基于循证评价与数据挖掘构建我国肺癌筛查高风险人群决策树模型的研究

Lung cancer remains an important health problem in mainland China. The number of new cases and mortality will be substantially increasing. There is none evidence on lung cancer screening high risk population selection in China. In order to provide evidence for clinicians, address the satisfactions among participants and assistant policy-makers decide smart healthcare policy, we design to make high quality risk prediction model based on our national lung cancer screening data. Our study will comprehensively evaluate evidence and make the strength of recommendations. The study will include four components: (1) Using the approach of systematic review, data from literatures associated with screening benefits and harms will be reviewed and the natural history of lung cancer and its high risks will be characterized; based on these findings, evaluation of evidence quality by international scale such as AMSTAR and AGREE II checklists will be performed, and then we scale the evidence based on GRADE system; (2) This part will be conducted in 16 screening center based on “Cancer Screening Program in Urban China” using survey research approach to collect data with history, risks and others; (3) Using a data mining approach, a Decision Tree model of high risk population of lung cancer will be informed by data collected from high quality evidence (Component 1) and survey results (Component 2). Based on the validated models, we will predict the lung cancer incidence for various strategies from parameter evidence by receiver operator characteristic curve; (4) Using Delphi method to develop the recommendations of the Decision Tree model. We will popularize based on screening center and some conferences. The study is expected to assist the local healthcare policy decision-making for high risk population of lung cancer screening strategies, improve participants’ awareness and provide practice guidelines for screening clinicians in mainland China.

肺癌位居我国人群恶性肿瘤发病首位，疾病负担日益加剧。肺癌筛查在高风险人群中进行效果明显。目前我国尚未有高质量证据支持的特异性的肺癌筛查项目高风险人群评估模型，无法为医务人员提供规范化指导，也未能对人群提供高满意度普及，也不可协助卫生政策制定者进行决策。因此进行适合我国国情、全面科学和高质量的肺癌筛查高风险评估模型的研究非常必要。本项目采取循证评价与数据挖掘理念，基于国外高质量研究证据和我国多个现场的大数据资料，构建肺癌筛查高风险人群决策树模型。将用系统评价的方法收集和评估证据并进行证据质量分级推荐；依托“城市癌症早诊早治项目”平台16个筛查现场，开展相关指标问卷调查；采用决策树法构建高风险人群决策树模型，最后采用德尔菲法评价模型适用性并给出高强度推荐意见，预期制定出具有我国人群特异性的肺癌高风险人群决策模型，为我国肺癌筛查的一线医务工作人员、受试者和卫生决策者提供重要科学的参考依据。

肺癌位居我国人群恶性肿瘤发病首位，疾病负担日益加剧。肺癌筛查在高风险人群中进行效果明显。目前我国尚未有高质量证据支持的肺癌筛查高风险人群分级评估模型，无法精确选择筛查人群并进行精准化筛查，尚不能为医务人员提供规范化指导，也未能对人群提供高满意度普及，也不可协助卫生政策制定者进行决策。因此进行适合我国国情、全面科学和高质量的肺癌筛查高风险评估模型的研究非常必要。本项目采取循证评价与数据挖掘理念，通过全面检索和系统评估全球当前肺癌筛查流行病学高质量证据，共733篇相关研究，包括随机对照试验、观察性研究、系统评价和实践指南等，并根据相对应的证据质量的评估工具进行质量评价和证据质量严格分级与推荐，确定模型参数选择。基于现场的回顾性临床试验方法进行流行病学资料收集，依托“城市癌症早诊早治项目”平台的筛查现场选取了5个具备代表性的现场，共纳入1826例研究对象，中位年龄为50.56±9.87。随访3年发现5例肺癌病例。采用病例对照研究，使用SAS9.4软件，并使用决策树中CHAID计算方法构建肺癌筛查人群风险决策树模型。该模型可预测前瞻性5年内的肺癌风险度，并可根据风险因素分为5个层级，包括极低危、低危、中危、高危和极高危。构建决策树预测模型的方法同样适用于其他癌种的预测模型构建中。同时，本项目构建数学模型与真实研究相结合的方式，充分认识到构建创新型数学统计模型与真实世界癌症防治效果的密切关系，在今后的研究中，也会按照此思路，更加符合实际需要使用数学统计的方法解决问题。本项目获得了肺癌筛查领域中各种形式的证据，包括随机对照试验的证据、系统评价的证据及临床实践指南中的证据，并对这些证据进行了严格的质量评价，可用于肺癌筛查证据的直接使用。同时，可对肺癌发病风险进行精细判断，为一线工作人员提供了较为便捷且实用的参考依据，为今后为进一步制订风险度分层指导下的肺癌精准筛查方案提供前期基础。