Leibniz MMS Days 2019 - Abstract
Oswald, Marcus
Pneumonia remains a disease with considerable morbidity and mortality, and it is easily misdiagnosed and inappropriately treated. The short-term mortality for moderate to severe pneumonia reaches 9% Macrolides are one of the most widely used antibiotics and currently widely used in treating a broad range of common bacterial infections, however, they can be rather ineffective or even detrimental in certain cases, e.g. cardiotoxic side-effects. The aim is now to identify clinical characteristics serving as theranostic markers to support the decision of using macrolides or other antibiotics as initial pneumonia treatment. Since in our available data the decision to give macrolide treatment was not performed in a random way, mathematical methods like propensity score analysis are required to artificially randomize the data. Several methods were implemented and compared, unfortunately the following machine learning part and the quality of the decision-rule heavily depend on the former method. Therefore criteria are required to test whether the modified data behaves like a randomized trial.