[Logistic regression analysis of clinical and ultrasonic features of breast nodules].

ABSTRACT

OBJECTIVE: To create a breast nodule estimation model based on grayscale and color Doppler ultrasonography using Logistic regression that can screen out the specific features for distinguishing breast malignancy from benignancy. METHODS: From July, 2009 to May, 2010, 217 patients were enrolled in the study in peking university first hospital. Clinical data and ultrasonic features were evaluated in 219 breast nodules of 217 patients confirmed by surgical pathology. Logistic regression model was established to screen out significant ultrasonic indexes for differentiating breast malignancy from benignancy. A receiver operating characteristics curve was made to assess diagnostic value of the Logistic regression model. Correlation was analyzed between the Logistic regression model and surgical pathology. RESULTS: Logistic regression model Logit(p) = -16.884 + 0.037 × age + 3.228 × longitudinal-transverse axis ratio + 1.412 × border + 2.663 × halo + 1.813 × microcalcium + 1.157 × resistance index + 2.204 × enlarged axillary lymph node (χ(2) = 167.107, P = 000). The areas of ROC curve for probability and identification of breast malignant and benign nodule were 0.948 and 0.882 respectively. Diagnostic sensitivity, specificity and accuracy were 91.6%, 84.9% and 88.9%. Logistic regression model positively correlated with surgical pathology (r = 0.768, P = 0.000). CONCLUSION: Our Logistic regression model can effectively differentiate malignant breast nodules from benign and can identify the ultrasonic features associated with breast cancer.