Long‐Term Glucagon‐Like Peptide 1 Receptor Agonist Use Is Not Associated With Increased Risk of Thyroid Cancer in Adults With Type 2 Diabetes

ABSTRACT

Background

Glucagon‐like peptide‐1 receptor agonists (GLP‐1RAs) are widely used for type 2 diabetes (T2DM) and obesity, but their potential association with thyroid cancer remains a concern. This study assessed thyroid cancer risk with long‐term GLP‐1RA use in a large real‐world cohort.

Methods

We conducted a propensity score matched cohort study using electronic health records from TriNetX, including 89,646 adults with T2DM who initiated GLP‐1RA therapy between 2014 and 2020, and demonstrated continued use for at least 1 year. Active comparator controls included users of insulin, metformin, sodium‐glucose transporter‐2 inhibitors, dipeptidyl peptidase‐4 inhibitors, sulfonylureas, and thiazolidinediones. The primary outcome was the incidence of thyroid cancer.

Results

During a median follow‐up of 4.5 ± 2.3 years, GLP‐1RA use was not associated with an increased risk of thyroid cancer compared with any of the other antidiabetic medications. As expected, GLP‐1RA use was associated with a greater reduction in HbA1c levels, while the negative control outcome remained unaffected. Findings remained consistent across subgroups stratified by sex, age, obesity‐status, glycaemic‐control, and GLP‐1RA type and in multiple sensitivity analyses.

Conclusions

In this large real‐world cohort study, long‐term GLP‐1RA use was not associated with an increased risk of thyroid cancer. These findings provide reassurance for the safety of GLP‐1RAs and support their continued evidence‐based use in clinical practice.

1. Introduction

Glucagon‐like peptide‐1 receptor agonists (GLP‐1RAs) have become a cornerstone in the management of type 2 diabetes mellitus (T2DM) and obesity, offering significant benefits in glycaemic control, weight loss, and cardiovascular risk reduction [1, 2, 3]. However, concerns have been raised regarding a potential association with an increased risk of thyroid malignancy [4, 5, 6, 7].

Preclinical studies in rodents have linked GLP‐1RAs to thyroid C‐cell hyperplasia, a precursor to medullary thyroid carcinoma (MTC), prompting the FDA to issue a black box warning against their use in patients with a personal or family history of MTC or multiple endocrine neoplasia type 2 (MEN2) [8, 9, 10]. However, significant interspecies differences limit the direct applicability of these findings to humans [8, 10]. While GLP‐1 receptors are present in human thyroid tissues, studies investigating their role in tumourigenesis have yielded inconsistent results [11].

Beyond MTC, the rising incidence of differentiated thyroid carcinoma (DTC) has led to investigations into potential contributing factors, including environmental exposures and medications [12]. The possible link between GLP‐1RA therapy and thyroid cancer remains inconclusive. Retrospective observational studies and pharmacovigilance database analyses, including reports from the FDA and the World Health Organization's VigiBase, have suggested an association between GLP‐1RAs and DTC risk [4, 6, 7, 13]. Although GLP‐1 receptors are more highly expressed in thyroid cancer cells than in normal thyroid tissue, direct evidence supporting a causal relationship between GLP‐1RA therapy and thyroid tumourigenesis is lacking [11].

The clinical evidence regarding the potential association between GLP‐1RA therapy and thyroid cancer is complicated by methodological limitations in observational studies, including unaccounted confounders and inconsistent findings. While concerns about thyroid malignancy remain unproven, they may nonetheless deter patients and healthcare providers from using GLP‐1RAs despite their well‐established health benefits. To address these uncertainties, we conducted a propensity score‐matched analysis using data from a multi‐institutional federated healthcare network to evaluate the risk of thyroid cancer in adults with type 2 diabetes treated with long‐term GLP‐1RA therapy.

2. Materials and Methods

2.1. Data Source

This study utilised TriNetX, a federated health network based in Cambridge, USA, which provides access to electronic health records (EHRs) from healthcare organisations (HCOs). TriNetX standardises data across institutions through automated preprocessing and mapping to a unified clinical data model, ensuring consistency and reducing missing data bias. The database has been externally validated and widely used in a wide range of research areas including the field of diabetes pharmacotherapy and cancer risk [14, 15, 16]. At the time of analysis in January 2025, the Global Collaborative Network within TriNetX included 141 HCOs with data from 156,626,093 patients across 18 countries. This study adheres to the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) guidelines [17]. Due to privacy restrictions imposed by TriNetX, direct access to individual patient level data is not possible, rather only to group aggregates and analyses using the TriNetX system.

