gms | German Medical Science

73. Jahrestagung der Deutschen Gesellschaft für Neurochirurgie (DGNC)
Joint Meeting mit der Griechischen Gesellschaft für Neurochirurgie

Deutsche Gesellschaft für Neurochirurgie (DGNC) e. V.

29.05. - 01.06.2022, Köln

Machine learning methods may improve the prediction of shunt dependency in aneurysmal subarachnoid haemorrhage compared to established prediction scores

Machine Learning Anwendungen können die Vorhersagegenauigkeit des Shunt-abhängigen Hydrocephalus bei der aneurysmatischen Subarachnoidalblutung verbessern

Meeting Abstract

  • presenting/speaker Anton Früh - Charité – Universitätsmedizin Berlin, Klinik für Neurochirurgie, Berlin, Deutschland
  • Adam Hilbert - Charité – Universitätsmedizin Berlin, CLAIM - Charité Lab for AI in Medicine, Berlin, Deutschland
  • Vince Istvan Madai - Charité – Universitätsmedizin Berlin, CLAIM - Charité Lab for AI in Medicine, Berlin, Deutschland; Birmingham City University, School of Computing and Digital Technology, Faculty of Computing, Engineering and the Built Environment, Birmingham, Vereinigtes Königreich
  • Julia Kiewitz - Charité – Universitätsmedizin Berlin, Klinik für Neurochirurgie, Berlin, Deutschland
  • Meike Unteroberdörster - Charité – Universitätsmedizin Berlin, Klinik für Neurochirurgie, Berlin, Deutschland
  • Stefan Wolf - Charité – Universitätsmedizin Berlin, Klinik für Neurochirurgie, Berlin, Deutschland
  • Peter Vajkoczy - Charité – Universitätsmedizin Berlin, Klinik für Neurochirurgie, Berlin, Deutschland
  • Dietmar Frey - Charité – Universitätsmedizin Berlin, Klinik für Neurochirurgie, Berlin, Deutschland; Charité – Universitätsmedizin Berlin, CLAIM - Charité Lab for AI in Medicine, Berlin, Deutschland
  • Nora F. Dengler - Charité – Universitätsmedizin Berlin, Klinik für Neurochirurgie, Berlin, Deutschland

Deutsche Gesellschaft für Neurochirurgie. 73. Jahrestagung der Deutschen Gesellschaft für Neurochirurgie (DGNC), Joint Meeting mit der Griechischen Gesellschaft für Neurochirurgie. Köln, 29.05.-01.06.2022. Düsseldorf: German Medical Science GMS Publishing House; 2022. DocV122

doi: 10.3205/22dgnc122, urn:nbn:de:0183-22dgnc1229

Published: May 25, 2022

© 2022 Früh et al.
This is an Open Access article distributed under the terms of the Creative Commons Attribution 4.0 License. See license information at http://creativecommons.org/licenses/by/4.0/.


Outline

Text

Objective: Early and reliable prediction of shunt dependent hydrocephalus (SDHC) after aneurysmal subarachnoid hemorrhage (aSAH) may shorten the overall length of intensive care and hospital stay and may reduce the risk of catheter-associated meningitis and overall treatment costs. Scoring systems are externally validated and may allow for easy calculations in clinical routine but do not reach excellent prediction performances regarding the occurrence of SDHC. We designed this study to test the comparability of predictive methods and whether inclusion of additional variables in a machine learning (ML) framework may improve the performance of SDHC prediction.

Methods: Local ethics committee approval was obtained (EA1/291/14). Combined clinical, radiographic, and laboratory features as well as standard scores (CHESS, SDASH) available on patient admission were analyzed using a consecutive single-center database of patients that presented with aSAH (n = 408). Conventional receiver operating curves with calculation of the area under the curve (AUC) of established scores was performed using IBM SPPS Statistics 27. Classification models (GLM; Lasso; Elastic Net) were trained in a supervised ML paradigm using scores and combined features. Train and test sets were split randomly (4:1 ratio) 50 times, in each iteration hyper-parameters were tuned in a ten-fold cross-validation using the train set. AUC was calculated for ML models for training and test sets. ML paradigm data is presented for the best performing model as training and (test).

Results: 295 patients survived the initial phase of the disease and were eligible for analysis. 20.7 % of patients developed SDHC. Conventional prediction of SDHC by CHESS and SDASH resulted in an AUC of 0.77 and 0.78, respectively. ML model calculations of variables included in CHESS and SDASH resulted in a maximum AUC of 0.73 and 0.75 (0.68 and 0.71), respectively. The inclusion of more clinical, radiographic, and laboratory parameters available on admission to ML paradigms increased the predictive performance to 0.83 (0.75).

Conclusion: The inclusion of clinical, radiographic, and laboratory parameters in an ML setup improves the predictive performance of SDHC after aSAH. More work is needed on method comparability and potential inclusion of more variables to allow for individualized SDHC prediction.