Logistic回归模型在卵巢肿瘤良恶性鉴别诊断中的应用

目的 探讨Logistic回归模型在鉴别卵巢肿瘤良、恶性中的临床应用价值.方法 选择经手术病理证实的601例卵巢肿瘤病例,术前记录包括其一般资料、二维灰阶超声、彩色多普勒超声在内的共计35个指标.其中400例作为模型训练组用于模型的创建,另201例作为模型验证组用于模型的评判.结果 筛选引入方程的指标包括:卵巢癌个人史、肿瘤的最大直径、肿瘤实质部分的最大直径、多房性囊实性肿块、实性肿块、腹水、乳头上血流、不规则囊壁、肿瘤血管评分4分及声衰减;模型对验证组病例诊断的受试者工作特征曲线下面积、灵敏度、特异度分别为0.963、93.9%和93.1%.结论 术前应用Logistic 回归模型对卵巢肿瘤良恶性有良好的鉴别诊断能力.

Preoperative diagnosis of ovarian tumors using Logistic model

Objective To assess the diagnostic value of Logistic model in differentiating between malignant and benign ovarian lesions. Methods Thirty-five indexes of clinical and ultrasound data were recorded in 601 ovarian lesions confirmed by surgical pathology. The Logistic model was developed on a training set( n - 400) and tested on a test set( n = 201). Results Variable selection resulted in a set of 10 variables for the models: personal history of ovarian cancer, maximal diameter of the lesion, maximal diameter of the solid component, multilocular-solid tumor, solid tumor, ascites, flow within papillation, irregular walls, very strong intratumoral blood flow (i. e. color score 4) and acoustic shadows. Test set area under the receiver-operating characteristics curve was 0.963 with a sensitivity 93.9% and a specificity 93. 1 %. Conclusions Logistic model can accurately separate malignant from benign ovarian masses.