基于抗中性粒细胞胞质抗体的列线图模型对川崎病患儿并发冠状动脉病变风险的预测作用
Nomogram for predicting the risk of coronary artery lesions in patients with Kawasaki disease based on anti-neutrophil cytoplasmic antibodies
Corresponding authors: SHEN Jie, E-mail:she6t@163.com。
Received: 2024-08-05 Accepted: 2024-12-18 Online: 2025-04-21
作者简介 About authors
陈蓉(1999—),女,住院医师,硕士;电子信箱:
目的·评估抗中性粒细胞胞质抗体(anti-neutrophil cytoplasmic antibody,ANCA)对川崎病(Kawasaki disease,KD)患儿并发冠状动脉病变(coronary artery lesion,CAL)的预测价值。方法·回顾性收集2018年1月至2024年5月上海交通大学医学院附属上海儿童医学中心收治的340例KD患儿的临床资料,按7∶3的比例随机分为训练集(n=237)和验证集(n=103)。通过单因素分析、最小绝对收缩和选择算法(least absolute shrinkage and selection operator,LASSO)筛选出CAL的危险因素,并将其纳入多因素Logistic回归分析,构建列线图预测模型。分别采用受试者操作特征(receiver operating characteristic,ROC)曲线、校准曲线及Hosmer-Lemeshow拟合优度检验、决策曲线分析(decision curve analysis,DCA)评价模型的区分度、校准度和临床适用性。根据Logistic回归方程中的自变量系数对各变量赋分,得到一个预测评分系统,并将其与目前3个常用评分系统(Kobayashi评分、Egami评分和Sano评分)的预测效能进行比较。结果·男性、低白蛋白血症、ANCA阳性和静脉注射免疫球蛋白抵抗是KD患儿发生CAL的危险因素,据此构建列线图预测模型。模型在训练集和验证集中的ROC曲线下面积分别为0.747(95%CI 0.667~0.821)和0.645(95%CI 0.500~0.794),表明模型预测效能良好;模型经校准曲线和Hosmer-Lemeshow拟合优度检验(训练集χ2 =5.105,P=0.746;验证集χ2 =13.549,P=0.094)验证,预测准确性良好;DCA显示模型具有一定的临床适用性。根据Logistic回归方程系数建立CAL的预测评分体系,与Kobayashi评分、Egami评分、Sano评分模型相比,其灵敏度(58.4%)和特异度(78.7%)均较高。结论·研究基于ANCA建立了一个可有效预测KD患儿发生CAL风险的评分模型,可为临床上早期识别高危患儿、制定个性化治疗方案和管理策略提供参考。
关键词:
Objective ·To evaluate the predictive value of anti-neutrophil cytoplasmic antibodies (ANCA) in Kawasaki disease (KD) complicated with coronary artery lesions (CALs) and to construct a nomogram prediction model. Methods ·A retrospective study was conducted to collect the clinical data of 340 children with KD admitted to Shanghai Children's Medical Center from January 2018 to May 2024. All patients were randomly divided in a 7:3 ratio into a training set (n=237) and a validation set (n=103). Univariate analysis and least absolute shrinkage and selection operator (LASSO) were applied to screen the risk factors of CALs, which were incorporated into multifactorial Logistic regression analysis to develop the nomogram model. The model's discrimination, calibration and clinical practicability were evaluated using the receiver operating characteristic (ROC) curve, calibration curve, Hosmer-Lemeshow goodness-of-fit test, and decision curve analysis (DCA). A new predictive scoring system was obtained by assigning scores to each variable based on the coefficients of the independent variables in the Logistic regression equation, and its predictive efficacy was then compared with that of three commonly used scoring systems, Kobayashi, Egami, and Sano scoring models. Results ·Male, low serum albumin level, ANCA positivity, and intravenous immunoglobulin resistance were risk factors for the development of CALs in children with KD, based on which a nomogram model was constructed. The area under the ROC curve for the nomogram in the training set and validation set were 0.747 (95%CI 0.667‒0.821) and 0.645 (95%CI 0.500‒0.794), respectively, indicating good effectiveness. The model was verified to have good predictive accuracy through the calibration curve and Hosmer-Lemeshow goodness-of-fit test (training set: χ2 =5.105, P=0.746; validation set:χ2 =13.549, P=0.094). The DCA showed its clinical usefulness. A predictive scoring system for CALs was developed based on the coefficients of the Logistic regression equation, which demonstrated higher sensitivity (58.4%) and specificity (78.7%) compared to the Kobayashi, Egami, and Sano scoring models. Conclusion ·This study developed a new scoring model based on ANCA to effectively predict the risk of CALs in KD patients. The model provides valuable reference for clinicians to identify high-risk patients early, and to formulate personalized treatment plans and management strategies.
