小细胞肺癌营养风险决策支持框架：基于不完整临床数据填补策略的时间序列模型
Decision-Support Framework for Nutritional Risk in Small Cell Lung Cancer: A Time-Series Model Using Imputation Strategies for Incomplete Clinical Data.
作者
作者单位
- Department of Nursing, Peking Union Medical College Hospital, Chinese Academy of Medical Sciences and Peking Union Medical College, Beijing, China.
- Clinical Research Validation Platform, National Facility for Translational Medicine, Peking Union Medical College Hospital, Chinese Academy of Medical Sciences and Peking Union Medical College, Beijing, China.
- Department of Pulmonary and Critical Care Medicine, Peking Union Medical College Hospital, Chinese Academy of Medical Sciences and Peking Union Medical College, Beijing, China.
- Peking Union Medical College Hospital, State Key Laboratory for Complex Severe and Rare Diseases, Institute of Clinical Medicine, Chinese Academy of Medical Sciences, Beijing, China.
摘要
中文
在小细胞肺癌（SCLC）患者中，营养状况是疾病进展、治疗耐受性和预后的关键决定因素。预后营养指数（PNI）反映免疫和营养状况，广泛用于评估预后风险，但其纵向监测常因真实世界临床数据不完整而受限。本研究旨在提出一种决策支持框架，以预测SCLC患者的未来营养状况并改进风险筛查，尤其在数据缺失情况下。利用前四个随访时间点的PNI值及相关变量预测第五个时间点的PNI（PNI5），并评估不同缺失数据填补策略对预测性能的影响。采用均值填补、多重填补（MI）、卡尔曼滤波、K近邻（KNN）和XGBoost填补法处理缺失PNI值。使用两种机器学习算法（随机森林RF和XGBoost）建立预测模型。RF模型整体性能更优，MI联合RF取得最佳预测精度（MAE：2.952；RMSE：3.727）。预测PNI值通过预设阈值（PNI=45）进一步转化为风险类别，总体准确率为93.33%，阳性预测值为95.24%，阴性预测值为92.59%。重要的是，该模型能够对实验室数据不完整的患者进行风险评估，覆盖此前无法评估的病例，减少监测缺口。本研究强调了适当填补策略的重要性，并为SCLC患者的持续营养风险评估和早期识别高危患者提供了实用工具。
English
In patients with small cell lung cancer (SCLC), nutritional status is a key determinant of disease progression, treatment tolerance, and prognosis. The prognostic nutritional index (PNI), reflecting both immune and nutritional conditions, is widely used to evaluate prognostic risk, but its longitudinal monitoring is often limited by incomplete clinical data in real-world settings. This study aimed to propose a decision-support framework for predicting future nutritional status and improving risk screening in SCLC patients with missing data. Using PNI values and related variables from the first four follow-up time points to predict the PNI at the fifth time point (PNI5) and evaluated the impact of different missing data imputation strategies on predictive performance. Missing PNI values were imputed using mean imputation, multiple imputation (MI), Kalman filtering, k-nearest neighbors (KNN), and XGBoost imputation. Predictive models were developed with two machine learning algorithms: random forest (RF) and XGBoost. The RF model demonstrated better overall performance, and MI combined with RF achieved the best predictive accuracy (MAE: 2.952; RMSE: 3.727). Predicted PNI values were further translated into risk categories using a predefined threshold (PNI = 45), with overall accuracy of 93.33%, PPV 95.24%, and NPV 92.59%. Importantly, the model enabled risk assessment in patients with incomplete laboratory data, covering previously unassessable cases and reducing monitoring gaps. This study highlights the importance of appropriate imputation strategies in and provides a practical tool for continuous nutritional risk assessment and early identification of high-risk patients in SCLC.
分类与指标
- 研究类型
- AI/ML
- 病种
- 肺癌
- JCR 分区
- Q2
- 影响因子
- 2.6
- 新锐分区
- 4区