The Association Between Thyroid Differentiation Score and Survival Outcomes in Papillary Thyroid Carcinoma

Abstract

Context

Thyroid differentiation score (TDS), calculated based on mRNA expression levels of 16 genes controlling thyroid metabolism and function, has been proposed as a measure to quantify differentiation in papillary thyroid carcinoma (PTC).

Objective

The objective of this study is to determine whether TDS is associated with survival outcomes across patient cohorts.

Methods

Two independent cohorts of patients with PTC were used: (1) The Cancer Genome Atlas (TCGA) thyroid cancer study (N = 372), (2) MD Anderson Cancer Center (MDACC) cohort (N = 111). The primary survival outcome of interest was progression-free interval (PFI). Association with overall survival (OS) was also explored. The Kaplan–Meier method and Cox proportional hazards models were used for survival analyses.

Results

In both cohorts, TDS was associated with tumor and nodal stage at diagnosis as well as tumor driver mutation status. High TDS was associated with longer PFI on univariable analyses across cohorts. After adjusting for overall stage, TDS remained significantly associated with PFI in the MDACC cohort only (adjusted hazard ratio [aHR] 0.67, 95% CI 0.52-0.85). In subgroup analyses stratified by tumor driver mutation status, higher TDS was most consistently associated with longer PFI in BRAFV600E-mutated tumors in the MDACC cohort after adjusting for overall stage (TCGA: aHR 0.60, 95% CI 0.33-1.07; MDACC: aHR 0.59, 95% CI 0.42-0.82). For OS, increasing TDS was associated with longer OS in the overall MDACC cohort (aHR = 0.78, 95% CI 0.63-0.96), where the median duration of follow-up was 12.9 years.

Conclusion

TDS quantifies the spectrum of differentiation status in PTC and may serve as a potential prognostic biomarker in PTC, mostly promisingly in BRAFV600E-mutated tumors.

The rise in thyroid cancer incidence has stabilized in recent years. However, thyroid cancer–specific incidence-based mortality rates have increased in the United States (1). Papillary thyroid carcinoma (PTC) accounts for approximately 85% of newly diagnosed thyroid cancer cases and is the main histologic subtype influencing thyroid cancer incidence and mortality trends (2). A significant proportion of PTCs are low risk with excellent long-term survival outcomes, but a subset of tumors manifests aggressive clinical behavior with increased rates of disease progression and mortality (2, 3). Pretreatment risk stratification of PTC is currently based on a limited set of clinicopathologic characteristics, namely age, gender, and TNM stage, which cannot accurately predict aggressive behavior (4, 5). As a result, undertreatment of aggressive tumors can occur in clinical practice, which may lead to increased morbidity and mortality.

Dedifferentiation refers to a process whereby thyroid cancer cells lose specialized properties associated with their cell of origin, the thyroid follicular cell. These properties include those related to iodine metabolism and thyroid hormone regulation (2, 6, 7). Dedifferentiation has long been recognized as a biological phenomenon in highly lethal histologic subtypes of thyroid cancers including poorly differentiated and anaplastic thyroid carcinomas (7-9). In these tumors, dedifferentiation is identifiable based on the presence of canonical histopathologic features such as multiple mitoses, necrosis, and pleomorphic tumor cells (7-9). While it is known that PTC can transform into poorly differentiated and anaplastic thyroid carcinomas through terminal dedifferentiation in rare cases, accumulating evidence suggests that milder degrees of loss of differentiation occurs in a larger subset of PTCs (6, 10-13). Clinically, a manifestation of this is resistance to radioactive iodine (RAI) (6). As a result, these PTCs behave aggressively with high risks of disease recurrence and/or progression. Despite our current conceptualization of dedifferentiation in thyroid cancer as a continuum, these earlier states of loss of differentiation are difficult to recognize in PTC based on histopathology alone as cellular morphologic changes are not typically apparent and subtle changes may be varied across tumors (14).

