Evaluation of Sepsis Algorithmic Alerts Among Hospitalized Pediatric Patients in Qatar: A Retrospective Cohort Study.

BACKGROUND AND AIMS: Sepsis is a leading cause of morbidity and mortality among children globally and poses a significant challenge for early detection, particularly in pediatric populations where symptoms can be subtle. An automated system for screening algorithm for sepsis is built into the electronic health record (EHR) to assist in identifying sepsis ahead of time. This study aimed to evaluate the accuracy of an algorithmic sepsis alert system for early recognition of true or suspected sepsis among pediatric patients admitted to the Pediatric Intensive Care Unit (PICU) and Pediatric Inpatient Unit at Al Wakra Hospital in Qatar.

MATERIALS AND

METHODS

A retrospective cohort study was conducted using electronic health records of pediatric patients admitted between 01st Jan and 31st Dec 2019. Patients up to 14 years old with algorithmic alerts or clinical suspicion of sepsis were included. The study assessed the performance of algorithmic alerts, clinical suspicion, and physician judgment against defined sepsis criteria using sensitivity, specificity, predictive values, and ROC curve analysis.

RESULTS

Among 1922 patients, with an average age of 30.2 ± 37.0 months and an average hospital stay of 3.7 ± 3.3 days, 56% were male, and 85.2% were admitted to the pediatric inpatient unit. Algorithmic alerts for potential sepsis were positive in 659 patients (34.3%) upon admission, while 35 patients (1.8%) had positive clinical suspicion. Physicians judged 114 patients (5.9%) as having sepsis, and 220 (11.4%) met the criteria for true sepsis. The algorithmic alert had 9.7% sensitivity and 87.6% specificity, while clinical suspicion had 34.3% sensitivity and 89% specificity. Management as sepsis showed 21.5% sensitivity; physician judgment had 26.3% sensitivity. Without alerts, clinical suspicion and management as sepsis had 11.1% sensitivity.

CONCLUSION

The current pediatric sepsis alert system demonstrates low sensitivity but relatively high specificity. Clinical suspicion and physician judgment showed moderate improvements in sensitivity. Adjusting the algorithmic criteria may enhance early detection and reduce delays in management. A combined approach integrating algorithmic alerts with clinical evaluation is recommended to improve sepsis identification in pediatric patients.