Related Experiment Videos
Mining high-dimensional administrative claims data to predict early hospital readmissions
Danning He1, Simon C Mathews, Anthony N Kalloo
1Division of Health Sciences Informatics, Johns Hopkins University School of Medicine, Baltimore, Maryland, USA.
Journal of the American Medical Informatics Association : JAMIA
|October 1, 2013
Summary
This study developed a new algorithm using administrative billing codes to predict 30-day hospital readmissions. The model shows improved performance over existing methods, offering a widely applicable and portable solution for healthcare institutions.
Area of Science:
- Health Informatics
- Predictive Analytics in Healthcare
- Hospital Readmission Reduction Strategies
Background:
- Current readmission prediction models often rely on administrative and clinical data, but frequently exhibit suboptimal predictive performance (Area Under the Curve [AUC] < 0.70).
- There is a need for more accurate and accessible methods to identify patients at high risk of 30-day hospital readmission.
Purpose of the Study:
- To develop and validate an administrative claim-based algorithm for predicting 30-day hospital readmissions.
- To utilize standardized billing codes and essential admission characteristics available prior to discharge for prediction.
- To assess the algorithm's performance across different patient cohorts and healthcare institutions.
Main Methods:
- An algorithm was developed to leverage high-dimensional administrative claims data, automatically selecting empirical risk factors.
- The algorithm was applied to index admissions for general medical patients and patients with chronic pancreatitis (CP).
- Model training and validation were conducted using data from two distinct hospital settings (The Johns Hopkins Hospital and Johns Hopkins Bayview Medical Center).
Main Results:
- The algorithm identified predictive models using a subset of available billing codes (International Classification of Diseases, 9th Revision, Clinical Modification [ICD-9-CM] diagnoses and procedures, Current Procedural Terminology [CPT] codes).
- Models comprised 18 attributes for the medical cohort and five for the CP cohort.
- Achieved Area Under the Curve (AUC) values of ≥0.75 for the medical cohort and ≥0.65 for the CP cohort in both within-site and across-site validations.
Conclusions:
- A widely applicable and institutionally portable algorithm for predicting 30-day readmissions has been developed.
- The algorithm demonstrates comparable performance to existing state-of-the-art models that necessitate clinical data.
- This approach offers a promising tool for proactive readmission risk management in diverse healthcare settings.