The thyroid differentiation score (TDS), calculated based on the mRNA expression levels of 16 genes related to thyroid metabolism and function, was proposed by The Cancer Genome Atlas (TCGA) PTC study to characterize differentiation status in PTCs (13). Across tumors, TDS was associated with global gene expression patterns, mitogen-activated protein kinase (MAPK) driver mutation status, and the presence of TERT promoter mutations. The primary objective of the present study was to further evaluate the utility of TDS as a biomarker in PTC by determining whether it is associated with survival outcomes in 2 independent cohorts of patients with PTC.

Materials and Methods

TCGA Cohort

The TCGA cohort comprised 496 patients diagnosed with PTC from the TCGA PTC study (13). Patients who did not have TDS calculated using RNAseq data were excluded (n = 105). Because MD Anderson also contributed to the TCGA PTC study, we removed 19 patients who were in the MDACC cohort from the TCGA cohort. One patient, who had a follow-up duration less than 1 month, was excluded. Clinicopathologic information and survival outcomes were obtained retrospectively. Clinicopathologic information including age at diagnosis, gender, staging information, duration of follow-up, and survival outcomes was downloaded from the cBioPortal (http://cbioportal.org) (15) and the TCGA Pan-Cancer Clinical Data Resource (16) for each patient. The American Joint Committee on Cancer TNM classification 7th edition was utilized for staging. All patients were treated with primary surgery and were staged according to their pathological TNM stage. Tumor molecular data including MAPK driver mutation status from whole exome sequencing data were downloaded from the supplementary materials published in the TCGA PTC study (13). Tumors were classified as BRAF-mutated if a somatic BRAFV600E mutation was identified, RAS-mutated if any mutation in NRAS, HRAS, or KRAS were identified, and as “other” if no driver mutations were identified in either BRAFV600E or RAS family genes.

MDACC Cohort

The MDACC cohort comprised 111 patients with PTC treated at the University of Texas MD Anderson Cancer Center (MDACC) between 1993 and 2012. Approval for the study was obtained from the MD Anderson Institutional Review Board. All included patients had confirmation of a PTC diagnosis by a head and neck pathologist and were treated with primary surgery. MAPK driver mutation status was ascertained based on whole exome sequencing consensus mutation calls generated using MuSe and MuTect2 (17, 18). Tumors were classified as BRAFV600E, RAS, or “other” based on criteria listed above for the TCGA cohort. Clinicopathologic information including age at diagnosis, gender, staging information, duration of follow-up, and survival outcome data were collected via retrospective chart review. RAI sensitivity was determined via retrospective chart review by an oncologic endocrinologist after review of clinical notes, RAI scans, and positron emission tomography scans, based on previously published criteria (19, 20). Patients with not enough information available, including lack of post-treatment RAI scans or lost to follow-up after RAI administration, were excluded from the RAI analysis. Patients were staged using the American Joint Committee on Cancer TNM classification, 7th edition (4).

Thyroid Differentiation Score

TDS was calculated based on mRNA expression levels of 16 thyroid metabolism and function genes using tumor RNAseq data. Genes included in the score were DIO1, DIO2, DUOX1, DUOX2, FOXE1, GLIS3, NKX2-1, PAX8, SLC26A4, SLC5A5, SLC5A8, TG, THRA, THRB, TPO, and TSHR. For the TCGA cohort, TDSs for included patients were downloaded from the supplementary materials published in the TCGA PTC study (13). For the MDACC cohort, all 111 patients had their primary tumors profiled using RNAseq. We calculated the TDS using the method described by the TCGA study (13). In summary, after completing sequence alignment to GRCh37/hg19, filtered reads were further processed for variant calling using best practice of GATK (https://software.broadinstitute.org/gatk/best-practices/). Gene expression was quantified and normalized using RSEM (21). Log2 transformed counts were then centered by the median across 111 tumor samples. The mean value across the 16 genes was then obtained for each tumor and utilized as the TDS. Negative scores represent patients who have loss of expression in these genes, suggesting dedifferentiation. Positive scores indicated retained thyroid-specific gene expression.

