<![CDATA[Electronic Health Records (EHR) are a complete digital store of medical information which contains valuable information about a person’s health status. These data are an important tool for health care professionals to decide the management of the patient, i.e. the need for hospitalization (inpatient care) or outpatient treatment. The importance of EHR data underscores the need for hospitals to act quickly and develop appropriate follow-up for patients. The aim of this study is to assess and compare the performance of two deep learning models, Long Short Term Memory (LSTM) and Gated Recurrent Unit (GRU) for accurate classification of EHR data to predict whether a patient is hospitalized (inpatient or outpatient) based on laboratory test results. This study aims to compare these models to identify the best way to classify patient treatment needs. The data set used in this study is EHR data from a private hospital in Indonesia which contains haematocrit, haemoglobin, erythrocytes, leukocytes, thrombocytes, MCH, MCHC, MCV, age, gender, and source (inpatient or outpatient). The data was pre-processed by labelling, variable selection and handling missing and duplicate values. The data set is split for training and testing the models, LSTM and GRU. Performance evaluation is based on accuracy, training time and other relevant metrics to evaluate both the classification precision and computational efficiency of the models. The classification accuracies of the two models were close to each other with LSTM being 75.57% and GRU being 75.34%. However, GRU showed better training efficiency, executing faster than LSTM model. In the cases where the training time efficiency is the priority, the GRU model is the better choice for its faster speed. However, LSTM performs slightly better in terms of accuracy, especially in classifying outpatient cases, making it more appropriate for applications where accuracy is of high importance.]]>
