Beyond the Alarm: Proactive Predictions for Cardiac Arrest Incidents in Hospitals Using Interpretable Machine Learning Models

Journal: Journal of the Operational Research Society

Synopsis: Code Blue is an emergency alert code, activated when any in-hospital patient faces cardiac or respiratory arrest. Many hospitals have a Code Blue team, which responds as soon as the Code Blue is initiated and performs cardiopulmonary resuscitation (CPR). In a hospital, cardiac arrest varies from 1 to 5 events per 1000 patients. Globally, Cardiovascular diseases are the leading cause of death. In the United States, more than 290,000 adults suffer from cardiac arrests in hospitals. Code Blue calls for immediate action based on standard protocols by the entire hospital administration including the emergency room physician, anesthetist, floor nurse, and doctor.

In a hospital, a cardiac arrest patient’s survival depends on the early identification of signs like unresponsiveness or pulselessness and, the initiation of CPR at the earliest. Thus, early prediction of Code Blue will help in reducing the burden on hospital administration.

In this paper, the authors build a machine-learning model to predict code blue incidents. Patients under continuous monitoring show at least some deteriorating vital signs before cardiac arrest.  The authors observed that the patients who had cardiac arrest showed higher variance in vital signs as compared to other patients. This study uses the electronic medical record data of the last 24 hours along with the doctor’s clinical notes. They use the technique of Natural Language Processing (NLP) to extract features from the doctor’s clinical notes. The extracted features combined with other data are used to build the final prediction model. Although machine learning predictive models have high predicting power, they lack explaining ability. It becomes difficult for healthcare providers to understand these models as healthcare involves a lot of complexity and decision-making at different stages. They use Explainable Artificial Intelligence (XAI) methods to interpret the model. The enhanced explaining ability of the model will help healthcare professionals better understand the cause of Code Blue. The prompt alert from an early warning signal ensures that patients receive immediate care from medical practitioners, thereby substantially increasing the potential for saving lives. Based on these findings, hospital managers will be able to reframe the policy regarding the announcement of Code Blue.