2.2. Subjects

The study cohort consisted of adults (≥ 18 years old) diagnosed with type 2 diabetes mellitus (T2DM) (ICD‐10 E11) who initiated long‐term treatment with a glucagon‐like peptide‐1 receptor agonist (GLP‐1RA; ATC A10BJ) between January 2015 and January 2020. As GLP‐1RA treatment has a very high discontinuation rate during the first year [18], we defined long‐term GLP‐1RA users as patients with at least one additional report of a GLP‐1RA prescription during the subsequent 1–3 years following GLP‐1RA initiation.

Active comparator control cohorts included adults (≥ 18 years old) with T2DM who have no history of GLP‐1RA use and initiated long‐term anti‐diabetic medications during the same period, including insulin (ATC A10A, VA HS501, CPT 1011517), metformin (RxNorm 6809), dipeptidyl peptidase 4 inhibitors (DPP4i; ATC A10BH), sodium glucose transporter 2 inhibitors (SGLT2i; ATC A10BK), sulfonylureas (SU; ATC A10BB) or thiazolidinediones (TZD; ATC A10BG). Similar to GLP‐1RA, the control cohorts were similarly required to show evidence of at least one prescription of the same medication category during the subsequent 1–3 years following initiation of treatment.

Exclusion criteria for both the study and control cohorts included preexisting diagnoses of neoplasms (C00‐D49), prior oncology treatments, or a diagnosis of multiple endocrine neoplasia syndrome type 2 (E31.22–23). Participants in the active comparator groups who initiated GLP‐1RA during follow‐up were excluded from the primary analysis but were included in a separate sensitivity analysis.

2.3. Propensity Score Matching

The index date for collection of baseline characteristics was defined as the date of the long‐term prescription of either GLP‐1RA or active comparator (1–3 years after initiation). Baseline characteristics including age, sex, race, ethnicity, body mass index (BMI), sociodemographic factors, medical comorbidities (ICD‐10 codes), medications, laboratory test values, and healthcare engagement, were determined using EHR data from a time window extending 5 years prior to and up until the index date. The baseline characteristics were reported as means with standard deviations (SDs) for continuous variables and as counts with percentages for categorical variables. The chi‐square test was used to compare categorical variables, while independent two‐sample t‐tests were used for continuous variables. Study and control cohorts were balanced using TriNetX's built‐in 1:1 propensity score matching (PSM) incorporating all baseline characteristics listed in Supporting Information S1: Tables S1A–S1F. The TriNetX PSM algorithm uses logistic regression implemented by the function LogisticRegression of the scikit‐learn package in Python version 3.7. based on a greedy nearest neighbour matching approach, with a calliper distance of 0.1 pooled SDs of the logit of the propensity score. To eliminate the influence of record ordering, the covariate matrix is randomised before matching. No specific imputation is implemented for missing data. Because of patient privacy concerns imposed by TriNetX, site‐specific stratification could not be included in the PSM model.

2.4. Outcomes

The primary outcome was incidence of thyroid cancer (ICD‐10 C73) during the follow‐up period, defined as the time from the index date to a maximum of 10 years. Patients were censored at the first occurrence of an outcome or if the last recorded event in their EHR fell within the analysis time window. Outcomes were compared between PSM cohorts using Kaplan‐Meier analysis with the log‐rank test calculated using R's survival package v3.2.3. Hazard ratios (HR) were calculated using a proportional hazard model wherein the cohort to which the patient belonged was used as the independent variable. HR were assessed for proportionality using the generalised Schoenfeld approach where a p‐value ≤ 0.05 was considered significant for non‐proportionality.

To test the robustness of the primary analysis, we repeated individual PSM analyses stratified by subgroups according to age (above or below 60 years), sex, race, BMI (above or below 30 kg/m²), HbA1C (above or below 8%), type of GLP‐1RA, and type of HCO (academic/non‐academic). We also performed three additional sensitivity analyses for the main outcome by changing the main analysis parameters: (1) Extension of the follow‐up period to a maximum of 20 years (by including patients who started treatment before 2015). (2) Removing the exclusion of patients in the active comparator group who received GLP1‐RA during follow‐up. (3) Using drug‐initiation as the index‐date for PSM and follow‐up.

