Estimation of ventilation, length of stay, and mortality using artificial neural networks

Abstract
The APACHE II severity of disease classification system has been used for many years to predict probability of patient mortality; however, this approach is manual and was developed for use in a specific patient population. In this study, we trained feed-forward neural networks to estimate various outcomes of patients in a local Intensive Care Unit (ICU). The outcomes estimated include: patient mortality, length of stay and artificial ventilation requirements. The results compare favourably with those previously discussed in the literature. The eventual goal is to incorporate these neural networks into a larger patient management system for use in the ICU; it is anticipated that such a system will aid in medical decision making and will have potential use as a teaching tool.