Researchers at UC San Diego Health tested COMPOSER, an AI system that uses electronic health record data to flag possible sepsis before clinicians typically recognize it. The 2022 study found earlier detection and reported lower in-hospital mortality among patients flagged by the system, while noting that broader testing is needed to establish how well the approach works elsewhere.
TLDR: A UC San Diego hospital research team tested COMPOSER, an AI system that scans patient records for early signs of sepsis. The 2022 study found it could flag cases hours before usual recognition and reported lower in-hospital mortality among alerted patients. Further research must test its benefits across hospitals and populations.
A hospital research team in California tested an artificial intelligence system designed to spot sepsis before clinicians formally recognize it. The system, called COMPOSER, analyzed routinely collected patient information at UC San Diego Health, including vital signs, laboratory results and medical history. In a study published in Nature Medicine in 2022, researchers reported that the tool could alert clinicians to possible sepsis hours before traditional recognition. The work offers a practical test of how AI might support time-sensitive care inside a busy hospital, rather than simply producing predictions in a laboratory setting. Because the records are updated during routine care, the system could assess changing patterns over time rather than depend on a single measurement. That distinction matters when deterioration emerges gradually.
Sepsis is a dangerous reaction to infection that can lead to organ damage and death. Its early signs, such as changes in temperature, heart rate or blood pressure, can also occur in many less serious conditions. That overlap makes early identification difficult, especially when information is spread across different parts of a patient’s electronic record. COMPOSER was built to combine multiple signals and estimate whether a patient might be developing sepsis. Its output is an alert for medical staff, not a diagnosis or an instruction to begin treatment.
The team trained and evaluated the system using electronic health record data from UC San Diego hospitals. Researchers then assessed its performance in clinical use, where alerts could be reviewed by the care team. The model was designed to account for patterns in the data and reduce alerts triggered by misleading or incomplete information. Such safeguards matter because an alert that is too noisy can distract staff and undermine trust. The researchers reported that the system detected many cases earlier than usual clinical recognition, creating an opportunity for a clinician to reassess a patient’s condition. This kept review within routine clinical work for staff.
The study also examined outcomes after implementation. Researchers reported that patients whose sepsis was identified through the alert process had lower in-hospital mortality than patients whose sepsis was not flagged by the system. The finding is encouraging, but it does not prove that the algorithm alone caused the difference. Earlier recognition may prompt faster evaluation and treatment, while other changes in care, differences among patients, or the way clinicians responded to alerts can also affect outcomes. The study’s results are evidence from a real hospital, not a guarantee that the same effect will occur everywhere.
Moving an algorithm into clinical practice requires more than strong test results. A model can behave differently when patient populations, documentation habits or hospital workflows change. Clinicians need to understand what an alert means, decide whether it fits the patient’s situation, and retain authority over care. Researchers also need to monitor false alarms and missed cases, and check whether performance varies across groups. The UC San Diego team developed the system for local records and clinical processes, so other hospitals would need careful evaluation and adaptation before relying on it.
The study shows how hospital-based AI can be evaluated as part of a care pathway: the system watches for risk, staff interpret the alert, and treatment decisions remain clinical. The next steps include testing the approach across more hospitals and patient populations, measuring long-term effects on outcomes and workload, and determining which responses to an alert make the greatest difference. If those evaluations confirm benefit, tools like COMPOSER could help care teams notice deterioration sooner. Their value will depend not only on prediction accuracy, but also on whether alerts lead to timely, appropriate care.
Susan Carter serves as a Senior Correspondent for Just Right News, where she leads the network’s comprehensive coverage of Health, Medicine, and Public Policy. With a career dedicated to dissecting the complexities of the American healthcare system, Susan brings a principled perspective to the most pressing debates of the day. Her reporting is characterized by a commitment to individual liberty, fiscal responsibility, and a healthy skepticism of government overreach, making her a vital voice for audiences seeking clarity in an increasingly regulated landscape.
Raised in Minneapolis, Minnesota, Susan’s professional outlook is deeply informed by the values of the American heartland. Growing up in the Twin Cities, she witnessed firsthand the importance of community-driven solutions and the resilience of the individual. This Midwestern foundation instilled in her a firm belief that the best answers to public policy challenges often come from local innovation and personal responsibility rather than top-down federal mandates. Her background allows her to translate complex legislative jargon into stories that resonate with families who value common sense and local autonomy.
Now based in Baltimore, Maryland, Susan operates from a unique vantage point at the intersection of medical innovation and urban policy. Living and working in a city renowned for its world-class medical institutions and significant public health challenges, she has a front-row seat to the realities of the modern healthcare landscape. Her proximity to these hubs of research and policy allows her to scrutinize how government intervention affects the quality of care and the freedom of medical professionals. In Baltimore, she sees the tangible results of policy decisions, using the city as a lens through which to examine the broader national health debate and the impact of federal spending on local communities.
As the lead for the acclaimed feature series “The Cost of Care,” Susan dives deep into the economic realities facing patients and providers alike. Her work focuses on the necessity of price transparency, the benefits of market competition, and the preservation of the sacred doctor-patient relationship. She is a staunch advocate for the idea that a well-informed public is the best defense against bureaucratic inefficiency. Through this series, she has exposed the hidden drivers of rising medical expenses, always seeking to empower the individual consumer over the massive institution.
Susan’s reporting on public policy is characterized by a commitment to rigorous inquiry and a focus on constitutional limits. Whether she is analyzing new pharmaceutical regulations, investigating the impact of federal health mandates, or exploring the future of medical technology, her goal remains the same: to provide Just Right News audiences with the facts they need to navigate a complex world. By blending her Midwestern sensibilities with the insights gained from her base in Maryland, Susan Carter has become a trusted authority for those who believe that personal freedom and public health are inextricably linked. Her work continues to challenge the status quo, advocating for a system that respects the taxpayer and prioritizes the patient.