Abstract
Background
Initiation of Glucagon-Like Peptide-1 receptor agonists (GLP-1RA) in patients with type 2 diabetes (T2D) treated with levothyroxine may decrease the required levothyroxine dose due to weight loss or enhance levothyroxine absorption through delayed gastric emptying. These changes may cause thyroid hormone over-replacement and increased risk of atrial fibrillation/flutter (AF/Aflutter) and stroke. Our study aims to investigate the impact of GLP-1RA initiation on risks of AF/Aflutter and stroke in patients with T2D treated with levothyroxine, compared to sodium-glucose cotransporter 2 (SGLT2) inhibitors.
Methods
Leveraging the target trial emulation framework, we conducted a retrospective study using observational data to emulate a new user, active comparator trial examining the effects of initiating GLP-1RA (exposure group) versus SGLT2 inhibitors (control group), with random treatment assignment emulated by propensity score matching with 1:1 ratio. We used a 15% nationally representative sample of U.S. Medicare beneficiaries to identify participants > 65 years, continuously on stable dose of levothyroxine for ≥ 6 months before the index date (i.e., GLP-1RA or SGLT2 inhibitor initiation), with continuous Medicare enrollment, without malignant cancer or palliative care during 1 year before the index date. The primary outcome was AF/Aflutter, and secondary outcome was stroke, including ischemic stroke or transient ischemic attack. We assessed the per-protocol effects of GLP-1RA vs. SGLT2 inhibitors using inverse-probability-censoring weighted Cox proportional hazards models.
Results
After matching, the study cohort included 2,384 participants in both GLP-1RA and SGLT2 groups with mean age (SD): 73.3 (5.9) vs. 73.2 (5.8), and 71.5% and 71.8% of female. The median follow-up time was 1.05 years. Compared to SGLT2 inhibitors, initiation of GLP-1RA was associated with higher risk of AF/Aflutter (HR: 1.46; 95% CI: 1.28–1.67), while no statistically significant difference was observed between the two groups for stroke (HR: 1.17; 95% CI: 0.98–1.39). Sensitivity analyses showed consistent results, including restricting outcomes to inpatient visits, conducting an intention-to-treat analysis, applying a prevalent new user design, and substituting SGLT2 inhibitors with dipeptidyl peptidase-4 (DPP4) inhibitors as the active comparator.
Conclusions
In patients with T2D historically treated with stable doses of levothyroxine, GLP-1RA initiation was associated with a higher risk of AF/Aflutter. Further research is warranted to investigate the potential roles of weight loss, TSH fluctuations, and levothyroxine dose adjustment after GLP-1RA in mediating the cardiovascular risk.
Supplementary Information
The online version contains supplementary material available at 10.1186/s12933-025-02971-7.
Keywords: GLP-1RA, levothyroxine, SGLT2 inhibitors, atrial fibrillation or flutter, stroke, target trial emulation
Research insights
What is currently known about this topic?
T2D and hypothyroidism commonly coexist in older adults.
GLP-1RA reduces cardiovascular risk in T2D patients but may affect thyroid hormone needs.
Levothyroxine over-replacement increases AF/Aflutter and stroke risk.
What is the key research question?
Does GLP-1RA initiation increase AF/Aflutter or stroke risk in T2D patients treated with levothyroxine?
What is new?
This is the first nationwide target trial emulation examining GLP-1RA safety in levothyroxine users.
We demonstrate that GLP-1RA initiation significantly increases the risk of AF/Aflutter.
This study uncovers a new drug–drug interaction between GLP-1RA use and levothyroxine therapy.
How might this study influence clinical practice?
Highlights need to monitor thyroid status and adjust levothyroxine after GLP-1RA initiation.
Introduction
Glucagon-like peptide-1 receptor agonists (GLP-1RA) are effective therapies for patients with type 2 diabetes (T2D) and/or obesity, particularly for those with coexisting cardiovascular disease and/or high cardiovascular risk [1]. These agents provide multiple benefits, including weight reduction, cardio-renal protection, improved heart failure symptoms and hepatic steatohepatitis, and overall reduced mortality [2–9]. Prescriptions of GLP-1RA have surged by 300% from 2020 to 2022, reaching 9 million annually [10], reflecting the high prevalence of T2D (15%) and obesity (42%) among U.S. adults [10–13]. Hypothyroidism is also common in this population with higher prevalence in women and older adults [14], and in general affecting 30% and 21% of patients with T2D and obesity, respectively [14–17]. Levothyroxine is the standard treatment for hypothyroidism, effectively replacing endogenous thyroid hormone to restore levels and relieve symptoms [18]. Treatment adequacy is assessed with thyroid-stimulating hormone (TSH) testing, a standard biomarker [19–22]. Clinical guidelines recommend at least annual TSH testing, or more frequently with clinical changes, to maintain euthyroid status and avoid complications from excess or inadequate thyroid hormone replacement.
Importantly, initiation of GLP-1RA may decrease levothyroxine dose requirements in patients with hypothyroidism through weight loss, enhance levothyroxine absorption due to delayed gastrointestinal (GI) transit, and have direct effects on the hypothalamic-pituitary-thyroid axis [23–29], potentially increasing the risk of thyroid hormone over-replacement (i.e., iatrogenic hyperthyroidism). Thyroid hormone over-replacement is associated with increased risks of atrial fibrillation or flutter (AF/Aflutter) and stroke, especially in older individuals with T2D and obesity—populations already at elevated cardiovascular risk [30–35]. For example, a study of 643,687 levothyroxine users found that over-replacement was associated with higher AF risk (OR: 1.31, 95% CI: 1.24–1.38), while another study showed increased stroke risk (OR: 1.33, 95% CI: 1.24–1.43) [19, 36].
With increased use of GLP-1RA overall, it is imperative to examine whether initiation of GLP-1RA increases the risk of thyroid hormone over-replacement and ultimately elevated risks of AF/Aflutter and stroke among levothyroxine users with T2D. Although case reports have described levothyroxine over-replacement and subsequent AF/Aflutter following GLP-1RA initiation [37], no largescale study has evaluated the real-world effect of GLP-1RA initiation on cardiovascular outcomes. We address this knowledge gap by examining the impact of GLP-1RA initiation on the risk of new-onset AF/Aflutter and stroke among patients with T2D who continuously received levothyroxine therapy.
