Prediction of general medical admission length of stay with natural language processing and deep learning: a pilot study
- PDF / 370,609 Bytes
- 2 Pages / 595.276 x 790.866 pts Page_size
- 104 Downloads / 185 Views
IM - COMMENTARY
Prediction of general medical admission length of stay with natural language processing and deep learning: a pilot study Greta Falavigna1 Received: 28 January 2020 / Accepted: 30 January 2020 © Società Italiana di Medicina Interna (SIMI) 2020
The article by [1] focuses on a very interesting topic for EDs, i.e., predicting the length of stay of patients. The relevance of this topic is widely recognized in the literature, since it is linked to a twofold problem: on one hand, physicians clearly care about people’s health; on the other hand, they have to take budget constraints into account [2, 3]. Starting from these considerations, models for assessing the pressure of the medical care system are very welcome. The use of complex systems in clinical evaluations is not so widespread, although in recent years, researchers have turned their attention to how these can be applied [4]. However, given their effectiveness in generalization and prediction [5], these methodologies should be employed more broadly, also by integrating them with more traditional ones, such as, for instance, multivariate regression or fuzzy models. The paper by [1] suggests applying machine learning and deep learning to predict the length of stay (i.e., ≤ 2 days or > 2 days) and discharge destination (i.e., home or nonhome destination) of patients, considering Natural Language Processing. The aim of this work is to improve the management of hospitals and, in particular, of Emergency Departments. Indeed, the authors analyze a sample of about 300 patients admitted to different hospital divisions, based on what the physicians decide upon arrival at the Emergency Department. In detail, patients can be hospitalized in the Acute Medical unit (AMU) if the ED doctors determine that their LOS is shorter than or equal to 2 days; otherwise, they are sent to the General Medicine ward. It is clear that more accurate evaluations by physician result in better organization and more effective hospital management. * Greta Falavigna [email protected] 1
Research Institute on Sustainable Economic Growth of Italian National Council of Research (IRCrES-CNR), Via Real Collegio 30, 10024 Moncalieri, TO, Italy
The authors use four different Machine Learning methodologies, which are convolutional neural networks, artificial neural networks, logistic regressions, and random forest techniques. The main idea is to assess a patient’s length of stay and discharge destination considering the free text compiled by the physician at the time of admission. After a pre-processing phase, necessary for model implementation, these complex models should be able to forecast the length of stay and discharge destination of each patient. The results are evaluated considering the area under the curve (AUC), as well as the accuracy and standard values of correct and incorrect classifications (i.e., true positive, true negative, false positive, and false negative). As for the length of stay (LOS), the authors find that artificial neural networks perform better
Data Loading...