Statistical Analysis

Descriptive statistics were used to summarize demographic and clinicopathologic characteristics. Linear regression was used to assess the association between TDS clinicopathologic variables and tumor mutation status. The Wilcoxon test was used to compare TDS for responder vs nonresponders to RAI. The primary outcome of interest was progression-free interval (PFI), which was defined as the time interval between the date of diagnosis and the date of the patient having a new tumor event, including progression of disease, local recurrence, distant metastasis, new primary tumors all sites, or died with the cancer without a new tumor event. PFI was a recommended outcome endpoint for the TCGA PTC study as described in the TCGA Pan-Cancer Clinical Data Resource (16). Overall survival (OS) defined as the time interval between the date of diagnosis and the date of death from any cause. OS was evaluated as a secondary outcome of interest. We also performed exploratory analyses with the following outcomes in the MDACC cohort where additional survival data were available: (1) recurrence-free survival (RFS) defined as date of diagnosis to the first documented recurrence (local, regional, and/or distant) or death due to disease, and (2) cancer-specific survival (CCS) defined as the interval between date of diagnosis to death due to thyroid cancer. Duration of follow-up was calculated as the time interval from the date of diagnosis until date of last known follow-up or contact.

The Kaplan–Meier method was used to estimate survival times. The log rank test was used to compare estimated survival times. Cox proportional hazard models were used to determine the association between TDS and survival outcomes. Models estimated hazard ratios with 95% CIs. Multivariable models were built based on univariate analyses. The proportional hazards assumption was assessed using a test of Schoenfeld residuals for covariates in all final models. Wald test statistics were carried out to evaluate the association between survival outcomes and TDS while controlling for other covariates. Heteroscedasticity-consistent SEs were used.

We calculated the median, mean, and SD of TDS for each cohort. For Kaplan–Meier analyses, TDS was then categorized as low if it was less than or equal to the mean TDS in each cohort. For linear regression, TDS was coded as a continuous variable. For Cox regression analyses, TDS was coded both as a categorical variable and as a continuous variable.

Linear regression analysis was performed using the “lm” function from base R (v2021.09.2, build 382). Survival analyses were performed using the R package survival (v. 3.4-0; Therneau, 2022) and Kaplan–Meier curves were generated using R package ggpubr (v. 0.4.0; Kassambara, 2020). All statistical analyses were performed using RStudio (v2021.09.2, build 382) with a significance threshold of .05.

Results

Study Cohorts

The TCGA and MDACC cohorts comprised 372 and 111 patients, respectively. The clinicopathologic characteristics of the TCGA and MDACC cohorts are summarized in Table 1. The 2 cohorts were similar in terms of age at diagnosis; however, the MDACC cohort had a higher proportion of male patients than the TCGA cohort (37.8% vs 25.5%). The TCGA cohort had a higher portion of T1 disease (28.8% vs 11.7%) and a lower proportion of T4 disease (2.9% vs 29.7%) than the MDACC cohort. The MDACC cohort also had a higher proportion of patients with nodal and/or distant metastases. In terms of tumor mutation status, the TCGA cohort had higher proportions of patients with BRAFV600E-mutated and RAS-mutated tumors. The RAS mutation was only detected in 4 patients in the MDACC cohort. The MDACC cohort had a larger range of TDS. Descriptive statistics of TDS by cohort is summarized elsewhere (Table S1 (22)).

Table 1.

