Forecasting childhood adversities from conditions of birth.

Abstract

Childbirth presents an optimal time for identifying high-risk families to commence intervention that could avert various childhood health and social adversities. We sought to establish the minimum set of exposures required to accurately predict a range of adverse childhood outcomes up to the age of 13years, from a set of 14 individual and familial risk exposures evident at the time of birth. Participants were 72,059 Australian children and their parents drawn from a multi-register population cohort study (data spanning 1994-2018). Risk exposures included male sex, young mother (aged ≤21years), no (or late first; >16weeks) antenatal visit, maternal smoking during pregnancy, small for gestational age, preterm birth, pregnancy complications (any of hypertension, diabetes mellitus, gestational diabetes or pre-eclampsia), >2 previous pregnancies of ≥20weeks, socio-economic disadvantage, prenatal child protection notification, and maternal or paternal mental disorder or criminal offending history. Individual outcomes included early childhood developmental vulnerability (age 5years), sustained educational underachievement (age 8 and 10years), mental disorder diagnoses, substantiated childhood maltreatment, and contact with the police as a victim or person-of-interest up to age 13-14years. Risk exposures at birth predicted individual childhood outcomes with fair to excellent accuracy: the area under the receiver operating characteristic curves ranged between 0.60 (95% CI 0.58, 0.62) for childhood mental disorder and 0.83 (95% CI 0.82, 0.85) for substantiated child maltreatment. The presence of five or more exposures characterised 12-25% of children with one or more adverse outcomes and showed high predictive certainty for models predicting multiple outcomes, which were apparent in 9% of the population. Up to a quarter of the neonatal population at risk of multiple adverse outcomes can be detected at birth, with implications for population health screening. However, cautious implementation of these models is warranted, given their relatively low positive predictive values.