IntroductionThyroid nodules are commonly encountered in clinical practice and are detected in up to 65 % of the general population (Dean and Gharib, 2008[15]). At the same time, most nodules are benign, and approximately 10-15 % harbor malignancy (Kamran et al., 2013[29]). Fine needle aspiration (FNA) cytology is the standard diagnostic test for thyroid nodules, and the “Bethesda System for Reporting Thyroid Cytopathology” is a widely accepted framework that classifies thyroid FNA biopsy results into six diagnostic categories, with each category reflecting a distinct likelihood of malignancy (Haugen et al., 2016[27]). In brief, the categories include I) non-diagnostic or unsatisfactory, II) benign, III) atypia of undetermined significance or follicular lesion of undetermined significance (AUS/FLUS), IV) follicular neoplasm or suspicious for a follicular neoplasm, V) suspicious for malignancy, and VI) malignant. Categories III, IV, and V are considered indeterminate and pose challenges in clinical management because they do not clearly distinguish between benign and malignant lesions. Consequently, these categories often require further diagnostic procedures, such as molecular testing, to guide clinical decisions (Haugen et al., 2016[27]).
The risk of malignancy for Bethesda III nodules ranges from 6-48 %, while Bethesda IV carries a 16-73 % risk, and Bethesda V has a 60-75 % risk (Singh and Wang, 2011[45]). Due to this uncertainty, most patients with indeterminate nodules are referred for diagnostic thyroid surgery; however, only 20-37 % prove to be malignant on final pathology (Yang et al., 2007[54]). This leads to potential overtreatment and unnecessary surgical risks.
Various molecular markers have been studied to improve the preoperative diagnosis of indeterminate nodules, including BRAF and RAS point mutations, RET/PTC rearrangements, and galectin-3 immunostaining (Bartolazzi et al., 2018[4]; Ferrari et al., 2018[20]; Lu et al., 2023[32]; Patel et al., 2017[40]). Studies have demonstrated up to 70 % of papillary thyroid cancer harbors a mutation (Oishi et al., 2017[39], Prete et al., 2020[41]). Among these, RAS mutations (in NRAS, HRAS, and KRAS genes) are found in 10-20 % of papillary thyroid cancers (Gilani et al., 2022[21]). Several studies have evaluated RAS testing specifically in nodules with indeterminate cytology, with mutation rates of 8.5 %-72 % reported (An et al., 2015[1]; Macerola et al., 2019[34]; Marotta et al., 2021[35]).
While several individual studies have been published, the diagnostic utility and clinical significance of RAS mutations in indeterminate nodules remain unclear. In this sense, we performed a systematic review and meta-analysis to synthesize the existing evidence on 1) the frequency of RAS mutations in indeterminate thyroid nodules and 2) the association between RAS mutation status and risk of malignancy. Additionally, we examined the relationship between RAS mutations and histologic subtypes of malignancy. Our findings aim to clarify the diagnostic utility of RAS mutation testing for triaging indeterminate nodules.
Materials and MethodsSearch strategyThis meta-analysis followed the Preferred Reporting Items for Systematic Review and Meta-Analyses (PRISMA) guidelines. We systematically searched PubMed and Embase to cover the published articles between 2010-2024, using a combination of relevant keywords related to indeterminate thyroid nodules, fine needle aspiration, RAS mutations, and thyroid cancer. We manually screened the reference lists of included studies and relevant reviews for additional eligible studies.
Selection criteriaStudies were included if they met the following criteria: (1) evaluated indeterminate thyroid nodules (Bethesda category III, IV, or V) diagnosed by FNA, (2) assessed RAS mutational status, (3) had surgical histopathological confirmation of nodule diagnosis, (4) reported data to calculate the proportion of RAS mutations and/or the association between RAS status and cancer risk. Case reports, conference abstracts, editorials, and non-English studies were excluded.
Three investigators independently screened the titles, abstracts, and full texts of retrieved studies against the eligibility criteria. Disagreements were resolved by consensus or consultation with a fourth investigator if needed.
Data extraction Three investigators used a standardized form to extract data from the included studies on study characteristics (author, year, country, design), patient characteristics (age, sex, nodule size), FNA details (Bethesda category, sample size), number of nodules tested for RAS, number of RAS mutations, final surgical pathology diagnosis, and data for 2x2 tables of RAS status (positive/negative) versus malignancy (positive/negative).
