Poster #52 - Developing an individualized risk calculator for psychopathology among young adults victimized during childhood

• In Event: 1-115 - Poster Session 03
  In Poster Session: PS 03 Section - Developmental Psychopathology

Integrative Statement

Background: Childhood victimization – including physical abuse/neglect, sexual abuse, emotional abuse/neglect, exposure to domestic violence, and bullying victimization – has been associated with elevated risk for psychopathology. However, growing evidence suggests that a significant proportion of children do not go on to develop psychopathology following victimization. Therefore, better understanding of individual risk for psychopathology in victimized children is important for rational allocation of child protection resources. Conventional explanatory approaches, such as linear or logistic regression, only offer an estimate of average risk within the population under investigation (i.e., victimized children). In contrast, prediction modelling techniques are optimized to determine individual risk, and to measure predictive accuracy in new or future observations. Prediction modelling techniques therefore have the potential to provide important analytical evidence to inform clinical decision-making.

Aims: To develop and internally validate a prediction model for individual risk of psychopathology in victimized children.

Methods: Participants were drawn from the Environmental Risk (E-Risk) Longitudinal Twin Study, a birth cohort of 2,232 British children followed throughout childhood. Victimization was assessed prospectively between the ages of 5 and 12 years. Children were defined as victimized if they had experienced severe exposure to at least one of the following: domestic violence between mother and partner; frequent bullying by peers; physical maltreatment by an adult; sexual abuse; emotional abuse or neglect; and physical neglect. Psychopathology was assessed at age 18 using private structured interviews during home visits. Selection of 22 potential individual-, family-, and community-level predictors of psychopathology in victimized children was informed by a recent systematic review (Meng et al., 2018). To reduce the variance in model coefficients and promote generalizability to new data, we applied the Least Absolute Shrinkage and Selection Operator (LASSO) penalty as part of regularized logistic regression. Internal validity was tested via ten-fold nested cross-validation, and quantified using measures of discrimination and calibration.

Results: 28.48% (n=591) of study members had been exposed to at least one form of severe victimization in childhood. Although these children had significantly higher proportions of age-18 psychopathology compared to their non-victimized peers, two out of five victimized children did not satisfy diagnostic criteria for any psychiatric disorder. For each psychiatric outcome (‘any psychiatric disorder’, ‘any externalizing disorder’, and ‘any internalizing disorder’), LASSO regularization selected a subset of variables to maximize predictive accuracy and parsimony. All models achieved good discrimination and were well-calibrated.

Conclusions: These findings highlight the potential clinical utility of prediction modelling techniques in improving evidence-based decision-making for victimized children. External validation of these models in independent samples is required before any clinical implementation.

Authors

• Alan J. Meehan, King's College London
  Presenting Author

• Rachel M Latham, King's College London
  Non-Presenting Author

• Louise Arseneault, King's College London
  Non-Presenting Author

• Daniel Stahl, King's College London
  Non-Presenting Author

• Helen L. Fisher, King's College London
  Non-Presenting Author

• Andrea Danese, King's College London
  Non-Presenting Author