The prevention of hospital readmissions, particularly preventable or unnecessary readmissions, has become an area of interest for public policy makers, health insurers, and providers. Hospital and state pediatric readmission rates vary widely, with some conditions, such as premature birth, varying by over 600%. Identifying patients at highest risk of readmission allows providers to develop interventions to reduce or eliminate the chance of a hospital readmission. However, most methods to predict the risk of readmission for both adult and pediatric patients frequently misclassify patients. Most of these algorithms rely on medical factors identified in hospital administrative data, such as the reason for hospitalization and the presence of co- existing medical conditions, as a key component of a risk calculation model. This feature ignores (1) the condition of the child at both the time of admission and the time of discharge; (2) features of the outpatient management, such as access to health care and the quality of the outpatient provider; and (3) more detailed measures of familial structure, support, and resources to care for a sick child that may be additional risk factors for readmission. As a result, pediatric models for both all-cause readmissions and specific models in the prematurely-born infant have poor discrimination, with c-statistics between 0.6 and 0.7, and consequent high misclassification. The principal goal of this study is to develop a real-time predictor of readmission risk for pediatric patients. This proposal will use two innovative approaches to develop and validate this tool. First, this project will link inpatien hospitalization data from the Pediatric Health Information System, with inpatient hospital records from 43 Children's Hospitals that care for 22% of pediatric hospitalizations from across the United States, with outpatient insurance data either from the Medicaid Analytic Extract files for children with Medicaid insurance or from The Health Care Cost Institute for children with private insurance or managed-care Medicaid. This broad cohort of patients will provide information not only on medical risk, but improved information on illness severity, severity of co-existing health conditions, and access to quality outpatient care. Second, this project will adapt the Psychosocial Assessment Tool, a 7-scale tool to assess family risks and resources, including family structure, emotional and behavioral concerns, marital/family problems, beliefs, and other stressors, from oncology to the general pediatric population. With the help of a patient/provider advisory committee, we will then develop readmission risk prediction models to allow providers in real-time to identify those children at highest risk for hospital readmission, and to target interventions to reduce this risk.
The prevention of hospital readmissions, particularly preventable or unnecessary readmissions, has become an area of interest for public policy makers, health insurers, and providers. However, most methods to predict the risk of readmission for both adult and pediatric patients frequently misclassify patients. The principal goal of this study is to develop a real-time predictor of readmission risk for pediatric patients.
