The Problem
Each year 50,000 children die in the U.S.,and more than half of those deaths occur in inpatient hospital settings(1). It is estimated that 10% of inpatient childhood deaths are preventable(2).Hospitalized children, particularly infants less than 1 year of age, can decompensate quickly which can lead to a rapid response team (RRT) activation and cardiopulmonary arrest. Predicting which child might decompensate is often difficult. Hospitalized children are often non-verbal,cognitively impaired,or simply lack sufficient insight to alert providers to their clinical deterioration. Providers sometimes can rely on parental/caregiver input,but this is not always available or reliable.Early recognition of clinical deterioration and rapid escalation of care are critically important to prevent “failure to rescue”, defined as mortality after a treatable condition. Optimal strategies for early recognition of pediatric clinical deterioration are not known. Machine learning models may identify markers that can be used to assist health care systems rapidly identify and treat these situations when they occur(3).
Our Solution
We aim to:
- Combine electronic health records(EHR)data with electronically captured, bedside physiologic data to develop a dynamic machine learning based model to facilitate earlier recognition of clinical deterioration;
- Pilot the model on a single pediatric unit for complex pediatric inpatients and develop the workflow to optimize opportunities for early rescue;
- Iterate upon the model for quality improvement and implement and disseminate the model to other DUHS pediatric units.
Impact
The overarching theme of this initiative is to leverage bedside data combined with clinical, diagnostic, demographic, and laboratory values from the EHR to develop novel machine learning methods that will identify, in real time, children at risk for clinical deterioration. We will then develop a process and workflow to enable earlier identification of these children prior to the requirement for RRT activation or cardiopulmonary arrest that will facilitate safer and more effective deployment of escalated care such as transfer to ICU, initiation of inotropic support, renal replacement therapy, mechanical ventilation, or ECMO/VAD.
References
- When Children Die: Improving Palliative and End-of-Life Care for Children and Their Families. Institute of Medicine (US) Committee on Palliative and End-of-Life Care for Children and Their Families; Field MJ, Behrman RE, editors. Washington (DC):National Academies Press (US); 2003.
- Bamber AR, Mifsud W, Wolfe I, Cass H, Pryce J, Malone M, Sebire NJ.Potentially preventable infant and child deaths identified at autopsy; findings and implications.Forensic Sci Med Pathol. 2015 Sep;11(3):358-64
- https://psnet.ahrq.gov/primers/primer/38/failure-to-rescue
