Document Type
Article
Publication Date
4-22-2025
Abstract
Electronic Health Records (EHRs), which include demographic information, clinical notes, vital signs, laboratory test results, and others, provide rich information for clinical outcome prediction. In this work, we propose a novel attention embedded residual long short-term memory (LSTM) fully Convolutional Network (FCN) to perform the clinical predictions of inpatients’ length of stay (LoS) and mortality. The proposed model is uniquely composed of a convolutional neural network (CNN) layer, three residual blocks, an LSTM unit, an FCN module, and a self-attention module. This innovative architecture allows for comprehensive feature extraction, where the CNN and residual blocks enhance clinical data features, the FCN and LSTM separately extract spatial and temporal features, and the self-attention mechanism focuses on pertinent information while filtering out noise. By optimizing the loss function to address class imbalance and overfitting, our model ensures robust and accurate predictions. Experimental results demonstrate that the proposed model outperforms state-of-the-art methods, validating its effectiveness and feasibility in inpatient length of stay and mortality prediction.
Recommended Citation
J. Chen, M. Li, M. Milosevich, T. Le, A. Bahsoun and Y. Wen, "Inpatient Length of Stay and Mortality Prediction Utilizing Clinical Time Series Data," in IEEE Access, vol. 13, pp. 74724-74738, 2025, https://doi.org/10.1109/ACCESS.2025.3563199.
Copyright
The authors
Creative Commons License
This work is licensed under a Creative Commons Attribution 4.0 License.
Comments
This article was originally published in IEEE Access, volume 13, in 2025. https://doi.org/10.1109/ACCESS.2025.3563199.