Methods
The steps and key elements of target trial emulation are shown in Supplement Fig. 1. In accordance with the target trial emulation framework, we first specified the protocol of a hypothetical randomized controlled trial (RCT) comparing the effects of initiating a GLP-1RA versus an SGLT2 inhibitor on the outcomes of AF/Aflutter and stroke among patients with T2D who have been receiving stable dose of levothyroxine. Then, we emulated this target RCT using observational data from Medicare claims database. The protocol of the target RCT and the details of emulation are summarized in Supplement Table 1. When rigorously implemented, this approach can produce results comparable to those of the target RCT [38–41]. Methodologically akin to a retrospective cohort analysis, our study differs in that it emulates a target trial by prespecifying the protocol, aligning index date, and using propensity score matching to emulate randomization, thereby reducing immortal-time bias, improving transparency, and aligning real-world evidence generation with clinical trial standards.
We used an active-comparator, incident new-user design to emulate the target trial. Compared with a placebo or non-user comparator, this design minimizes immortal time bias and confounding by balancing disease severity or treatment eligibility in two groups, providing a more reliable real-world estimate of relative effects. SGLT2 inhibitors were chosen as the comparator given they are guideline-recommended second-line glucose-lowering medication and share similar cardiovascular benefits with GLP-1RA, which helps reduce confounding by indication. While SGLT2 inhibitors do promote modest weight loss (1–3 kg), their effect is less pronounced than that of GLP-1RA, and they have less impact on thyroid hormone absorption [42, 43]. No association has been observed between SGLT-2 inhibitors and thyroid dysfunction, making iatrogenic hyperthyroidism unlikely. To quantify potentially avoidable AF/Aflutter through timely TSH testing and levothyroxine dose titration, we conducted an exploratory analysis that chronologically outlines the sequence of these events. The study protocol received approval from the Institutional Review Board at the University of Florida, USA (Approval code: IRB201900262).
Data Source: 15% nationally representative sample of Medicare fee-for-service (FFS) administrative claims data, encompassing hospitalization claims (Part A), outpatient services (Part B), and prescription drug records (Part D).
Eligibility criteria: First, we identified patients meeting claims-based criteria for T2D between 01/01/2011 and 12/31/2019, using International Classification of Disease, Ninth/Tenth Revision, Clinical Modification (ICD-9/10-CM) diagnosis codes occurred at the primary or secondary position from inpatient, outpatient or carrier claims (Supplement Table 2). Participants entered the base cohort only once, when diabetes criteria were met (i.e., on their first inpatient visit, or the second outpatient visit with a diagnosis of T2D).
Participants were excluded if they met any of the following criteria: (1) age < 65 years; (2) history of any malignancy (except non-melanoma skin cancer) within 1 year before/on the index date; (3) receipt of palliative care within 1 year before/on the index date; (4) history of thyroid neoplasm, hyperthyroidism, thyroiditis, or other thyroid diseases (excluding hypothyroidism) within 1 year before/on the index date; (5) use of medications suggestive of thyroid disease or those known to interfere with thyroid hormone levels (e.g., newly prescribed proton pump inhibitors [PPIs], prednisone, or prednisolone) within 6 months before/on the index date; (6) diagnosis of AF/Aflutter or stroke at any time before/on the index date; (7) lack of continuous enrollment in Medicare Parts A, B, and D for at least 1 year before the index date; and (8) without ≥ 6 months of continuous levothyroxine use at a stable dose or a medication possession ratio (MPR) ≥ 80% during the 1-year pre-index period. The cohort creation process is illustrated in Fig. 1.
Fig. 1.
Flowchart for the cohort creation process
Treatments: After entering the base cohort, participants joined the study cohort on the date of initiating GLP-1RA (exposure group) or SGLT2 inhibitors (control group), defined as the index date. Initiation was defined as having no prescriptions for either medication class in the one year before the index date. Participants initiated both GLP-1RA and SGLT2 inhibitors on the same date were excluded.
Outcomes: The primary outcome was atrial fibrillation or flutter (AF/Aflutter) given they share similar causes (i.e. thyroid overreplacement) and symptoms [35, 44, 45]. The secondary outcome was stroke, including ischemic stroke and/or transient ischemic attack (TIA). Events were identified using relevant ICD-9/10-CM codes (Supplement Table 2) at primary position from hospitalizations or emergency department (ED) visit records [46], as both AF/Aflutter and stroke are frequently first diagnosed in ED visits [47, 48], and excluding these encounters could underestimate incidence. The event date was determined by the date of admission or ED visit.
Follow-up: Consistent with the target trial protocol, patients were followed from the date of treatment initiation until the first occurrence of any of the following events: a study outcome, bariatric surgery, initiation of medications suggestive of thyroid disease or that interfere with thyroid hormone levels (e.g., PPIs, prednisone, or prednisolone), diagnosis of a thyroid neoplasm, thyroid cancer, hyperthyroidism, thyroiditis, or other thyroid diseases (excluding hypothyroidism), loss to follow-up (defined by not continuous enrolling in the Medicare Part A/B/D plan), discontinuation of assigned treatment (GLP-1RA or SGLT2), switch to the other studied treatment, death, or the end of the study period (12/31/2020).
Causal Contrasts of Interest: We assessed the per-protocol effect of initiated and continuously received GLP-1RA vs. SGLT2 inhibitors on the new-onset of AF/Aflutter or stroke among patients with T2D historically on stable doses of levothyroxine.
Statistical analysis: To emulate randomization, we applied propensity score (PS) matching with 1:1 ratio using the nearest neighbor approach without replacement, with a caliper of 0.25 of the logit of PS. The propensity score was estimated using logistic regression model including 30 baseline covariates, covering demographics (age, sex, race), behavioral factors (alcohol use, smoking status), index year, comorbidities, and medication use (Supplement Table 3).
