Developing a Pediatric Readmission (PERE) Risk Prediction Tool for Decision Support During Pre-Discharge Planning

点击翻译按钮获取中文摘要

Project Summary/Abstract Pediatric Potentially Preventable Readmissions (“PPRs”) take a toll on the quality of life of young patients and their families and also place extra strain on health care systems that care for children. Accurate readmission risk prediction algorithms can aid clinicians in identifying patients most at risk for PPRs and thereby facilitate improved identification of patients most in need of more intensive discharge planning. Such focused risk assessments help with allocation of staffing and resources. Unfortunately, existing PPR risk predictions rarely explicitly consider pediatric patients and are not highly accurate. With recent advances in machine learning and deep learning, assessment of a large range of risk factors for pediatric PPRs is possible. To that end, our primary objective is to develop a suite of Pediatric Readmission (“PERE”) risk prediction algorithms to predict risk of a PPR within three (3), seven (7), and thirty (30) days of discharge from a child’s current inpatient stay. We will do this by leveraging recent advances in statistical data science algorithms. We will also develop PERE predictions such that they can be used to assess risk for PPRs prior to and during the discharge planning phases of a patient’s current inpatient stay. In environments of limited staffing and resources, the PERE risk prediction suite will support clinicians with a means of identifying patients most at risk for PPRs who may be candidates for increased discharge planning related care. Relevance We will focus our PERE risk prediction efforts on an AHRQ priority population: Children. Children represent a vulnerable population due to their developing physiology, limited communication abilities, and reliance on a caregiver to appropriately understand instructions and administer care after discharge.1,2 Early identification of risk for PPRs in three (3), seven (7), and thirty (30) days in this vulnerable population will alert hospital nurses and care managers of the increased risk a particular patient may face that can be more fully attended to during discharge planning. By using deep-learning and other emerging machine learning methods to optimize risk prediction, this proposal also supports AHRQ’s mission to harness data and technology to improve health care quality and patient outcomes and to facilitate improved measurement, monitoring, and surveillance of patient risk. This technology will be implemented using data from a very large pediatric patient care database to develop and validate PERE risk predictions.