A logistic regression model incorporating TRVmax, frequent exacerbations, and non-invasive ventilation use achieved an AUC of 0.825 in external validation for predicting 90-day readmission in elderly patients with COPD and pulmonary heart disease.
Observational (n=588)
Yes
Can a prediction model accurately forecast 90-day COPD/PHD-related readmission in elderly patients with COPD complicated by pulmonary heart disease?
A logistic regression model incorporating TRVmax, frequent exacerbations, and non-invasive ventilation use accurately predicts 90-day readmission in elderly patients with COPD and pulmonary heart disease.
Effect estimate: AUC 0.825
Background Older individuals developing chronic obstructive pulmonary disease (COPD) and concurrent pulmonary heart disease (PHD) have a higher early readmission probability post-discharge. But validated and interpretable models for forecasting 90-day COPD/PHD-related readmission among such individuals remain scarce. Methods This retrospective two-center study recruited ≥ 65-year-old patients developing COPD and concurrent PHD from Affiliated Hospital of North Sichuan Medical College (development cohort) as well as Hospital of Integrated Traditional Chinese and Western Medicine of Dazhou City (external validation cohort). Candidate predictors were selected from pre-discharge routine data during index hospitalization, encompassing demographics, comorbidities, laboratory tests during the first 24 h post-admission, primary transthoracic echocardiographic parameters obtained during index hospitalization, and in-hospital non-invasive ventilation (NIV) use. We excluded troponin T and procalcitonin out of analysis because of massive missingness. We utilized multiple imputation for processing other missing data via chained equations. LASSO logistic regression was carried out for predictor selection by 10-fold cross-validation. Five models, consisting of logistic regression (LR), XGBoost, linear support vector machine, naive Bayes, and decision tree, were constructed, followed by performance comparison. Following calibration of model probabilities, the threshold acquired based on the development cohort was applied in internal bootstrap and external validation. Additionally, model performance was evaluated by calculating area under the receiver operating characteristic (ROC) curve (AUC), accuracy, sensitivity, specificity, as well as Brier score. At last, SHAP analysis and nomogram development were used for the final model. Results There were altogether 588 patients enrolled, involving 433 and 155 in the development and external validation cohorts separately. XGBoost had the best apparent performance in the development cohort, however, LR showed the highest external discrimination (AUC 0.825) and stable calibration for 90-day COPD/PHD-related readmission. The SHAP analysis-identified predictors were TRVmax, frequent exacerbations, and NIV use. Conclusion LR demonstrates the highest robustness and generalizability in terms of its performance, meanwhile, it maintains its clinical interpretability and transparency. Therefore, it can be applied in discharge planning for 90-day COPD/PHD-related readmission among old COPD patients with PHD.
Zhang et al. (Thu,) conducted a observational in COPD complicated by pulmonary heart disease (n=588). Logistic regression prediction model was evaluated on 90-day COPD/PHD-related readmission (AUC 0.825). A logistic regression model incorporating TRVmax, frequent exacerbations, and non-invasive ventilation use achieved an AUC of 0.825 in external validation for predicting 90-day readmission in elderly patients with COPD and pulmonary heart disease.