| dc.description.abstract | In Sweden, the number oflong-term sick-listed has increased by about 30% per year during the period 1997-2001, and the cost for health insurance is 108 billion SEK in the state budget (2002). Thus, the prediction of work resumption is very important. It is also important to identify the influence of different factors. In this thesis, an approach for prediction of future work resumption is proposed. The suggested method takes into account the dependency structure of the predictors in a flexible way.
In the first paper (1) an approach based on Bayes theorem is proposed for predicting a binary outcome conditionally on the values of a set of discrete predictors. Point- and interval estimators are derived for this probability, and the properties of these estimators are examined in detail by theoretical results and simulations. It is found that the variance of the estimated probability is heavily dependent on the situation. It was also found that a sample size of at least 400 was required to obtain reliable estimates of the prediction probabilities.
The second paper (2) is an application of the suggested approach in paper (1) to predict the outcome of work resumption 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 has been used to identify independent groups of predictors such that predictors within groups are dependent but independent of predictors in other groups. In a first step, point- and interval estimates of the probability of 'no work resumption' given the value of a set of predictors were calculated from the data set. In a second step, new observations were generated with the same characteristics as those in the first step. Predictive- and relative predictive values as well as proportions of correct classifications were calculated. The predictive values and the proportions of correct predictions ranged from 0.59 to 0.81 and 0.70 to 0.86, respectively. | sv |
