**5. AI and ML for vital sign monitoring**

Today's healthcare, especially in the hospital setting, is complex, fast-paced, and busier than ever. Physicians make many individual decisions and treatment plans that are influenced by copious amounts of data that are collected and available for review in the EHR. Hospitalized patients are monitored frequently through vital signs and lab tests for signs of deterioration or instability. There is now a desperate need to automate this essential job and quickly alert clinicians if there are any signs of deterioration.

With the advancement of technology, artificial intelligence (AI) and machine learning (ML) algorithms are being used to analyze vital sign data and detect signs of disease in real-time, improving the accuracy and speed of diagnosis [39]. Conditions, such as sepsis, are commonly managed in the hospital setting and are the leading cause of inhospital death [40]. Traditionally, clinicians have relied on scoring systems such as the modified early warning score (MEWS), SIRS, Rothman index, sequential organ failure assessment score (SOFA), and quick SOFA (qSOFA) to identify patients at risk of deterioration. These scores utilize several data points from the patient's record to predict the risk of deterioration. However, due to their high sensitivity and low discriminatory ability, these scores may identify a larger number of patients at risk than present [39, 41].

Studies have concluded that individual machine learning models can predict sepsis onset ahead of time and with more accuracy compared directly with the traditional sepsis screening tools such as SIRS, MEWS, and SOFA scores [39, 41]. From a clinical perspective, ML models are particularly useful as they could trigger earlier detection of sepsis and allow for early antibiotic administration leading to decreased mortality. Some additional studies have also highlighted earlier predictions of severe deterioration in sepsis utilizing only vital signs. For instance, Mao et al. developed a gradient tree boosting model using data from only six vital signs: systolic BP, diastolic BP, heart rate, respiratory rate, peripheral capillary oxygen saturation, and temperature. This model was able to predict sepsis at the onset with high AUC (0.92) and septic shock 4hours in advance with a high AUC (0.96). The model was also able to predict severe sepsis 4hours in advance with a higher AUC (0.85) than the onset time for statistically calculated SIRS AUC (0.75) [42].

Additionally, AI-based monitoring incorporated into the EHR can facilitate the use of large volumes of data for the prediction of mortality in hospitalized patients. Shickel et al. used a modified recurrent neural network model on temporal intensive