To verify the internal validity of the results, we included both positive and negative control outcomes. To confirm GLP‐1RA use during follow‐up, we examined the proportion of an additional prescription of GLP‐1RA in the EHR. As a surrogate marker for treatment compliance, we used the most recent HbA1C during follow‐up (mean compared by t‐test) as well as the change in mean HbA1C from the index date. For a negative control outcome (i.e., not expected to be affected by treatment type), we used the composite incidence of burns (ICD‐10 codes T20–T32) and contact with a dog (ICD‐10 code W54) compared by HR with the log‐rank test, similar to the primary outcome.

3. Results

A total of 89,646 adult patients with T2DM and evidence of long‐term GLP‐1RA use were identified (Figure 1A). On average, patients prescribed GLP‐1RA were younger and had a higher BMI than those prescribed alternative anti‐diabetic medications. Following PSM, all the significant differences in baseline characteristics between the cohorts were balanced (Table 1).

FIGURE 1.

(A) Participant selection process (number of participants). (B) Study timeline. CA, cancer; DPP4i, dipeptidyl peptidase 4 inhibitors; GLP‐1RA, glucagon like peptide 1 receptor agonists; MEN2, multiple endocrine neoplasia type 2; SGLT2i, sodium glucose transporter 2 inhibitors; SU, sulfonylureas; T2DM, type 2 diabetes mellitus; TZD, thiazolidinediones.

TABLE 1.

Baseline characteristics before and after propensity‐score matching for GLP‐1RA versus insulin (for the other comparisons see Supporting Information S1: Table S1).

