3271 - THE ROLE OF PERINATAL PSYCHOLOGICAL AND BEHAVIORAL FACTORS IN PSYCHOPATHOLOGY RISK PATHWAYS: A MACHINE LEARNING APPROACH

Session: 3269 - PERINATAL MENTAL HEALTH IN MOTHERS AND FATHERS: DEVELOPMENTAL, CLINICAL, AND CROSS-CULTURAL PERSPECTIVES
AUTHORS:
Antonucci Linda Antonella (Department of Translational Biomedicine and Neuroscience, University of Bari Aldo Moro ~ Bari ~ Italy) , Rollo Simone (Department of Translational Biomedicine and Neuroscience, University of Bari Aldo Moro ~ Bari ~ Italy) , Fanizza Antonella (Department of Translational Biomedicine and Neuroscience, University of Bari Aldo Moro ~ Bari ~ Italy) , Franciosa Giuseppe (Department of Translational Biomedicine and Neuroscience, University of Bari Aldo Moro ~ Bari ~ Italy) , Dispoto Eleonora (Department of Translational Biomedicine and Neuroscience, University of Bari Aldo Moro ~ Bari ~ Italy) , Asselti Martina Grazia (Department of Translational Biomedicine and Neuroscience, University of Bari Aldo Moro ~ Bari ~ Italy) , Pennacchio Teresa Claudia (Department of Translational Biomedicine and Neuroscience, University of Bari Aldo Moro ~ Bari ~ Italy) , Pergola Giulio (Department of Translational Biomedicine and Neuroscience, University of Bari Aldo Moro ~ Bari ~ Italy) , Infurna Maria Rita (Department of Psychology, Educational Science and Human Movement, University of Palermo ~ Palermo ~ Italy)
Abstract text:
Introduction: The perinatal period is characterized by burdening physical and psychological changes which may lead to psychopathology. Many psychosocial factors modulate the coping strategies that pregnant women might rely on during pregnancy. Amongst those, reflective function (RF) seems to act as a protective factor in pregnant women. We aimed (i) at training and validating a machine learning (ML) algorithm discriminating higher vs. lower perinatal depressive symptoms (PDS), (ii) at investigating its prognostic relevance by verifying associations between ML decisions and post-partum depression, and (iii) at investigating the mediating role of RF in the relationship between ML decisions and PDS.
Method: We recruited 851 pregnant women, which were split in a Discovery Sample (DS, N=509 form Palermo), and three Validation Samples (VS, N=59 from Bari, N=228 from Lecce, N=55 from Messina). In 41 women, post-partum depression was assessed 6 months after delivery. We trained in DS and validated in VS a ML algorithm discriminating higher vs. lower PDS based on item-wise information on attachment style, stressful life events, social support, childhood trauma, and coping strategies. Prognostic relevance of the algorithm to post-partum depression was investigated via ANOVA, while moderation analysis testing the role of RF in the relationship between ML decisions and PDS via R.
Results: The algorithm discriminated higher vs.lower PDS with 71.8% accuracy in DS, and with accuracies ranging between 69.3% and 81.6% in VS. Higher ML decision scores were associated with subsequent post-partum depression (p=0.001). RF significantly mediated the association between ML decisions and PDS.
Conclusion: We identified specific psychosocial impacting PDS levels, which might therefore specifically be targeted in early intervention programs. Our mediation findings suggest that such findings should involve personalized actions aimed at boosting RF to make pregnant women able to better cope with the challenges of the perinatal period and prevent post-partum psychopathology.