Innovations in Sepsis Detection

Next peer review manuscript on sepsis surveillance was published in collaboration with our Karolinska Institute colleagues. "Predicting sepsis onset using a machine learned causal probabilistic network algorithm based on electronic health records data" was title of article. Cohort of 82,852 hospital admissions and 8038 sepsis episodes was used. Compared to the National Early Warning Score (NEWS2), which is an established method to identify sepsis, the SepsisFinder triggered earlier and had a higher area under receiver operating characteristic curve (AUROC) (0.950 vs. 0.872), as well as area under precision-recall curve (APR) (0.189 vs. 0.149). A machine learning comparator based on a gradient-boosting decision tree model had similar AUROC (0.949) and higher APR (0.239) than SepsisFinder but triggered later than both NEWS2 and SepsisFinder. #Sepsis #ML #alert

Bayesian Health publishes the results of a 2-year trial of its #AI algorithm for early detection of #sepsis, covering over 760,000 patient encounters of which 17,000 developed sepsis in 5 sites over 2 years, showing an 18% reduction in sepsis mortality. In addition to saving people's lives - Sepsis kills 270,000 Americans every year - the importance of these publications is by placing a high bar on: 1. How trials should evaluate medical AI algorithms 2. How clinical workflow integration is a critical component of success 3. How provider adoption is as important as the algorithm/model itself https://lnkd.in/e25iV9P4 #healthcareai #healthai #ethicalai #research

🚀 Improving Neonatal Care with AI 🧠 Proud to share our latest research published in Pediatric Research where we introduce a groundbreaking application of artificial intelligence in neonatal care. We have developed an AI-based graph representation learning (GRL) model to accurately distinguish between culture-negative sepsis (C-NS) and rule-out sepsis (R-OS) in newborns. We utilized electronic health record (EHR) data from over 3,200 antibiotic courses in neonatal intensive care units (NICUs). By encoding clinical events as nodes and using temporal data to create connections, our GRL model captures the intricate, time-dependent relationships within the data. This method allows for an unbiased representation of the clinical course, enabling us to identify subtle differences between C-NS and R-OS cases. Our model demonstrated that temporal patterns, rather than just the presence of certain clinical attributes, are key to distinguishing C-NS from R-OS. The AI-driven approach outperformed traditional methods, which often rely on manual labeling or non-consensus definitions, in identifying these critical distinctions. This scalable model reduces the need for manual data review and opens new avenues for real-time decision support in NICUs. By accurately identifying cases of C-NS, our model has the potential to reduce unnecessary antibiotic use, which is crucial for preventing antibiotic resistance and improving outcomes for vulnerable newborns. Our method could be adapted to tackle other complex conditions in neonatal and pediatric care. https://lnkd.in/dW_73xPK #AI #HealthcareInnovation #NeonatalCare #MachineLearning #MountSinai #Pediatrics

AI tackles antimicrobial resistance in intensive care (ICU), enabling same-day assessment critical to preventing life-threatening sepsis. King's College London. 01 November 2024. Excerpt: Antimicrobial resistance, poses a challenge to healthcare. It is estimated to cause 1.2 million deaths globally. The cost to UK NHS at least £180 million per year. Bloodstream infections can become antibiotic resistant leading to life-threatening sepsis. Once infection has reached sepsis stage there is high probability patients will rapidly develop organ failure, shock, and even death. Some patients have more antimicrobial resistance, due to previous use of antibiotics. Genetics and diet can also alter the microbiome. Kings College London’s Faculty of Life Sciences & Medicine and clinicians at Guy’s and St Thomas’ NHS Foundation Trust collaborated in an interdisciplinary study – to improve outcomes of critically ill patients. Note: The team showed AI and machine learning can provide same-day triaging for ICU patients, in environments with limited resources. The technology was shown more cost-effective than manual testing. Current assessments of ICU patients require lengthy lab tests, requiring bacteria lab cultured, up to five days. This can impact outcomes, given fragility of ICU patients, who may suffer life-threatening illnesses. Earlier access to information would enable clinicians to make quicker, informed decisions. Proper use of antibiotics correlates with positive outcomes. Per Davide Ferrari, King’s College London, “Our study provides further evidence on the benefits of AI in healthcare, relating to antimicrobial resistance and bloodstream infections. It comes at an important time, NHS is investing in shared data resources, to make patient care more collaborative and efficient. Dr Lindsey Edwards, expert in microbiology at King’s College London: “An important way to tackle grave threat of antimicrobial resistance is to protect antibiotics we have with urgent need for fast diagnostics. Often patients with drug-resistant infection present to ICU in critical condition and may not survive long enough for current gold standards of diagnostics to determine specifics of infection. In this event, clinicians prescribe ‘blinded' a broad-spectrum antibiotic to save the patient. “However, this will kill many beneficial microbes without killing the harmful pathogen leading to drug pathogen resistance. “Study findings are promising. Using AI to speed up diagnostics to allow prescription of the correct antibiotic could have a huge impact on survival and care outcomes; preserve antibiotics developed and prevent further antibiotic resistance." Data from 1,142 patients at Guy’s and St Thomas’ NHS Foundation Trust were used in the study, paving the way for ongoing research using datasets of more than 20,000 patients. Additional information enclosed https://lnkd.in/eRcsBGgv

