A Statistical-Learning Model for Unplanned 7-Day Readmission in Pediatrics

Area of Science:

• Pediatric healthcare quality improvement
• Clinical informatics
• Machine learning in medicine

Background:

• Pediatric 7-day unplanned readmissions are a critical indicator of healthcare quality.
• High readmission rates necessitate improved care strategies and patient management.
• Machine learning offers a powerful tool for analyzing complex health data to identify risk factors.

Purpose of the Study:

• To utilize machine learning on electronic health records to identify significant predictors of pediatric 7-day unplanned readmissions.
• To rank these predictors by clinical significance using least absolute shrinkage and selection operator (LASSO) regression coefficients.

Main Methods:

• A dataset of 50,241 inpatient and observation encounters from a tertiary pediatric hospital was analyzed.
• A LASSO regression model was developed using 50% of the data and validated on the remaining 50%.
• Variables included demographics, social determinants of health, illness severity, diagnoses, medications, and psychosocial factors.

Main Results:

• The study identified previous hospitalizations, readmissions, medications, multiple comorbidities, and longer lengths of stay as significant predictors.
• Previous emergency department use and specific diagnoses also modified readmission risk.
• The predictive model achieved an area under the curve of 0.778, indicating good performance.

Conclusions:

• Key predictors of readmission include medications, healthcare resource utilization, prior readmissions, illness severity, and psychosocial factors.
• Many identified predictors are unmodifiable, emphasizing the need for patient- and parent-centered discussions.
• Intervention strategies should focus on identifying and addressing modifiable underlying causes of readmissions in high-risk pediatric patients.