We used weighted Cox proportional hazards models to estimate the hazard ratio (HR) and 95% confidence interval (CI) for primary and secondary outcomes. Inverse probability censoring weight (IPCW) was estimated by pooled logistic regression to account for informative censoring [49]. The proportional hazards assumption was evaluated using Schoenfeld residuals. We created Kaplan–Meier curves to better visualize the results. In the exploratory analysis, we described the percentage and time sequence of thyroid-stimulating hormone (TSH) test, levothyroxine dose adjustment, and new-onset of AF/Aflutter within 18 months after the index date.
We conducted five sensitivity analyses to assess the robustness of our findings. First, to address potential outcome misclassification, we restricted the primary outcome by capturing inpatient visits only (i.e., not including ED visits). Second, an intention-to-treat analysis was performed to evaluate the effect of GLP-1RA initiation compared to SGLT2 inhibitors without censoring for treatment discontinuation or crossover during follow-up. Third, we assessed residual confounding by examining the association between GLP-1RA and car accident, defined by the inpatient and ED visits with related ICD-9/10-CM codes. Car incidents are a well-adopted negative control outcome because they are not expected to be associated with the exposure but share similar risk factors with AF (e.g., epilepsy, hypertension, obesity and sleeping disorders) [50, 51]. Other factors to car accidents, weather conditions and driving behaviors, are assumed to be similar between the comparison groups. Fourth, to expand the sample size, we conducted prevalent new user design by allowing previous use of SGLT2 inhibitors before index date for both groups. Control patients were selected from time-based exposure sets by exactly matching the length between the base cohort entry date and index date. We also exactly matched the duration of SGLT2 inhibitors’ use before index date in month level. Lastly, we replaced SGLT2 inhibitors with dipeptidyl peptidase-4 (DPP4) inhibitors as the active comparator to evaluate the robustness of our findings given that DPP4 inhibitors have less cardioprotective effects than SGLT2 inhibitors [52–54]. Patients initiating DPP4 inhibitors were identified using the same criteria as SGLT2 inhibitor initiators, with the index date defined as the first DPP4 inhibitor prescription. Individuals with prior DPP4 inhibitor use within one year were excluded. All other eligibility criteria and analytic approaches were identical to those in the cohort of SGLT2 inhibitors.
Results
Population characteristics
After 1:1 PS matching, we identified 2,384 patients in both the GLP-1RA and SGLT2 group, respectively. The propensity scores between the two groups demonstrated substantial overlap (Supplement Fig. 2) and baseline characteristics were well balanced, with standardized mean differences (SMD) < 0.1 for all covariates (Table 1, Supplement Fig. 3). The mean age (standard deviation [SD]) was 73.3 (5.9) vs. 73.2 (5.8) years in the GLP-1RA vs. SGLT2 group, and women comprised 71.5% and 71.8% of the two groups, respectively. Most participants were white (86.2% vs. 85.6%), without history of tobacco use (82.5% vs. 83.8%) or alcohol disorder (99.3% vs. 99.4%). The mean (SD) Charlson comorbidity index was 2.6 (1.6) vs. 2.5 (1.5), and the prevalence of heart failure was 10.1% vs. 12.1%, hypertension 86.6% vs. 85.7%, myocardial infarction 3.6% vs. 3.1%, coronary artery disease 23.9% vs. 23.8%, and chronic kidney disease 23.2% vs. 20.7%, respectively. The median follow-up time was 1.05 years in both groups.
Table 1.
Baseline characteristics of patients in GLP-1RA and SGLT2 groups before and after PS matching
| Characteristics | Before match | After match | ||||
|---|---|---|---|---|---|---|
| GLP-1RA (n = 3,544) | SGLT2 (n = 4,419) | SMD | GLP-1RA (n = 2,384) | SGLT2 (n = 2,384) | SMD | |
| Age, years | -0.04 | 0.02 | ||||
| Mean (SD) | 73.2 (5.9) | 73.4 (5.8) | 73.3 (5.9) | 73.2 (5.8) | ||
| [65,70) | 1,136 (32.1%) | 1,324 (30.0%) | 759 (31.8%) | 776 (32.6%) | ||
| [70,75) | 1,194 (33.7%) | 1,484 (33.6%) | 794 (33.3%) | 782 (32.8%) | ||
| [75,80) | 712 (20.1%) | 956 (21.6%) | 482 (20.2%) | 496 (20.8%) | ||
| [80,85) | 298 (8.4%) | 416 (9.4%) | 209 (8.8%) | 205 (8.6%) | ||
| [85,90) | 151 (4.3%) | 177 (4.0%) | 104 (4.4%) | 87 (3.6%) | ||
| ≥ 90 | 53 (1.5%) | 62 (1.4%) | 36 (1.5%) | 38 (1.6%) | ||
| Female gender | 2,597 (73.3%) | 3,054 (69.1%) | 0.09 | 1,705 (71.5%) | 1,712 (71.8%) | -0.01 |
| Race/ethnicity | -0.08 | -0.02 | ||||
| White | 3,059 (86.3%) | 3,738 (84.6%) | 2,054 (86.2%) | 2,041 (85.6%) | ||
| Black | 192 (5.4%) | 189 (4.3%) | 112 (4.7%) | 114 (4.8%) | ||
| Hispanic | 105 (3.0%) | 159 (3.6%) | 76 (3.2%) | 78 (3.3%) | ||
| Others | 188 (5.3%) | 333 (7.5%) | 142 (6.0%) | 151 (6.3%) | ||
| Year of the index date | -0.01 | 0.02 | ||||
| 2013 | 91 (2.6%) | 27 (0.6%) | 20 (0.8%) | 17 (0.7%) | ||
