Researchers have developed artificial intelligence models that can predict with high accuracy which children are at risk of developing sepsis, a life-threatening condition caused by the body’s extreme response to an infection. The new models, created by scientists at Northwestern University and Ann & Robert H. Lurie Children’s Hospital of Chicago, use data routinely collected in the first few hours of a child’s visit to the emergency department to identify those at high risk for sepsis within 48 hours, allowing for early, preemptive care. This development is a significant step toward precision medicine for pediatric sepsis, which is a leading cause of death in children globally.
The study, published in JAMA Pediatrics, is the first to use AI models to predict sepsis in children based on the new Phoenix Sepsis Criteria. Before this research, predictive models had not been successful in improving early diagnosis. The new models demonstrated a strong ability to identify children who would later develop sepsis without incorrectly flagging those who were not at risk, a crucial balance for avoiding unnecessary and aggressive treatments. This is a major advancement in the field, as previous efforts to predict sepsis using machine learning had not been specifically applied to the broader pediatric inpatient population using electronic health record (EHR) data.
A Multi-Center Approach to Model Development
The research team utilized a large and diverse dataset to develop and validate the AI models. The study included five health systems that contribute to the Pediatric Emergency Care Applied Research Network (PECARN), which gave the scientists access to a wide range of patient data. The researchers retrospectively analyzed data from emergency department visits between January 2016 and February 2020 to discover the machine-learning models. They then tested the models on data from 2021 to 2022 to confirm their performance. In both phases of the study, the models used data from the first four hours of care to predict the likelihood of sepsis developing in the following two days.
Utilizing Electronic Health Records
A key aspect of the new AI models is their reliance on EHR data that is routinely collected. This means that the models can be implemented in hospital settings without the need for additional, specialized tests or data collection, which could create extra costs. The use of existing EHR data is a common and effective approach in developing machine learning models for sepsis prediction, as it allows the algorithms to analyze a wealth of information, including clinical notes and physiological data, to identify patterns that may not be apparent to human observers. The models in this study focused on data from the initial hours of an emergency department visit, a critical window for intervention.
The Phoenix Sepsis Criteria
A novel aspect of this research is the use of the new Phoenix Sepsis Criteria. These criteria were developed to provide a more accurate and consistent definition of sepsis in children, which has been a challenge in the past. The successful application of these criteria in the AI models represents a significant advancement. By training the models on these new, more precise criteria, the researchers were able to create a tool that is better attuned to the specific ways that sepsis presents in children. This helps to overcome some of the limitations of previous sepsis prediction tools that were often based on less specific criteria, such as the Systemic Inflammatory Response Syndrome (SIRS) criteria.
Implications for Clinical Practice
The successful development of these predictive models has the potential to transform how pediatric sepsis is managed in emergency departments. By identifying high-risk patients early, clinicians can initiate preemptive care, which could include closer monitoring and earlier administration of antibiotics and other treatments. This is crucial because the rapid progression of sepsis means that even a short delay in treatment can have severe consequences. The models’ ability to distinguish between high-risk and low-risk patients is also vital for avoiding the overuse of antibiotics, which is a major contributor to the growing problem of antimicrobial resistance.
A Step Toward Precision Medicine
Dr. Elizabeth Alpern, the corresponding author of the study, described the predictive models as “a huge step toward precision medicine for sepsis in children”. Precision medicine aims to tailor medical treatment to the individual characteristics of each patient. In the context of sepsis, this means identifying which children are most likely to benefit from aggressive early treatment, while sparing others from the potential harms of unnecessary interventions. The use of AI to create more granular categories of patients is a growing trend in sepsis research, as it can help to better define the condition and develop more targeted therapies. These new models are a prime example of how AI can be used to achieve this goal.
Challenges and Future Directions
While the results of this study are promising, there are still challenges to be addressed before these AI models can be widely implemented in clinical practice. One major challenge is ensuring that the models perform well across different hospitals and patient populations. Although this study used data from multiple health systems, further validation in a wider range of settings will be necessary. Additionally, the integration of these models into clinical workflows will require careful planning and training to ensure that they are used effectively by healthcare providers. Future research will likely focus on refining the algorithms with even more diverse datasets and exploring the use of real-time data to provide continuous risk assessment for patients.