Advancing Sepsis Care: Understanding and Managing Sepsis-Induced Coagulopathy (SIC) This week’s featured article in our newsletter is an important review by Iba and colleagues titled “Sepsis-induced coagulopathy (SIC) in the management of sepsis” (Annals of Intensive Care, 2024). The authors present an updated framework to identify and manage coagulation abnormalities early in sepsis—a critical step given the high mortality associated with sepsis complicated by disseminated intravascular coagulation (DIC). Highlights include: 1) Pathophysiology Insight: SIC represents an early phase of coagulation dysfunction driven by thromboinflammation involving activated leukocytes, platelets, and endothelial injury, which promotes microthrombi formation leading to organ dysfunction. 2) Practical Scoring System: The SIC scoring system, adopted by the ISTH, uses platelet count, prothrombin time (PT-INR), and SOFA score to provide a simple, rapid bedside tool for early detection of coagulopathy before overt DIC develops. 3) Prevalence and Prognosis: Approximately 25% of septic patients develop SIC early, and these patients have significantly higher morbidity and mortality, emphasizing the need for prompt diagnosis. 4) Clinical Utility: SIC criteria help identify patients at high risk of progression to overt DIC and can guide timely therapeutic interventions, including anticoagulant therapies, though further trials are needed to confirm efficacy. 5) Research and Trials: SIC provides a standardized framework for clinical trials, aiding in patient stratification and outcome prediction, which is vital for advancing personalized sepsis management. Early recognition and monitoring of SIC can be a game-changer in sepsis care, offering clinicians actionable insights to improve patient outcomes. As always, don’t forget to like, share, and subscribe. See you on the other side! Interprofessional Critical Care Network (ICCN) Julie Helms, Jerrold Levy

AI That Listens to Nurses? Now We're Talking. For years, nurses have noticed when something feels “off” with a patient—subtle changes that aren’t yet reflected in vitals or labs. We see it in their pallor, their breathing, the way they answer a simple question. But too often, those observations don’t trigger action fast enough. Now, AI is helping to bridge that gap. A study from Columbia University found that the CONCERN Early Warning System—an AI tool that analyzes nursing documentation—identified patient deterioration nearly two days earlier than traditional methods. That’s two days of opportunity to prevent a crisis. 🔗 https://lnkd.in/gB7TsS8P The results? ✅ 35% reduction in mortality risk ✅ Shorter hospital stays ✅ Lower sepsis risk ✅ More timely ICU transfers Here’s why this matters: AI isn’t replacing nursing judgment—it’s amplifying it. The system works because it recognizes the patterns in how nurses assess and monitor patients. It takes our expertise—the instincts honed over years of bedside care—and makes it visible to the entire care team in real time. This is the kind of AI I support in healthcare. Not automation for automation’s sake, but technology that enhances clinical decision-making and improves patient outcomes. The question now: How do we scale innovations like this responsibly? Because if AI can truly help nurses catch deterioration sooner, it shouldn’t be a luxury—it should be the standard. #healthcareinnovation #aiinhealthcare #nurseinformatics #nursingtechnology #patientoutcomes #clinicaldecisionmaking #earlywarningsystems #hospitalcare #nursingdata #machinelearning #sepsisprevention #patientdeterioration #nurseadvocacy #nursesintech #nursingleadership #nursinginformatics #aiethics #healthcarepolicy #evidencebasedpractice #healthcareai #nursesonlinkedin

