Multitask Multilayer Model for Predicting HR, BP Inside ICU as Well as Patient Death

Abstract

Predicting vital signs in real-time has critical impact on predicting life-threatening events such as mortality, morbidity, etc. To ensure the importance of the chosen study, we conduct a bibliometric study that analyzes the utilization of vital signs in mortality prediction. We analyze the authors, keywords, organizations, studies, and citations from 2020 to 2024. Our proposal in this paper is a two-layer multilayer system. To anticipate the patient's heart rate and blood pressure during the following three hours, we start by analysing past vital data. In addition to these critical signs prediction, the patient's data should be used to produce an early death prediction. The MIMIC-III database was used for this study. Along with other metrics like vital signs, lab results, imaging reports, etc., it contains patient data like age, weight, and height. To estimate a patient's heart rate and blood pressure, we suggest a technique that uses a Multitask multilayer LSTM_GRU model to extract signal and statistical information from the patient's vital signs. For the regression task, LSTM_GRU gives the best performance of 22.906, 2.91, and 0.921 for MSE, MAE, and R2 score, respectively. For the classification task, the stacking ensemble model gives the best performance of <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\text{A C C} = \text{0. 9 4 3}$</tex>, <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\mathrm{F}$</tex>-measure <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$=0.955$</tex>, and AUC <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$=0.943$</tex>.