黑婷婷,李广益,刘维红.列线图与支持向量机模型预测卒中后营养不良风险的比较及验证研究[J].中华物理医学与康复杂志,2026,48(5):417-423
扫码阅读全文
|
| 列线图与支持向量机模型预测卒中后营养不良风险的比较及验证研究 |
|
| |
| DOI：10.3760/cma.j.cn421666-20250625-00531 |
| 中文关键词: 脑卒中 营养不良 风险分层 支持向量机 列线图 |
| 英文关键词: Stroke Malnutrition Risk stratification Support vector machine modeling Nomograms |
| 基金项目:天津市卫生健康委员会中医中西医结合科研项目（2023218） |
|
| 摘要点击次数: 436 |
| 全文下载次数: 430 |
| 中文摘要: |
| 目的 基于脑卒中患者营养不良的独立危险因素，构建列线图及支持向量机（SVM）两种模型，并对比列线图及SVM模型对卒中后营养不良的预测性能及临床适用性。 方法 将前期研究确定的6个卒中后营养不良独立危险因素［分别是年龄、糖尿病史、C反应蛋白（CRP）、美国国立卫生研究院卒中量表（NIHSS）评分、洼田饮水试验（WST）评分及汉密尔顿抑郁量表（HAMD）评分］作为预测变量，按7∶3比例将入选脑卒中患者随机分为训练集（n=280）与验证集（n=120）。基于训练集数据分别构建列线图模型与SVM模型，采用Bootstrap内部验证、验证集测试、校准曲线及决策曲线分析（DCA）全面对比、评估两模型性能；并利用风险评分对入选患者进行低、中、高危分层。 结果 风险分层结果显示，低、中、高危组患者的营养不良发生率分别为18.8%、43.3%和86.1%，其风险梯度显著（P<0.001）。对于训练集，SVM与列线图模型的曲线下面积（AUC）分别为0.835和0.822；对于验证集，两模型的AUC分别为0.817和0.813，两者相当。校准曲线提示SVM模型的预测概率与实际发生率一致性较好。决策曲线分析（DCA）提示在广泛的阈值概率范围内，应用SVM模型可能获得更高的临床净获益。 结论 本研究构建并验证了卒中后营养不良的列线图与SVM预测模型，其中列线图模型具有直观、可视的优势，便于临床快速评估；SVM模型则在校准度及决策效用方面展现出潜在优势，上述两种模型均为临床早期识别卒中后营养不良高危患者提供了有效工具。 |
| 英文摘要: |
| Objective To develop a nomogram and a support vector machine (SVM) model based on previously-identified independent risk factors for predicting the risk of post-stroke malnutrition, and to compare their predictive power and their clinical applicability. Methods Six independent risk factors determined by prior research-age, history of diabetes, C-reactive protein level, National Institutes of Health Stroke Scale score, water swallowing test score, and Hamilton Depression Scale score-served as the predictor variables. A dataset of 400 patients was randomly split into a training set (n=280) and a validation set (n=120) at a 7∶3 ratio. The nomogram and SVM models were constructed using the training set. Patients were stratified into low-, medium-, and high-risk groups based on model-derived risk scores. Model performance was comprehensively evaluated and compared using bootstrap internal validation, validation set testing, calibration curves, and decision curve analysis. Results Risk stratification revealed malnutrition incidence rates of 18.8%, 43.3%, and 86.1% in the low-, medium-, and high-risk groups, respectively, demonstrating a significant gradient. In the training set, the areas under the curves for the SVM and nomogram models were 0.835 and 0.822, respectively. However, in the independent validation set, the areas under the curves were 0.817 and 0.813, respectively. The calibration curve suggested better agreement between predicted probabilities and observed outcomes for the SVM model. Moreover, the decision curve analysis indicated that applying the SVM model might yield higher clinical net benefit across a wide range of threshold probabilities. Conclusions A nomogram and an SVM model for predicting post-stroke malnutrition were successfully developed and validated. The former offers the advantage of visual intuitiveness for rapid clinical assessment, while the latter shows potential advantages in calibration and decision utility. Both models provide effective tools for the early identification of high-risk patients. The choice of model in clinical practice may depend on the specific need for interpretability versus predictive precision. |
|
查看全文
查看/发表评论 下载PDF阅读器 |
| 关闭 |
|
|
|