| 2014 | 111 (3.1%) | 120 (2.7%) | 70 (2.9%) | 78 (3.3%) | ||
| 2015 | 123 (3.5%) | 222 (5.0%) | 89 (3.7%) | 111 (4.7%) | ||
| 2016 | 234 (6.6%) | 443 (10.0%) | 176 (7.4%) | 170 (7.1%) | ||
| 2017 | 910 (25.7%) | 1,183 (26.8%) | 680 (28.5%) | 672 (28.2%) | ||
| 2018 | 1,007 (28.4%) | 1,008 (22.8%) | 619 (26.0%) | 608 (25.5%) | ||
| 2019 | 1,068 (30.1%) | 1,416 (32.0%) | 730 (30.6%) | 728 (30.5%) | ||
| Tobacco use | 0.06 | 0.03 | ||||
| Never | 2,878 (81.2%) | 3,710 (84.0%) | 1,966 (82.5%) | 1,998 (83.8%) | ||
| Used before | 494 (13.9%) | 500 (11.3%) | 302 (12.7%) | 276 (11.6%) | ||
| Using now | 172 (4.9%) | 209 (4.7%) | 116 (4.9%) | 110 (4.6%) | ||
| Medical history | ||||||
| Acid reflux or gastro-esophageal reflux disease | 594 (16.8%) | 671 (15.2%) | 0.04 | 378 (15.9%) | 387 (16.2%) | -0.01 |
| Alcohol disorder | 28 (0.8%) | 27 (0.6%) | 0.02 | 17 (0.7%) | 14 (0.6%) | 0.02 |
| Bariatric surgery | 28 (0.8%) | 28 (0.6%) | 0.02 | 18 (0.8%) | 16 (0.7%) | 0.01 |
| Celiac disease | < 11 | < 11 | -0.02 | < 11 | < 11 | 0.01 |
| Chronic kidney disease | 1,090 (30.8%) | 816 (18.5%) | 0.29 | 553 (23.2%) | 494 (20.7%) | 0.06 |
| Coronary artery disease | 900 (25.4%) | 1,058 (23.9%) | 0.03 | 569 (23.9%) | 567 (23.8%) | 0.002 |
| Crohn’s disease | < 11 | 18 (0.4%) | -0.02 | < 11 | < 11 | 0.01 |
| Food allergies | 11 (0.3%) | 15 (0.3%) | -0.01 | < 11 | < 11 | -0.01 |
| Gastroparesis | 16 (0.5%) | 25 (0.6%) | -0.02 | 11 (0.5%) | 12 (0.5%) | -0.01 |
| Heart failure | 449 (12.7%) | 517 (11.7%) | 0.03 | 241 (10.1%) | 289 (12.1%) | 0.06 |
| Hyperlipidemia | 2,855 (80.6%) | 3,601 (81.5%) | -0.02 | 1,921 (80.6%) | 1,947 (81.7%) | -0.03 |
| Hypertension | 3,107 (87.7%) | 3,759 (85.1%) | 0.08 | 2,064 (86.6%) | 2,042 (85.7%) | 0.03 |
| Irritable bowel syndrome | 51 (1.4%) | 56 (1.3%) | 0.01 | 37 (1.6%) | 31 (1.3%) | 0.02 |
| Myocardial infarction | 145 (4.1%) | 130 (2.9%) | 0.06 | 85 (3.6%) | 74 (3.1%) | 0.03 |
| Obesity | 986 (27.8%) | 750 (17.0%) | 0.26 | 522 (21.9%) | 466 (19.5%) | 0.06 |
| Overweight | 65 (1.8%) | 85 (1.9%) | -0.01 | 49 (2.1%) | 50 (2.1%) | -0.003 |
| Sleep apnea | 588 (16.6%) | 540 (12.2%) | 0.13 | 334 (14.0%) | 318 (13.3%) | 0.02 |
| Ulcerative colitis | < 11 | 11 (0.2%) | 0.01 | < 11 | < 11 | 0 |
| Charlson comorbidity index | 0.30 | 0.08 | ||||
| Mean (SD) | 2.9 (1.7) | 2.4 (1.4) | 2.6 (1.6) | 2.5 (1.5) | ||
| 1 | 775 (21.9%) | 1,269 (28.7%) | 631 (26.5%) | 653 (27.4%) | ||
| 2 | 1,077 (30.4%) | 1,583 (35.8%) | 786 (33.0%) | 831 (34.9%) | ||
| 3 | 525 (14.8%) | 663 (15.0%) | 353 (14.8%) | 350 (14.7%) | ||
| 4 | 570 (16.1%) | 510 (11.5%) | 327 (13.7%) | 310 (13.0%) | ||
| 5 | 330 (9.3%) | 223 (5.0%) | 148 (6.2%) | 139 (5.8%) | ||
| 6 | 165 (4.7%) | 108 (2.4%) | 81 (3.4%) | 63 (2.6%) | ||
| 7 | 66 (1.9%) | 39 (0.9%) | 37 (1.6%) | 23 (1.0%) | ||
| 8 | 36 (1.0%) | 24 (0.5%) | 21 (0.9%) | 15 (0.6%) | ||
| Qrisk-3 score | -0.002 | 0.03 | ||||
| 10%-20% | 433 (12.2%) | 534 (12.1%) | 294 (12.3%) | 309 (13.0%) | ||
| 20%-30% | 1,174 (33.1%) | 1,470 (33.3%) | 793 (33.3%) | 807 (33.9%) | ||
| > 30% | 1,937 (54.7%) | 2,415 (54.7%) | 1,297 (54.4%) | 1,268 (53.2%) | ||
| Family history of cardiovascular disease | 37 (1.0%) | 59 (1.3%) | -0.03 | 25 (1.0%) | 25 (1.0%) | 0 |
| Number of all-cause hospitalizations | 0.17 | 0.03 | ||||
| Mean (SD) | 1.3 (0.7) | 1.2 (0.5) | 1.2 (0.5) | 1.2 (0.5) | ||
| 0 | 3,021 (85.2%) | 3,993 (90.4%) | 2,097 (88.0%) | 2,128 (89.3%) | ||
| 1 | 411 (11.6%) | 353 (8.0%) | 243 (10.2%) | 211 (8.9%) | ||
| ≥2 | 112 (3.2%) | 73 (1.7%) | 44 (1.8%) | 45 (1.9%) | ||
| Statin | 445 (12.6%) | 545 (12.3%) | 0.01 | 287 (12.0%) | 295 (12.4%) | -0.01 |
| Number of antihypertensive medications—Mean (SD) | 2.3 (1.3) | 2.1 (1.3) | 0.15 | 2.2 (1.3) | 2.2 (1.3) | 0.03 |
| Number of antidiabetic medications—Mean (SD) | 1.4 (1.0) | 2.0 (1.1) | -0.51 | 1.6 (0.9) | 1.6 (0.9) | 0.05 |
| Number of antilipidemic medications—Mean (SD) | 0.3 (0.7) | 0.3 (0.7) | -0.01 | 0.3 (0.7) | 0.3 (0.7) | -0.01 |
AF/Aflutter and stroke
The GLP-1RA group had higher incidence rates of AF/Aflutter (0.74 vs. 0.71) and stroke (0.74 vs. 0.49) per 100 person-years compared to the SGLT2 group (Table 2). Although the number of crude event for AF/Aflutter were the same in both groups (13 each), they had different follow-up time (1,758.20 vs. 1,834.42 person-years), resulting in higher incidence rate of AF/Aflutter in GLP-1RA group. In the Cox regression models, we applied IPCW to account for informative censoring and unequal follow-up time. The enlargement of the relative difference suggested the existence of informative censoring, and this pattern was consistent between the crude estimates and the IPCW-adjusted estimates in the sensitivity analyses. The results showed that GLP-1RA initiation was associated with a significantly increased risk of AF/Aflutter vs. SGLT2 inhibitors (HR: 1.46; 95% CI: 1.28–1.67), with no statistically significant difference for stroke (HR: 1.17; 95% CI: 0.98–1.39) (Fig. 2). Kaplan–Meier curves demonstrated consistent results (Fig. 3).
