Predicting adverse infant health outcomes using routine screening variables: modelling the impact of interdependent risk factors

This paper sets out a methodology for risk assessment of pregnancies in terms of adverse outcomes such as low birth-weight and neonatal mortality in a situation of multiple but possibly interdependent major dimensions of risk. In the present analysis, the outcome is very low birth-weight and the observed risk indicators are assumed to be linked to three main dimensions: socio-demographic, bio-medical status, and fertility history. Summary scores for each mother under each risk dimension are derived from observed indicators and used as the basis for a multidimensional classification to high or low risk. A fully Bayesian method of implementation is applied to estimation and prediction. A case study is presented of very low birth-weight singleton livebirths over 1991-93 in a health region covering North West London and parts of the adjacent South East of England, with validating predictions to maternities in 1994