基于病证结合的放射性肺损伤早期预测模型的建立和验证

Radiation-induced lung injury (RILI) is the major toxicity of radiotherapy for lung cancer. The early prediction of RILI will make us intervene early to prevent adverse consequences. It is a hot issue to construct a multi-factor model for predicting RILI now and there is no report for predictive models based on the combination of disease and syndrome. Our previous studies first found the Traditional Chinese medicine (TCM) syndrome associated with RILI and found a variety of risk factors associated with RILI in Western medicine. We also established a database of patients with lung cancer received radiotherapy which had complete data. Our project will further include patients into the database and extract the basic elements of the TCM syndrome based on the information collected by the TCM four diagnostic methods. We will also collect clinical information and radiation dose-volume parameters and detect the protein and genetic markers associated RILI. Then, a modified LKB NTCP model and a Support Vector Machine model will be used to establish two predictive models of RILI respectively, which verified and compared in another cohort of patients and finally to establish an efficient predictive models. The model based on the combination of disease and syndrome will help early predict the occurrence of RILI and guide the prevention and treatment of RILI by the means of Integrated Traditional Chinese and Western Medicine.

放射性肺损伤(radiation-induced lung injury，RILI)是肺癌放疗的主要毒性反应，早期预测其发生有助于早期预防，避免不良后果。构建多因素RILI预测模型是目前研究热点，尚未见病证结合RILI预测模型的报道。前期研究首次确立了与RILI发生相关的肺癌中医证型，明确了多种RILI西医风险因素，并初步建立了具有完备资料的肺癌放疗患者数据库。本项目在前期工作基础上继续入组病例，根据患者放疗前四诊信息提取证候要素，收集临床信息和放疗剂量体积参数，并检测蛋白和基因水平相关指标。以放疗前中医证候要素和西医风险因素为预测因子，以发生RILI为结局，分别通过优化传统LKB NTCP模型的方式和支持向量机技术建立模型，并对各模型进行验证和对比，最终确立高效力的RILI预测模型。基于病证结合的RILI模型不但有助于RILI的准确预测，而且对RILI的中西医结合防治具有指导意义。

放射治疗在肺癌治疗中起着不可或缺的作用，放射性肺损伤(radiation-induced lung injury，RILI)是其主要的毒性反应之一，有症状RILI的发生率可达27.3%。RILI会引起一系列不良后果，包括：延误治疗进程，降低生活质量，影响预后等。早期预测RILI发生可以帮助医生对高危病例早期采取措施预防，避免不良后果。构建多因素RILI预测模型是目前研究热点，尚未见病证结合RILI预测模型的报道。团队首次提出肺癌患者放疗前的中医证候可预测RILI的假说，并在前期研究中证实中医证型（阴虚热毒和气阴两虚证）是≥2级RILI的独立危险因素。考虑到把复杂的证候归纳为几个简单的证型会造成丰富的四诊信息的丢失，本项目首先在前期研究基础上扩充样本量至452例，应用证候要素代替证型，同时联合放射物理参数及其他一般临床情况等作为自变量，以≥2级RILI的发生情况作为分类标签，采用Logistic回归单因素筛选和多因素分析确定中医证候要素中的阴虚证和热毒证以及西医临床物理因素中的年龄和全肺MLD是≥2级RILI的独立危险因素。之后利用筛选出的独立预测因子构建了预测RILI风险的列线图模型。该模型在经过1000次Bootstrap自抽样内部验证后，其C-index值为0.739，提示此模型具有较高准确度。校准曲线结果显示，此模型预测值与真实值之间符合度的平均绝对误差为0.029，预测风险接近于实际风险，符合度好。在此基础上，研究进一步入组199例作为外验证组，应用已建立模型预测外验证组RILI发生风险，其C-index值为 0.714，同时，校准曲线结果显示，此模型预测值与验证真实值之间符合度的平均绝对误差为0.019。外验证结果证实模型的预测准确度高，符合度好。该病证结合模型的建立不但有助于RILI的准确预测，而且对RILI的中西医结合防治具有指导意义。