Table 2.
Incidence rates for outcomes by interventions in the primary analysis and sensitivity analyses
| Outcome | GLP-1RA group (N = 2,384) | SGLT2 group (N = 2,384) | Incidence rate ratio (IRR), 95% CI# | ||||
|---|---|---|---|---|---|---|---|
| Number of events | Exposure person-years | Event rate, per 100 person-years | Number of events | Exposure person-years | Event rate, per 100 person-years | ||
| Primary analysis | |||||||
| AF/Aflutter | 13 | 1,758.20 | 0.74 | 13 | 1,834.42 | 0.71 | 1.04 (0.48—2.25) |
| Stroke | 13 | 1,758.64 | 0.74 | <11 | Suppression | 0.49 | 1.51 (0.64—3.52) |
| Analysis used only inpatient visits to define the outcomes | |||||||
| AF/Aflutter | <11 | Suppression | 0.54 | <11 | Suppression | 0.52 | 1.04 (0.42—2.56) |
| Stroke | 13 | 1,758.64 | 0.71 | <11 | Suppression | 0.47 | 1.51 (0.64—3.52) |
| Intention-to-treat analysis | |||||||
| AF/Aflutter | 19 | 3,035.90 | 0.63 | 18 | 3,224.53 | 0.56 | 1.12 (0.59—2.14) |
| Stroke | 22 | 3,033.21 | 0.73 | 26 | 3,216.71 | 0.81 | 0.90 (0.51—1.58) |
| Analysis based on prevalent new user design*, GLP-1RA group (N = 2,863), SGLT2 group (N = 2,863) | |||||||
| AF/Aflutter | 17 | 1,949.51 | 0.87 | 16 | 2,246.72 | 0.71 | 1.22 (0.62—2.42) |
| Stroke | 14 | 1,953.28 | 0.72 | 12 | 2,238.92 | 0.54 | 1.34 (0.62—2.89) |
| Analysis using DPP4 as comparator, GLP-1RA group (N = 2,236), DPP4 group (N = 2,236) | |||||||
| AF/Aflutter | 15 | 1,737.09 | 0.86 | 21 | 2,197.88 | 0.96 | 0.90 (0.47, 1.75) |
| Stroke | 14 | 1,733.14 | 0.81 | 29 | 2,187.96 | 1.33 | 0.61 (0.32, 1.15) |
| Negative control outcomes analysis | |||||||
| Car accident | 11 | 1,759.54 | 0.63 | 12 | 1,827.12 | 0.66 | 0.95 (0.42—2.16) |
*Prevalent new user design included incident new users of GLP-1RA and prevalent users of SGLT2 inhibitors
#IRR and 95%CI in this table were crude estimation directly calculated from event counts and person-time
Fig. 2.
Forest plots for primary analysis and sensitivity analysis
Fig. 3.
Kaplan–Meier curves for AF/Aflutter and stroke by treatment group. A Kaplan–Meier curves for AF/Aflutter by treatment group (n = 2,384 per group). B Kaplan–Meier curves for stroke by treatment group (n = 2,384 per group)
In the exploratory analysis, within 18 months after the index date, nearly 60% of AF/Aflutter events occurred before patients received a TSH test in the GLP-1RA group, whereas the corresponding percentage was only <40% in the SGLT2 group. Around 25% AF/Aflutter events occurred after a TSH test but before a subsequent levothyroxine dose adjustment in GLP-1RA group (Supplement Fig. 4).
Sensitivity analysis
All sensitivity analyses showed consistent results with the primary analysis. When we captured outcomes using inpatient visits only, GLP-1RA group showed significant higher risk of AF/Aflutter (HR: 1.38, 95% CI: 1.18 – 1.61) and no statistically significant difference on the risk of stroke (HR: 1.17, 95% CI: 0.98 – 1.39). Compared to the per-protocol analysis, in the intention-to-treat analysis, the associations between GLP-1RA and AF/Aflutter remained consistent in direction and magnitude but were statistically insignificant (HR: 1.11, 95% CI: 0.58– 2.11), likely attributable to mediocre adherence of medication, and treatment switching over time, which may have attenuated the observed effect estimates. The results for stroke still showed no significant difference between the two groups (HR: 0.91, 95% CI: 0.52 – 1.61). In addition, negative control outcome, car accidents, (HR: 0.96, 95% CI: 0.42 – 2.17) were not associated with the initiation and continued use of GLP-1RA (Table 2, Fig. 2).
Using prevalent new user design, we included 2,863 patients in each study group. After propensity score matching, the propensity scores showed substantial overlap between the GLP-1RA and SGLT2 groups and the baseline characteristics were well balanced between the two groups, with SMD < 0.1 for all the baseline covariates (Supplement Figs. 5, 6). Initiation and continued use GLP-1RA was associated with an increased risk of AF/Aflutter (HR: 1.47, 95% CI: 1.32–1.65), but not stroke (HR: 1.16, 95% CI: 0.98–1.38). Using DPP4 inhibitors as the comparator, 2,236 participants per group were included. After propensity score matching, the propensity scores showed substantial overlap between the GLP-1RA and DPP4 groups and the baseline characteristics were well balanced between the two groups, with SMD < 0.1 for all the baseline covariates (Supplement Table 4, Figs. 7, 8). GLP-1RA use was associated with higher AF/Aflutter risk (HR: 1.58, 95% CI: 1.45–1.72) but lower stroke risk (HR 0.65, 95% CI: 0.60–0.71).
