Fellow: John M. Morrison, John Hopkins All Children's Hospital
Article: Leary JC, Price LL, et. al. Developing Models for 30-Day Unplanned Readmission Among Children With Medical Complexity. Hospital Pediatrics. 2019. 9(3):201-208
Summary: Children with medical complexity including those with multiple complex chronic conditions (CCCs) have a greater risk of readmission to the hospital upon discharge and disproportionate use of medical services. Creating methods for identifying the most at-risk patients could lead to targeted interventions for reducing the burden of readmissions on families and health care systems. In this article, Leary et. al. developed a multivariate model combining sociodemographic, admission-related, and discharge-related factors. Their findings suggest that a recent admission or ED visit (within 6 months), number of CCCs, and a non-surgical admission place a child at the greatest risk for readmission (Adjusted OR 1.7-2.3; c-statistic 0.65). This model was further modified to create a more clinically useful points-based system that could be used to identify patients most at-risk for readmission in real time.
What are the key strengths of the article? This study utilized a large dataset of unplanned readmissions to identify unique factors associated with an increased risk of readmission. Study authors then condensed the results of complex statistical methodology into a simplified clinician-friendly scoring system that could theoretically be used by clinicians at the time of patient admission. This study also included other potentially relevant factors related to social determinants of health such as income and public insurance status in their initial model. However, these factors were ultimately not associated with an increased risk of readmission.
Are there any limitations or flaws in the article? Study authors identified that the prediction model offered only a modest level of discrimination. The study is further limited by the model’s reliance on information available only in the electronic health record (EHR). There likely are additional data not currently captured in the EHR that would be valuable to the readmission model.
What is the major takeaway message? Predictive modeling can stratify an individual’s risk for readmission in a clinically meaningful manner. Simplified versions of this modeling could alert clinicians in real-time prior to discharging a patient of a heightened risk for readmission. This timely information could be used in future scenarios to develop readmission mitigation strategies and allocate limited hospital resources to efforts that reduce overall health system use.
Describe how this article should impact our practice: The full impact of predictive modeling of readmissions is currently unknown as interventions incorporating these powerful tools are lacking. Early models such as that shared by Leary et. al. open the door for multiple scholarly endeavors. Future work identifying modifiable factors upon which clinicians and health care systems can act is necessary for our field to fully realize the potential of real-time risk prediction tools.
Article: Leary JC, Price LL, et. al. Developing Models for 30-Day Unplanned Readmission Among Children With Medical Complexity. Hospital Pediatrics. 2019. 9(3):201-208
Summary: Children with medical complexity including those with multiple complex chronic conditions (CCCs) have a greater risk of readmission to the hospital upon discharge and disproportionate use of medical services. Creating methods for identifying the most at-risk patients could lead to targeted interventions for reducing the burden of readmissions on families and health care systems. In this article, Leary et. al. developed a multivariate model combining sociodemographic, admission-related, and discharge-related factors. Their findings suggest that a recent admission or ED visit (within 6 months), number of CCCs, and a non-surgical admission place a child at the greatest risk for readmission (Adjusted OR 1.7-2.3; c-statistic 0.65). This model was further modified to create a more clinically useful points-based system that could be used to identify patients most at-risk for readmission in real time.
What are the key strengths of the article? This study utilized a large dataset of unplanned readmissions to identify unique factors associated with an increased risk of readmission. Study authors then condensed the results of complex statistical methodology into a simplified clinician-friendly scoring system that could theoretically be used by clinicians at the time of patient admission. This study also included other potentially relevant factors related to social determinants of health such as income and public insurance status in their initial model. However, these factors were ultimately not associated with an increased risk of readmission.
Are there any limitations or flaws in the article? Study authors identified that the prediction model offered only a modest level of discrimination. The study is further limited by the model’s reliance on information available only in the electronic health record (EHR). There likely are additional data not currently captured in the EHR that would be valuable to the readmission model.
What is the major takeaway message? Predictive modeling can stratify an individual’s risk for readmission in a clinically meaningful manner. Simplified versions of this modeling could alert clinicians in real-time prior to discharging a patient of a heightened risk for readmission. This timely information could be used in future scenarios to develop readmission mitigation strategies and allocate limited hospital resources to efforts that reduce overall health system use.
Describe how this article should impact our practice: The full impact of predictive modeling of readmissions is currently unknown as interventions incorporating these powerful tools are lacking. Early models such as that shared by Leary et. al. open the door for multiple scholarly endeavors. Future work identifying modifiable factors upon which clinicians and health care systems can act is necessary for our field to fully realize the potential of real-time risk prediction tools.