Keywords:
本文引用格式
陈蓉, 张锰, 朱荻绮, 郭颖, 沈捷.
CHEN Rong, ZHANG Meng, ZHU Diqi, GUO Ying, SHEN Jie.
川崎病(Kawasaki disease,KD)又称黏膜皮肤淋巴结综合征,是一种儿童常见的急性发热出疹性疾病,以超过5 d的持续性发热、多形性皮疹、颈部淋巴结非化脓性肿大、双侧球结膜充血、口唇及口腔改变、手足硬肿及指端脱皮为主要临床特征,可伴发全身多系统表现[1]。KD以全身性血管炎为主要病理改变,累及全身中、小动脉,以冠状动脉最显著[2-3]。冠状动脉病变(coronary artery lesion,CAL)是川崎病最重要的并发症,早期主要表现为冠状动脉扩张和冠状动脉瘤形成;随着病情进展,后期形成血栓和内膜增生、钙化,出现冠状动脉狭窄或闭塞;最终导致心肌缺血、梗死甚至猝死,严重危害儿童生命健康[4],是部分国家和地区儿童后天性心脏病的主要病因之一[5-6]。
中性粒细胞胞质抗体(anti-neutrophil cytoplasmic antibody,ANCA)是一种以中性粒细胞胞质成分为靶抗原的自身抗体,可以诱导中性粒细胞活化,介导血管内皮细胞损伤,与血管炎症性疾病关系密切[7]。本研究旨在探讨ANCA对KD合并CAL的预测价值,构建并验证列线图预测模型,为临床早期识别和干预高危患儿提供依据。
1 对象与方法
1.1 研究对象
图1
1.2 资料收集
通过医院电子病历系统收集患儿临床资料。①一般资料:性别、年龄、IVIG使用前发热时间、IVIG治疗反应。②IVIG治疗前实验室检查指标:C反应蛋白(C reactive protein,CRP)、中性粒细胞百分比(neutrophil percentage,NEUT%)、血红蛋白(hemoglobin,HGB)、血小板计数(platelet,PLT)、红细胞沉降率(erythrocyte sedimentation rate,ESR)、铁蛋白(ferritin,Ferr)、血钠(serum sodium,Na+)、丙氨酸氨基转移酶(alanine amino-transferase,ALT)、天冬氨酸氨基转移酶(aspartate amino-transferase,AST)、白蛋白(albumin,ALB)、总胆红素(total bilirubin,TBIL)、氨基末端脑钠肽前体(N-terminal pro-brain natriuretic peptide,NT-proBNP)、白介素-6(interleukin-6,IL-6)、CD3+CD4+T细胞计数、抗核抗体(antinuclear antibodies,ANA)、ANCA。③影像学资料:超声心动图、冠状动脉造影结果等。
1.3 血清ANCA检测
采集肝素促凝血3 mL,静置至血液凝固,2 683×g离心5 min后取血清,2~8 ℃冰箱保存。采用间接免疫荧光法测定血清ANCA,所用试剂均购于欧蒙医学实验诊断有限公司,荧光显微镜下出现特异性荧光判为阳性。
1.4 结局定义与分组
1.4.1 IVIG抵抗
IVIG抵抗定义为标准治疗结束后36 h体温仍高于38 ℃或用药后2周内再次发热,并伴有KD的其他临床表现[1]。
1.4.2 CAL诊断标准
CAL的诊断符合2012年版《川崎病冠状动脉病变的临床处理建议》[8],包括冠状动脉扩张、冠状动脉瘤、冠状动脉狭窄和闭塞等。满足以下任意一项即可诊断为冠状动脉扩张性病变:冠状动脉内径>3 mm(<5岁),或冠状动脉内径>4 mm(≥5岁);病变节段内径与邻近正常节段内径的比值≥1.5;经体表面积校正后的冠状动脉内径Z值≥2。
1.4.3 病例分组
将训练集和验证集分别分为CAL组(合并CAL)和NCAL组(不合并CAL)。
1.5 统计学分析
采用SPSS 27.0和R 4.3.1软件进行统计分析。符合正态分布的定量资料用x±s表示,非正态分布的定量资料用M(Q1,Q3)表示,组间比较分别采用t检验和Mann-Whitney U检验;定性资料以频数(百分率)表示,组间比较采用χ2 检验。将单因素分析、最小绝对收缩和选择算法(least absolute shrinkage and selection operator,LASSO)筛选的危险因素纳入多因素Logistic回归分析,得到CAL的预测因素并构建列线图模型。进一步采用pROC软件包绘制受试者操作特征(receiver operating characteristic,ROC)曲线,评价模型区分度;采用rms软件包绘制校准曲线,评价模型校准度;采用rmda软件包进行决策曲线分析(decision curve analysis,DCA),评价模型临床适用性。将模型指标中的连续型数据根据正常参考范围转化为分类数据,并根据Logistic回归方程中的自变量系数对各指标进行赋分,得到评分模型。采用Medcalc 23.0.1软件对比新评分系统与Kobayashi评分[9]、Egami评分[10]、Sano评分[11]的ROC曲线,并计算曲线下面积(area under the curve,AUC)、灵敏度、特异度、约登指数等。P<0.05为差异具有统计学意义。