Baseline characteristics of the TCGA and MDACC cohorts

	TCGA cohort		MDACC cohort
Patient characteristics	n	Mean (range) or %	n	Mean (range) or %
Age at diagnosis, years	372	46.8 (15.0-89.0)	111	48.6 (18.0-87.0)
Sex
Female	277	74.5	69	62.2
Male	95	25.5	42	37.8
Tumor stage
T1	107	28.8	13	11.7
T2	121	32.5	39	35.1
T3	113	30.4	24	21.6
T4	11	2.9	33	29.7
TX	20	5.4	2	1.8
Nodal stage
N0	169	45.4	40	36.0
N1	148	39.8	69	62.2
NX	55	14.8	2	1.8
Metastasis stage
M0	184	49.5	97	87.4
M1	5	1.3	12	10.8
MX	183	49.2	2	1.8
Overall pathologic stage
I	217	58.3	65	58.6
II	40	10.7	20	18.0
III	75	20.2	9	8.1
IV	38	10.2	15	13.5
NA	2	0	2	1.8
Tumor driver mutation
BRAFV600E	209	56.2	54	48.6
OTHER	111	29.8	53	47.8
RAS	52	14.0	4	3.6
TDS	372	0.033 (−4.08-2.59)	111	0.266 (−9.696-3.057)
Vital status
Alive	363	97.6	72	64.9
Dead	9	2.4	39	35.1
Duration of follow-up, years		3.2 (0.07-14.8)		12.9 (0.005-28.134)

Abbreviations: MDACC, MD Anderson Cancer Center; NA, not available; TCGA, The Cancer Genome Atlas; TDS, thyroid differentiation score.

Association Between TDS, Tumor Mutation Status, Clinicopathologic Characteristics, and RAI

Table 2 shows estimates of the associations between TDS and patient characteristics. In both cohorts, TDS was significantly associated with tumor stage and nodal stage where an inverse relationship is observed. In general, advanced stage tumors had lower differentiation status as estimated by TDS. While overall TNM stage was a significant predictor of TDS in the TCGA cohort, the association did not reach statistical significance in the smaller MDACC cohort. TDS was not significantly associated with age and gender.

Table 2.

Results of linear regression analyses assessing the relationship between TDS and other patient characteristics in TCGA and MDACC cohorts

	TCGA cohort				MDACC cohort
Characteristic	n	β	SE	P	n	β	SE	P
Age	372	<0.01	<0.01	.75	111	<−0.01	<0.01	.46
Gender	372			.83				.58
Female	277	1			69	1
Male	95	0.03	0.14		42	−0.17	0.31
Tumor stage	352			<.01	109			.03
T1	107	1			13
T2	121	−0.23	0.15		39	−0.59	0.49
T3	113	−0.67	0.15		24	−0.79	0.52
T4	11	−1.52	0.35		33	−1.39	0.50
Nodal stage	317			<.01	109			<.01
N0	169	1			40	1
N1	148	−0.73	0.12		69	−0.83	0.30
M stage	189			.64	109			.63
M0	184	1			97	1
M1	5	−0.24	0.50		12	−0.24	0.49
Overall stage	370			<.01	110			.42
I	217	1			65	1
II	40	0.37	0.19		20	0.65	0.40
III	75	−0.38	0.15		9	0.28	0.56
IV	38	−0.74	0.19		15	−0.03	0.45
Driver mutation	372			<.01	112			<.01
Other	111	1			53	1
BRAFV600E	209	−1.41	0.10		54	−0.89	0.29
RAS	52	0.44	0.14		4	0.59	0.78

P value calculated based on the Wald test. Bold values indicate significant P values <0.05.

Abbreviations: M, distant metastases; MDACC, MD Anderson Cancer Center; MX, distant metastases not assessed; NX, pathologic nodal stage not assessed; TCGA, The Cancer Genome Atlas.

Tumor driver mutation was significantly associated with TDS in the TCGA cohort, where, on average, BRAFV600E-mutated tumors were less differentiated than the “other” subgroup while RAS-mutated tumors all had higher TDSs (Tables S2 and S3 (22)) (13). This finding was consistent in the MDACC cohort, where TDS was also significantly associated with driver mutation status and BRAFV600E-mutated tumors had lower differentiation on average, compared to the “other” subgroup.

In the MDACC cohort, in patients where RAI was utilized with available records to determine sensitivity (n = 25), tumors that were sensitive to RAI (n = 12) on average had higher TDS than those that were refractory (n = 13) (Fig. 1).

