Key result
Machine learning model predicts prolonged hospital stay after PFO closure with ~0.71 AUC.
Why the study?
Can machine learning models utilizing routine clinical and procedural predictors accurately predict prolonged hospital stay (>7 days) in patients undergoing transthoracic echocardiography- or X-ray-guided patent foramen ovale closure?
Observational (n=216)
No
Can machine learning models utilizing routine clinical and procedural predictors accurately predict prolonged hospital stay (>7 days) in patients undergoing transthoracic echocardiography- or X-ray-guided patent foramen ovale closure?
An exploratory random forest model incorporating 10 routine clinical and procedural variables demonstrated moderate discrimination (AUC 0.711) for predicting prolonged hospitalization (>7 days) in patients undergoing TTE- or X-ray-guided PFO closure.
Answers are grounded in this paper and cite their sources.
Exploratory model offers moderate discrimination for prolonged stay after PFO closure; leaves open prospective validation before clinical use.
Patent foramen ovale (PFO) can be managed using different image-guided treatment modalities; however, the determinants of prolonged hospitalization remain insufficiently defined. This retrospective study aimed to identify clinical predictors associated with length of hospital stay and to develop a machine learning model for predicting prolonged hospitalization in patients with PFO. Clinical data from 216 patients with PFO treated at Gansu Provincial People’s Hospital were retrospectively analyzed. Among them, 106 patients underwent transthoracic echocardiography (TTE)-guided treatment and 110 underwent X-ray-guided treatment. According to the median length of stay, patients were divided into a short-stay group (≤7 days) and a prolonged-stay group (>7 days). Baseline variables were compared between groups. Least absolute shrinkage and selection operator (LASSO) regression was used for feature selection. Five machine learning algorithms were then developed and compared in the training cohort, followed by validation in an independent validation cohort. Thirteen baseline variables showed significant between-group differences. LASSO regression retained 10 candidate predictors for model construction. Among the evaluated models, the random forest model showed the best overall performance. In the validation cohort, the model achieved an accuracy of 0.692, a precision of 0.708, a recall of 0.850, an F1-score of 0.773, and an AUC of 0.711, indicating moderate discrimination. The random forest model demonstrated moderate predictive discrimination for prolonged hospital stay. Given the limited sample size and single-center retrospective design, the model should be considered exploratory and requires external validation in larger independent cohorts before clinical use.
No takes yet. Share an insight, caveat, or question.
Zhang et al. (2026) conducted an observational in Patent foramen ovale (PFO) undergoing transcatheter closure (n=216). Transthoracic echocardiography (TTE)-guided PFO closure and machine learning prediction (Random Forest model) vs. X-ray-guided PFO closure (and alternative machine learning models: XGBoost, SVM, Logistic Regression, Naive Bayes) was evaluated on Prolonged hospital stay (>7 days) (AUC 0.711 (Random Forest validation cohort), p=0.042). A random forest model using 10 clinical predictors showed moderate discrimination for prolonged hospital stay (>7 days) after PFO closure.
Synapse has enriched 8 closely related papers on similar clinical questions. Consider them for comparative context: