A0307
Title: Distributional regression for lung function of cystic fibrosis patients with a special focus on center-specific effects
Authors: Elisabeth Bergherr - Georg-August-Univerität Göttingen (Germany) [presenting]
Colin Griesbach - Georg-August-University Goettingen (Germany)
Marisa Lane - Georg-August-University Goettingen (Germany)
Abstract: Rapid lung function decline is a severe problem for cystic fibrosis patients throughout their whole life. We have access to the data from the German cystic fibrosis registry, which includes thousands of patients with hundreds of thousands of observations repeatedly over each year and hundreds of variables, like sociodemographic information, biomarkers, but also gene expression data. We plan to estimate a prediction model for the lung volume measured by the \%FEV1-value. The latter is one of the key indicators for healthy functionality of the lung. Since not only the expectation of the volume but also the variation is of major importance for the patients, a Gaussian distributional regression model will be used. The vast amount of possible explanatory variables calls for a strong selection algorithm, which is one of the key features of gradient boosting. We will develop a way of including the information on the center in which the individual patients are treated. The algorithm will hence either detect a spatial pattern, or account for the center specific variation in terms of (possibly clustered) random effects.
