Hier finden Sie wissenschaftliche Publikationen aus den Fraunhofer-Instituten.

An Explainable Multimodal Neural Network Architecture for Predicting Epilepsy Comorbidities Based on Administrative Claims Data

: Linden, Thoma; Jong, J. de; Lu, C.; Kiri, V.; Haeffs, K.; Fröhlich, H.

Volltext urn:nbn:de:0011-n-6351839 (2.8 MByte PDF)
MD5 Fingerprint: 9b60b1d5f83b9ca7bb9c668007ddaa78
(CC) by
Erstellt am: 27.5.2021

Frontiers in artificial intelligence 4 (2021), Art. 610197, 14 S.
ISSN: 2624-8212
Zeitschriftenaufsatz, Elektronische Publikation
Fraunhofer SCAI ()
precision medicine; personalized medicine; P4 medicine; comorbidity prediction; administrative claims data; epilepsy; neural networks; machine learning

Epilepsy is a complex brain disorder characterized by repetitive seizure events. Epilepsy patients often suffer from various and severe physical and psychological comorbidities (e.g., anxiety, migraine, and stroke). While general comorbidity prevalences and incidences can be estimated from epidemiological data, such an approach does not take into account that actual patient-specific risks can depend on various individual factors, including medication. This motivates to develop a machine learning approach for predicting risks of future comorbidities for individual epilepsy patients. In this work, we use inpatient and outpatient administrative health claims data of around 19,500 U.S. epilepsy patients. We suggest a dedicated multimodal neural network architecture (Deep personalized LOngitudinal convolutional RIskmodelDeepLORI) to predict the time-dependent risk of six common comorbidities of epilepsy patients. We demonstrate superior performance of DeepLORI in a comparison with several existing methods. Moreover, we show that DeepLORI-based predictions can be interpreted on the level of individual patients. Using a game theoretic approach, we identify relevant features in DeepLORI models and demonstrate that model predictions are explainable in light of existing knowledge about the disease. Finally, we validate the model on independent data from around 97,000 patients, showing good generalization and stable prediction performance over time.