COPD and asthma: longitudinal and cross-sectoral real-world data for machine learning application for quality improvement and knowledge acquisition – CALM-QE (EAACI Congress 31.05.2024)

Autoren: A. Hoheisel1; D. Stolz1; H. Binder2; C. Taube3; I. Lehmann4; AM. Dittrich5, 6; T. Ganslandt7; G. Rohde8; K. Bluemchen9; K. Marquardt10; S. Kuhnert11; K. Strauch12; J. Kraus13; H. Schneider14; H. Kestler13; F. Ueckert15; B. Schmeck16; R. Bals17; C. Gundler15; H. Renz18

Affiliation: 1 Medical Center and Faculty of Medicine, University of Freiburg, Freiburg (Brsg.), Germany; 2 Institute of Medical Biometry and Statistics, Freiburg (Brsg.), Germany; 3 University Hospital Essen-Ruhrlandklinik, Essen, Germany; 4 Humboldt-Universität zu Berlin, and Berlin Institute of Health (BIH), Berlin, Germany, Berlin, Germany; 5 Hannover Medical School, Hannover, Germany; 6 German Center for Lung Research (DZL), Hannover Medical School, Hannover, Germany; 7 Friedrich-Alexander Universtät Erlangen-Nürnberg, Erlangen, Germany; 8 Goethe University Frankfurt, University Hospital, Medical Clinic I,  Frankfurt/Main, Germany; 9 University Hospital Frankfurt, Goethe University Frankfurt, Frankfurt/Main, Germany; 10University Hospital of Gießen and Marburg, Giessen, Germany; 11Cardio-Pulmonary Institute (CPI), Justus Liebig University, Giessen, Germany; 12University Medical Center of the Johannes Gutenberg University Mainz, Mainz, Germany; 13Institute of Medical Systems Biology, Ulm University, Ulm, Germany; 14Technische Hochschule Mittelhessen, Giessen, Germany; 15Applied Medical Informatics, University Medical Center Eppendorf, Hamburg, Germany; 16Philipps-University Marburg, German Center for Lung Research (DZL), Marburg, Germany; 17Saarland University, Kirrberger Strasse 1, Homburg/Saar, Germany; 18and the Universities of Giessen and Marburg Lung Center (UGMLC), Philipps University Marburg, Marburg, Germany

Background: Chronic obstructive pulmonary disease (COPD) and bronchial asthma (BA) are the most common non-communicable pulmonary diseases with a major socio-economic impact, which worldwide affected more than 200 million COPD patients and 250 million asthma patients in 2019 (Vos et al. 2020). Development, severity and exacerbations (even allergytriggered) are the result of complex gene-environment interactions (Reséndiz-Hernández and Falfán-Valencia 2018, D’Amato et al. 2013). Recent research revealed a broad heterogeneity in terms of phenotypes and endotypes with some degree of overlap between COPD and BA (Ray et al. 2015). In parallel to this novel concept more specific medications have been developped, leading to the challenge to stratify patients for this precision medicine approach. The major challenge in the field of COPD and BA is how to translate this trait treatable approach to the individual patient.

Method: CALM-QE is a multicentre project involving 12 German university hospitals, their outpatient clinics and private practices. In the previous Medical Informatics Initiative (MII) project, the participating project partners successfully made pneumology patient data accessible. This made it possible to extract large data sets of up to >25,000 patients at five sites, analyse them locally and combine the results into a joint analysis over a period of >10 years. The objective of this follow-on project is to develop, train and test predictive models for important clinical outcomes based on multidimensional “real-world-datasets” for COPD and BA patients across sectoral boundaries. They include the already established clinical core dataset, expansion by lung function, medication, chest imaging, addition of local climate and air pollutant data, and biosignals obtained via wearables and furthermore including the paediatric perspective.

Results: All of the abovementioned data will be used to identify and stratify disease trajectories. In addition, as a first proof of concept, we include data from research studies, including longitudinal data from the German COPD cohort. Imaging biomarkers be extracted from CT thorax using Coreline software and analyses based on neural networks and deep learning methods.

Conclusion: Based on a systematic big data approach from multidimensional large data sets for clinically relevant endpoints, this project could develop novel predictive models that serve the 4P medicine approach (personalised, participatory, predictive and preventive).