ROLE OF ARTIFICIAL INTELLIGENCE FOR DIAGNOSING EARLY SEPSIS IN ICU SETTING

Abstract

Background: The evidence-based treatments of patients diagnosed with sepsis are a key aspect of healthcare today; however, early detection of the sepsis process and/or identification of clinical signs and symptoms is necessary to provide timely, appropriate sepsis treatment. Early detection of the sepsis process is challenging due to the subtleties associated with the presenting signs and symptoms, making it difficult to identify patients with developing sepsis based on small physiological changes over time. Machine-Learning Models Recently, advanced Machine-Learning models utilizing Artificial Intelligence to analyze large volumes of clinical data have provided an exciting new way to identify patterns and predict patients who are at high risk of developing sepsis complications before there are any clinical findings. The addition of Artificial Intelligence to the ICU monitoring systems provides a valuable enhancement to clinical decision-making, allowing for more timely initiation of therapy. Aim: To evaluate the effectiveness of artificial intelligence–based predictive algorithms in identifying early sepsis among patients admitted to the intensive care unit.

Methods: A large Indian hospital's Department of Medicine and Critical Care performed an observational, analytical, prospective study on 150 of their adult patients who had been admitted to the ICU within a 24 month time frame with an infection/risk factor(s) for sepsis. Clinical information, such as vital signs, laboratory test results, and organ function, was tracked over time using an electronic medical record (EMR) system. They also tracked the patient's clinical data in a computerized AI algorithm to estimate the likelihood of sepsis based on real-time clinical data and to generate early alerts for physicians when patients were suspected of developing sepsis. The study evaluated the results of the AI algorithm by analyzing sensitivity, specificity, predictive values, and the time lost between the prediction and the clinical confirmation of sepsis. The accuracy of the AI algorithm was also evaluated by comparing it to the existing standards used for diagnosing sepsis.

Results: Of 150 total patients included in this study, 68 patients were identified with clinically confirmed sepsis during their stay in the ICU. The artificial intelligence algorithm was able to identify early sepsis in 61 out of these 68 patients presented with a sensitivity of around 89.7%, which means that it had identified nearly all instances of early sepsis in these patients prior to clinical diagnosis. Additionally, the specificity of the artificial intelligence system for identifying non-septic patients was 87.8% and, therefore, approximately 87.8 percent of the patients who did not have sepsis were accurately diagnosed as being non-septic by the algorithm. In most cases, the early alert generated by the algorithm was approximately five hours earlier than the clinical diagnosis made by the treating physician. This early identification was helpful in allowing for the rapid initiation of antimicrobial therapy and supportive care, which contributed to greater stabilization of the hemodynamic status of many patients and prevented a number from progressing into septic shock.

Conclusion: The use of artificial intelligence to support predictive models for early identification of sepsis in an intensive care unit (ICU) setting has great promise. AI algorithms use a patient's clinical and laboratory information to continuously analyze for evidence of early physiological changes that indicate deterioration in condition and enable healthcare providers to provide timely treatment. Incorporating artificial intelligence into existing ICU monitoring systems can improve early identification of sepsis, help optimize the management of sepsis and result in a decrease in the incidence of sepsis caused morbidity and mortality.