Using discriminant analysis to predict attacks of complicated pneumoconiosis in coalworkers

The application of the location model method of discriminant analysis to the problem of early identification of cases of complicated pneumoconiosis among coalworkers in the UK is described. Using a small set of predictor variables, the resulting allocation rule achieved an estimated 84% of cases and 60% of non-cases currently classified. The performance obtained using three other methods of discrimination was essentially the same as for the location model. It is concluded that prediction of such cases on the basis of the variables used for this analysis, is in principle feasible but of limited use in practice, because of the unacceptably high false-positive rate.