Study outcomesThe primary outcomes were the pooled proportion of RAS mutations in indeterminate nodules and the relative risk (RR) of malignancy associated with RAS mutation positivity. Secondary outcomes included the odds ratio for the histological subtype of malignancy (classical/follicular variant papillary thyroid cancer).
Single proportion meta-analysisWe calculated the pooled proportion of 1) Bethesda category III/IV/V nodules and 2) the rate of malignancy among surgically resected indeterminate nodules using the Freeman-Tukey double arcsine transformation to stabilize the variances and then summarized the proportions using a random effects meta-analysis model. For the rate of RAS mutations in indeterminate nodules, we used an arcsine square-root transformation to account for studies with RAS mutation rates of zero. We pooled the transformed proportions using a random effects model.
Pairwise comparison meta-analysisWe used a Mantel-Haenszel random effects model to estimate the relative risk of malignancy associated with RAS mutation positivity (Dettori et al., 2022[17]). The between-study variance was estimated using the DerSimonian-Laird method (Bakbergenuly et al., 2020[2]). We added a continuity correction of 0.5 to studies with zero events.
For the secondary outcome of the histological subtype, we used a Mantel-Haenszel random effects model to calculate the odds ratio of classical variant/follicular variant papillary carcinoma associated with RAS mutations. We used the same methods as above to quantify and test for heterogeneity.
Heterogeneity analysisFor all proportion meta-analyses, we quantified heterogeneity using the I2 statistic and tested for heterogeneity using Cochran's Q test (West et al., 2010[52]). We used the Knapp-Hartung modification to account for uncertainty in the estimated variance of each study (Jackson et al., 2017[28]). Confidence intervals for individual studies were calculated using the Clopper-Pearson method (Jackson et al., 2017[28]).
Evaluation of publication biasWe assessed publication bias through visual inspection of funnel plots and quantitatively using Egger's regression test (Egger et al., 1997[18]). For estimates with evidence of publication bias, we performed trim-and-fill analysis to adjust for hypothetical missing studies.
Sensitivity analyses and meta-regression analysisWe conducted sensitivity analyses excluding studies with a high risk of bias, studies with inadequate sample size, and other subgroup analyses to evaluate the robustness of findings. All analyses were done in R version 4.0.2 using the meta R and metafor packages.
ResultsLiterature searchA systematic literature search identified 318 potentially relevant studies. After screening titles and abstracts, 52 articles were retrieved for full-text review. Thirty studies with 13,328 nodules in 12,338 patients met the final inclusion criteria (Figure 2(Fig. 2)).
Study characteristicsThe 30 included studies comprised 16 prospective and 14 retrospective designs, conducted across 10 countries from 2010 to 2022, including 10 studies from the USA (Beaudenon-Huibregtse et al., 2014[5]; Belovarac et al., 2022[6]; Guan et al., 2020[24]; Gupta et al., 2013[25]; Lupo et al., 2020[33]; Moses et al., 2010[37]; Nikiforov et al., 2011[38]; Shrestha et al., 2016[44]; Stence et al., 2015[47]; Valderrabano et al., 2017[50]), five from Italy (Cantara et al., 2010[8]; Censi et al., 2017[9]; Colombo et al., 2021[13]; De Napoli et al., 2016[14]; Macerola et al., 2019[34]), and four from China (Liu et al., 2014[30]; Lu et al., 2021[31]; Song et al., 2020[46]; Wu et al., 2019[53]), among others (An et al., 2015[1]; Bardet et al., 2015[3]; Chen et al., 2020[10]; Cho et al., 2020[11]; Decaussin-Petrucci et al., 2017[16]; Eszlinger et al., 2014[19]; Gill et al., 2015[22]; Grimmichova et al., 2022[23]; Ravella et al., 2020[43]; Tolaba et al., 2021[49]; Vishwanath et al., 2022[51]). Sample sizes ranged from 42 to 2,306 nodules evaluated by FNA. About 5,307 nodules were diagnosed as indeterminate, of which 3,327 underwent surgical resection with histopathological confirmation (Table 1(Tab. 1); References in Table 1: An, 2015[1]; Bardet, 2015[3]; Beaudenon-Huibregtse, 2014[5]; Belovarac, 2022[6]; Cantara, 2010[8]; Censi, 2017[9]; Chen, 2020[10]; Cho, 2020[11]; Colombo, 2021[13]; De Napoli, 2016[14]; Decaussin-Petrucci, 2017[16]; Eszlinger, 2014[19]; Gill, 2015[22]; Grimmichova, 2022[23]; Guan, 2020[24]; Gupta, 2013[25]; Liu, 2014[30]; Lu, 2021[31]; Lupo, 2020[33]; Macerola, 2019[34]; Moses, 2010[37]; Nikiforov, 2011[38]; Ravella, 2020[43]; Shrestha, 2016[44]; Song, 2020[46]; Stence, 2015[47]; Tolaba, 2021[49]; Valderrabano, 2017[50]; Vishwanath, 2022[51]; Wu, 2019[53]).
