Our study included 159 patients with RAIR-TC and 759 matched-controls. The characteristics of the patients with RAIR-TC were as expected according to previous literature [7, 8], with lungs being the most frequent metastatic location followed by neck and bone metastases. Men (38%) and follicular cancers (28%) were over-represented in comparison with the whole DTC population in the literature [19]. The large number of RAIR patients included in our study allowed the identification of seven independent clinical correlated factors for the evolution towards RAIR-disease, which were: age at diagnosis ≥ 55 years; synchronous metastases at initial work-up in neck, lungs and bones; metastatic persistence or recurrence in neck and lungs.

The strength of our work relied on the original and powerful methodology. First, the nested case–control study design made it possible to control potential confusion bias as the radioiodine sensitive patients were comparable to the cases in terms of management, including the use of radioiodine, follow-up as they came from the same cohort of patients. Furthermore, matching the cases and the controls on main criteria such as patient sex, histological type and tumor size, made it possible to investigate other important potential risk factors that otherwise could have been hidden. Second, our methodology took into account the time between DTC and RAIR diagnosis in the analysis. Indeed, as RAIR-disease can occur from the beginning of patient management to 29 years later, comparing RAIR and radioiodine sensitive patients with the same duration of follow-up is an important factor in the methodology. This is the major limitation of all previously published studies on the same topic and predictive factors for becoming RAIR can, therefore, be misinterpreted. Third, our study is one of the first to have investigated the role of both synchronous and metachronous distant metastases in predicting the occurrence of RAIR status.

Furthermore, very few studies have addressed the question of risk factors for developing RAIR-disease in DTC [13,14,15,16] and the methodology as well as the choice of the control was disputable. Li et al. included 112 cases and 224 unmatched controls randomly selected among patients who had undergone total thyroidectomy and RAI therapy. The independent risk factors for developing RAIR-disease in this work were smoking, tumor histological type (FTC), ETE, number of lymph node metastases (≥ 4), lymph node metastases ratio ≥ 53% and pN stage (N1) while age and multifocality were not independent risk factors [13]. Shobab et al. studied DTC patients with DM and with a minimum follow-up of 3 years. This study included 54 RAIR-TC for 22 sex and age-matched controls and found no statistically significant differences between the RAIR and the controls in tumor size, ETE and histology [15]. Only the cumulative RAI dose and number of RAI-therapies were significantly higher in the cases. Liu et al. in a recent retrospective study focused on DTC with DM evaluated by 18F-FDG PET/CT. They included 223 RAIR-TC and 181 non RAIR-DTC and found that age at diagnosis ≥ 48, recurrence between the thyroidectomy and the first RAI-therapy (mean time between thyroidectomy and RAI-therapy of 9 and 26 months, respectively, for the radioiodine sensitive and RAIR groups) and 18FDG uptake on the metastatic site were independently associated with RAIR-DTC [16]. However, RAIR-DTC and non RAIR-DTC were not matched in this study and the median time of follow-up was short (20 months).

A meta-analysis by Luo et al. in 2020 including 13 studies concluded that ETE and high-risk histological type (including tall cell PTC, diffuse sclerosing variant of PTC, hobnail PTC, FTC and PDTC) increased the risk of developing RAIR-disease [14]. They found no difference regarding sex, age, tumor size, multifocality, or lateral lymph node metastases. However, the 13 studies included in the meta-analysis were significantly heterogeneous in terms of methodology and studied population. Some studies had a small sample size (n = 40 to 336) and others had a low level of evidence as they were not case–control studies but case series with unmatched controls [13, 20]. Finally, Kersting et al. focused on the subgroup of PDTC and found that tumor size > 40 mm, ETE and age > 55 were significant predictors of RAIR-disease in a total of 51 patients [21].

Based on histological and clinical variables of interest, we proposed a score for predicting the occurrence of RAIR-disease calculable after total thyroidectomy and RAI therapy with a sensitivity of 86% (CI 80–91%) and a specificity of 92% (CI 90–94%). Owing to the statistical methodology used, as cases and controls were matched on follow-up duration, the time to onset of RAIR-disease prediction was not possible.

In the literature, two prediction scores of RAIR-disease are available based on clinical and histological factors identified to be independent risk factors [13, 16]. The first one proposed by Li et al., discussed above, and including smoking, tumor histological type (FTC), ETE, number of lymph node metastases (≥ 4), lymph node metastases ratio ≥ 53%, and pN stage (N1) had an AUC of 0.876 with a sensitivity and specificity of 77.7% and 81.2% [13]. The second one proposed by Liu et al. and including age ≥ 48, recurrence between the operation and iodine-131 treatment, 18F-FDG uptake on the metastatic sites had an AUC of 0.898 with a sensitivity and specificity of 76% and 93% [16]. In a different way, a biological prediction model based on the preoperative thyroglobulin (pre-TG) was proposed by Cheng et al. including 90 RAIR-TC and 786 controls [22]. They found that elevated pre-TG was correlated with RAIR-disease. A cut-off value of 70.05 ng/ml was retained to distinguish between both groups with 62.2% sensitivity and 77.1% specificity. AUC was 0.76. In total, with an AUC of 0.95 and a maximal Youden index of 0.78, our model has the strength of efficiently and clearly discriminating between radioiodine sensitive and RAIR-TC patients, compared to previous scores.

Nevertheless, our work has some limitations. Matching the cases and the controls on few key criteria offered the advantages described above, but a classical cohort study design could have allowed us to adjust these factors for the whole cohort. Another limitation laid in not considering histological subtypes, especially aggressive subtypes that were not systematically available on the histology reports for our patients. At last, we could not test our score on an external validation cohort due to the rarity of the RAIR-disease. However, the results obtained by cross-validation (AUC = 0.93, maximum absolute calibration error = 0.46, Brier score = 0.16) and by bootstrap (AUC = 0.95, maximum absolute calibration error = 0.45, Brier score = 0.16) showed high internal validity of the score.

Genotyping RAIR-TC in our cohort is the subject of ongoing research. Some previous works have specifically examined the impact of mutations in the telomerase reverse transcriptase (TERT) promoter and in the BRAF oncogene, belonging to the MAPK pathway, especially the V600E mutation. While the association of BRAF and TERT mutations seems to be clearly associated with RAIR-disease [20, 23, 24], the impact of each mutation alone remains subject to debate [20, 23, 24]. TC somatic genotyping will probably help to better predict the evolution toward RAIR status.