Characteristic name	Characteristic ID	Before propensity score matching					After propensity score matching
		GLP‐1RA (n = 89,646)		Insulin (n = 354,939)		SMD	GLP‐1RA (n = 62,494)		Insulin (n = 62,494)		SMD
		Patients (%)	Mean ± SD	Patients (%)	Mean ± SD		Patients (%)	Mean ± SD	Patients (%)	Mean ± SD
Demographics
Age at index (years)	AI	89,646 (100)	57.07 ± 11.65	354,939 (100)	60.04 ± 14.70	0.223882	62,494 (100)	58.05 ± 11.74	62,494 (100)	58.18 ± 14.33	0.009754
Female	F	45,554 (50.81)		163,272 (46.00)		0.096469	31,737 (50.78)		31,915 (51.07)		0.005698
Male	M	39,643 (44.22)		181,403 (51.11)		0.138209	28,037 (44.86)		27,934 (44.70)		0.003314
Unknown sex	UN	23,997 (26.77)		96,375 (27.15)		0.008652	16,968 (27.15)		16,808 (26.90)		0.005765
Race
Asian	2028–9	3333 (3.72)		14,682 (4.14)		0.021548	2267 (3.63)		2231 (3.57)		0.003093
Black/African American	2054–5	15,806 (17.63)		79,089 (22.28)		0.116563	11,429 (18.29)		11,267 (18.03)		0.006724
White	2106–3	56,352 (62.86)		208,220 (58.66)		0.086034	39,025 (62.45)		39,353 (62.97)		0.010854
Unknown race	UNK	10,725 (11.96)		38,354 (10.81)		0.036462	7376 (11.80)		7288 (11.66)		0.004376
Ethnicity
Hispanic or Latino	2135–2	8553 (9.54)		31,978 (9.01)		0.018320	5876 (9.40)		6037 (9.66)		0.008773
Not Hispanic or Latino	2186–5	57,096 (63.69)		226,586 (63.84)		0.003068	39,650 (63.45)		39,649 (63.44)		0.000033
Unknown ethnicity	UN	23,997 (26.77)		96,375 (27.15)		0.008652	16,968 (27.15)		16,808 (26.90)		0.005765
Co‐morbidities
Ischaemic heart diseases	I20‐I25	17,084 (19.06)		129,153 (36.39)		0.394621	14,014 (22.43)		14,013 (22.42)		0.000038
T2DM kidney complications	E11.2	16,361 (18.25)		93,577 (26.36)		0.195826	12,022 (19.24)		12,168 (19.47)		0.005913
T2DM ophthalmic complications	E11.3	9965 (11.12)		46,477 (13.09)		0.060681	7179 (11.49)		7295 (11.67)		0.005801
Overweight and obesity	E66	47,472 (52.95)		116,521 (32.83)		0.415337	29,906 (47.85)		29,926 (47.89)		0.000641
Goitre	E04	4788 (5.34)		13,235 (3.73)		0.077542	3085 (4.94)		3030 (4.85)		0.004080
Labs and measurements
BMI (kg/m2)	9083	59,914 (66.83)	36.04 ± 8.01	244,338 (68.84)	31.34 ± 8.17	0.581633	41,074 (65.72)	35.45 ± 8.05	41,620 (66.60)	34.87 ± 8.12	0.071695
Haemoglobin (g/d)	9014	59,699 (66.59)	13.44 ± 1.83	269,966 (76.06)	12.25 ± 2.32	0.569149	41,532 (66.46)	13.26 ± 1.88	41,557 (66.50)	13.05 ± 2.05	0.105692
HbA1c (%)	9037	67,224 (74.99)	7.95 ± 2.05	242,226 (68.24)	8.07 ± 2.37	0.051922	44,348 (70.96)	7.96 ± 2.13	45,128 (72.21)	8.13 ± 2.20	0.078790
eGFR by MDRD	8001	69,545 (77.58)	79.03 ± 28.03	286,348 (80.67)	66.16 ± 36.95	0.392297	47,460 (75.94)	76.57 ± 28.86	47,586 (76.14)	76.61 ± 32.42	0.001405
TSH (mIU/L)	9040	46,653 (52.04)	3.77 ± 35.46	168,607 (47.50)	4.09 ± 33.56	0.009346	30,605 (48.97)	4.03 ± 37.69	30,534 (48.86)	3.86 ± 33.74	0.004783
Medications
Metformin	6809	71,040 (79.25)		151,890 (42.79)		0.805793	45,800 (73.29)		46,170 (73.88)		0.013429
Sulfonylureas	A10BB	35,797 (39.93)		82,361 (23.20)		0.365864	22,558 (36.10)		22,802 (36.49)		0.008120
DPP‐4 inhibitors	A10BH	22,077 (24.63)		49,395 (13.92)		0.274077	13,698 (21.92)		13,904 (22.25)		0.007947
SGLT‐2 inhibitors	A10BK	29,360 (32.75)		15,199 (4.28)		0.787716	11,910 (19.06)		11,387 (18.22)		0.021491
Thiazolidinediones	A10BG	10,712 (11.95)		14,370 (4.05)		0.294378	5875 (9.40)		5953 (9.53)		0.004264
Thyroid hormones	HS851	14,699 (16.40)		55,683 (15.69)		0.019311	10,278 (16.45)		10,294 (16.47)		0.000690

Abbreviations: BMI: body mass index; eGFR by MDRD: estimated glomerular filtration rate by modification of diet in renal disease; HbA1c: haemoglobin A1c; HR: hazard ratio; NCO: negative control outcome; PSA: prostate‐specific antigen; TSH: thyroid stimulating hormone.

The median follow‐up period, starting 1–3 years after the initial GLP‐1RA prescription was an additional 4.5 ± 2.3 years (Figure 1B). Long‐term prescription of GLP‐1RA was not associated with a significantly increased risk of thyroid cancer compared to any of the other anti‐diabetic medication categories: versus insulin (HR 0.949; 95% CI 0.745–1.208 p = 0.6716; n = 62,494 per PSM cohort); versus metformin (HR 1.065; 0.853–1.331 p = 0.5776; n = 73,107 per PSM cohort); versus DPP4i (HR 1.097; 0.824,1.46; p = 0.5249; n = 42,955 per PSM cohort); versus SGLT2i: (HR 0.904; 0.589,1.388; p = 0.6456; n = 23,532 per PSM cohort); versus SU: (HR 1.114; 0.855–1.451; p = 0.4258; n = 54,634 per PSM cohort); versus TZD: (HR 1.148; 0.966–1.365; p = 0.9744; n = 18,765 per PSM cohort). Survival curves for each of the comparisons are shown in Figure 2.