2 结果
2.1 训练集和验证集临床资料比较
本研究共纳入340例KD患儿。训练集237例,其中58例KD患儿发生CAL(24.47%);验证集103例,其中19例KD患儿发生CAL(18.45%);2组差异无统计学意义(P=0.281)。2组各项基线资料比较,差异均无统计学意义(均P>0.05),具有可比性(表1)。
表1 训练集和验证集基线资料比较
Tab1
| Variable | Training set (n=237) | Validation set (n=103) | P value① | |||||
|---|---|---|---|---|---|---|---|---|
| CAL (n=58) | NCAL (n=179) | P value | CAL (n=19) | NCAL (n=84) | P value | |||
| Gender/n(%) | <0.001 | 0.682 | 0.897 | |||||
| Male | 45 (77.59) | 85 (47.49) | 12 (63.16) | 46 (54.76) | ||||
| Female | 13 (22.41) | 94 (52.51) | 7 (36.84) | 38 (45.24) | ||||
| Age/month | 30.00 (12.25,47.50) | 33.00 (18.00,56.50) | 0.148 | 29.00 (15.00,49.50) | 31.00 (18.75,52.50) | 0.702 | 0.781 | |
| Fever duration before IVIG/n(%) | 0.228 | 0.316 | 0.941 | |||||
| <5 d | 13 (22.41) | 26 (14.53) | 5 (26.32) | 13 (15.48) | ||||
| ≥5 d | 45 (77.59) | 153 (85.47) | 14 (73.68) | 71 (84.52) | ||||
| CRP/(mg·L-1) | 64.80 (37.95,103.03) | 63.40 (34.90,95.10) | 0.601 | 81.70 (51.60,109.70) | 61.60 (39.78,101.92) | 0.161 | 0.420 | |
| NEUT%/% | 66.26 (55.21,76.89) | 65.70 (53.70,76.78) | 0.895 | 68.21 (55.82,80.45) | 67.32 (52.11,74.67) | 0.280 | 0.933 | |
| HGB/(g·L-1) | 107.00 (101.00,115.00) | 110.00 (103.00,117.00) | 0.209 | 107.00 (101.00,112.50) | 110.50 (102.75,116.00) | 0.139 | 0.462 | |
| PLT/(×109·L-1) | 332.00 (227.50,452.50) | 336.00 (259.00,416.00) | 0.702 | 253.00 (219.50,338.00) | 329.50 (262.75,417.00) | 0.038 | 0.170 | |
| Na+/(mmol·L-1) | 135.60 (134.00,137.25) | 137.00 (135.00,138.65) | 0.010 | 136.00 (133.80,137.55) | 136.55 (135.00,138.15) | 0.190 | 0.922 | |
| ESR/(mm·h-1) | 65.50 (44.75,75.00) | 65.00 (48.00,81.00) | 0.531 | 66.00 (54.50,74.00) | 71.00 (54.50,87.00) | 0.655 | 0.120 | |
| Ferr/(ng·mL-1) | 171.25 (110.67,270.36) | 175.52 (130.20,241.20) | 0.712 | 252.60 (152.65,291.90) | 167.55 (136.88,272.72) | 0.241 | 0.177 | |
