基于循证大数据的肛瘘复发风险评估及预测模型的构建研究

Anal fistula is one of the most common anorectal disease threatening people’s health. Its high postoperative recurrence rate has always been a tough problem in clinics. It is of high significance to use mathematical models for the prediction of the outcome of disease which can early intervene and reduce the occurrence of adverse events. Our previous research has showed that anal fistula recurrence is associated with multidimensional factors. In this project, we will collect large sample cohort data of anal fistula patients using evidence-based approach to establish a clinical research database based on evidence-based literature search for anal fistula patients, which will be used to analyze, screen and validate the effect indicators and risk prediction model for anal fistula recurrence. The following three aspects of research will be conducted. (1) Based on decision tree model, random forest method, the nomogram model and other data mining methods, we will establish mathematical models of the anal fistula recurrence risk assessment and prediction. (2) We will analyse the clinical data (patient baseline characteristics, traditional Chinese medicine (TCM) syndrome, blood biochemical test and immunity indicators, imaging examinations, etc.) and patient follow-up data that can influence the postoperative recurrence of anal fistua. (3) We will analyze the risk factors of postoperative anal fistula recurrence and other adverse events from multilevel and multiple approach. We will also explore the correlation between the related indicators obtained from evidence-based literature search and the recurrence of anal fistula to establish the risk assessment and prediction model of anal fistula recurrence. Moreover, we will evaluate the model to provide early warning and precise prevention mechanism for anal fistula recurrence, and to provide scientific basis for clinical research and application.

肛瘘是危害健康的常见肛肠疾病，手术后高复发率一直是临床上的棘手难题。运用数学模型预测疾病的转归早期干预不良事件的发生意义重大。前期研究发现肛瘘复发与多维因素相关。本项目应用循证医学方法，收集肛瘘患者的大样本研究数据，建立基于循证检索的肛瘘临床数据库，分析、筛选和验证肛瘘复发的效应指标和建立风险预测模型，研究内容包括：（1）基于决策树模型、随机森林法、列线图等数据挖掘方法，建立肛瘘复发风险评估和预测模型。（2）分析纳入患者的临床（基线资料、中医证候、血液生化及免疫指标、影像学检查等）和随访资料等信息，探讨其对肛瘘复发的影响，筛选肛瘘复发的危险因素。（3）从多途径、多层次对肛瘘复发及其他不良事件的危险因素进行分析，探讨基于循证文献检索的相关指标与肛瘘复发风险的相关性，建立肛瘘复发的风险评估和预测模型，并对模型进行评价，以提供肛瘘复发的早期预警和精准预防机制，并为临床科研推广应用提供科学依据。

明确影响肛瘘术后复发因素对理解病因，进行病因预防和个体化精准干预意义重大。本项目基于对肛瘘复发危险因素的文献行系统评价和证据分级，进行Delphi专家咨询形成共识得到肛瘘术后复发的核心危险因素集。同时构建肛瘘队列，筛选肛瘘术后复发的预测因素构建预测模型，通过ROC曲线下面积对模型进行评价，采用Hosmer-Lemeshow拟合优度检验评价模型的性能。（1）对于患者相关因素，肛门手术史与肛瘘术后复发显著相关(RR=1.52; 95％CI，1.04-2.23)，为中等质量（II类）证据；对于瘘管和手术相关因素，肛瘘类型为高位经括约肌间瘘(RR=4.77;95％CI, 3.83-5.95）、内口位置不明确（RR=8.54,95％CI,5.29-13.80）、马蹄形扩展（RR，1.92;95％CI, 1.43-2.59）与复发显著相关，为高质量（I类）证据；而是否行肛瘘挂线手术（RR=2.97,95％CI，1.10-8.06）及多发瘘管（RR=4.77;95％CI，1.46-15.51）也与复发显著相关，为中等质量（II类）证据。（2）Delphi专家咨询共有49个危险因素纳入评估，在3个领域的14种危险因素上达成共识。（3）基于回顾性资料收集我院1998例连续性肛瘘患者信息，共45例肛瘘患者确定为术后复发。在全分析集中，应用Logistic回归确定的复发危险因素建立预测模型，ROC显示模型具有良好的区分能力(AUC=0.768)，并采用不同的内部验证方法（bootstrap重采样和6:4比例随机分割为训练集和验证集）均具有良好的区分能力。校准曲线显示出良好的拟合优度。基于机器学习和分类决策树建模得到的区分度也较为一致。（4）前瞻性收集本院564例肛瘘患者的信息，随访6-12个月，建立肛瘘复发的高危人群的临床信息数据库。随访期间发生复发病例90例；分别应用多元Logistic回归、列线图、机器学习、LASSO回归、分类决策树分别建立肛瘘复发的预测模型，并融合中医证候类型构建中医证候的肛瘘复发模型。发现这些模型对于肛瘘复发的区分度均较为良好（AUC范围 0.752-0.768）；以bootstrap重采样和校准曲线对模型进行评估和拟合优度检测，均具有良好的性能和一致性；决策曲线也显示模型对肛瘘患者有较好的临床获益，本项目对肛瘘术后复发的精准化判断和提前干预起到了很好的作用。