Figure 1.

In the MDACC cohort, 25 patients were evaluable for response to radioactive iodine. Tumors sensitive to radioactive iodine on average had higher TDS than those that were refractory.

TDS and Survival Outcomes

We first examined TDS as a categorical variable where patients were categorized as having low vs high TDS based on their respective cohort means. Figure 2 depicts estimates for PFI by TDS category. In the MDA cohort, PFI was significantly longer among participants in the high TDS category than those in the low TDS category (Fig. 2B). In the TCGA cohort, no significant difference in PFI was detected between TDS high vs low groups (Fig. 2A). In the MDA cohort where detailed survival data were available, we also assessed the association between TDS and RFS, as well as CCS. We found that low TDS was significantly associated with both shorter RFS (Fig. S1 (23)) and CCS in the MDA cohort (Fig. S2 (23)). Across the 2 cohorts, no significant associations between TDS category and OS were observed (Table 3 and Fig. 3).

Figure 2.

Progression-free interval estimates by TDS category. (A) Kaplan–Meier estimates by TDS category in the TCGA cohort. (B) Estimates in the MDACC cohort.

Table 3.

Multivariable analyses of TDS and survival outcomes in all patients

	Multivariable model
	PFI			OS
	aHRa	95% CI	P	aHRb	95% CI	P
TCGA
TDScat			.28			.83
High	Ref	—	—	—	—	—
Low	1.47	0.71-3.05		0.85	0.19-3.82
TDScont	0.78	0.57-1.07	.12	1.13	0.57-2.26	.73
MDACC
TDScat			.06			.93
High	Ref	—		—	—	—
Low	1.97	0.96-4.01		1.03	0.52-2.04
TDScont	0.67	0.52-0.85	<.01	0.78	0.63-0.96	.02

Bold values indicate significant P values <0.05.

Abbreviations: aHR, adjusted hazard ratio; MDACC, MD Anderson Cancer Center; TCGA, The Cancer Genome Atlas; TDScat, TDS category.

TDScat defined as follows: TDS low if TDS ≤ cohort mean, TDS high if TDS > cohort mean; TDScont, thyroid differentiation score treated as a continuous variable.

^a Adjusted for overall stage.

^b Adjusted for age and overall stage.

Figure 3.

Overall survival estimates by TDS category. (A) Kaplan–Meier estimates by TDS category in the TCGA cohort. (B) Estimates in the MDACC cohort.

We next examined TDS as a continuous variable. In univariable analyses, TDS was significantly associated with PFI in both cohorts (Tables S1 and S2 (22)). After adjusting for overall stage, increasing TDS was associated with longer PFI (Table 3). This relationship achieved statistical significance in the MDACC cohort, where each unit increase in TDS was associated with a 33% risk reduction in terms of disease progression. While continuous TDS was not significantly associated with OS in the TCGA cohort, it was significantly associated with OS in the MDACC cohort where increasing TDS was associated with lower risk of death from any cause (Table 3). Subsequent survival analyses were performed treating TDS as a continuous variable.

TDS and Survival Outcomes in Patients Stratified by Driver Mutation Status

Next, we performed subgroup analyses by tumor driver mutation status to determine whether the association between TDS and survival outcomes varied across tumor subgroups. Results of multivariable analyses are shown in Tables 4 to 6. In summary, the association between TDS and PFI differed across tumor subtypes defined by tumor driver mutation status. A statistically significant association was only detected within the BRAFV600E-mutated subtype for PFI in the MDACC cohort where increasing TDS was significantly associated with lower risks of disease progression. A similar nonsignificant trend was observed in the TCGA cohort. No significant associations were identified in terms of PFI in the BRAF and RAS wildtype subgroup. While a RAS-specific analysis was performed in the TCGA cohort, which was not significant, it was not performed in the MDACC cohort given only 4 patients had RAS-mutated PTC. TDS was not significantly associated with OS across mutation strata for both cohorts.

