Development of a Temporary Custody Re-Entry Risk Prediction Model using Longitudinal Cohort Data in Japan

The number of temporary custody cases due to child maltreatment and other reasons at child guidance centers in Japan is increasing, prompting imperative scrutiny of re-entry risk assessment. We aimed to derive and internally validate a risk prediction model for re-entry into temporary custody. The risk prediction model was developed from an analysis of data from 725 children recruited in this retrospective single-center longitudinal cohort study conducted in Japan. The anticipated outcome was re-entry into temporary custody. Predictor variables were selected from 15 prospective variables concerning information on children and their familial contexts. A risk prediction model was developed using stepwise logistic regression. The final risk model was validated via C-statistic using cross-validation and bootstrap resampling methods. Calibration was assessed using the Hosmer-Lemeshow test. Of 725 children under temporary custody, 178 (24.6%) experienced re-entry into temporary custody. The predictors in the conclusive risk model were a history of temporary custody (p < 0.001), age at first birth of < 25 years (p < 0.002), single mother or stepfamily (p < 0.001), reasons for temporary custody including child abuse (p = 0.076) and child sex (p = 0.152), and child disability certificate (p = 0.252). Calibration scrutiny via the Hosmer-Lemeshow test revealed no discernible irregularities (p = 0.320). The naïve C-statistic of the model was 0.70, whereas the optimism-corrected C-statistics was 0.67-0.69. The presented risk prediction model showed acceptable calibration and discriminatory capability. The model can optimize limited human resources by providing valid risk estimates of the likelihood of re-entry to temporary custody within one year.