MEWS: a clinical model to predict probability of maternal sepsis trigger in obstetrics patients

Unmet Need

Sepsis is a life-threatening emergency characterized by an overactive and improper inflammatory response by the body to an infection or injury leading to tissue damage, organ failure and death if not treated promptly. Maternal sepsis is defined as sepsis that occurs during pregnancy, delivery, post birth or abortion. In 2019, maternal sepsis was estimated to occur in 0.04% of deliveries and be involved in 23% of all maternal deaths in the US. Successful treatment of sepsis depends on administering antibiotics as soon as possible, ideally within a few hours of detection. Surgery may be needed if damaged tissue needs to be removed. However, various factors such as non-specific symptoms like fever, chills, fatigue, atypical presentations associated with pregnancy, and clinical complexity can delay detection leading to failure to treat patients in time. There is a need for solutions that help identify patients at high risk for sepsis and facilitate prompt treatment improving patient outcomes and mortality.

Technology

Duke inventors have developed a clinical predictive model software called Maternal Early Warning System (MEWS) Predicting Sepsis Trigger that predicts patient risk of sepsis starting from entry to an obstetric hospital unit. This is intended to be used by clinicians to identify high-risk patients for further sepsis assessment. Specifically, the model uses patient electrical health record data as input and runs on an hourly basis from the start of an admission to the obstetric unit to estimate the probability of the patient meeting sepsis trigger criteria within 4 hours. The model stops running if sepsis has been triggered or the patient has been discharged from the OB unit. This has been demonstrated using training data from patients aged 9-60 years admitted to the Duke University Hospital (DUH) and Duke Regional Hospital (DRH) between January 2016 and December 2020 and evaluated on patients in the same age range admitted to DUH and DRH between January to December 2022. This evaluation demonstrated high model accuracy with high performance metrics of 0.924 and 0.927 AUC respectively (model accuracy increases as AUC approaches 1).

Other Applications

This technology could also be adapted to predict sepsis probability in patients in other hospital units.

Advantages

• Early alerts to clinicians about high-risk patients, reducing the time to sepsis detection and treatment.
• Provides continuous patient monitoring without increasing clinician burden.
• High accuracy model offering a robust of identification of high-risk patients.
• Improves hospital resource efficiency by focusing sepsis assessments on high-risk patients.