Discussions
We emulated a target trial to investigate the risk for adverse cardiovascular events (incident AF/Aflutter and stroke) after GLP-1RA initiation among patients with T2D treated with levothyroxine, using SGLT2 inhibitors as a comparator (and DPP4 inhibitors in a sensitivity analysis). Our results, supported by multiple sensitivity analyses, showed that GLP-1RA initiation was associated with a significantly increased risk of new-onset AF/Aflutter. The incidence rate of stroke was also higher but not statistically significant. These findings suggest that the initiation and continued use of GLP-1RA may result in decreased levothyroxine requirements or increased levothyroxine absorption – expected due to weight loss and slowed gastric motility. Without timely levothyroxine dose adjustment to reflect this clinical change, the over-replacement of thyroid hormone can lead to serious adverse events. Consistent with our population-level findings: case reports have described levothyroxine over-replacement and subsequent AF following GLP-1RA initiation [37, 55]. Our findings support American Diabetes Association (ADA) recommendations to monitor thyroid hormone levels in patients initiating and uptitrating GLP-1RA therapy [56] to ensure safe use of levothyroxine.
Indeed, both GLP-1RA and SGLT2 inhibitors have demonstrated protective effects for general cardiovascular outcomes, including AF/Aflutter and stroke. Several large RCTs, including LEADER (liraglutide) [57] and REWIND (dulaglutide) [58], have demonstrated that GLP-1RA can significantly reduce cardiovascular risk (i.e., composite of nonfatal stroke, myocardial infarction, and cardiovascular-cause death) by 12–13% in patients with T2D. A meta-analysis of RCTs showed that semaglutide treatment resulted in a 17% reduction in AF incidence rate [59], while another reported a 13% reduction in non-fatal stroke [60]. Similarly, another meta-analysis reported that GLP-1RAs reduce the risk of major adverse cardiovascular events (MACE) by 13% [61]. Likewise, SGLT2 inhibitors have shown significant reductions in cardiovascular risk by 14% in RCTs such as EMPA-REG OUTCOME (empagliflozin) [62] and CANVAS (canagliflozin) [63]. In DECLARE-TIMI 58, dapagliflozin reduced the risk of AF/Aflutter events by 19% compared with placebo [64]. In a meta-analysis of 38 RCTs, the pooled analysis showed that SGLT2 inhibitors significantly reduced the incidence of atrial fibrillation by 13% [65]. In addition, another meta-analysis suggested that the combination therapy of GLP-1RA and SGLT2 inhibitors had a significantly lower risk of MACE than treatment as usual [66].
However, in patients receiving levothyroxine, the cardiovascular benefits of GLP-1RA may be attenuated compared with SGLT2 inhibitors. Two mechanisms may predispose to thyroid hormone over-replacement and thus increase AF/Aflutter risk with GLP-1RA: weight loss and delayed gastric emptying. First, weight loss decreases the required dose of levothyroxine. In a real-world study, patients with hypothyroidism who lost > 5 pounds after initiating GLP-1RA experienced a TSH decrease of 0.36–0.55 mIU/L [29], and similar associations have been reported after bariatric surgery [25, 27, 67]. Second, delayed gastric emptying may enhance levothyroxine absorption: pharmacokinetic (PK) study showed the concurrent use of oral semaglutide and levothyroxine increased the AUC of total and free T4 level by 33% [28]. In comparison, no PK study indicated that the use of SGLT2 inhibitors could impact the required dose of levothyroxine to treat hypothyroidism.
We did not identify a statistically significant increase in stroke risk among GLP-1RA users compared to SGLT2 users. Although uncontrolled hypothyroidism is associated with increased stroke risk via mechanisms, such as increased vascular resistance, endothelial dysfunction, dyslipidemia, and a pro-thrombotic state [68], the increased stroke risk may be attenuated by the potential cerebrovascular protective effects of GLP-1RA [69]. Indeed, our sensitivity analysis partially supports the above hypothesis. We observed a lower stroke risk among patients initiated GLP-1RA than DPP4 inhibitors (HR: 0.65, 95% CI: 0.60–0.71), which is consistent with existing evidence. Several RCTs including SAVOR‑TIMI 53 (saxagliptin) [52], EXAMINE (alogliptin) [53], and TECOS (sitagliptin) [54] have shown that DPP4 inhibitors did not significantly reduce stroke risk [1].
Differences in GLP-1RA formulations may further influence the interaction with levothyroxine. Short-acting GLP-1RA (e.g., exenatide, lixisenatide) more prominently delays gastric emptying [70] but generally produces more modest and earlier-plateauing weight loss [71]. In contrast, long-acting GLP-1RA has a less pronounced effect on gastric emptying over time due to tachyphylaxis [72] but yields more significant and consistent weight loss [71]. In addition, injectable semaglutide has greater weight reduction than oral semaglutide [73]. Regarding gastric emptying, injectable GLP-1RA generally produces a more pronounced and sustained delay, whereas oral semaglutide shows a modest, early-phase effect that attenuates at steady state [74]. These pharmacologic distinctions underscore the importance of formulation-specific research of GLP-1RA initiation in patients treated with levothyroxine.
