How AI Can Help in Clinical Decision Support: Early Sepsis Detection and Prevention

By Abhishek Ray, Founder & CEO, Finarb Analytics Consulting
LinkedIn: Finarb
LinkedIn: Abhishek Ray

Sepsis is a life-threatening medical emergency triggered by an infection, causing an extreme response in the body. This response can lead to widespread inflammation, tissue damage, and organ failure. It can be caused by bacterial, viral, or fungal infections. Early symptoms include faster heart rate, reduced urine output, fever, chills, and psychosis. Immediate medical attention is crucial as delayed treatment can lead to severe complications.

According to the CDC, at least 1.7 million adults in the US develop sepsis each year, with at least 350,000 adults who develop sepsis dying during hospitalization. Cases of sepsis are very common among children, with an estimated 20 million cases and 2.9 million global deaths in children under 5 years of age. Moreover, about 85% of sepsis related deaths occurred in low- and middle-income countries.

As such, the economic impact of sepsis is staggering, with the average cost of hospital admission per patient estimated at $18,400. This results in a combined total cost of over $20.3 billion each year for sepsis hospitalizations alone.

Given the high mortality rate and the significant economic burden, early detection and continuous monitoring of sepsis are of paramount importance. Early detection can lead to timely intervention, potentially saving lives and reducing the length and cost of hospital stays. Continuous monitoring can ensure that any changes in a patient’s condition are promptly addressed, further improving patient outcomes.

AI modeling for early sepsis detection and prevention

Artificial Intelligence (AI) solutions can play a pivotal role in this regard. AI can analyze vast amounts of data quickly and accurately, identifying patterns and predicting outcomes. An AI-enabled clinical decision support system that leverages data from various sources, such as electronic health records, laboratory results, and patient monitoring devices, can continuously analyze the patient’s condition and generate real-time risk scores indicating the likelihood of developing sepsis.

The system also provides alerts and notifications to the healthcare professionals, enabling them to intervene proactively and initiate appropriate treatments. The system is dynamic and can be tailored to different prediction windows such as 72 hours, 48 hours, 24 hours, 12 hours, and 6 hours ahead of the anticipated onset of sepsis. This can allow healthcare professionals to intervene earlier, potentially preventing the onset of sepsis and improving patient outcomes.

A solution covering the entire spectrum of data-driven healthcare, from data collection and integration to data analytics and predictive modelling, to data visualization and a dashboard that displays the most critical indicators will be of immense value to healthcare professionals. One can use data from a secure data pipeline that integrates data from multiple sources using standardized protocols, such as HL7 to train their models. By applying advanced machine learning algorithms, one can build accurate predictive models that can identify high-risk patients and provide actionable insights. A dashboard with a user-friendly and interactive interface, displaying patient’s status, vitals, risk scores, and top predictors of sepsis in real-time, will allow healthcare professionals to monitor multiple patients and track their progress and outcomes.

In collaboration with a leading Texas hospital, the sepsis detection application, with the Sepsis Early Detection Algorithm (SEDA) at the heart of it was developed. It predicts sepsis risk dynamically based on collected data. For a time, window of 24 hours, that is, making sepsis prediction 24 hours before its onset, the model was able to achieve an AUC-ROC score of 90%, and for narrower time windows, the AUC-ROC score achieved was as high as 94%.

The model demonstrates a high level of accuracy in predicting sepsis, particularly in the critical hours leading up to its onset. This could potentially allow for earlier intervention and improved patient outcomes.

Such a solution can be applied to other clinical scenarios as well where early detection and intervention are crucial such as cardiac arrest, stroke, kidney failure, and more. The solution can be easily customized and adapted to different use cases, data sources, and clinical parameters, and can provide a comprehensive and holistic view of the patient’s health and risk factor. In addition, one can also integrate it with other existing systems and platforms such as electronic health records, hospital information systems, and telemedicine applications, to enhance the efficiency and effectiveness of healthcare delivery.

AI models can be immensely helpful in managing sepsis effectively, from its early onset to timely intervention during treatment. Doctors, and hospital rapid response teams can make timely intervention by continuously monitoring real-time patient data, and probability scores of patients.