FIGURE 2.

Risk of thyroid cancer with GLP‐1RA versus other anti‐diabetic medications. (A) GLP1‐RA versus insulin: n = 62,494 per PSM cohort; mean follow‐up years 4.5 ± 2.3 versus 3.9 ± 2.5; 137 versus 127 patients with an outcome. (B) GLP1‐RA versus metformin: n = 73,107 per PSM cohort; mean follow‐up years 4.5 ± 2.3 versus 4.0 ± 2.5; 169 versus 143 patients with an outcome. (C) GLP1‐RA versus DPP4i: n = 42,955 per PSM cohort; Mean follow‐up years 4.6 ± 2.4 versus 4.2 ± 2.5; 104 versus 86 patients with an outcome. (D) GLP1‐RA versus SGLT2i: n = 23,532 per PSM cohort; mean follow‐up years 4.3 ± 2.3 versus 3.9 ± 2.2; 42 versus 42 patients with an outcome. (E) GLP1‐RA versus SU: n = 54,634 per PSM cohort; mean follow‐up years 4.6 ± 2.4 versus 4.1 ± 2.5; 123 versus 99 patients with an outcome. (F) GLP1‐RA versus TZD: n = 18,765 per PSM cohort; mean follow‐up years 4.9 ± 2.5 versus 1790 ± 963; 35 versus 30 patients with an outcome. CI, confidence interval; DPP4i, dipeptidyl peptidase 4 inhibitors; GLP‐1RA, glucagon like peptide 1 receptor agonists; HR, hazard‐ratio; NCOs, negative control outcomes; PSM: propensity‐score matched; SGLT2i, sodium glucose transporter 2 inhibitors; SU, sulfonylurea; TZD, thiazolidinedione.

Analyses stratified by sex, race, age, obesity status, HBA1c, type of GLP‐1RA and healthcare organization type consistently demonstrated no increased risk of thyroid cancer associated with long‐term GLP‐1RA use compared with other anti‐diabetic medications (Figure 3).

FIGURE 3.

Risk of thyroid cancer up‐ to 10 years of follow up associated with GLP‐1RA versus other anti‐diabetic medications, stratified by subgroups (n = per PSM cohort). BMI, body mass index; CI, confidence interval; DPP4i, dipeptidyl peptidase 4 inhibitors; GLP‐1RA, glucagon like peptide 1 receptor agonists; HbA1c, haemoglobin A1c; HR, hazard‐ratio; PSM, propensity‐score matched; SGLT2i, sodium glucose transporter 2 inhibitors; SU, sulfonylurea; TZD, thiazolidinedione.

Overall, the control outcomes behaved as expected. The proportion of participants in the GLP‐1RA cohort with an additional prescription for GLP‐1RA during follow‐up was 87%. Long‐term GLP‐1RA prescription was associated with greater reductions as well as lower mean HbA1C than with the other anti‐diabetic medications during follow‐up (Table 2) with no change in the incidence of the negative control composite outcome (Table 3). Sensitivity analyses including extended follow‐up to 20 years as well as changes in analysis parameters were consistent with the primary analysis, finding no increase in risk of thyroid cancer with GLP‐1RA versus any of the active comparators (Supporting Information S1: Tables S2–S4).

TABLE 2.

Positive control outcome: Change in HbA1C with GLP‐1RA versus active comparators.

	GLP‐1RA		Active comparator		p‐value
	Most recent HbA1C% ± SD	ΔHbA1C%	Most recent HbA1C% ± SD	ΔHbA1C%
Insulin	7.54 ± 1.8	−0.41	7.78 ± 1.9	−0.33	< 0.0001
Metformin	7.56 ± 1.8	−0.26	7.63 ± 1.9	−0.21	< 0.0001
DPP4i	7.50 ± 1.7	−0.25	7.60 ± 1.8	−0.15	< 0.0001
SGLT2i	7.63 ± 1.7	−0.21	7.72 ± 1.7	−0.01	< 0.0001
SU	7.54 ± 1.8	−0.26	7.75 ± 1.9	−0.15	< 0.0001
TZD	7.49 ± 1.6	−0.17	7.48 ± 1.7	−0.11	0.720

Abbreviations: DPP4i: dipeptidyl peptidase 4 inhibitors; GLP‐1RA: glucagon like peptide 1 receptor agonists; HbA1c: haemoglobin A1C; SD: standard deviation; SGLT2i: sodium glucose transporter 2 inhibitors; SU: sulfonylureas; TZD: thiazolidinediones.