Pooled rates in indeterminate nodulesThe pooled proportion of Bethesda category III/IV/V nodules was 93 % (95 % CI, 73-99 %) (Figure 3(Fig. 3); References in Figure 3: An, 2015[1]; Bardet, 2015[3]; Beaudenon-Huibregtse, 2014[5]; Belovarac, 2022[6]; Cantara, 2010[8]; Censi, 2017[9]; Chen, 2020[10]; Cho, 2020[11]; Colombo, 2021[13]; De Napoli, 2016[14]; Decaussin-Petrucci, 2017[16]; Eszlinger, 2014[19]; Gill, 2015[22]; Grimmichova, 2022[23]; Guan, 2020[24]; Gupta, 2013[25]; Liu, 2014[30]; Lu, 2021[31]; Lupo, 2020[33]; Macerola, 2019[34]; Moses, 2010[37]; Nikiforov, 2011[38]; Ravella, 2020[43]; Shrestha, 2016[44]; Song, 2020[46]; Stence, 2015[47]; Tolaba, 2021[49]; Valderrabano, 2017[50]; Vishwanath, 2022[51]; Wu, 2019[53]). The frequency ranged from 13 % to 100 %.
For indeterminate subtypes, atypia of undetermined significance or follicular lesion of undetermined significance (AUS/FLUS) occurred in 44 % (95 % CI, 31-59 %), follicular neoplasm (FN) in 42 % (95 % CI, 28-57 %), and suspicious for malignancy (SUSP) in 18 % (95 % CI, 13-24 %) (Figure 4(Fig. 4); References in Figure 4: An, 2015[1]; Bardet, 2015[3]; Beaudenon-Huibregtse, 2014[5]; Belovarac, 2022[6]; Cantara, 2010[8]; Censi, 2017[9]; Chen, 2020[10]; Cho, 2020[11]; Colombo, 2021[13]; De Napoli, 2016[14]; Decaussin-Petrucci, 2017[16]; Eszlinger, 2014[19]; Gill, 2015[22]; Grimmichova, 2022[23]; Guan, 2020[24]; Gupta, 2013[25]; Liu, 2014[30]; Lu, 2021[31]; Lupo, 2020[33]; Macerola, 2019[34]; Moses, 2010[37]; Nikiforov, 2011[38]; Ravella, 2020[43]; Shrestha, 2016[44]; Song, 2020[46]; Stence, 2015[47]; Tolaba, 2021[49]; Valderrabano, 2017[50]; Vishwanath, 2022[51]; Wu, 2019[53]).