| ALT/(IU·L-1) | 27.50 (18.00,48.75) | 24.00 (15.50,59.50) | 0.356 | 30.00 (20.00,134.50) | 28.00 (18.75,60.00) | 0.540 | 0.215 | |
| AST/(IU·L-1) | 33.00 (27.00,47.75) | 33.00 (28.00,46.50) | 0.731 | 33.00 (28.00,70.00) | 34.00 (26.00,42.50) | 0.507 | 0.579 | |
| ALB/(g·L-1) | 34.62±4.36 | 3 6.66±3.98 | 0.002 | 35.22±2.87 | 36.04±4.59 | 0.326 | 0.586 | |
| TBIL/(μmol·L-1) | 8.80 (6.12,12.15) | 8.20 (5.60,11.45) | 0.352 | 9.10 (4.85,24.05) | 8.75 (6.65,11.72) | 0.953 | 0.264 | |
| NT-proBNP/(pg·mL-1) | 652.00 (185.50,2460.25) | 302.00 (104.00,984.00) | 0.014 | 291.00 (94.50,1528.50) | 406.00 (186.50,1189.75) | 0.792 | 0.691 | |
| IL-6/(pg·mL-1) | 74.51 (25.09,246.56) | 71.00 (33.11,169.65) | 0.402 | 74.39 (35.95,150.71) | 80.64 (24.84,167.78) | 0.875 | 0.905 | |
| Count of CD3+CD4+/n | 1 173.84 (694.43,1 779.17) | 1 183.81 (635.24,2 016.62) | 0.857 | 862.66 (396.05,1 269.99) | 1 110.01 (670.87,2 006.53) | 0.208 | 0.724 | |
| ANA/n(%) | 0.934 | 0.547 | 1.000 | |||||
| Positive | 12 (20.69) | 40 (22.35) | 5 (26.32) | 17 (20.24) | ||||
| Negative | 46 (79.31) | 139 (77.65) | 14 (73.68) | 67 (79.76) | ||||
| ANCA/n(%) | 0.007 | 0.002 | 0.946 | |||||
| Positive | 8 (13.79) | 6 (3.35) | 5 (36.32) | 2 (2.38) | ||||
| Negative | 50 (86.21) | 173 (96.65) | 14 (73.68) | 82 (97.62) | ||||
| IVIG resistance/n(%) | 0.008 | 0.031 | 0.846 | |||||
| Yes | 14 (24.14) | 17 (9.50) | 6 (31.58) | 9 (10.71) | ||||
| No | 44 (75.86) | 162 (90.50) | 13 (68.42) | 75 (89.29) | ||||
2.2 KD患儿发生CAL的预测因素
根据是否发生CAL,将训练集分为CAL组和NCAL组,进行单因素差异性分析。结果显示2组在性别、Na+、ALB、NT-proBNP、ANCA阳性率、IVIG抵抗方面的差异均具有统计学意义(均P<0.05),其余指标的差异均无统计学意义(均P>0.05),详见表1。
图2
图2
训练集的LASSO回归模型风险因素筛选
Note: A. LASSO coefficient profiles for 6 variables. B. A 9-fold cross-validation used in the LASSO regression. Dotted vertical lines represent the optimal values, determined using the minimum criteria (left dotted line) and the 1 standard error criterion (right dotted line). λmin=0.000 6, λ1se=0.064 0.
