目的 探索老年营养风险指数(GNRI)作为早期肺癌手术患者预后标志物的价值及决策树预测模型的构建。 方法 回顾性分析的 138 例早期肺癌手术患者均在四川大学华西医院 2021 年 1 月至 2022 年 12 月期间收集,根据患者预后情况分 为两组,即预后不佳组(n = 21),预后良好组(n = 117),比较两组一般资料,经二元 Logistic 回归分析、决策树分析影响预后独 立因素,并利用 Bootstrap 法对模型进行内验证。 结果 预后良好组患者 GNRI 高于预后不佳组(P<0. 05)。 经二元 Logistic 回归 分析,年龄≥75 岁、胸膜受累、术前贫血史、GNRI 值≤97. 35 是早期肺癌术预后的独立影响因素(P<0. 05)。 同时,再以单因素 分析有统计意义的 4 个变量为预测因素,通过 SPSS 软件构建决策树模型,其中,以年龄为主,检测预测值为 91. 1%;以胸膜受 累为主,检测预测值为 88. 6%;以术前贫血史为主,检测预测值为 90. 5%;以 GNRI 值为主,检测预测值为 73. 2%,利用 Bootstrap 法对模型进行内验证,模拟区分度,AUC 为 0. 872,95%CI 为 0. 805 ~ 0. 938。 而经 ROC 曲线模型分析,GNRI 面积图为 0. 903,95%CI 为 0. 842~ 0. 963,敏感度为 90. 5%,特异度为 79. 5%。 结论 GNRI 能够预测早期肺癌手术患者预后情况,且预后 还受到患者年龄、胸膜受累、术前贫血史影响,通过建立操作简便、可视化效果强的决策树模型,能够尽早预测预后情况,为治 疗方案拟定、预后评估提供科学依据。

Objective To explore the value of geriatric nutritional risk index as a prognostic marker for early lung cancer patients and the construction of a decision tree prediction model. Method The retrospective analysis of 138 patients with early lung cancer were collected in West China Hospital from January 2021 to December 2022 and divided into two groups according to the prognosis of patients namely poor prognosis n = 21 and good prognosis n = 117 . Comparing the general data binary Logistic regression analysis and decision tree analysis affected the prognosis independent factors and the model was verified internally using Bootstrap method. Result The GNRI of patients with good prognosis was higher than that of patients with poor prognosis P<0. 05 . According to binary logistic regression analysis age ≥ 75 years old pleural involvement preoperative anemia history and GNRI value ≤ 97. 35 were independent factors affecting the prognosis of early lung cancer surgery P< 0. 05 . At the same time using four statistically significant variables from single factor analysis as predictive factors a decision tree model was constructed using SPSS software. Among them age was the main factor and the detected predictive value was 91. 1% Mainly involving pleura with a predicted detection value of 88. 6% Based on a history of preoperative anemia the predicted detection value is 90. 5% Based on GNRI values the predicted detection value is 73. 2%. The model was internally validated using Bootstrap. Result To simulate discrimination with an AUC of 0. 872 and a 95%CI of 0. 805-0. 938. According to the ROC curve model analysis the GNRI area plot is 0. 903 with a 95%CI of 0. 842-0. 963 a sensitivity of 90. 5% and a specificity of 79. 5%. Conclusion The nutritional risk index can predict the prognosis of early lung cancer patients undergoing surgery and the prognosis is also influenced by the patient ' s age pleural involvement and preoperative anemia history. By establishing a decision tree model that is easy to operate and visually effective the prognosis can be predicted as early as possible providing scientific basis for treatment plan formulation and prognosis evaluation.