Poster #218 - Development of an individualized risk calculator for poor functioning in young adults victimized during childhood

• In Event: 3-206 - Poster Session 13
  In Poster Session: PS 13 Section - Social Relationships

Integrative Statement

Background: Individuals who experience childhood victimization – including abuse, neglect, bullying and domestic violence – are at greater risk of developing functional impairment later in life compared to their non-victimized peers. However, not all victimized children demonstrate poor outcomes; these average group differences fail to capture the resilience that many victimized children display. Research examining predictors of functioning following victimization typically use logistic regression models to generate a probability of being classified as having a particular outcome. However, such explanatory models describe the average probability of an outcome for a group rather than the prediction for an individual. Moreover, the ability of these models to make accurate predictions in new cases has rarely been examined.

Aim: We sought to advance existing knowledge by moving from the averaged associations of logistic regression models to individualized risk prediction using regularized regression. This approach can prevent overfitting of a multivariable model to a specific sample by applying additional penalties to reduce the variability of the regression estimates. We aimed to build and internally validate the predictive performance of an evidence-based risk calculator that identifies those victimized children who are at greatest risk of developing poor functional outcomes in young adulthood.

Method: We utilized prospectively-collected data from the Environmental Risk (E-Risk) Longitudinal Twin Study, a nationally-representative birth cohort of 2,232 British twin children. Exposure to childhood victimization was assessed when the children were 5, 7, 10 and 12 years old. Factor analysis of functional outcomes assessed at age 18 revealed two factors, conceptualized as (i) ‘poor psychosocial outcome’ (social isolation; low life satisfaction; loneliness; low sleep quality; adolescent poly-victimization) and (ii) ‘poor economic outcome’ (low educational achievement; not in education, employment or training; parenthood; criminal cautions and convictions). Thus, we developed and evaluated a prediction model for each of these two outcomes. We selected candidate predictor variables based on a recent meta-analysis of the literature (Meng, Fleury, Xiang, Li & D’Arcy, 2017) which we mapped to E-Risk Study data collected between ages 5 and 12. To achieve parsimonious prediction models, we used the Least Absolute Shrinkage and Selection Operator (LASSO) logistic analytic approach which performs variable selection and regularization to enhance prediction accuracy. Nested 10-fold cross-validation was used for internal validation.

Results: We found heterogeneity in the psychosocial and economic outcomes of individuals exposed to childhood victimization. To understand this variability, LASSO regularized regression models retained a subset of predictor variables which together predicted an individual’s risk probability. Internal validation showed both prediction models to have good discrimination (the ability to distinguish between victimized children who do and do not develop the poor outcome) and calibration (the agreement between model-predicted probabilities and observed outcomes).

Conclusion: Our results demonstrate the applicability of prediction modelling techniques to individual risk for poor functional outcomes following childhood victimization. Risk prediction tools – whilst common in medicine – remain relatively novel in psychology yet have the potential to assist practitioners in decision-making regarding the targeting of interventions, which is particularly useful in a climate of scarce resources.

Authors

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

• Alan J. Meehan, 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

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

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