gms | German Medical Science

66. Jahrestagung der Deutschen Gesellschaft für Medizinische Informatik, Biometrie und Epidemiologie e. V. (GMDS), 12. Jahreskongress der Technologie- und Methodenplattform für die vernetzte medizinische Forschung e. V. (TMF)

26. - 30.09.2021, online

Macrolide beta-lactam combination, Fluoroquinolone or beta-lactam mono therapy for patients hospitalized with moderate CAP? Developing an individual three-way treatment rule using machine learning

Meeting Abstract

Search Medline for

  • Marcus Oswald - Universitätsklinikum Jena, Jena, Germany
  • Rainer König - Universitätsklinikum Jena, Jena, Germany

Deutsche Gesellschaft für Medizinische Informatik, Biometrie und Epidemiologie. 66. Jahrestagung der Deutschen Gesellschaft für Medizinische Informatik, Biometrie und Epidemiologie e. V. (GMDS), 12. Jahreskongress der Technologie- und Methodenplattform für die vernetzte medizinische Forschung e.V. (TMF). sine loco [digital], 26.-30.09.2021. Düsseldorf: German Medical Science GMS Publishing House; 2021. DocAbstr. 213

doi: 10.3205/21gmds062, urn:nbn:de:0183-21gmds0625

Published: September 24, 2021

© 2021 Oswald et al.
This is an Open Access article distributed under the terms of the Creative Commons Attribution 4.0 License. See license information at



Introduction: The type of antibiotic therapy in community-acquired pneumonia (CAP) of moderate severity is subject for discussion. According to the German S3-guidelines of 2016 [1] as primary therapy „an individual decision“ between beta-Lactam-monotherapy and Macrolide-beta-Lactam-combination is recommended. Fluoroquinolones are mentioned as alternative therapy.

In order to support this individual clinical decision, we developed an algorithm to find score-based treatment rule that uses clinical parameters that were known before the treatment decision was done. In former work we presented a treatment rule choosing between beta-Lactam-mono- and Macrolide-beta-Lactam-combination-therapy by a simple decision tree [2]. Nevertheless, score-based rules turned out to be more robust and easier to extend to a third treatment type – the Fluoroquinolones.

Methods: We constructed a score system where pros and cons of certain patient subgroups for all three treatment types are collected and the optimal combination – in terms of maximizing 30- or 180-days survival – is computed. These optimal scores are cross-validated in a 10xCV scheme on 7453 hospitalized patients with moderate severity (non-ICU patients) from the observational, prospective, multinational CAPNETZ [3] study. Since the treatment decision in this data was not done in a random way, methods for balancing were required. Combining propensity weighting with machine learning turned out to be challenging. We developed new balancing methods and adaptions to existing ones.

Results: We found scores of different lengths consisting of typically available parameters at the day of hospital admission. The cross-validated results were promising. All the score-based treatment rules lead to lower mortality than the observed standard of care.

Conclusion: Machine learning has a huge potential to support clinical decisions. Nevertheless, there are big challenges like combining machine learning algorithms with balancing methods. As next steps we plan to validate the treatment rules on an external patient set. Afterwards a randomized clinical trial should confirm our results in order to transfer to the clinical routine.

The authors declare that they have no competing interests.

The authors declare that an ethics committee vote is not required.


Deutsche Gesellschaft für Pneumologie und Beatmungsmedizin e.V. (DGP); Deutsche Gesellschaft für Infektiologie e.V. (DGI); Paul-Ehrlich-Gesellschaft für Chemotherapie e.V. (PEG). Leitlinie Behandlung von erwachsenen Patienten mit ambulant erworbener Pneumonie. 31.12.2015 [updated 2021 Apr 14; cited 2021 May 5].
König R, Cao X, Oswald M, Forstner C, Rohde G, Rupp J, et al. Macrolide combination therapy for patients hospitalised with community-acquired pneumonia? An individualised approach supported by machine learning. Eur Respir J. 2019;54(6):1900824.
Suttorp N, Welte T, Marre R, Stenger S, Pletz M, Rupp J, et al. [CAPNETZ. The competence network for community-acquired pneumonia (CAP)]. Bundesgesundheitsblatt Gesundheitsforschung Gesundheitsschutz. 2016;59(4):475-81. DOI: 10.1007/s00103-016-2318-7. External link