TABLE 3.

Negative control outcome: GLP‐1RA versus active comparators.

Active comparator	HR for NCO (95% CI)
Insulin	0.99 (0.91–1.09)
Metformin	0.97 (0.90–1.06)
DPP4i	1.07 (0.96–1.20)
SGLT2i	1.08 (0.92–1.28)
SU	0.96 (0.87–1.05)
TZD	1.14 (0.97–1.34)

Abbreviations: DPP4i: dipeptidyl peptidase 4 inhibitors; GLP‐1RA: glucagon like peptide 1 receptor agonists; HR: hazard ratio; NCO: negative control outcome; SGLT2i: sodium glucose transporter 2 inhibitors; SU: sulfonylureas; TZD: thiazolidinediones.

4. Discussion

In this large, multicenter retrospective cohort study, we investigated the association between long‐term GLP‐1RA use and thyroid cancer risk in adults with T2DM. After controlling for potential confounding baseline characteristics through PSM, our analysis did not identify an increased thyroid cancer risk in long‐term GLP‐1RA users compared with other diabetes medications including insulin, metformin, SUs, TZDs, DPP4i, or SGLT2i during a follow‐up period of up‐to 10 years.

Secondary analyses from randomised controlled trials (RCTs) and metanalyses have rarely reported thyroid cancer cases among GLP‐1RA‐treated participants, resulting in imprecise effect estimates and no conclusive evidence of increased risk. Observational studies have yielded mixed results. For example, a retrospective study using the TriNetX database examined the association between GLP‐1RAs and 13 obesity‐associated cancers in patients with T2DM and found no difference in thyroid cancer incidence between GLP‐1RA users and those treated with insulin [19]. Similarly, Dore et al. observed no increased thyroid cancer risk among GLP‐1RA users compared with those using metformin or glyburide [20]. A large registry‐based study from Scandinavia found no increased risk of thyroid cancer among GLP‐1RA users compared with DPP‐4 inhibitor users, even after accounting for detection bias and competing risks [21]. In a separate Korean nationwide cohort, Bea et al. evaluated thyroid cancer risk among GLP‐1RA and DPP‐4i users compared with SGLT2i, which served as a neutral comparator [22]. Both GLP‐1RAs and DPP‐4 inhibitors were not associated with increased thyroid cancer risk. Most recently, an international multisite cohort study using linked data from Denmark, Norway, and Scotland found no association between GLP‐1RA use and thyroid cancer risk when compared to DPP‐4 inhibitors [14]. However, a modest effect was observed when compared with SUs, which may be attributed to an imbalance in unmeasured obesity. Importantly, these studies were limited by relatively short median follow‐up durations, ranging from approximately 2–4 years, potentially limiting their ability to assess long‐term cancer risk.

Other studies have suggested potential concerns. A case‐control analysis by Bezin et al. reported a modestly elevated hazard ratio for thyroid cancer with a cumulative use of GLP‐1RA of 1–3 years compared to other second‐line diabetes medications [5]. However, this study had notable limitations, including its non‐randomised design, reliance on medical claims data with unusually high rates of MTC, a short exposure lag, and an inability to adequately control for detection bias. Additionally, the GLP‐1RA group had a higher prevalence of obesity and thyroid disease, which are known risk factors for thyroid cancer. Pharmacovigilance data have also suggested a higher frequency of thyroid cancer in GLP‐1RA users, though these findings are limited by the voluntary nature of reporting systems and lack of robust comparator groups [6, 23, 24, 25]. A recent study by Brito et al. supports the notion that the observed increase in thyroid cancer among GLP‐1RA users may be due to enhanced detection and surveillance rather than a true increase in cancer risk, as the increase was limited to the first year after GLP‐1RA initiation [26].

