This Patient Looks Like That Patient: Prototypical Networks for Interpretable Diagnosis Prediction from Clinical Text

Proceedings of the 2nd Conference of the Asia-Pacific Chapter of the Association for Computational Linguistics and the 12th International Joint Conference on Natural Language Processing (Volume 1: Long Papers), pages 172–184 November 20–23, 2022. ©2022 Association for Computational Linguistics 172 This Patient Looks Like That Patient: Prototypical Networks for Interpretable Diagnosis Prediction from Clinical Text Betty van Aken1, Jens-Michalis Papaioannou1, Marcel G. Naik2, Georgios Eleftheriadis2, Wolfgang Nejdl3, Felix A. Gers1, Alexander Löser1 1 Berliner Hochschule für Technik (BHT), 2 Charité Berlin, 3 Leibniz University Hannover {bvanaken,michalis.papaioannou,gers,aloeser}@bht-berlin.de, {marcel.naik,georgios.eleftheriadis}@charite.de, nejdl@L3S.de Abstract The use of deep neural models for diagnosis prediction from clinical text has shown promis- ing results. However, in clinical practice such models must not only be accurate, but provide doctors with interpretable and helpful results. We introduce ProtoPatient, a novel method based on prototypical networks and label-wise attention with both of these abilities. ProtoPa- tient makes predictions based on parts of the text that are similar to prototypical patients– providing justifications that doctors understand. We evaluate the model on two publicly avail- able clinical datasets and show that it outper- forms existing baselines. Quantitative and qual- itative evaluations with medical doctors further demonstrate that the model provides valuable explanations for clinical decision support. 1 Introduction Medical professionals are faced with a large amount of textual patient information every day. Clinical decision support systems (CDSS) aim to help clinicians in the process of decision-making based on such data. We specifically look at a sub- task of CDSS, namely the prediction of clinical diagnosis from patient admission notes. When clinicians approach the task of diagnosis predic- tion, they usually take similar patients into account (from their own experience, clinic databases or by talking to their colleagues) who presented with typical or atypical signs of a disease. They then compare the patient at hand with these previous en- counters and determine the patient’s risk of having the same condition. In this work, we propose ProtoPatient, a deep neural approach that imitates this reasoning process of clinicians: Our model learns prototypical char- acteristics of diagnoses from previous patients and PRESENT ILLNESS: complained headache of his life since yd PRESENT ILLNESS: 62yo male patient presented with worst of her life. Transferred to hospital upoon neepisodes xt ICU where she had episodes problems keeping food iin and of nausea transferred for a CT which showed signs of a large [...] CHIEF COMPLAINT: headaches PRESENT ILLNESS: 62yo male patient presented with severe headaches. He had nausea and vomited multiple times. MEDICATION ON ADMISSION: levoxil, spirvia, pempril, multi-vit, trazadone FAMILY HISTORY: fhx significant for DMII & HTN, father with [...] This Patient newly admitted to the ICU That Patient with prototypical signs of intracerebral hemorrhage Looks Like ... severe headaches. He felt nauseous and vomited multiple times... ... complained of worst headache of her life ... ... episodes of nausea and problems keeping food down ... Figure 1: Basic concept of the ProtoPatient method. The model makes predictions for a patient (left side) based on the comparison to …

Content truncated to 500 words