed to investigate the effects of parental psychopathologies on social and emotional problems in the age range of 1-3 years and to determine children at risk. Methods The study data were obtained from the 2009 Early Childhood Mental Health Profile taking population distribution into consideration with the properties representing Turkey. The primary caregiver of the child completed the Psychiatric Evaluation Form for 1-3 years, the Brief Infant-Toddler Social Emotional Assessment (BITSEA), the Ages and Stages Questionnaire (ASQ), and the Brief Symptom Inventory (BSI) for themselves. Machine learning models used for prediction. The performance of prediction models was evaluated with the ten-fold cross-validation method. Area Under Curve (AUC) values were calculated with Receiver Operating Characteristic (ROC) curves to evaluate the performance of each model. Results The evaluation was made of the data of 2775 children, comprising 1507 (54.3%) males and 1268 (45.7%) females with a mean age of 26.19±9.11 months (range, 10-48 months). A total of 106 children were identified as at risk, as they were above the clinical cut-off point (1.5 standard deviations) of the BITSEA points and below the cut-off points of any one of the developmental areas of the ASQ. Modeling was applied to the data of these 106 children. The Support Vector Machines (SVM) model was selected for prediction with the automatically optimized highest AUC value. Weighting for the SVM algorithm showed mothers' BSI scores, fathers' education and health problems, duration of breastfeeding, unplanned pregnancy are significant for predicting BITSEA-problem scores in the model. Conclusion To be able to understand the complex relationship with parental psychopathology and behavioral problems, machine learning methods were used successfully in this study. Further studies with more massive data sets, more extended follow-up periods, and stronger algorithms will be able to identify risk groups earlier and allow early interventions to be implemented.