目的 探讨Ⅲ—Ⅳ期肺癌免疫检查点抑制剂（immune checkpoint inhibitors，ICIs）治疗后30 d内非计划再入院（unplanned patient readmission， UPR）的危险因素，构建并验证相应的风险预测模型。 方法 收集2023年1月至2024年5月在重庆市中医院接受ICIs治疗的Ⅲ—Ⅳ期肺癌患者的资料。应用Boruta算法初筛风险因素，采用logistic回归识别独立危险因素，构建列线图预测模型。通过受试者工作特征曲线（ROC）评估模型区分能力，以校准曲线验证模型一致性，利用决策曲线分析（DCA）评估其临床实用性。采用限制性立方样条回归（RCS）联合SHAP可解释分析明确危险因素与非计划再入院之间的剂量反应关系及阈值点。 结果 最终纳入284例患者，UPR发生率为30.63%。logistic回归分析显示住院天数、NRS 2002评分、介入治疗及KPS评分是非计划再入院的独立风险因素（P < 0.05）。模型训练集的ROC曲线AUC = 0.88，95%CI： 0.84 ~ 0.93，灵敏度84%，特异度80%，验证集的ROC曲线AUC = 0.87，95%CI： 0.79 ~ 0.95，灵敏度82%，特异度70%，模型的校准曲线接近于对角线，显示良好的准确性；DCA分析显示模型阈值在10% ~ 90%之间具有净获益，SHAP分析显示住院天数是UPR的关键因素，SHAP-RCS分析发现住院天数> 6.43 d、NRS 2002 > 2.05、KPS < 79.01且有介入手术史的患者再入院风险更高。 结论 基于4个风险因素构建的预测模型能有效预测Ⅲ—Ⅳ期肺癌ICIs患者30 d内非计划再入院的风险，具有良好的临床适用性，对于高风险患者应予以重点关注及提前干预。

Objective To identify risk factors for unplanned 30-day readmission （UPR） following immune checkpoint inhibitors （ICIs） treatment in stage Ⅲ—Ⅳ lung cancer patients and to develop/validate a predictive model. Methods We retrospectively analyzed clinical data from stage Ⅲ—Ⅳ lung cancer patients treated with ICIs at our institution （January 2023-May 2024）. Risk factors were preliminarily screened using the Boruta algorithm； independent predictors were identified via logistic regression. A nomogram prediction model was subsequently developed. Model performance was evaluated by： discrimination （receiver operating characteristic curves， ROC）， calibration （calibration plots）， and clinical utility （decision curve analysis， DCA）. Restricted cubic spline （RCS） regression combined with SHapley Additive exPlanations （SHAP） analysis further explored dose-response relationships and threshold effects of key risk factors on UPR. Results Among 284 included patients， the UPR incidence was 30.63%. Independent risk factors identified by logistic regression were： hospital length of stay， Nutritional Risk Screening 2002 （NRS 2002） score， invasive procedures， and Karnofsky Performance Status （KPS） score （all P < 0.05）. The model showed strong discrimination： training set AUC = 0.88 （95% CI： 0.84 ~ 0.93）， sensitivity 84%， specificity 80%； validation set AUC = 0.87 （95% CI： 0.79 ~ 0.95）， sensitivity 82%， specificity 70%. Calibration curves indicated good model fit. Decision curve analysis demonstrated positive net benefit at threshold probabilities of 10%~90%. SHAP analysis prioritized length of stay as the most influential predictor； SHAP-RCS analysis revealed increased UPR risk when： hospital stay > 6.43 days， NRS 2002 > 2.05， KPS < 79.01， or prior invasive procedures. Conclusion The nomogram model incorporating four key risk factors effectively predicts 30-day unplanned readmission risk in stage Ⅲ—Ⅳ lung cancer patients receiving ICI therapy. With robust performance and clinical utility， it may facilitate early identification and intervention for high-risk individuals.