Multilabel classification of disease prediction in patients presenting with dyspnea
Justus Baarts, Stéphanie Giezendanner, Giorgia Lüthi-Corridori, Michael Brändle, Thomas Dieterle, Luca Gabutti, Angelika Hammerer‐Lercher, Paul Hasler, Katja Henny-Fullin, Jürgen Muser, Gregor Leibundgut, Anne Barbara Leuppi-Taegtmeyer, Corinne Punsap Marbet, Christian Schraner, Jörg D. Leuppi, Fabienne Jaun · 2021
Introduction: Dyspnea is related to a wide range of health conditions. The underlying causes of dyspnea are not mutually exclusive; patients could have symptoms of multiple different diseases at the same time. The aim of this study is to identify patterns in patient data that indicate a probability associated with COPD, pneumonia, asthma, pulmonary embolism or heart insufficiency. Methods: The study was based on retrospective data from the Cantonal Hospital of Baselland in Switzerland from 2014. All adult patients with chief complaints of dyspnea, shortness of breath and air hunger were screened and the final diagnosis was evaluated independently by two General Internal Medicine residents. Predictors encompassed demographics, diagnostic tests and laboratory findings (using CREATE system, a tool for big data research). We applied a multilabel classification algorithm (which transforms the multilabel problem into binary classification problems, one for each label). The dataset was split into training (80%) and testing (20%) datasets and 10-fold cross validation was used. Results: Out of 4891 patients hospitalized at the internal medicine ward, 602 patients were included with chief complaints of dyspnea. Patients without any diagnosis of COPD, pneumonia, asthma, pulmonary embolism or heart failure (n=84) were excluded from further analyses. Preliminary results showed that classifier chains achieved good performance that is 78% precision, 65% recall, 67% F1 score. Conclusion: Multilabel classification might be useful to predict the risks of these five main diagnosis of dyspnea. However, we need further dedication to feature selection from the hospital patient records to improve predictions.