Predicting occupational lung diseases

This thesis aims at demonstrating the development, validation, and application of prediction models for occupational lung diseases. Prediction models are developed to estimate an individual’s probability of the presence or future likelihood of occurrence of an outcome (i.e. disease of interest or its related condition). These models are used to assist clinical decision making for individuals, or to stratify individuals into risk groups with a different likelihood for developing disease or with disease severity. In this thesis the development of a diagnostic model is described to detect sensitization to wheat allergens among bakery workers. This model made use of questionnaire items only. Secondly a more generic model was developed for sensitization to high molecular weight allergens in bakery workers (exposed to wheat and or fungal alpha amylase allergens) and laboratory animal workers (exposed to rat and or mouse urinary allergens). The third diagnostic model included questionnaire items and lung function test results to predict the probability of having chest X-ray changes indicative for pneumoconiosis in Dutch construction workers exposed to silica dust. All diagnostic models were transformed into easy-to-use scoring systems to facilitate their application in practice. These diagnostic models enable objective and standardized quantification of the probability of having or developing disease without performing (invasive) advanced and costly reference test. We also developed prognostic models for occurrence of occupational sensitization and respiratory symptoms in apprentices in animal health technology. The prognostic value of questionnaire items, skin-prick tests, and bronchial challenge to methacholine, as a single test or in combination with others, was assessed. Another element in the thesis was to assess the generalizability of a prediction model. Different statistical approaches were used to externally validate a questionnaire model for sensitization to LA allergens -developed in Dutch workers- in Canadian animal health technology apprentices. A new model was eventually developed in Canadian apprentices and compared to the original Dutch model. Model revision was done to evaluate if inclusion of new predictors from the Canadian setting could improve the performance of the original model. Finally the application of a diagnostic model for occupational sensitization is described in a surveillance program for respiratory diseases in baking and flour-producing industries in the Netherlands. The diagnostic model was applied in 5,325 Dutch bakery and flour and enzyme exposed workers. This chapter illustrates how a diagnostic model may improve the efficiency of surveillance programs. After assuring that a model is valid and produces accurate predictions, it is important to determine probability cut-off points to stratify individuals into risk categories. The choice for a cut-off point must be based on the balance between the proportion of missed cases and reduction of unnecessary diagnostic tests. This thesis shows that predicting lung diseases in the context of occupational health care and practice is possible. The use of the prediction tools assists the decision making process and would hopefully reduce expenses. Application of prediction models has not been fully explored but efforts to increase the use of predictive models deserve strong support