64H—Creating a Readmission Risk Predictor With AI and Cloud Computing
Tuesday, March 29th
About the Session
Children’s Hospital of Orange County (CHOC) developed a pediatric readmission predictive model in 2017 to assist clinicians with reducing unplanned readmissions and to enhance the understanding of risk factors leading to readmissions. The rate of pediatric 7-day and 30-day unplanned readmissions are closely tied to quality of care, with high rates indicative of needed improvements. CHOC applied machine learning algorithms to the electronic health records of approximately 40,000 hospital admissions and subsequent unplanned readmissions to study predictors of both 7-day and 30-day readmissions in pediatrics. The algorithm was able to identify patients at high, moderate and low risk of readmissions. CHOC gained an opportunity for providers to proactively take measures to reduce the likelihood of readmission based on this predicted likelihood of occurrence. The algorithm results were initially viewed in a spreadsheet and were subsequently reported every hour in the electronic medical record (EMR). Knowing which patients are at a higher risk to return and the risk factors contributing to this risk, helps providers decide whether clinical interventions are needed to improve quality of care and prevent readmission. In other words, the predictive model advanced CHOC toward targeted interventions for readmission risks. The probability of readmission score is assessed daily during the morning huddle on 100% of patients and can be reassessed at any time, for example, prior to discharge to ensure any potential condition changes are acted upon to prevent a readmission. The readmission transition of care committee meets monthly to review the current state of readmissions and to monitor for areas of improvement. After years of maintaining an average 7-day unplanned readmission rate around 4%, CHOC was able to decrease the 7-day unplanned readmission rate to approximately 3% with the utilization of the near-real-time readmission predictive score. Similar decreases in the 30-day rate were measured from 2019, decreasing from 12.3% to 11%, both highly significant changes. In this session, attendees will learn how CHOC utilized the EHR, dashboards, population health data tools, and on-demand cloud computing platforms and APIs to successfully predict and reduce readmissions.
- Describe how to utilize machine learning to significantly improve the process for creating predictive algorithms and data management.
- Explain how to push a readmission predictive score to the point of care and to ultimately decrease unplanned readmissions.