风湿性心脏瓣膜病手术风险预测模型

Establishment of risk prediction model and risk score for in-hospital mortality after adult rheumatic heart valve sur- gery

目的利用我国心脏瓣膜外科治疗数据库，对我院20年间成人心脏瓣膜手术病例资料进行回顾性分析，建立成人风湿性心脏瓣膜病手术风险预测模型。方法选择我院1991年1月至2010年12月间年龄≥18岁的单纯风湿性心脏瓣膜病患者3889例为研究总体，将人选患者随机分为建模亚库（占全体60％）和验证亚库（占全体40％），按手术部位分为二尖瓣手术组、主动脉瓣手术组及联合瓣膜手术组，应用建模亚库数据建立模型。通过Hosmer—Lemeshowχ2检验反映模型预测的符合程度，通过ROC曲线下面积反映模型预测的鉴别效度。根据模型中各危险因素的权重系数β及其变量类型共同构建风险预测的评分标准。结果在院病死率为4．2％（165／3889）。所建立的模型共发现7个独立危险因素：心功能NYHA分级≥Ⅱ级（OR=3．36，95％CI：2．42—4．67）、术前肌酐值〉110mmol／L（OR=2．69，95％CI．1．51～4．79）、既往胸痛史（OR=2．33，95％CI：1．07～5．11）、手术状态（OR=2．32，95％CI：0．94～5．73）、既往高血压史（OR=2．24，95％CI：1．19～4．23）、术前危重状态（OR=2．14，95％CI：1．27—3．60）及年龄〉50岁（OR=1．57，95％CI：1．18～2．09）。建模亚库、验证亚库及三个亚组患者的Hosmer—LemeshowX2检验的P值均大于0．05且ROC曲线下面积均大于0．70，提示已建立的风险预测模型具有较满意的预测符合程度及鉴别效度，其预测准确性良好。评分方法：年龄51—60岁：1分、61～70岁：2分、〉70岁：3分；高血压史：1分；肌酐〉110umol／L：4分；NYHA分级Ⅱ级：2分、Ⅲ级：4分、Ⅳ级：6分；既往胸痛史：1分；术前危重状态：2分；限期手术：2分、急诊手术：4分。结论成功建立了一套适合自身的成人风湿性心脏瓣膜手术风险预测模型，确定了相应的在院死亡相关的危险因素及术前风险评估的评分方法，同时也为今后多中心手术风险预测模型的建立提供一定的参考。

Objective To establish a surgical risk prediction model for in-hospital mortality of adult rheumatic heart dis- ease. Methods The study sample comprised of 3 889 patients with adult （ is, or older than 18 years） rheumatic heart valve surgery only. All patients were divided into three subgroups according to the surgery site of left atrioventricular valve： mitral valve surgery group; aortic valve surgery group; and mitral and aortic valve surgery group. The data was splited into develop- ment（60% ） and validation（40% ） data sets, and then the risk model was developed by using a logistic regression model ac- cording to the data in development data set. Model calibration was analyzed by Hosmer-Lemeshow goodness-of-fit statistic, and model discrimination was tested by calculating the area under the receiver operating characteristic（ROC） curve. Risk score was finally set up according to the coefficient β and rank of variables in logistic regression model. Results The general in-hospital mortality of the whole group is 4.2% （ 165/3 889）. We established a risk prediction model and found seven risk factors： heart function in NYHA functional class ≥ⅡI grade （ OR = 3.36, 95% CI： 2.42 - 4.67 ） , preoperative creatinine 〉 110 mmol/L （ OR = 2.69, 95 % CI： 1.51 - 4.79 ） , history of previous chest pain （ OR = 2.33,95 % CI： 1.07 - 5.11 ） , surgical status （ OR = 2.32, 95% CI：0. 94 - 5.73） , previous history of hypertension（ OR = 2.24, 95% CI： 1.19 -4.23 ） , preoperative critical state （ OR = 2.14, 95% CI： 1.27 - 3.60） and age 〉 50 years （ OR = 1.57, 95% CI： 1.18 - 2.09 ）. Our risk model showed good calibration and discriminative power for the development data set, validation data set, and three subgroup in which Hos-mer-Leme-show test＇ s P value were greater than 0.05 and the area under the ROC curve were greater than 0.70. Scoring meth- ods ： age 51 - 60years ： 1 point, age 61 - 70 yeas ： 2 points, age 〉 70 years ： 3 points ; history of hypertension ： 1 point; creati- nine 〉 110 umol/L： 4 points; NYHA class stage Ⅱ ： 2 points, NYHA class stage Ⅲ： 4 points; NYHA class stage Ⅳ： 6 points; history of previous chest pain ： 1 point ; preoperative critical condition ： 2 points ; urgent surgery ： 2 points ： emergency surgery： 4 points. Conclusion We have created a new risk prediction model and risk score, which can accurately predicts outcomes in patients undergoing heart valve surgery for our center. Furthermore, our risk model can also enable beuchmarking and comparisons between multicenter in a meaningful way in the future.