Development of Multivariable Prediction Models for 30-Day Risk of Readmission After COPD Hospital Admission: A...

Approximately 20% of patients who are discharged from hospital for an acute exacerbation of COPD (AECOPD) are readmitted within 30 days. Prediction scores are helpful to identify those who are at higher risk of readmission, such that they can be prioritized for readmission-reducing interventions.

To develop and determine the accuracy and precision of a clinical prediction model using data available in electronic medical records to predict 30-day readmission in patients discharged after a hospitalization with an AECOPD.

A dataset was created using all admissions to General Internal Medicine from 2012 to 2018 at seven hospitals in Toronto, Canada. We fit and internally validated models with six algorithms.

Of the 16,314 patients admitted with an exacerbation of COPD, 15.4% were readmitted at 30 days. Top-performing models included LASSO, logistic regression, linear discriminant analysis, and XGBoost with C-statistics of 0.688 ± 0.024, 0.690 ± 0.026, 0.687 ± 0.023, and 0.686 ± 0.022. The four top models had similarly high specificity (96%-98%) with poor sensitivity (14%-20%) at a decision threshold of 50%. At a more aggressive decision threshold of 20%, specificity was less (69%-73%) with a modest improvement in sensitivity (55%-59%). The most important predictor of readmission risk was the number of hospitalizations in the previous year.

We generated clinical prediction models to predict all-cause 30-day readmissions after an acute exacerbation using data from 7 hospitals' electronic medical records. Further work should be done to improve performance, especially sensitivity.