# Machine learning-based prediction of 30-day unplanned readmission risk in day surgery lung cancer patients after lobectomy or sublobectomy: a real-world study

## TL;DR

This study developed a machine learning model to predict 30-day unplanned readmissions after lung cancer surgery in day surgery patients, showing strong performance and identifying key risk factors.

## Contribution

The study introduces a high-performing random forest model for predicting readmission risk in day surgery lung cancer patients, validated in a real-world setting.

## Key findings

- The random forest model achieved a ROC-AUC of 0.939 and accuracy of 0.825 in predicting readmission risk.
- Key predictors included white-blood-cell count, prothrombin time, and intraoperative blood loss.
- The model showed favorable net benefit across a wide range of threshold probabilities and was well-calibrated.

## Abstract

Unplanned readmission within 30 days after lobectomy or sublobectomy for early stage lung cancer adversely affects patient recovery and healthcare costs. While machine-learning (ML) approaches offer potential for improved prediction, few models have been developed for day-surgery settings. This study aimed to develop and validate an ML-based model to predict 30-day unplanned readmission in lung cancer patients undergoing ambulatory lung resection.

We included patients who underwent lobectomy or sublobectomy in a day-surgery pathway between December 2022 and January 2025. The least absolute shrinkage and selection operator (LASSO) was used for feature selection. Data were split into training (70%) and validation (30%) sets. Nine ML algorithms were trained and evaluated using area under the receiver-operating-characteristic curve (ROC-AUC), precision-recall AUC (PR-AUC), accuracy, decision-curve analysis (DCA), and calibration curves. Model interpretability was assessed with SHapley Additive exPlanations (SHAP).

After propensity-score matching, 380 patients were analyzed, including 111 with unplanned readmission. LASSO identified 12 predictive features: age, payment category, prothrombin time (PT), white-blood-cell count (WBC), hemoglobin, intraoperative blood loss, surgical approach, pathological diagnosis, tumor count, tumor size, occupational category, and forced expiratory volume in 1 s (FEV1). The random forest (RF) model performed best in the validation set (ROC-AUC = 0.939, accuracy = 0.825), showed favorable net benefit across threshold probabilities of 10–80%, and was well-calibrated. SHAP analysis indicated WBC, PT, hemoglobin, intraoperative blood loss, and “unknown” occupational category as the top five predictors; WBC, PT, and blood loss were positively associated with readmission risk.

An RF-based model effectively predicted 30-day unplanned readmission after lung-cancer day surgery. The identified risk factors provide a basis for early stratification and targeted intervention, supporting optimized perioperative care in ambulatory settings.