Statistics analysis

Statistical Package for Social Sciemces (SPSS, Inc, Chicago, IL, USA) was used for Statistics analysis. Univariate analysis was performed using chi-square criterion while multivariate analysis was performed using logistic regression analysis. Multiple testing issue was considered when age and tumor size were compared in several groups. A scoring model was constructed to further display the relationships among factors and probability of CLNM. The score of each risk factor was weighted according to the beta coefficient obtained from the logistic regression model. For convenience, all the beta coefficient divided the least one and then rounded to the nearest whole number to keep the scoring model simple. The total score for each patient represented the sum of scores for each risk factor and a risk-scoring model was built to predict the stratified CLNM in PTC patients. To evaluate the predictive performance of the scoring model and find a appropriate cut-off point, we adopted the receiver operating characteristic curve (ROC curve) and evaluated it by the area under the ROC curve (AUROC). Kaplan Meier curves were performed before Cox proportional hazard models. The log-rank test was used to tell the differences of recurrence statistically. A Cox regression model was used to determine prognostic factors. A difference was considered statistically significant when p  < 0.05 and in factors compared.When comparing multiple groups, the new P value was considered to be 0.05/group number.