UCSD Health study reveals AI’s potential in reducing sepsis mortality
Sepsis, a critical and often fatal response to infection that causes widespread inflammation and organ damage, is a major health concern, claiming the lives of approximately 350,000 Americans annually. Timely detection is crucial for effective treatment, involving prompt administration of antibiotics and intravenous fluids to stabilise the patient.
Researchers at UC San Diego Health have been exploring the potential of artificial intelligence (AI) to enhance early sepsis diagnosis. Their internally developed AI system, named COMPOSER, utilises machine learning and is trained on over 100,000 digital patient records from previous sepsis cases. A recent study published in the journal npj Digital Medicine has demonstrated the potential of COMPOSER in reducing mortality rates.
COMPOSER operates by analysing electronic health records of emergency patients at UCSD every hour. It evaluates various parameters, including medication histories and recent vital statistics, to identify individuals who may be in the initial stages of sepsis. This approach is particularly beneficial in ambiguous cases, where symptoms do not distinctly indicate sepsis.
Dr. Gabriel Wardi, a co-author of the study and a specialist in emergency medicine and sepsis, highlights the algorithm’s significance in situations where diagnostic clarity is lacking. The system acts as an additional tool for medical professionals, suggesting further examination of patients at risk of developing sepsis, thereby facilitating timely intervention.
The study involved 6,217 emergency patients at UCSD’s Hillcrest and La Jolla emergency departments. Comparing the outcomes of 5,000 patients treated before the implementation of COMPOSER with 1,152 patients during its active phase, researchers observed a reduction in sepsis mortality from 11.39% to 9.5%. While these results are promising, they represent correlations rather than direct cause-and-effect, due to the non-randomised nature of the trial.
Karin Molander, director of the Sepsis Alliance, acknowledges the potential benefits of AI in healthcare, particularly in continuous monitoring without human limitations. However, she emphasises the importance of verifying AI-generated recommendations.
The implementation of COMPOSER required careful calibration to minimise false alerts, ensuring that it aids rather than burdens healthcare providers. The system, while capable of predicting sepsis, cannot replace medical professionals, who are responsible for all patient care decisions.
Shamim Nemati, a co-author and UCSD associate professor, notes the challenges in training the algorithm to differentiate between sepsis and similar conditions. The ongoing development of COMPOSER includes enhancements to request additional diagnostic tests when necessary.
UCSD is expanding the use of COMPOSER to include patients admitted to the hospital, and its application is expected to extend to the new East Campus. The integration of advanced wearable sensors and large-language models like ChatGPT into the system is also underway, aiming to improve data accuracy and reduce false alarms.
UC San Diego Health is actively incorporating AI into various aspects of patient care, having appointed its first chief AI officer and collaborating with Microsoft Inc. to utilise AI technologies like ChatGPT in routine patient communications. This initiative is part of a broader strategy to centralise data integration and maximise the benefits of AI in healthcare.