Fig 2
Risk factors selection using a LASSO regression model in the training set
表2 训练集KD患儿合并CAL影响因素的Logistic回归分析
Tab 2
| Variable | B | SE | P value | OR | 95%CI |
|---|---|---|---|---|---|
| Male | -1.259 | 0.362 | 0.001 | 0.283 | 0.135‒0.564 |
| ALB | -0.115 | 0.041 | 0.005 | 0.891 | 0.820‒0.964 |
| ANCA | 1.292 | 0.620 | 0.037 | 3.639 | 1.078‒12.740 |
| IVIG resistance | 0.653 | 0.437 | 0.135 | 1.922 | 0.800‒4.496 |
2.3 列线图预测模型的构建
相比于由性别、ALB、ANCA构建的CAL预测模型,由性别、ALB、ANCA、IVIG抵抗4个因素构建的模型赤池信息准则(Akaike information criterion,AIC)值更小,模型拟合更好。根据AIC准则构建的模型方程为Logit P=4.508-1.259×性别-0.115×ALB+1.292×ANCA+0.653×IVIG抵抗;式中ALB单位为g/L。绘制列线图,将回归方程可视化(图3)。通过下方的刻度尺,可以得出每个变量在不同取值下所对应的单项分数,单项分数相加为总得分,与总得分相对应的预测概率即为KD患儿发生CAL的风险。
图3
图3
KD患儿合并CAL的列线图预测模型
Fig 3
Nomogram for prediction of CALs risk in patients with KD
2.4 列线图预测模型的验证与评价
2.4.1 区分度
采用Bootstrap重抽样法在训练集和验证集的ROC曲线中对模型进行验证(重抽样次数为500),AUC分别为0.747(95%CI 0.667~0.821)和0.645(95%CI 0.500~0.794),表明模型区分度良好 (图4)。
图4
图4
预测模型在训练集和验证集中的ROC曲线
Note: A.Training set. B.Validation set.
Fig 4
ROC curve of the nomogram in the training set and validation set
2.4.2 校准度
模型在训练集和验证集中的校准曲线显示,预测曲线及Bootstrap重抽样500次的校准曲线贴近对角线,预测概率与实际概率一致性良好。进一步行Hosmer-Lemeshow拟合优度检验,在训练集和验证集中结果分别为χ2=5.105(P=0.746)和χ2=13.549(P=0.094),提示模型拟合良好(图5)。
图5
图5
预测模型在训练集和验证集中的校准曲线
Note: A.Training set. B.Validation set.
Fig 5
Calibration curve of the nomogram in the training set and validation set
2.4.3 临床适用性
训练集和验证集的DCA显示,当阈概率值分别在5%~61%和1%~36%范围时,采用该预测模型预测KD患儿CAL发生风险的净获益较高,表明该模型具有一定临床应用价值(图6)。
图6
图6
预测模型在训练集和验证集中的决策曲线
Note: A.Training set. B.Validation set.
Fig 6
Decision curve of the nomogram in the training set and validation set
2.5 预测评分模型性能的比较
表3 KD合并CAL影响因素的Logistic回归分析
Tab 3
| Variable | B | SE | P value | OR | 95%CI | Score |
|---|---|---|---|---|---|---|
| Male | -0.972 | 0.299 | 0.001 | 0.378 | 0.206‒0.670 | 5 |
| ALB<35 g·L-1 | 0.687 | 0.279 | 0.014 | 1.986 | 1.149‒3.441 | 3 |
| ANCA positivity | 1.550 | 0.507 | 0.002 | 4.711 | 1.758‒13.140 | 8 |
| IVIG resistance | 0.792 | 0.365 | 0.030 | 2.208 | 1.064‒4.481 | 4 |
表4 4种评分系统对KD合并CAL的预测性能比较
Tab 4
| Scoring model | AUC | 95%CI | Sensitivity | Specificity | PPV | NPV | Youden index |
|---|---|---|---|---|---|---|---|
| New score | 0.716 | 0.665‒0.764 | 0.584 | 0.787 | 44.6 | 86.6 | 0.372 |