Although the majority of patients received a TSH test after initiating GLP-1RA therapy, there remains an opportunity to reduce the risk of AF/Aflutter by optimizing the timing of TSH monitoring. In our exploratory analysis, most AF/Aflutter events (nearly 60%) in the GLP-1RA group occurred among patients who had not undergone any TSH testing. Around 25% AF/Aflutter events occurred after a TSH test but before a subsequent levothyroxine dose adjustment. These patterns suggest insufficient monitoring and lack of timely dose adjustment following GLP-1RA initiation, potentially leading to thyroid hormone over-replacement. Clinical recommendations emphasize frequent TSH evaluation and levothyroxine dose reassessment in the setting of significant weight loss [19]. In practice, current guidelines recommend monitoring TSH levels approximately 6 to 8 weeks after initiating or adjusting levothyroxine therapy. Once the appropriate dose is achieved, TSH should be reassessed after 4 to 6 months and then monitored every 12 months [75]. Future research should evaluate whether a close-monitoring strategy is beneficial when GLP-1RA is initiated among those who have been receiving levothyroxine. Such a strategy could include thyroid function assessment at GLP-1RA initiation, and every 3 months during active weight loss period (or more frequent, e.g., every month when 5% weight reduction is achieved), and timely levothyroxine dose titration to optimize thyroid hormone replacement. These considerations are hypothesis-generating and highlight the need for prospective studies that integrate longitudinal weight, and thyroid hormone measurements to determine the optimal monitoring approach.
In our study, the incidence rates of AF/Aflutter (0.74 vs. 0.71/100 person-years) were relatively low, compared with 1.97/100 person-years reported among general age-comparable patients receiving primary care at Massachusetts General Hospitals [76]. Nonetheless, our observed association between thyroid hormone over-replacement and new-onset AF/Aflutter aligns with a nested case–control study of 183,360 patients: compared to low levothyroxine exposure, high and medium exposure was associated with 29% (95% CI 1.23–1.35) and 8% (95% CI 1.04–1.11) increased odds of AF, respectively [36]. Besides, the crude AF/Aflutter counts were small (13 events in each arm), raising concerns about instability of estimates, and confidence intervals appear narrow. We used inverse-probability-of-censoring weights to address differential follow-up, and stabilized censoring weights were well behaved (few extremes), which helps explain why the confidence intervals are not excessively wide. Nevertheless, some overfitting remains possible, and we should interpret the results cautiously. In addition, despite rigorous propensity score matching, residual confounding remains possible. Our negative control outcome analyses (car accidents) showed no association between GLP-1RA initiation and new-onset AF/Aflutter, which is reassuring, though unmeasured confounders like variation on weight reduction and TSH may remain. These results underscore the need for future studies with larger sample sizes, and measurements of potential confounders, especially weight and biomarkers (i.e., HbA1c) to confirm our findings.
Our study has multiple strengths that support the robustness of our findings. First, following the principles of trial emulation, we outlined the protocol of the target trial and detailed emulation steps using Medicare data. This enhances the transparency of the study design and supports its analytic rigor [38]. Second, we applied PS matching rigorously and reported findings with strong transparency. For instance, we exactly matched the length between the base and study cohort entry date to avoid time-related biases. Third, we adopted a new-user design to minimize bias from prior drug exposure and to better emulate an RCT. Fourth, we used IPCW to address informative censoring and the different follow-up time between the two study groups. Next, we estimated the per-protocol effect by accounting for treatment changes and discontinuation, reflecting real-world drug exposure. Last, we conducted a series of sensitivity analyses to evaluate the robustness of our findings.
This study also has several limitations. First, Medicare claims database lacks weight data, limiting evaluation of its potential mediating role in the increased risk of AF/Aflutter or stroke. Additionally, the absence of laboratory values (e.g., HbA1c, TSH) limits adjustment for disease severity and assessment of over-replacement. Further research is warranted using datasets that include weight, TSH, and HbA1c measurements (such as EHR data) to formally disentangle the mechanisms linking GLP-1RA initiation, weight loss, optimal levothyroxine dosing, and cardiovascular outcomes. Second, using diagnosis and procedure codes to define covariates and outcomes may cause misclassification. However, these coding algorithms have been validated and commonly used in research [77, 78], and any misclassification is likely to be non-differential, which would bias results toward the null rather than create spurious associations. Third, medication exposure was based on prescription fills, not direct consumption, which may overestimate the exposure to GLP-1RA and bias estimates, though this is a well-established pharmacoepidemiologic method. The calculation of daily dose relied on the recorded days of supply; thus, dose reduction strategies (e.g., taking 100 mcg from Monday to Saturday but skipping Sunday) may not have been accurately captured. Last, restricting the cohort to patients with good adherence to levothyroxine (MPR > 80%) limits the generalizability of our findings to individuals with lower adherence. In addition, this study used Medicare claims data, which provide large-scale, nationally representative coverage of older adults but exclude individuals under 65 years. This may limit the generalizability to younger populations, who have a lower baseline risk of AF/Aflutter and stroke [79, 80].
Conclusion
Our study suggests that among older adults with type 2 diabetes treated with levothyroxine, initiation and continued use of GLP-1RA was associated with an increased risk of atrial fibrillation or flutter compared to SGLT2 inhibitors, without an elevation on stroke risk. These findings highlight the need for prospective studies that integrate longitudinal assessments of weight, thyroid function, and levothyroxine dosing to validate these results and clarify the mechanisms linking GLP-1RA use to cardiovascular outcomes, including atrial fibrillation/flutter and stroke.
Supplementary Information
Below is the link to the electronic supplementary material.
Abbreviations
- GLP-1RA
Glucagon-Like Peptide-1 Receptor Agonists
- SGLT2
Sodium–glucose cotransporter 2
- DPP4
Dipeptidyl Peptidase-4
- T2D
Type 2 diabetes
- AF/Aflutter
Atrial fibrillation or flutter
- TSH
Thyroid-stimulating hormone
- MPR
Medication possession ratio
- PS
Propensity score
- IPCW
Inverse probability censoring weight
- SMD
Standardized mean differences
Author contributions
TJ and NMSO conceived and designed the study. FD, QL and YC conducted data analysis and created tables and figures. FD, TJ and NMSO interpreted the data and drafted the manuscript. JPB, SS, RGM, BC and HS reviewed and substantively revised the manuscript.
Funding
No funding.
Availability of data and materials
The data that support the findings of this study are available from Centers for Medicare & Medicaid Services (CMS), but restrictions apply to the availability of these data, which were used under license for the current study, and so are not publicly available. Data are, however, available from the authors upon reasonable request and with permission of CMS.
Declarations
Ethics approval and consent to participate
The study protocol received approval from the Institutional Review Board at the University of Florida, USA.
Consent for publication
Not applicable.
Competing interests
The authors declare no competing interests.
