Using discriminant function analysis to estimate the impact of a therapeutic intervention in burn care

Abstract

INTRODUCTION DURING the past few years, attempts have been made to develop baseline results of therapy for burned patients. These investigations have attempted to detect the influence of selected patient variables on the survival and morbidity of burn victims (Bull, 1971; Waisbren et al., 1975; Feller et al., 1976). Many of these studies have used probit analysis to determine the relationship between age, sex, body surface area burned, type of topical therapy used and others, as predictors of survival. McCoy et al. (1968) suggested that discriminant analysis (DA) might be useful in that it allows the incorporation of several predictor variables into a probability model. In 1975, Moores et al. described several discriminant functions (DF) and discussed the implications of using such a statistical tool. More recently, Clark et al. (1978) developed several discriminant functions to predict the survival of burned patients. They emphasized that discriminant function analysis offers more flexibility for examining and including potential predictor variables than does probit analysis. In addition, they point out that when using discriminant function analysis fewer cases are required to obtain accurate predictions. Finally, they suggest that discriminant analysis may be very useful in comparing patients receiving different treatment regimens, enabling medical personnel to make better decisions with regard to the management of burned patients. This investigation attempted to examine the value of discriminant analysis in measuring the relative impact of individualizing gentamicin dosage regimens in burned patients with Gramnegative septicaemia.