| Kobayashi score | 0.549 | 0.494‒0.602 | 0.260 | 0.840 | 32.3 | 79.5 | 0.100 |
| Egami score | 0.556 | 0.502‒0.610 | 0.455 | 0.658 | 28.0 | 80.5 | 0.113 |
| Sano score | 0.537 | 0.483‒0.591 | 0.546 | 0.510 | 24.6 | 79.3 | 0.055 |
将3种目前常用的评分体系分别应用于340例KD患儿,对是否合并CAL进行预测。结果显示:Kobayashi评分特异度最高,灵敏度最低;Sano评分灵敏度最高,特异度最低;3种评分体系的约登指数均低于0.3。与以上3种评分体系相比,本研究建立的评分体系约登指数最高,灵敏度和特异度较高(表4)。
3 讨论
江志贵等[20]通过检测46例KD患儿的血清ANCA,发现KD患儿ANCA阳性率高于对照组健康儿童,但在不完全性KD和完全性KD之间的差异无统计学意义。本研究与类似研究比较,KD患儿的ANCA阳性率存在一定差异,考虑可能原因如下。①各研究样本量不同。②间接免疫荧光法结果主要由人工判读,主观性较强,误差较大[21]。③IVIG使用和ANCA检测时间顺序不统一。有研究[22-23]表明,静脉使用IVIG,被动输注自身抗体可导致在给药后的数周内,针对感染性或自身免疫性疾病进行的一系列血清学试验出现假阳性,如乙型肝炎、梅毒血清学和ANCA检测等。赵建美等[24]研究发现,合并CAL的KD患儿ANCA阳性率高于无CAL的KD患儿。可见,ANCA在KD相关冠状动脉病变的病理生理过程中可能发挥重要作用。传统的CAL预测模型多基于炎症指标和治疗响应参数,缺乏基因、代谢组学等分子层面信息和免疫学标志物等,不能完全适用于所有地区、种族的KD群体。本研究全面纳入包括ANCA在内的可获取的20项临床指标,构建新的预测模型并验证其效能,希冀为KD患儿的早期风险评估和精准管理提供参考。
除ANCA以外,本研究还发现男性、ALB水平低、IVIG抵抗也是CAL的危险因素。KD具有明显的性别、年龄特征,既往已有多项研究[25-26]报道了发病年龄<1岁、男性是CAL发生的独立危险因素,男性患儿并发CAL的危险性约为女性患儿的2倍。一项来自日本的为期20年的回顾性研究[27]也证实患者的年龄和性别与CAL的发生显著相关。ALB是一种由肝细胞合成的炎症急性期反应蛋白,可以作为多种物质的转运体,参与急性和慢性炎症过程,在感染和慢性炎症时其水平降低[28]。KD急性期ALB水平下降,考虑与肝功能损伤、毛细血管渗漏和肾脏受累导致正常排泄屏障受损及大量ALB排出有关。有研究[29]发现,合并CAL的KD患儿ALB水平低于正常KD患儿,合并冠状动脉瘤的KD患儿ALB水平低于冠状动脉扩张患儿,提示ALB可作为冠状动脉病变严重程度的评价指标。低蛋白血症也是许多心血管疾病的独立预后指标[30]。大剂量IVIG联合阿司匹林是预防急性期KD患儿发生CAL的一线治疗方案,早期应用IVIG可将CAL发病率由25%降至5%;但部分患儿对该治疗无反应,其CAL发生率明显升高[31]。有研究[32]显示,IVIG应用时间延迟和IVIG无反应都可能引起血管炎症的持续状态,造成血流动力学改变,从而导致CAL;与本研究结果一致。
综上所述,ANCA可以作为一项新的预测指标,用于建立CAL预测模型;与经典的Kobayashi评分、Egami评分、Sano评分比较,灵敏度和特异度均较高,具有一定临床推广价值,可实现对KD患儿CAL发生风险的个体化预测。但本研究为单中心回顾性研究,缺乏多中心外部验证评估模型泛化能力。因部分患儿资料不完整、未完善血清ANCA检测、在外院已接受初始治疗等,未纳入研究队列,导致样本量和结局事件数量偏少,可能导致模型过拟合。未来仍需开展多中心、大样本的前瞻性研究进一步验证。
作者贡献声明
陈蓉、沈捷参与研究方案设计、论文写作和修改;陈蓉、张锰、朱荻绮参与数据收集和数据分析;所有作者参与病例随访。所有作者均阅读最终稿件并同意提交。
AUTHOR's CONTRIBUTIONS
The study was designed by CHEN Rong and SHEN Jie. The paper was drafted and revised by CHEN Rong and SHEN Jie. The data were collected and analyzed by CHEN Rong, ZHANG Meng and ZHU Diqi. Case follow-up was completed by all the authors. All authors have read the last version of paper and consented to submission.
利益冲突声明
所有作者声明不存在利益冲突。
COMPETING INTERESTS
All authors declare no relevant conflict of interests.
参考文献
/
| 〈 |
|
〉 |