Table 4.

Multivariable analyses of TDS and survival outcomes in BRAFV600E-mutated PTC

		PFI			OS
	N	aHRa	95% CI	P	aHRb	95% CI	P
TCGA cohort
TDS	208	0.60	0.33-1.07	.08	0.16	0.02-1.21	.07
MDACC cohort
TDS	52	0.59	0.42-0.82	<.01	0.71	0.49-1.04	.08

Bold value indicate significant P value <0.05.

Abbreviations: aHR, adjusted hazard ratio; MDACC, MD Anderson Cancer Center; OS, overall survival; PFI, progression-free interval; PTC, papillary thyroid carcinoma; TCGA, The Cancer Genome Atlas; TDS, thyroid differentiation score.

^a Adjusted for overall stage.

^b Adjusted for age and overall stage.

Table 6.

Multivariable analyses of TDS and survival outcomes in patients in RAS-mutated PTC

		PFI			OS
	n	aHRa	95% CI	P	aHRb	95% CI	P
TDS	52	0.29	0.04-1.99	.21	NA	NA	NA

Abbreviations: aHR, adjusted hazard ratio; NA, not available; PFI, progression-free interval.

^a Adjusted for overall stage

^b Adjusted for age and overall stage.

Table 5.

Multivariable analyses of TDS and survival outcomes in patients who are BRAF and RAS wildtype

		PFI			OS
	n	aHRa	95% CI	P	aHRb	95% CI	P
TCGA cohort
TDS	110	1.08	0.44-2.69	.86	1.07	0.21-4.11	.93
MDACC cohort
TDS	53	0.77	0.54-1.10	.15	0.70	0.44-1.13	.14

Abbreviations: aHR, adjusted hazard ratio; MDACC, MD Anderson Cancer Center; PFI, progression-free interval; TCGA, The Cancer Genome Atlas; TDS, thyroid differentiation score.

^a Adjusted for overall stage.

^b Adjusted for age and overall stage.

Discussion

In the current study, we evaluated TDS as a potential prognostic biomarker in PTC. We found that dedifferentiation as measured by TDS is associated with higher probability of disease progression, with the strongest associations observed in the BRAFV600E-mutated subtype.

Our findings are consistent with prior studies demonstrating that patients with aggressive clinical phenotype have primary tumors that show evidence of dedifferentiation (14, 24-27). Moreover, our findings in the MDACC cohort were consistent with those from the TCGA demonstrating that TDS differs by tumor driver mutation status and that BRAFV600E-mutated tumors were less differentiated on average (13). Interestingly, TDS range was the widest in BRAFV600E-mutated tumors in both cohorts, demonstrating that the greatest amount of heterogeneity exists in terms of tumor differentiation in these tumors. However, given that TDS was calculated from RNAseq data generated from 2 separate experiments, batch effect prohibits the direct comparison of TDS values across cohorts.

Although BRAFV600E-mutated PTC can display aggressive clinical behavior, detection of the mutation alone is not predictive of survival outcomes (28). Accumulating evidence suggests that the unclear prognostic role of the BRAFV600E mutation is likely because significant heterogeneity exists among these tumors (2). Our findings complement recently published studies demonstrating that BRAFV600E-mutated PTC with low TDS correspond to the subset of tumors with aggressive behavior with high rates of metastases, resistance to RAI, and unfavorable immune microenvironment (29, 30). These studies, however, did not evaluate TDS in terms of survival outcomes. For the first time, we showed that TDS can be used to differentiate BRAFV600E-mutated PTC in terms of PFI. A major strength of the present study is that the follow-up in the MDACC cohort is long with a mean duration of 12 years compared with 3 years in the TCGA cohort. In the study of survival outcomes in PTC, it has been observed that event rates are typically low and disease recurrence can occur many years after diagnosis, with risk reaching a peak at 12 years (31). As such, the MDACC cohort, with both molecular sequencing data as well as long-term clinical follow-up data, is a unique cohort for investigating associations between TDS and survival outcomes.