Mutation rate of RAS genes in indeterminate nodulesOut of the whole population, 91 % (95 % CI, 76 %-97 %) of patients with indeterminate nodules underwent surgery. Of these, 29 % (95 % CI, 18 %-43 %) had positive RAS mutation (Figure 5(Fig. 5); References in Figure 5: An, 2015[1]; Bardet, 2015[3]; Beaudenon-Huibregtse, 2014[5]; Belovarac, 2022[6]; Cantara, 2010[8]; Censi, 2017[9]; Chen, 2020[10]; Cho, 2020[11]; Colombo, 2021[13]; De Napoli, 2016[14]; Decaussin-Petrucci, 2017[16]; Eszlinger, 2014[19]; Gill, 2015[22]; Grimmichova, 2022[23]; Guan, 2020[24]; Gupta, 2013[25]; Liu, 2014[30]; Lu, 2021[31]; Lupo, 2020[33]; Macerola, 2019[34]; Moses, 2010[37]; Nikiforov, 2011[38]; Ravella, 2020[43]; Shrestha, 2016[44]; Song, 2020[46]; Stence, 2015[47]; Tolaba, 2021[49]; Valderrabano, 2017[50]; Vishwanath, 2022[51]; Wu, 2019[53]). Among RAS mutations, NRAS mutations predominated at 67 % (95 % CI, 55-79 %), followed by HRAS 24 % (95 % CI, 19-29 %) and KRAS 12 % (95 % CI, 5-22 %).
Rates of malignancy in RAS positive indeterminate nodulesIn surgically resected indeterminate nodules, 31 % (95 % CI 19-44 %) harbored malignancy. Specifically, in indeterminate nodules with positive RAS mutation, the malignancy rate was 58 % (95 % CI, 47 %-69 %) (Figure 6(Fig. 6); References in Figure 6: An, 2015[1]; Bardet, 2015[3]; Beaudenon-Huibregtse, 2014[5]; Belovarac, 2022[6]; Cantara, 2010[8]; Censi, 2017[9]; Chen, 2020[10]; Cho, 2020[11]; Colombo, 2021[13]; De Napoli, 2016[14]; Decaussin-Petrucci, 2017[16]; Eszlinger, 2014[19]; Gill, 2015[22]; Grimmichova, 2022[23]; Guan, 2020[24]; Gupta, 2013[25]; Liu, 2014[30]; Lu, 2021[31]; Lupo, 2020[33]; Macerola, 2019[34]; Moses, 2010[37]; Nikiforov, 2011[38]; Ravella, 2020[43]; Shrestha, 2016[44]; Song, 2020[46]; Stence, 2015[47]; Tolaba, 2021[49]; Valderrabano, 2017[50]; Vishwanath, 2022[51]; Wu, 2019[53]).
Risk of malignancy RAS mutation positivity conferred a 1.68-fold higher risk of malignancy (RR 1.68, 95 % CI, 1.21-2.34, p=0.002) (Figure 7(Fig. 7); References in Figure 7: An, 2015[1]; Bardet, 2015[3]; Beaudenon-Huibregtse, 2014[5]; Belovarac, 2022[6]; Cantara, 2010[8]; Censi, 2017[9]; Chen, 2020[10]; Cho, 2020[11]; Colombo, 2021[13]; De Napoli, 2016[14]; Decaussin-Petrucci, 2017[16]; Eszlinger, 2014[19]; Gill, 2015[22]; Grimmichova, 2022[23]; Guan, 2020[24]; Gupta, 2013[25]; Liu, 2014[30]; Lu, 2021[31]; Lupo, 2020[33]; Macerola, 2019[34]; Moses, 2010[37]; Nikiforov, 2011[38]; Ravella, 2020[43]; Shrestha, 2016[44]; Song, 2020[46]; Stence, 2015[47]; Tolaba, 2021[49]; Valderrabano, 2017[50]; Vishwanath, 2022[51]; Wu, 2019[53]).
Evaluation of bias and heterogeneity analysisThere was evidence of publication bias by Egger's test (p=0.03) and funnel plot asymmetry (Figure 8(Fig. 8)). Sensitivity analysis and Baujat plot indicated that heterogeneity was partly explained by one outlier study (Nikiforov et al., 2011[38]). This study was removed in a revised analysis for the relative risk of malignancy in the presence of RAS mutation (RR:1.75, 95 % CI 1.54-1.98) (Figure 9(Fig. 9); References in Figure 9: An, 2015[1]; Bardet, 2015[3]; Beaudenon-Huibregtse, 2014[5]; Belovarac, 2022[6]; Cantara, 2010[8]; Censi, 2017[9]; Chen, 2020[10]; Cho, 2020[11]; Colombo, 2021[13]; De Napoli, 2016[14]; Decaussin-Petrucci, 2017[16]; Eszlinger, 2014[19]; Gill, 2015[22]; Grimmichova, 2022[23]; Guan, 2020[24]; Gupta, 2013[25]; Liu, 2014[30]; Lu, 2021[31]; Lupo, 2020[33]; Macerola, 2019[34]; Moses, 2010[37]; Nikiforov, 2011[38]; Ravella, 2020[43]; Shrestha, 2016[44]; Song, 2020[46]; Stence, 2015[47]; Tolaba, 2021[49]; Valderrabano, 2017[50]; Vishwanath, 2022[51]; Wu, 2019[53]).
