*2.4. Statistical Analysis*

SPSS Statistics version 23.0 (IBM Corp., Armonk, NY, USA) and R 3.5.3 (R Development Core Team, Vienna, Austria) were used for statistical analyses. Continuous data were compared by Student's t-tests. The comparison of dichotomous data was performed by Pearson chi-squared tests. To minimize potential confounding effects and selection bias, we performed 3:1 propensity score matching [21]. We selected 4 factors that could affect the recurrence as follows: tumor size, microscopic ETE, LN metastasis, and coexisting

Hashimoto thyroiditis. RFS was assessed by Kaplan–Meier survival analysis and log-rank test. As the RFS curves for lobectomy met those for total thyroidectomy, we checked the proportionality of the hazards to use a Cox proportional hazards regression model. The log(-log(survival)) plot and the scaled Schoenfeld residuals test for assessment of proportionality of hazards were used to verify the proportional hazards assumption [22]. The log(-log(survival)) plot for operative extent gave rise to reasonably parallel lines and suggested proportionality (Figure S1). The scaled Schoenfeld residuals test also produced no evidence of a poor fit (*p* = 0.42). Univariable and multivariable Cox proportional hazards regression models, therefore, were used to identify risk factors that could affect recurrence. A *p*-value < 0.05 was considered to indicate a statistically significant difference.
