孤立性肺结节良恶性判断的数学预测模型建立与验证

Establishment and Verification of a Mathematical Model for Predicting Malignancy of Solitary Pulmonary Nodules

目的:通过多因素Logistic回归分析,建立CT判断孤立性肺结节(SPN)良恶性的数学预测模型,并与目前已知的国内外模型进行比较分析。方法:回顾性收集2012年1月至2013年1月在复旦大学附属肿瘤医院胸外科经手术切除并明确病理诊断的SPN患者的临床及CT资料共200例(A组),通过多因素Logistic回归分析进行筛选建立方程。另收集2013年2月至2013年7月经手术切除且明确病理诊断的SPN患者资料共89例(B组)用以验证。结果:A组200例SPN中良性64例,恶性136例,建立的数学预测方程为:Y=ex/(1+ex),X=-2.085+0.058×年龄-1.206×性别-2.157×钙化+0.505×短毛刺+1.729×长毛刺+1.782×分叶-1.005×边界。e为自然对数。B组数据进行验证:本组模型曲线下面积最大,为0.888±0.051。本组模型的特异性最高(94.4%)>Mayo Clinical模型(88.9%)>VA模型(72.2%)>国内模型(66.7%)。国内模型的敏感性最高(88.7%)>本组模型(83.1%)>VA模型(78.9%)>Mayo Clinical模型(45.1%),P<0.05。结论:本组数据建立的模型诊断效能较高,收集的临床及CT资料较以往任何一篇报道更全且全部为中国人,优于国内外公式单纯套用。

Purpose： To establish a CT mathematical model for diagnosis of the solitary pulmonary nodules （SPN） with multivariate Logistic regression analysis, and compared with other known models. Methods： A retrospective study was carried out in Fudan University Cancer Hospital, which included 200 patients with definite pathological diagnosis of SPNs from Jan 2012 to Jan 2013 （group A）. The mathematical prediction model was established with multivariate analysis. Other 89 SPN patients （group B） with definite pathological diagnosis in our hospital from Feb 2013 to Jul 2013 were used to validate this model. Results： In group A, 32% of the nodules were malignant, and 68% were benign. The mathematical model established by logistic regression was： Y=e^x/（1＋e^x）, X=-2.085＋0.058×age- 1.206×gender-2.157×calcification＋0.505×short spiculation＋1.729×long spiculation＋1.782×lobution-1.005×border. The data in group B were used to validate our model; the area under ROC curve was 0.888±0.051, which was greater than the others. The specificity of our mathematical model was 94.4%, which was higher than that of Mayo Clinical model （88.9%）, VA model （72.2%）, and domestic model （66.7%）; The sensitivity of domestic model was the highest （88.7%）, which was higher than that of our mathematical model （83.1%）, VA model （78.9%）, and Mayo Clinical model （45.1%）, P〈0.05. Conclusion： The pre-established mathematical prediction model in our study has a high clinical value for diagnosis of SPN. Our prediction model is sufficient and accurate to pretest the malignancy of patients with SPN.