A recent study investigating TDS in BRAFV600E-mutated PTC using TCGA data utilized a dichotomized TDS where high TDS was defined as TDS >0 and low TDS was defined as TDS <0 (29). In this study, we chose to model TDS as a continuous variable to capture the entire biological spectrum of tumor differentiation. Moreover, the utilized cut off of TDS = 0 does not have an absolute biological meaning as mRNA expression values from RNAseq are quantile normalized and centered around the median in individual cohorts, and therefore is a relative measure. In order to further develop TDS as a potential prognostic biomarker in PTC, future studies are needed to determine the optimal cut-off value for dividing patients into different differentiation categories. The incorporation of spike-in normals or reference genes in the RNAseq experiments for normalization will also be important to allow TDS values across different patients and study cohorts to be comparable (32, 33). This is an approach that has been successfully employed in the development of Oncotype Dx, which is an mRNA expression-based biomarker assay clinically used for breast cancer risk stratification, which contains 16 cancer-related genes and 5 reference genes (34).

Although TDS was significantly associated with PFI in both cohorts in univariate analyses, TDS was significant after adjusting for overall stage only in the MDACC cohort, but not in the TCGA cohort. A trend towards significance was observed in the BRAFV600E-mutated subgroup in the TCGA but statistical significance was not achieved. This is likely due to the longer duration of follow-up in the MDACC cohort, increasing the power to detect differences in PFI. Moreover, the MDACC cohort captured a higher proportion of patients with advanced disease with higher rates of disease progression. Prognostic markers in PTC should show improvement in PFI beyond currently used clinical characteristics (ie, stage). As such, independent cohorts with adequate duration of follow-up and/or events are needed to determine whether TDS can provide additional value in a predictive model for disease-specific survival outcomes beyond clinical variables. Moreover, additional studies are needed to determine whether TDS is associated with survival outcomes across all molecular subtypes of PTC.

Conclusions

Although it is known from prior studies that BRAFV600E-mutated PTC displays aggressive clinical behavior, survival outcomes vary. We found that tumor differentiation status, as measured by TDS, shows potential utility as a prognostic biomarker in BRAFV600E-mutated PTC where loss of differentiation is associated with shorter PFI. Additional studies are needed to explore this association in larger cohorts, ideally utilizing novel computational methods that will allow for qualitative comparison of thyroid differentiation gene expression across cohorts.

Abbreviations

aHR

adjusted hazard ratio

CCS

cancer-specific survival

MAPK

mitogen-activated protein kinase

MDACC

MD Anderson Cancer Center

OS

overall survival

PFI

progression-free interval

PTC

papillary thyroid carcinoma

RAI

radioactive iodine

RFS

recurrence-free survival

TCGA

The Cancer Genome Atlas

TDS

thyroid differentiation score

Funding

J.R.W., M.E.Z., M.E.C., S.L., M.D.W. are supported by the Petrick Thyroid Cancer Research Fund.

Author Contributions

J.R.W.: conceptualization (lead); writing—original draft and editing (lead); formal analysis (lead); methodology (lead). M.E.Z.: methodology; writing—review and editing. M.E.C.: review and editing. C.C.W.: methodology; writing–review. X.L.: methodology; statistical analyses, writing—editing. Y.D.: methodology; exploratory analysis; writing—review. W.W.: methodology; writing—review. S.Y.L.: conceptualization (supporting); writing—review and editing; Y.H.: writing—review and editing; methodology; L.E.: formal analysis; M.D.W.: methodology; writing—review. C.J.: methodology; formal analysis; writing—review and editing. D.R.: methodology; formal analysis; writing—review and editing.

Disclosures

The authors have declared no conflicts of interest to declare.

Data Availability

Restrictions apply to the availability of some generated and analyzed data during this study to preserve patient confidentiality and because they were used under permission from the TCGA. The corresponding author will on request detail the restrictions and any conditions under which access to some data may be provided.