目的 本研究旨在利用人体血液脂肪酸水平,建立可解释的预测模型来判断肺癌发生的风险,促进肺癌的预防和早 期诊断。 方法 回顾性收集 2023 年 6 月至 2025 年 6 月解放军总医院第一医学中心(1677 例)和解放军总医院第八医学中心 (160 例)收治的 1837 例肺癌(1194 例)和非肿瘤(643 例)患者临床和血液脂肪酸信息,通过设置随机数种子的方式,将解放 军总医院第一医学中心数据集按照 7 ∶ 3 随机拆分为训练集和内部验证集,将解放军总医院第八医学中心的数据集划为外部 验证集。 分别使用随机森林模型、极端梯度提升模型、支持向量机(SVM) 模型、分类与回归树模型,以及逻辑回归模型 5 种模 型构建了肺癌的风险预测模型。 采用受试者操作特征(ROC)曲线并计算曲线下面积(AUC)、临床决策曲线分析和校准曲线 比较了 5 种模型的预测性能,选出最优模型。 并使用 Shapley 加法解释 SVM 模型。 结果 建立了一个包括年龄、棕榈酸、棕榈 油酸、亚油酸、γ-亚麻酸、α-亚麻酸、花生四烯酸、二十碳五烯酸(EPA)、二十二碳六烯酸(DHA)、总单不饱和脂肪酸、总多不 饱和脂肪酸、AA 与二十碳五烯酸比值,以及 ω-6 多不饱和脂肪酸(ω-6 PUFAs)与 ω-3 多不饱和脂肪酸(ω-3 PUFAs)比值等 17 个预测因子的临床预测模型。 外部验证结果表明,随机森林模型、极端梯度提升模型、SVM 模型、分类与回归树模型,以及 逻辑回归模型的 ROC 曲线下面积分别为 0. 927、0. 931、0. 934、0. 840,以及 0. 912。 决策曲线表明,SVM 模型在大多数的概率 下往往相较于其他模型能产生更多的收益。 结论 基于 SVM 模型构建的肺癌风险预测模型性能最优, 可有效辅助医务人员早 期识别肺癌高危人群,结合风险因素实施精准营养干预,以降低发病风险。

Objective This study aimed to establish an interpretable predictive model for assessing the risk of lung cancer occurrence using human blood fatty acid levels thereby promoting the prevention and early diagnosis of lung cancer. Method Clinical and blood fatty acid data from 1 837 patients with lung cancer 1194 cases or non - tumorous 643 cases conditions were retrospectively collected from June 2023 to June 2025 at the First Medical Center 1677 cases and the Eighth Medical Center 160 cases of the Chinese PLA General Hospital. By setting a random number seed the dataset from the First Medical Center of the PLA General Hospital was randomly split into a training set and an internal validation set at a ratio of 7 ∶ 3 while the dataset from the Eighth Medical Center was designated as the external validation set. Five models—random forest extreme gradient boosting support vector machine SVM classification and regression tree and logistic regression—were developed to predict the risk of lung cancer. The predictive performance of the five models was compared using the area under the receiver operating characteristic curve decision curve analysis and calibration curves and the optimal model was selected. The SHapley additive explanations method was used to interpret the SVM model. Result A clinical prediction model was established incorporating 17 predictors including age palmitic acid palmitoleic acid linoleic acid γ - linolenic acid α - linolenic acid arachidonic acid eicosapentaenoic acid docosahexaenoic acid total monounsaturated fatty acids total polyunsaturated fatty acids the ratio of AA to eicosapentaenoic acid and the ratio of ω - 6 PUFAs to ω - 3 PUFAs. The external validation results showed that the area under the receiver operating characteristic curve of the Random Forest model Extreme Gradient Boosting model SVM model Classification and Regression Tree model and Logistic Regression model were 0. 927 0. 931 0. 934 0. 84 and 0. 912 respectively. Decision curve analysis indicated that across most probability thresholds the SVM model tended to yield greater net benefit compared to the other models. Conclusion The lung cancer risk prediction model constructed based on the SVM model exhibited the best performance. It can effectively assist medical personnel in early identification of high-risk populations for lung cancer and facilitate precise nutritional interventions targeting risk factors to reduce the incidence of the disease.