An optimization model for sequential decision-making applied to risk prediction after liver resection and transplantation.

The paper demonstrates a sequential decision procedure, which allows for optimal cost-effective or early decision-making while maintaining given error constraints. As a specific example the construction of a sequential decision procedure to determine if a patient was a high risk patient or not is used. The advantages of the procedure are demonstrated by a surgical problem of risk prediction from a clinical study on liver resection and transplantation. Data are available pre, peri- and postoperative, and form the basis of three clinical scores. The quality of the procedure is measured in terms of sensitivity and specificity, and the procedure will be optimized in a way, that a priori given error constraints will be maintained. A decision theoretic model is introduced and a robust procedure is developed. The approach is feasible for any fixed number of continuous clinical test scores obtained in a time sequence. Two sets of scores derived form linear discriminant analysis and artificial neural networks are compared.