Patients who underwent an initial surgery at the Asan Medical Center and were eventually diagnosed with MIFTC were identified. The medical records of these patients were retrospectively reviewed, and their clinicopathological characteristics and outcome data were recorded. Written informed consent for the clinical records to be used in this study was not obtained from the participants. Instead, patient identifiers were removed from all samples, and all data were de-identified prior to the analysis by a researcher who did not participate in this study. Distant metastasis included both synchronous and metachronous distant metastasis. Synchronous distant metastasis was defined as a metastasis detected at the time of the initial diagnosis or within 12 months after the initial operation.

After evaluating the selected risk factors for distant metastasis, patients were divided based on the number of risk factors: group 0 with no risk factors, group 1 with any one risk factor, group 2 with any 2 risk factors, and group 3 with all risk factors. Groups 1–3 were also subdivided into several subgroups based on the kind of involved risk factors. The relative risk for distant metastasis in each subgroup was estimated respective to the reference risk in the group without any risk factors. Moreover, DMFS rates between the aforementioned 4 groups were compared. Finally, risk groups were stratified with respect to DMFS rates.

Diagnostic hemithyroidectomy is the standard initial surgical procedure used at our institution to treat patients diagnosed with a follicular neoplasm by preoperative fine needle aspiration cytology. When a patient has another concomitant thyroid disease (i.e., Graves' disease or a nodular goiter of the opposite lobe), a total thyroidectomy is performed as the primary surgery. Central compartment node dissection is not routinely performed. However, ipsilateral or bilateral central compartment node dissection is performed if there are enlarged or suspicious lymph nodes. Completion thyroidectomy for MIFTC diagnosed after hemithyroidectomy is usually recommended for patients, especially when they had large primary tumor size (≥4.0 cm) or vascular invasion. Patients who refused completion thyroidectomy were provided sufficient information about the potential risks and received follow-up without further treatments.

All patients received follow-up examinations in the outpatient clinic at our institution. Patients were checked at 1, 6, and 12 months, and annually thereafter while receiving thyroid hormone treatment in order to decrease their serum thyroid-stimulating hormone (TSH) levels to subnormal levels without thyrotoxicosis. Thyroid function tests, including serum TSH, thyroglobulin (Tg), and anti-Tg antibody level, were routinely performed at every hospital visit, and neck ultrasonography and chest computed tomography (CT) scans were routinely performed once per year. Postoperative RAI remnant ablation was performed at 4–6 weeks after completion thyroidectomy or total thyroidectomy. Diagnostic whole body scan following thyroid hormone withdrawal or thyrogen administration was performed 6–12 months after remnant ablation with the simultaneous measurement of serum Tg.

Distant metastasis was diagnosed using whole body scan, chest CT, or ¹⁸F-deoxyglucose positron emission tomography/CT and confirmed by serial imaging or biopsy. We defined recurrence as the reappearance of pathologically proven malignant tissue and/or the appearance of structural lesions on imaging studies, regardless of the serum Tg levels. Biochemical recurrence with elevated serum Tg level but without clinical evidence of structural disease at 6 months after the initial operation was not classified as true recurrence.

Categorical variables are presented as numbers and percentages, and continuous variables as means±standard deviations. The risk factors for a distant metastasis were evaluated using univariate and multivariate Cox proportional hazard models, and the results are presented as hazard ratio (HR) with 95% confidence interval (CI). Multivariate analysis was performed after adjusting for all variables. To improve their clinical utility, continuous variables were converted into categorical variables on the basis of cut-off values, which were calculated using receiver operating characteristic (ROC) curve analysis. DMFS rates were calculated using the Kaplan-Meier method. Differences between the groups were analyzed for statistical significance using the log-rank test. All P values were 2-sided. Data were considered statistically significant if P<0.05. All statistical analyses were performed using SPSS (version 23.0; IBM Corp., Armonk, NY, USA).

Age, gender, surgical extent, primary tumor size, and extensive vascular invasion with ≥4 foci were potentially significant factors for distant metastasis by univariate analysis (Table 1). Age ≥45 years (HR=5.95; 95% CI=1.44–24.55; P=0.014), primary tumor size >4.0 cm (HR=4.79; 95% CI=1.18–19.33; P=0.028), and extensive vascular invasion (HR=19.63; 95% CI=5.18–74.39; P<0.001) were identified as independent risk factors by multivariate analysis. However, surgical extent and RAI ablation were not risk factors for prognosis in our study.

The relative risk of each risk factor for distant metastasis, based on its number and composition, to the reference risk (=1) in the group without any risk factors is described in Fig. 1 and Table 2. These risk factors differently influenced the development of distant metastasis according to their number and composition. Extensive vascular invasion with ≥4 foci was the most powerful risk factor for distant metastasis, irrespective of the presence or absence of other risk factors. The HRs of all subgroups with extensive vascular invasion were much higher than the others which did not have extensive vascular invasion.

As the number of involved risk factors increased, DMFS rates became significantly poorer (Fig. 2). Patients in groups 2 and 3 showed significantly worse DMFS rates compared to those of group 0 (P<0.001 for both) and group 1 (P<0.001 for both). The DMFS rates of group 3 were also significantly poorer than those of group 2 (P=0.004). Group 1 patients tended to have poorer DMFS rates compared to those of group 0 patients, but this difference was not significant (P=0.078). Therefore, these enrolled patients can be stratified into 3 risk groups with respect to DMFS rates: a low-risk group with ≤1 risk factor, an intermediate-risk group with any 2 risk factors, and a high-risk group with all 3 independent risk factors for distant metastasis.