Prediction of work resumption among men and women with lower back- and neck pain in a Swedish population

Abstract

An approach based on Bayes theorem is used to predict the binary outcome of work resumption X, where X = 1 if no work resumption and X = 0 otherwise, given a vector of discrete predictors Z for men and women with lower back- and neck pain in a Swedish population. In this application the predictors have a complex dependency structure. Hierarchical cluster analysis is used to create independent groups of dependent predictors such that predictors within groups are dependent while predictors in different groups are independent. The main purpose is to estimate the probability P(X = 1|z) and to calculate confidence intervals for this probability. Based on these estimates one may decide whether a given person should be predicted as healthy or as non-healthy, and predictive values are calculated in order to evaluate of the performance of the prediction analysis. The results are compared with the frequently used ordinary logistic regression method without interactions. It is found that ignoring the correlations between the predictors may give seriously misleading results. Also, the problem with missing values is discussed.