成人复杂性阑尾炎临床预测评分模型的建立与验证

Clinical predictive scoring model of adult complicated appendicitis: establishment and verification

目的:通过分析成人复杂性阑尾炎的高危因素,建立区分复杂性阑尾炎(CA)与非复杂性阑尾炎(UA)的评分预测模型,并验证模型从而辅助临床医师迅速判断急性阑尾炎的类型,并为治疗提供参考。方法:本研究采用回顾性病例对照研究方法,收集海南省第三人民医院2016-11—2018-04期间住院行阑尾切除术并经病理确诊的297例成人急性阑尾炎患者的临床资料。根据术中所见和病理类型,将297例急性阑尾炎患者分为UA组(207例,包括单纯和化脓性阑尾炎)和CA组(90例,包括穿孔和坏疽性阑尾炎)。将2组患者的术前资料按照一般情况、病史及体格检查、实验室检查及超声表现进行分类分析,应用单因素Logistic回归分析筛选出CA的12项高危因素后,分类别进行多因素Logistic回归分析进一步筛选变量。根据筛选结果最终有5个因子纳入最终回归方程,根据OR值进行变量赋值,预测模型的总分为12分,并得出回归方程的ROC曲线。为验证评分预测模型,进一步收集海南省第三人民医院2018-08—2019-11期间行阑尾切除术的180例急性阑尾炎病例的临床资料,计算该评分预测模型对CA的特异度、敏感度、准确度、阳性及阴性预测值、阳性及阴性似然比,并得出ROC曲线。结果:最终纳入回归方程的5个因子为超声见右髂窝积液(OR=7.567,赋值4分)、高热≥39℃或寒战(OR=6.449,赋值3分)、中性粒细胞绝对值>11.88×10^9/L(OR=3.469,赋值2分)、血淀粉酶≤56.9 U/L(OR=3.003,赋值2分)、恶心或呕吐(OR=2.117,赋值1分)。应用Medcalc 15.2软件得出ROC曲线下的面积为0.789(0.738~0.834),我们根据约登指数将判定复杂性阑尾炎的阈值设置为≥5分,因为此阈值对应着该模型最高的敏感度(63.33%)及特异度(79.23%)。在前瞻性验证队列中,该预测模型的敏感度为80.00%(67.70~89.20),特异度为79.17%(70.80~86.00),阳性预测值为65.75%,阴性预测值为88.79%,阳性似然比3.84(2.61~5.63),阴性似然比0.25(0.22~0.41),整体预测准确率为79.44%,ROC曲线下面积为0.796(0.729~0.852)。结论:在临床实践中,该评分预测模型可以较好的辅助临床医师判断CA或UA,并且简便、易行。

Objective: By analyzing the high risk factors of adult complex appendicitis, a score prediction model was established to distinguish complex appendicitis(CA) and uncomplicated appendicitis(UA), so as to assist clinicians to quickly judge the types of acute appendicitis and to provide references for treatment. Method: A retrospective case-control study was used to analyze the clinical data of 297 cases of adult acute appendicitis confirmed by pathology in the Third People’s Hospital of Hainan Province from November 2016 to April 2018.297 patients with acute appendicitis were divided into UA group(207 cases, including simple and suppurative appendicitis) and CA group(90 cases, including perforation and gangrene appendicitis). The preoperative data of the two groups were classified and analyzed according to the general situation, medical history, physical examination, laboratory examination and ultrasonic performance. After screening 12 high risk factors of CA by univariate logistic regression analysis, multivariate logistic regression analysis was conducted to further screen variables. According to the screening results, five factors were finally included in the final regression equation, and the variables were assigned according to the OR value. The total score of the prediction model was 12 points, and the ROC curve of regression equation was obtained. In order to verify the scoring prediction model, the clinical data of 180 cases of acute appendicitis who underwent appendectomy in the Third People’s Hospital of Hainan Province from August 2018 to November 2019 were further collected, and the specificity, sensitivity, accuracy, positive and negative predictive values, positive and negative likelihood ratios of the model for CA prediction were calculated, and the ROC curve was analyzed. Result: The final 5 factors included in the regression equation were the right iliac fossa effusion(OR=7.567, assigned 4 points), high fever ≥ 39 ℃ or shivering(OR=6.449, assigned 3 points), neutrophil absolute value>11.88 × 109/L(OR=3.469, assigned 2 points), blood amylase ≤ 56.9 U/L(OR=3.003, assigned 2 points), nausea or vomiting(OR=2.117, assigned 1 point). Using medcalc 15.2 software, the area under ROC curve was 0.789(0.738-0.834). We set the threshold value of complex appendicitis to ≥ 5 points according to Youden index, because the model has the highest sensitivity of 63.33% and specificity of 79.23%. In the prospective validation cohort, the sensitivity of the prediction model was 80.00%(67.70-89.20), the specificity was 79.17%(70.80-86.00), the positive prediction value was 65.75%, the negative prediction value was 88.79%, the positive likelihood ratio was 3.84(2.61-5.63), the negative likelihood ratio was 0.25(0.22-0.41), the overall prediction accuracy was 79.44%, and the area under ROC curve is 0.796(0.729-0.852). Conclusion: In clinical practice, the score prediction model can assist clinicians to judge CA or UA, and it is simple and easy to implement.