Histopathological subtypesAmong RAS-positive malignant thyroid nodules, the meta-analysis found that the follicular variant of papillary thyroid cancer (FV-PTC) constituted 38.60 % (95 % CI 23.40-56.30 %) of cases. The classical variant of papillary thyroid cancer (CV-PTC) represented 34.10 % (95 % CI 19.50-52.40 %) of cases. On the other hand, follicular thyroid cancer (FTC) accounted for 23.20 % (95 % CI 14.70-34.50 %) of RAS-positive malignant nodules. Significant heterogeneity was detected across the pooled estimates (I2>50 %) (Table 2(Tab. 2)).
Furthermore, pairwise comparisons between the subtypes revealed no statistically significant differences in the odds of harboring RAS mutations. Specifically, the odds ratio for FV-PTC versus CV-PTC was 1.06 (95 % CI 0.25-4.37; p=0.93). The FTC versus CV-PTC comparison yielded an odds ratio of 0.64 (95 % CI 0.22-1.82; p=0.40), and the FTC versus FV-PTC comparison showed an odds ratio of 0.51 (95 % CI 0.15-1.68; p=0.26) (Table 3(Tab. 3)).
DiscussionIn this systematic review and meta-analysis of 30 studies with over 13,000 thyroid nodules, we found that RAS mutations are prevalent in approximately 30 % of nodules with indeterminate fine needle aspiration cytology. A RAS mutation was associated with a 1.7-fold increased risk of malignancy. Among RAS-positive malignant nodules, the odds of classical variant papillary thyroid carcinoma were nearly ten times higher compared to follicular carcinoma. These findings suggest that RAS mutational testing may have clinical utility for refining the diagnosis of indeterminate nodules.
The 29 % pooled rate of RAS mutations we identified is within the 18-43 % generally reported for sporadic papillary thyroid cancers (Zhou et al., 2023[55]). However, prior studies in indeterminate nodules have reported RAS mutation frequencies of 26-34 %, more concordant with our estimate (Gill et al., 2015[22]; Macerola et al., 2019[34]; Song et al., 2020[46]). The prevalence likely reflects enrichment for RAS mutations among cytologically indeterminate nodules.
We found RAS positivity significantly predicts malignancy, conferring a 1.7-fold increased risk, although the data exhibited heterogeneity. This aligns with recent studies demonstrating that molecular profiling adds incremental diagnostic value over clinical features and ultrasound for thyroid nodules with indeterminate cytology (Morand et al., 2024[36]; Stewardson et al., 2023[48]). Specifically, a previous meta-analysis found that RAS-mutated indeterminate nodules have a positive predictive value of 78 % and a positive likelihood ratio of 4.23, suggesting further investigation into the RAS mutation in indeterminate nodules for diagnostic accuracy (Clinkscales et al., 2017[12]).
A recent multicenter, multinational retrospective study examined thyroid nodules with a wider range of Bethesda III to VI cytology through expansive molecular profiling, including ThyGenX/ThyGeNEXT and ThyroSeq V3 tests (Morand et al., 2024[36]). Notably, the study identified RAS-like alterations predominantly in Bethesda III and IV nodules, which aligns with the focus of our meta-analysis. These RAS-like mutations were associated with a lower likelihood of extrathyroidal extension, nodal disease, and aggressive histology, which further illuminates the diagnostic subtleties that RAS mutation testing can bring to the clinical evaluation of indeterminate thyroid nodules. However, while this study provides valuable insights into the potential for molecular testing to predict malignant behavior and guide patient management, it encompasses a broader range of nodules than our analysis, including Bethesda VI and a variety of molecular alterations beyond those in the RAS gene family. The inclusion of a mix of RAS-like and non-RAS mutations, and especially the inclusion of Bethesda VI category nodules, which are known to have a high likelihood of malignancy, suggest divergent clinical implications that are beyond the scope of our investigation, thus not meeting our eligibility criteria. Nevertheless, their findings complement our results by highlighting the importance and utility of RAS mutation testing among the array of molecular diagnostic tools. Future studies could build upon these results, examining the implications of conducting molecular profiling that targets specific mutation types across a tightly defined range of Bethesda categories, ultimately contributing to a more personalized approach to thyroid nodule management.
