食管癌患者术后抑郁风险预测模型的构建

Development of a risk prediction model for postoperative depression in patients with esophageal cancer

目的探讨食管癌患者术后发生抑郁的危险因素,并构建风险预测模型。方法选取江南大学附属医院南院区2022年9月至2023年3月接受食管癌手术的患者269例(抑郁组191例和非抑郁组78例)作为模型构建集,选取江南大学附属医院南院区2023年3至5月食管癌手术患者78例作为外部验证集。将贝克抑郁量表第二版评分≥5分且2位江南大学附属医院精神心理科主任医师诊断为抑郁症的患者纳入抑郁组,其余患者纳入非抑郁组。比较抑郁组与非抑郁组患者的一般资料、血液常规检查、超敏C反应蛋白(hs-CRP)、血电解质、血脂、临床症状(胃食管反流、睡眠障碍和食欲等)和抑郁评分。统计学方法采用独立样本t检验和Mann-Whitney U检验。采用二元logistic回归模型分析食管癌患者术后抑郁的独立危险因素,并构建风险预警模型。采用Hosmer-Lemeshow检验和受试者操作特征曲线(ROC)评价模型的拟合度和预测效能,并采用交叉验证方法验证模型的有效性。结果食管癌患者术后抑郁症发生率为71.0%(191/269)。抑郁组白细胞计数、hs-CRP、血磷、β_(2)微球蛋白水平和睡眠障碍患者占比均高于非抑郁组[1.3(1.1,5.4)×10^(9)/L比0.9(0.3,1.1)×10^(9)/L、75.8(54.8,102.1)mg/L比60.8(3.6,61.5)mg/L、(1.33±0.32)mmol/L比(1.02±0.19)mmol/L、(2.17±0.72)mg/L比(2.12±0.49)mg/L、84.3%(161/191)比33.3%(26/78)],差异均有统计学意义(Z=9.24、7.88,t=9.24,χ^(2)=67.87,t=1.98,均P<0.001];血红蛋白、血小板计数、高密度脂蛋白(HDL)水平和食欲不振患者占比低于非抑郁组[(119.91±24.51)g/L比(122.09±22.97)g/L、(203.43±58.45)×10^(9)/L比(311.55±83.54)×10^(9)/L、(1.04±0.30)mmol/L比(1.43±0.23)mmol/L、73.3%(140/191)比84.6%(66/78)],差异均有统计学意义(t=-2.00、-8.42、-8.48,χ^(2)=3.96,P=0.047、<0.001、<0.001、=0.047]。多因素logistic回归模型分析显示睡眠障碍[OR=3.976,95%置信区间(95%CI)1.601~9.872]、食欲不振[OR=0.271,95%CI 0.092~0.791]、白细胞计数[OR=31.808,95%CI 2.879~351.401]、hs-CRP[OR=1.031,95%CI 1.017~1.044]、血小板计数[OR=0.990,95%CI 0.982~0.997]和HDL[OR=0.017,95%CI 0.001~0.242]是食管癌患者术后抑郁的独立影响因素(P=0.003、=0.017、=0.005、<0.001、=0.008、=0.003)。得到风险预警模型公式:抑郁症发生概率=1-1/{1+exp[1.544+1.380×睡眠障碍(是=1,否=0)-1.307×食欲不振(是=1,否=0)-0.010×血小板计数(×10^(9)/L)-4.063×HDL(mmol/L)+0.030×hs-CRP(mg/L)+3.460×白细胞计数(×10^(9)/L)]}。Hosmer-Lemeshow检验显示预测模型的拟合度较好(χ^(2)=2.01,P=0.981),ROC曲线下面积为0.949,灵敏度为0.874,特异度为0.872。外部验证集交叉验证显示风险预测模型的准确度为67.9%。结论食管癌患者术后抑郁风险预测模型可能用于食管癌术后抑郁患者的早期筛查。

