User login
Critical Care Network
Sepsis/Shock Section
Early recognition is the linchpin of sepsis management, as mortality from sepsis increases by 4% to 9% for every hour that diagnosis and treatment are delayed.1,2 Artificial intelligence (AI) and machine learning (ML) are increasingly featured in discussions and publications about sepsis care. Already ML models are embedded in electronic medical records (EMR), driving best-practice advisories that are presented to users.3 Epic, the EMR that serves over half of patients in the US, offers its own proprietary cognitive computing model for early detection.
As ML permeates the critical care space, it is increasingly important that clinicians understand the limitations of these models. Recently Kamran et al (NEJM AI) evaluated the Epic sepsis model with disappointing results after excluding cases already recognized by clinicians. The model achieved a positive predictive value of 5%, and 80% of high-risk sepsis cases were missed.3
An application study by Lilly et al (CHEST) showed that an ML model for clinically actionable events was more accurate with less alarm burden when compared to biomedical monitor alarms or telemedicine systems.4 The clinical utility of this model, however, remains questionable; presumably by the time a patient monitor has alarmed, the term “early recognition” can no longer be applied. In this study a significantly elevated false-positive rate required clinicians to review all cases prior to action.
ML models seem to offer incredible potential to clinicians. How they fit into current practice, however, deserves careful consideration. It may be that we just are not there yet.
References
1. Sepsis Alliance. (2024, June 19). Septic shock. 2024. https://www.sepsis.org/sepsisand/septic-shock/. Accessed September 10, 2024.
2. Djikic M, Milenkovic M, Stojadinovic M, et al. The six scoring systems’ prognostic value in predicting 24-hour mortality in septic patients. Eur Rev Med Pharmacol Sci. 2024;28(12):3849-3859.
3. Kamran F, Tjandra D, Heiler A, et al. Evaluation of sepsis prediction models before onset of treatment. NEJM AI. 2024.
4. Lilly CM, Kirk D, Pessach IM, et al. Application of machine learning models to biomedical and information system signals from critically ill adults. CHEST. 2024;165(5):1139-1148.
Critical Care Network
Sepsis/Shock Section
Early recognition is the linchpin of sepsis management, as mortality from sepsis increases by 4% to 9% for every hour that diagnosis and treatment are delayed.1,2 Artificial intelligence (AI) and machine learning (ML) are increasingly featured in discussions and publications about sepsis care. Already ML models are embedded in electronic medical records (EMR), driving best-practice advisories that are presented to users.3 Epic, the EMR that serves over half of patients in the US, offers its own proprietary cognitive computing model for early detection.
As ML permeates the critical care space, it is increasingly important that clinicians understand the limitations of these models. Recently Kamran et al (NEJM AI) evaluated the Epic sepsis model with disappointing results after excluding cases already recognized by clinicians. The model achieved a positive predictive value of 5%, and 80% of high-risk sepsis cases were missed.3
An application study by Lilly et al (CHEST) showed that an ML model for clinically actionable events was more accurate with less alarm burden when compared to biomedical monitor alarms or telemedicine systems.4 The clinical utility of this model, however, remains questionable; presumably by the time a patient monitor has alarmed, the term “early recognition” can no longer be applied. In this study a significantly elevated false-positive rate required clinicians to review all cases prior to action.
ML models seem to offer incredible potential to clinicians. How they fit into current practice, however, deserves careful consideration. It may be that we just are not there yet.
References
1. Sepsis Alliance. (2024, June 19). Septic shock. 2024. https://www.sepsis.org/sepsisand/septic-shock/. Accessed September 10, 2024.
2. Djikic M, Milenkovic M, Stojadinovic M, et al. The six scoring systems’ prognostic value in predicting 24-hour mortality in septic patients. Eur Rev Med Pharmacol Sci. 2024;28(12):3849-3859.
3. Kamran F, Tjandra D, Heiler A, et al. Evaluation of sepsis prediction models before onset of treatment. NEJM AI. 2024.
4. Lilly CM, Kirk D, Pessach IM, et al. Application of machine learning models to biomedical and information system signals from critically ill adults. CHEST. 2024;165(5):1139-1148.
Critical Care Network
Sepsis/Shock Section
Early recognition is the linchpin of sepsis management, as mortality from sepsis increases by 4% to 9% for every hour that diagnosis and treatment are delayed.1,2 Artificial intelligence (AI) and machine learning (ML) are increasingly featured in discussions and publications about sepsis care. Already ML models are embedded in electronic medical records (EMR), driving best-practice advisories that are presented to users.3 Epic, the EMR that serves over half of patients in the US, offers its own proprietary cognitive computing model for early detection.
As ML permeates the critical care space, it is increasingly important that clinicians understand the limitations of these models. Recently Kamran et al (NEJM AI) evaluated the Epic sepsis model with disappointing results after excluding cases already recognized by clinicians. The model achieved a positive predictive value of 5%, and 80% of high-risk sepsis cases were missed.3
An application study by Lilly et al (CHEST) showed that an ML model for clinically actionable events was more accurate with less alarm burden when compared to biomedical monitor alarms or telemedicine systems.4 The clinical utility of this model, however, remains questionable; presumably by the time a patient monitor has alarmed, the term “early recognition” can no longer be applied. In this study a significantly elevated false-positive rate required clinicians to review all cases prior to action.
ML models seem to offer incredible potential to clinicians. How they fit into current practice, however, deserves careful consideration. It may be that we just are not there yet.
References
1. Sepsis Alliance. (2024, June 19). Septic shock. 2024. https://www.sepsis.org/sepsisand/septic-shock/. Accessed September 10, 2024.
2. Djikic M, Milenkovic M, Stojadinovic M, et al. The six scoring systems’ prognostic value in predicting 24-hour mortality in septic patients. Eur Rev Med Pharmacol Sci. 2024;28(12):3849-3859.
3. Kamran F, Tjandra D, Heiler A, et al. Evaluation of sepsis prediction models before onset of treatment. NEJM AI. 2024.
4. Lilly CM, Kirk D, Pessach IM, et al. Application of machine learning models to biomedical and information system signals from critically ill adults. CHEST. 2024;165(5):1139-1148.