Cross-lingual Approaches for the Detection of Adverse Drug Reactions in German from a Patient’s Perspective

Proceedings of the 13th Conference on Language Resources and Evaluation (LREC 2022), pages 3637–3649 Marseille, 20-25 June 2022 © European Language Resources Association (ELRA), licensed under CC-BY-NC-4.0 3637 Cross-lingual Approaches for the Detection of Adverse Drug Reactions in German from a Patient’s Perspective Lisa Raithel1,2,3, Philippe Thomas1, Roland Roller1, Oliver Sapina1, Sebastian M¨oller1,2, Pierre Zweigenbaum3 1Deutsches Forschungszentrum f¨ur K¨unstliche Intelligenz (DFKI) Berlin, 10559 Berlin, Germany 2Technische Universit¨at Berlin, 10623 Berlin, Germany 3Universit´e Paris-Saclay, CNRS, Laboratoire interdisciplinaire des sciences du num´erique (LISN), 91405 Orsay, France lisa.raithel@dfki.de Abstract In this work, we present the first corpus for German Adverse Drug Reaction (ADR) detection in patient-generated content. The data consists of 4,169 binary annotated documents from a German patient forum, where users talk about health issues and get advice from medical doctors. As is common in social media data in this domain, the class labels of the corpus are very imbalanced. This and a high topic imbalance make it a very challenging dataset, since often, the same symptom can have several causes and is not always related to a medication intake. We aim to encourage further multi-lingual efforts in the domain of ADR detection and provide preliminary experiments for binary classification using different methods of zero- and few-shot learning based on a multi-lingual model. When fine-tuning XLM-RoBERTa first on English patient forum data and then on the new German data, we achieve an F1-score of 37.52 for the positive class. We make the dataset and models publicly available for the community. Keywords: pharmacovigilance, text classification, adverse drug reactions 1. Introduction Adverse drug reactions (ADRs) are a major and in- creasing public health problem and exist all over the world, mostly under-reported (Alatawi and Hansen, 2017). They describe an unanticipated, negative reac- tion to a medication. Of course, new drugs are tested extensively when being developed, but certain vulner- able groups, like pregnant or elderly people, are rarely part of clinical trials and still might be in need of med- ication at some point. Furthermore, clinical trials and also physicians prescribing medications cannot cover every potential use case. Therefore, an efficient moni- toring of ADRs from different angles is important and necessary for improving public health and safety in medication intake. One of those angles is the classification of documents containing mentions of ADRs. While there exist var- ious resources for training NLP models on the task of ADR classification, for example scientific publica- tions or drug leaflets, the world wide web provides a more up-to-date and diverse source of information, es- pecially social media and patient fora (Sarker and Gon- zalez, 2015). Here, patients freely write about their ex- periences during medication therapy. By using their own non-expert language, e.g. using laymen terms for describing their situation, they create a rich data source for pharmacovigilance from a patient’s perspective in a wide variety of languages. Therefore, there is theoret- ically a huge amount of text to leverage. On the other side, however, this text is noisy (e.g. spelling mistakes), incomplete (e.g. deleted messages) …

Content truncated to 500 words