It is worth noting that the risk conferred by RAS mutations appears more modest than other markers like BRAFV600E, which carry higher specificity for papillary thyroid cancer (Zou et al., 2014[56]).
Our study estimates the distribution of RAS mutation subtypes in indeterminate nodules, with NRAS codominant at 67 %, followed by HRAS and KRAS. Previous reviews have suggested NRAS mutations predominate but lacked sufficient data to quantify the relative distribution (Clinkscales et al., 2017[12]). RAS subtype may have implications for prognostication, as NRAS and KRAS mutations have been associated with less aggressive disease, whereas HRAS mutations confer a higher probability of carcinoma outcome risk (Radkay et al., 2014[42]).
Regarding histopathological outcomes, our meta-analysis found RAS-positive malignant nodules harbored a distribution of 34 % classical variant PTC, 39 % follicular variant PTC, and 23 % follicular thyroid carcinoma. There was significant heterogeneity across studies for all subtype estimates. The predominance of classic and follicular variant PTC histologies likely reflects the known association between RAS mutations and papillary carcinoma, rather than follicular carcinoma, in thyroid malignant transformation (Cameselle-Teijeiro et al., 2020[7]). The lack of distinction in RAS mutation prevalence between subtypes indicates that cytological phenotype does not necessarily confer genotype specificity. Overall, RAS analysis can improve diagnostic accuracy without necessarily providing definitive subclassification capacity between PTC variants. Additional large-scale studies are still needed to clarify if subtle inter-subtype differences exist in the likelihood of harboring RAS mutations.
Currently, the ATA guidelines issue a weak recommendation for molecular testing to help guide the management of cytologically indeterminate thyroid nodules (Haugen, 2017[26]). Based on our findings, we suggest that RAS analysis is a useful diagnostic adjunct for nodules with indeterminate cytology. Testing can be readily performed on FNA cytology samples prior to surgical decision-making. Given that, nearly half of the indeterminate nodules harbor RAS mutations, which confer an increased risk of malignancy, a positive RAS test result may prompt consideration of surgical excision or more frequent ultRASound follow-up instead of continued observation. Negative RAS testing suggests lower malignancy risk, providing reassurance for conservative management.
Incorporating RAS analysis into indeterminate thyroid nodule evaluation may reduce unnecessary surgeries for benign nodules and allow earlier diagnosis of RAS-positive classical variant papillary cancers. There is potential value for RAS subtyping as well. Additional studies are still needed to determine if RAS mutations can stratify malignant risk within the AUS/FLUS, FN, and SUSP Bethesda categories. Future research should also examine the prognostic significance and predictive value of RAS mutations regarding response to adjuvant therapy in thyroid cancer patients.
Several limitations should be considered when interpreting our meta-analysis. There was significant between-study heterogeneity, which may reflect differences in geographic cohorts, molecular methods, and histopathological classification across studies. To mitigate this issue, we conducted sensitivity analyses, which achieved homogeneity after excluding an outlier study. Additionally, eligible studies were observational. Consequently, the potential for residual confounding factors cannot be ruled out.
Furthermore, while focusing on RAS status, we recognize that other germline and somatic mutations might coexist and influence the outcomes, but these factors were not accounted for in our analysis due to data limitations. A significant gap in our meta-analysis is the inability to assess the prognostic significance of RAS mutations, mainly because most of the included studies did not provide follow-up data, which limits our understanding of the long-term implications of these mutations in the studied population. Finally, providing a nuanced understanding of how RAS mutation status might correlate with malignancy risk across different genders could have valuable implications for personalizing the management of thyroid nodules and is an essential suggestion for progressing the field.