Want the inside story on what AI tools are being piloted at ScionHealth? This week I chatted with VP of Strategic Operations & Initiatives, Sarah Hughes, MSHA, MBA about how they’re using Eon, LUMINARE, and Operait Health to build out their AI toolkit. — 🟩 Q: What are your biggest tech-related priorities for 2024? 1) Workforce support—How can we deploy technology solutions that make our staff’s lives and daily work easier and more efficient? For example, we are currently deploying a technology platform, M7 Health, which has built a dynamic nursing workforce management system to help us understand and optimize each nurse’s work-life preferences with regard to scheduling and career goals. 2) Patient activation and engagement—How can we better engage and activate patients on both the front end (i.e., scheduling appointments and interacting with our organization) and the back end (i.e., continuing to engage with us to manage their health)? 🟩 Q: Where are you currently exploring AI-driven solutions? We’re running pilots to address a few different administrative and operational efficiency use cases: Incidental findings on radiology reads—We’re working with Eon to automatically scan radiology reads to identify and manage patients with incidental findings, and proactively reach out to them to schedule follow-up care. Sepsis early detection—We’re currently testing LUMINARE's platform, which implements the hospital’s sepsis protocols by automating intervention, streamlining communication, and simplifying reporting and analytics. Operating room scheduling—We’re also looking at a vendor called Operait Health which has built an AI-first solution to optimize surgical demand and right-size OR staffing by forecasting gaps in the OR schedule and placing cases unrestrained by blocks. 🟩 Q: Which areas are you most excited about the potential for new solutions? We’re excited about the potential for virtual remote care in the home on the long-term acute care (LTAC) side. We’ve already seen success with these models in our community hospitals, and we believe there’s potential to similarly deliver a better patient experience at lower cost by bringing care into the home for a subset of our LTAC patients. 🟩 Q: Any general advice for healthcare IT leaders out there? It can be easy to underestimate the legacy IT “tax” of bringing in any new vendor, given that any new solution must work with our existing tech stack. Vendors need to factor this in when discussing implementation timelines, and health systems need to ensure they include all the right internal stakeholders, including operational leaders, early in the conversation to appropriately plan for integrating and deploying solutions into day-to-day operations. — p.s. I loved this conversation because of all the practical wisdom Sarah shared. We're on the lookout for similarly insightful health IT leaders to feature. Who should talk to?

Integrating AI into healthcare is a must if we are to work smart in this healthcare crisis, despite some initial concerns from nurses about whether AI would supersede their decision-making processes. Since 2018, nurses at Aurora, Colorado-based UCHealth have been using AI to detect sepsis, saving thousands of patients' lives. "The statewide Virtual Sepsis program analyzes 2,000 patients a day for early signs of the complication, notifying nurses and physicians when they should take a closer look." By alerting nurses to at-risk patients two to four hours before this deadly complication, this AI tool reduces patient mortality by 30% or more. AI's predictive analytics capabilities are a game-changer in patient care and offer a promising future for improved patient outcomes. Nurses need all the assistance they can get to handle unrealistic workloads and insufficient support. AI has the potential to enhance the support available to nurses, helping to alleviate some of the challenges they and their patients face. The key is integrating AI in a way that supports and enhances nurses' skills rather than replacing their critical human touch. Thoughts? #nursesonlinkedin #nurseinnovation #nurseleaders #nurses #healthcareinnovation

New in Nature Communications, we report a fully printed, chip-less wearable neuromorphic system that integrates sensing and computing for real-time analysis of multiple physiological and biochemical signals. Demonstrated for sepsis diagnosis, this soft, standalone device offers a scalable solution for secure, on-body health monitoring without relying on rigid electronics or cloud-based processing. Congrats to YONGSUK CHOI, Peng Jin, Sanghyun Lee, and the team! Caltech Read the paper: https://lnkd.in/gPyNqQuc Yu Song Roland Tay Gwangmook Kim Jounghyun Yoo Hong Han Jeonghee Yeom Jeong Ho Cho Dong-Hwan Kim Yonsei University 성균관대학교(Sungkyunkwan University)

For patients with sepsis, quickly identifying the bacteria in their blood and what antibiotics they are susceptible to is critical. Jennifer Dionne and her collaborators at Stanford are developing methods using Raman spectroscopy to detect even small numbers of bacterial cells within a droplet of blood and identify those bacteria at the same time. In this clip, we're talking about gathering Raman spectra in the process of acoustic bioprinting where droplets are ejected and printed on a substrate. Those printed droplets can then be assessed by electron microscopy to visually confirm the presence of the bacteria detected as the droplet is in flight. You can listen to the full episode on cc: Life Science here: https://lnkd.in/dXycgfE7 #lifescience #ramanspectroscopy #sepsis #bioprinting