胃癌患者化疗后并发肺部感染的风险预测模型的建立

Establishment of a risk prediction model for pulmonary infection after chemotherapy in patients with gastric cancer

摘要:

目的 建立胃癌患者化疗后并发肺部感染的风险预测模型。方法 前瞻性选取2019年2月—2022年9月中国人民解放军南部战区海军第二医院收治的284例胃癌患者作为研究对象，按照8∶2随机分为训练集（227例）和验证集（57例）。患者均接受SOX化疗方案，随访4个周期，训练集患者根据化疗后是否并发肺部感染分为感染组与非感染组。对比两组患者临床资料，采用多因素逐步Logistic回归模型分析胃癌患者化疗后并发肺部感染的危险因素，建立列线图模型预测胃癌患者化疗后并发肺部感染的风险，并以Bootstrap法验证，计算C-index指数，绘制受试者工作特征（ROC）曲线，以曲线下面积（AUC）预测模型对胃癌患者化疗后并发肺部感染的效能。结果 两组患者性别、体质量指数、肿瘤部位、病理类型、合并高血压、吸烟史、饮酒史、白细胞计数、TNM分期比较，差异均无统计学意义（P >0.05）。感染组年龄≥ 60岁、合并糖尿病、合并低蛋白血症、住院时间≥ 20 d、化疗≥ 2个周期、化疗前KPS评分< 80 分比例大于非感染组（P <0.05），化疗前CRP、化疗前降钙素水平原高于非感染组（P <0.05），感染组化疗前Hb、化疗前血小板计数低于非感染组（P <0.05）。多因素逐步Logistic回归分析结果：年龄≥ 60岁［O^R=4.272（95% CI：1.878，9.717）］、合并低蛋白血症［O^R=5.333（95% CI：2.345，12.133）］、化疗≥ 2个周期［O^R=5.613（95% CI：2.467，12.767）］、化疗前KPS评分< 80分［O^R=2.732（95% CI：1.201，6.215）］为胃癌患者化疗后并发肺部感染的危险因素（P <0.05）。ROC曲线结果显示：列线图模型预测训练集胃癌患者化疗后并发肺部感染的敏感性为81.69%（95% CI：0.695，0.883），特异性为85.26%（95% CI：0.734，0.950），曲线下面积为0.897（95% CI：0.830，0.972）；列线图模型预测验证集胃癌患者化疗后并发肺部感染的敏感性为80.00%（95% CI：0.627，0.876），特异性为83.33%（95% CI：0.678，0.901），曲线下面积为0.889（95% CI：0.812，0.953）。临床影响曲线曲线结果显示，预测模型在> 0.4的风险阈值概率范围内总体净收益。结论 年龄较大、合并低蛋白血症、化疗周期较长、化疗前KPS评分较低的胃癌患者化疗后并发肺部感染的风险较高，建立风险预测模型有助于预测胃癌患者化疗后并发肺部感染风险，可进一步推广。

Abstract:

Objective To establish a risk prediction model for pulmonary infection in patients with gastric cancer after chemotherapy.Methods The 284 gastric cancer patients admitted to The Second Naval Hospital of the Southern Theater Command of the PLA from February 2019 to September 2022 were prospectively selected, and randomly divided into the training set (227 cases) and the validation set (57 cases) in an 8:2 ratio. All patients received chemotherapy with the SOX regimen and were followed up for 4 treatment cycles. Patients in the training set were divided into the infected group and the non-infected group according to whether they were complicated with pulmonary infection after chemotherapy. By comparing the infected group and the non-infected group, multivariable Logistic regression was performed to analyze the risk factors for pulmonary infection in patients with gastric cancer after chemotherapy, and a nomogram model was established to predict the risk of pulmonary infection in patients with gastric cancer after chemotherapy, which was verified by the Bootstrap method. The C-index was calculated, and the receiver operating characteristic (ROC) curve was plotted, where the area under the curve (AUC) was used to confirm the predictive efficacy of the model for pulmonary infection after chemotherapy in patients with gastric cancer.Results There was no difference in the sex composition, body mass index, tumor sites, histological types of tumors, percentage of hypertension, history of smoking, history of alcohol consumption, white blood cell count, or TNM stages of tumors between the two groups (P >0.05). The percentages of patients with age ≥ 60 years old, diabetes mellitus, hypoproteinemia, length of hospital stays ≥ 20 days, chemotherapy ≥ 2 cycles, and KPS scores < 80 before chemotherapy in the infected group were higher than those in the non-infected group (P < 0.05). The levels of C-reactive protein (CRP) and procalcitonin before chemotherapy in the infected group were higher than those in the non-infected group (P < 0.05). The level of hemoglobin (Hb) and the platelet count before chemotherapy in the infected group were lower than those in the non-infected group (P < 0.05). Multivariable Logistic regression analysis showed that age ≥ 60 years old [O^R = 4.272 (95% CI: 1.878, 9.717) ], hypoproteinemia [O^R = 5.333 (95% CI: 2.345, 12.133) ], chemotherapy ≥ 2 cycles [O^R = 5.613 (95% CI: 2.467, 12.767) ] and KPS scores < 80 before chemotherapy [O^R = 2.732 (95% CI: 1.201, 6.215) ] were risk factors for pulmonary infection in patients with gastric cancer after chemotherapy (P < 0.05). The ROC curve analysis exhibited that the sensitivity of the nomogram model based on the training set for predicting pulmonary infection after chemotherapy in patients with gastric cancer was 81.69% (95% CI: 0.695, 0.883), with a specificity of 85.26% (95% CI: 0.734, 0.950), and an AUC of 0.897 (95% CI: 0.830, 0.972). The sensitivity of the nomogram model based on the validation set was 80.00% (95% CI: 0.627, 0.876), with a specificity of 83.33% (95% CI: 0.678, 0.901), and an AUC of 0.889 (95% CI: 0.812, 0.953). The CIC demonstrated that the prediction model exhibited overall net benefits when the threshold probability was greater than 0.4.Conclusion Gastric cancer patients with advanced age, hypoproteinemia, more chemotherapy cycles and lower KPS scores before chemotherapy have higher risks for pulmonary infection after chemotherapy. The establishment of a risk prediction model is helpful for heralding the risks of pulmonary infection after chemotherapy in patients with gastric cancer.