ObjectiveTo explore the risk factors of postoperative depression in patients with esophageal cancer,and to develop a risk prediction model which providing a theoretical basis for the early detection of depression in high-risk groups by clinical staff.MethodsFrom September 2022 to March 2023,at the South Campus of Affiliated Hospital of Jiangnan University,269 hospitalized patients with esophageal cancer(191 in depression group,78 in non-depression group)were selected as the model construction set.From March to May 2023,at the South Campus of Affiliated Hospital of Jiangnan University,78 hospitalized patients with esophageal cancer were selected as the external validation set.The patients with Beck depression inventory-Ⅱscore≥5 and depression diagnosed by two experts(chief psychiatrists of the Department of Psychiatry of Affiliated Hospital of Jiangnan University)were considered as depression and included in the depression group,and the other patients were enrolled in the non-depression group.The general data,blood routine examination,high-sensitivity C-reactive protein(hs-CRP),blood electrolytes,blood lipids,clinical symptoms(gastroesophageal reflux,sleep disturbance,appetite,etc.)and depression score were compared between the depression group and the non-depression group.Independent sample t-test and Mann-Whitney U test were used for statistical analysis.Multiple logistic regression model was performed to analyze the independent risk factors of postoperative depression in patients with esophageal cancer,and a risk warning model was constructed.The Hosmer-Lemeshow test and receiver operating characteristic curve(ROC)were used to evaluate the fitting degree and predictive efficiency of the model,and the cross-validation method was used to verify the effectiveness of the model.ResultsThe incidence of postoperative depression in patients with esophageal cancer was 71.0%(191/269).The total white blood cell count,hs-CRP,blood phosphorusβ_(2)microglobulin and the proportion of sleep disorders of the depression group were higher than those of the non-depression group(1.3(1.1,5.4)×10^(9)/L vs.0.9(0.3,1.1)×10^(9)/L,75.8(54.8,102.1)mg/L vs.60.8(3.6,61.5)mg/L,(1.33±0.32)mmol/L vs.(1.02±0.19)mmol/L,(2.17±0.72)mg/L vs.(2.12±0.49)mg/L,84.3%(161/191)vs.33.3%(26/78),and the differences were statistically significant(Z=9.24,7.88,t=9.24,χ^(2)=67.87 t=1.98;all P<0.001);hemoglobin,total platelet count,high-density lipoprotein(HDL)and the proportion of poor appetite were lower than those of the non-depression group((119.91±24.51)g/L vs.(122.09±22.97)g/L,(203.43±58.45)×10^(9)/L vs.(311.55±83.54)×10^(9)/L,(1.04±0.30)mmol/L vs.(1.43±0.23)mmol/L,73.3%(140/191)vs.84.6%(66/78)),and the differences were statistically significant(t=-2.00,-8.42 and-8.48,χ^(2)=3.96;P=0.047,<0.001,<0.001,=0.047).The results of multifactorial logistic regression model analysis showed that sleep disorder(OR=3.976,95%confidence interval(95%CI 1.601 to 9.872)),loss of appetite(OR=0.271,95%CI 0.092 to 0.791),white blood cell count(OR=31.808,95%CI 2.879 to 351.401),hs-CRP(OR=1.031,95%CI 1.017 to 1.044),platelet count(OR=0.990,95%CI 0.982 to 0.997),and HDL(OR=0.017,95%CI 0.001 to 0.242)were independent influencing factors of postoperative depression in patients with esophageal cancer.The formula of risk warning model was probability of depression=1-1/{1+exp[1.544+1.380×sleep disturbance(yes=1,no=0)-1.307×loss of appetite(yes=1,no=0)-0.010×platelet count(×10^(9)/L)-4.063×HDL(mmol/L)+0.030×hs-CRP(mg/L)+3.460×white blood cell count(×10^(9)/L)]}.The results of Hosmer-Lemeshow test showed that the model has a good fit(χ^(2)=2.01,P=0.981),with an area under the ROC of 0.949,a sensitivity of 0.874,and a specificity of 0.872.The cross-validation of the external validation set showed that the accuracy of the risk warning model was 67.9%.ConclusionThis study is a preliminary study on the risk warning model of postoperative depression in patients with esophageal cancer,which provides a novel approach for screening depression in patients with esophageal cancer after surgery.