The present study overcomes many of the limitations of existing literature. We conducted robust matching to address factors such as healthcare use, cancer screening, and biases related to psychosocial or preventative medicine, which are particularly relevant for these cancers. We also matched for obesity and a history of nodular thyroid disease, both of which can influence outcomes. Furthermore, as patient medication adherence information is unavailable in EHRs, we required an additional prescription of GLP‐1 RA within 1–3 years of the initial prescription to suggest long‐term adherence to therapy. Though a case of initial discontinuation followed by a later prescription of the medication is theoretically possible, the lower mean HbA1c observed over follow‐up with GLP‐1RA compared to other anti‐diabetic medications provides additional support for long‐term adherence to the medication. A key strength of this study is the large cohort with a long‐term follow‐up period of up‐to 10 years in the main analysis and up to‐20 years in the sensitivity analysis, which provides valuable insights into the prolonged effects of GLP‐1RA therapy.

4.1. Limitations

While TriNetX applies standardised data‐quality procedures [27], investigators do not have access to raw patient‐level data or internal analytic pipelines. Therefore, while overall data integrity and completeness are ensured through TriNetX's validation processes as well as specific external validation studies [28, 29, 30], we could not independently verify these elements in our study beyond the platform's built‐in quality‐control mechanisms. This represents an inherent methodological limitation of all analyses conducted within TriNetX, affecting analytical transparency. Additional limitations of the current study stem from its retrospective, observational design, which relies on EHR codes and may result in unmeasured or uncontrolled confounders and biases including those related to differences in cancer screening. ICD‐codes for thyroid cancer have been shown to have very high sensitivity but lower specificity [31]. This may have led to the misclassification of some benign nodules as cancer but not to a systematic bias when comparing GLP‐1RA versus other medications. Additionally, due to limitations in data available through TriNetX, we could not directly adjust for variations in iodine intake that influence thyroid cancer incidence, though we did adjust for baseline presence of goitre, thyroid nodules, TSH and use of thyroid hormone, which are all affected by iodine intake. To assess for the presence of unaccounted confounders, we compared the incidence of a negative control outcome, which was not found to be different between the groups, though residual bias could never be fully ruled out. GLP‐1RA treatment has a notoriously high attrition rate during the first year. We attempted to overcome this limitation by including only patients with an additional long‐term prescription after initiation. The difference in a positive control outcome, a sustained modest yet significant long term reduction in HbA1C, is reassuring in this context, though true adherence can never be fully verified in a EHR based study. A specific limitation in the PSM was the inability to balance specific HCO in the TriNetX network. We addressed this limitation by adding a sub‐group stratification comparing academic and non‐academic HCOs separately. Though the analysis was performed on a global network, most of the data is US‐based, limiting the true global generalisability of the findings.

4.2. Conclusions

The clinical implications of our findings are significant. GLP‐1RAs are now integral to T2DM and obesity management, providing substantial benefits in glycaemic control, weight loss, and cardiovascular risk reduction. Misplaced concerns about thyroid cancer may deter their use, depriving patients of these well‐established benefits. Our findings, based on a decade of follow‐up, provide reassurance that long‐term GLP‐1RA use is not associated with an increased risk of thyroid cancer. These results should support evidence‐based decision‐making, emphasising that the benefits of GLP‐1RA therapy outweigh unsubstantiated concerns about thyroid cancer risk.

Author Contributions

Rena Pollack: conceptualisation, writing – original draft, writing – review and editing. Joshua Stokar: data curation, formal analysis, writing – original draft, writing – review and editing.

Funding

The authors have nothing to report.

Ethics Statement

TriNetX data are de‐identified in compliance with HIPAA regulations; thus, informed consent was not required. The study received an exemption from the Hadassah Medical Centre institutional ethics board approval.

Conflicts of Interest

The authors declare no conflicts of interest.

Peer Review

The peer review history for this article is available at https://www.webofscience.com/api/gateway/wos/peer-review/10.1002/dmrr.70104.

Supporting information

Supporting Information S1

Pollack, Rena , and Stokar Joshua. 2025. “Long‐Term Glucagon‐Like Peptide 1 Receptor Agonist Use Is Not Associated With Increased Risk of Thyroid Cancer in Adults With Type 2 Diabetes,” Diabetes/Metabolism Research and Reviews: e70104. 10.1002/dmrr.70104.

Data Availability Statement

Details required to replicate the network queries are provided in the manuscript. Access to TriNetX can be requested by contacting join@trinetx.com, although access may incur costs and require a data‐sharing agreement.