Footnotes
Publisher's Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
References
- 1.McCoy RG, Herrin J, Swarna KS, Deng Y, Kent DM, Ross JS, et al. Effectiveness of glucose-lowering medications on cardiovascular outcomes in patients with type 2 diabetes at moderate cardiovascular risk. Nat Cardiovasc Res. 2024;3(4):431–40. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 2.FDA Approves New Drug Treatment for Chronic Weight Management [https://www.fda.gov/news-events/press-announcements/fda-approves-new-drug-treatment-chronic-weight-management-first-2014]
- 3.ElSayed NA, Aleppo G, Aroda VR, Bannuru RR, Brown FM, Bruemmer D, et al. Pharmacologic Approaches to Glycemic Treatment: Standards of Care in Diabetes. Diabetes Care. 2023;46(1):S140–57. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 4.Bethel MA, Patel RA, Merrill P, Lokhnygina Y, Buse JB, Mentz RJ, et al. Cardiovascular outcomes with glucagon-like peptide-1 receptor agonists in patients with type 2 diabetes: a meta-analysis. Lancet Diabetes Endocrinol. 2018;6(2):105–13. [DOI] [PubMed] [Google Scholar]
- 5.Sattar N, Lee MMY, Kristensen SL, Branch KRH, Del Prato S, Khurmi NS, et al. Cardiovascular, mortality, and kidney outcomes with GLP-1 receptor agonists in patients with type 2 diabetes: a systematic review and meta-analysis of randomised trials. Lancet Diabetes Endocrinol. 2021;9(10):653–62. [DOI] [PubMed] [Google Scholar]
- 6.Wilding JPH, Batterham RL, Calanna S, Davies M, Van Gaal LF, Lingvay I, et al. Once-weekly semaglutide in adults with overweight or obesity. N Engl J Med. 2021;384(11):989–1002. [DOI] [PubMed] [Google Scholar]
- 7.Newsome PN, Buchholtz K, Cusi K, Linder M, Okanoue T, Ratziu V, et al. A placebo-controlled trial of subcutaneous semaglutide in nonalcoholic steatohepatitis. N Engl J Med. 2021;384(12):1113–24. [DOI] [PubMed] [Google Scholar]
- 8.Kosiborod MN, Abildstrøm SZ, Borlaug BA, Butler J, Rasmussen S, Davies M, et al. Semaglutide in patients with heart failure with preserved ejection fraction and obesity. N Engl J Med. 2023;389(12):1069–84. [DOI] [PubMed] [Google Scholar]
- 9.Kosiborod MN, Petrie MC, Borlaug BA, Butler J, Davies MJ, Hovingh GK, et al. Semaglutide in patients with obesity-related heart failure and type 2 diabetes. N Engl J Med. 2024;390(15):1394–407. [DOI] [PubMed] [Google Scholar]
- 10.Tichy EM, Hoffman JM, Suda KJ, Rim MH, Tadrous M, Cuellar S, et al. National trends in prescription drug expenditures and projections for 2022. Am J Health Syst Pharm. 2022;79(14):1158–72. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 11.Estimates of Diabetes and Its Burden in the United States. National Diabetes Statistics Report website [https://www.cdc.gov/diabetes/data/statistics-report/index.html]
- 12.Stierman B, Afful J, Carroll MD, Chen TC, Davy O, Fink S, Fryar CD, Gu Q, Hales CM, Hughes JP et al: National Health and Nutrition Examination Survey 2017-March 2020 Prepandemic Data Files-Development of Files and Prevalence Estimates for Selected Health Outcomes. Natl Health Stat Report 2021(158). [DOI] [PMC free article] [PubMed]
- 13.Adhikari R, Jha K, Dardari Z, Heyward J, Blumenthal RS, Eckel RH, et al. National trends in use of Sodium-glucose Cotransporter-2 inhibitors and Glucagon-like Peptide-1 receptor agonists by cardiologists and other specialties, 2015 to 2020. J Am Heart Assoc. 2022;11(9):e023811. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 14.Jali MV, Kambar S, Jali SM, Pawar N, Nalawade P. Prevalence of thyroid dysfunction among type 2 diabetes mellitus patients. Diabetes Metab Syndr. 2017;11(Suppl 1):S105-s108. [DOI] [PubMed] [Google Scholar]
- 15.Díez JJ, Iglesias P. Prevalence of diabetes in people with thyroid dysfunction. Med Clin (Barc). 2023;160(8):333–40. [DOI] [PubMed] [Google Scholar]
- 16.Hadgu R, Worede A, Ambachew S. Prevalence of thyroid dysfunction and associated factors among adult type 2 diabetes mellitus patients, 2000–2022: a systematic review and meta-analysis. Syst Rev. 2024;13(1):119. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 17.Wyne KL, Nair L, Schneiderman CP, Pinsky B, Antunez Flores O, Guo D, et al. Hypothyroidism prevalence in the United States: a retrospective study combining national health and nutrition examination survey and claims data, 2009–2019. J Endocr Soc. 2022;7(1):bvac172. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 18.Pearce SH, Brabant G, Duntas LH, Monzani F, Peeters RP, Razvi S, et al. 2013 ETA guideline: management of subclinical hypothyroidism. Eur Thyroid J. 2013;2(4):215–28. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 19.Jonklaas J, Bianco AC, Bauer AJ, Burman KD, Cappola AR, Celi FS, et al. Guidelines for the treatment of hypothyroidism: prepared by the American Thyroid Association task force on thyroid hormone replacement. Thyroid. 2014;24(12):1670–751. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 20.Ettleson MD, Papaleontiou M. Evaluating health outcomes in the treatment of hypothyroidism. Front Endocrinol (Lausanne). 2022;13:1026262. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 21.Chaker L, Razvi S, Bensenor IM, Azizi F, Pearce EN, Peeters RP. Hypothyroidism. Nat Rev Dis Primers. 2022;8(1):30. [DOI] [PubMed] [Google Scholar]
- 22.Duntas LH, Jonklaas J. Levothyroxine dose adjustment to optimise therapy throughout a patient’s lifetime. Adv Ther. 2019;36(Suppl 2):30–46. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 23.Tee SA, Tsatlidis V, Razvi S. The GLP-1 receptor agonist exenatide reduces serum TSH by its effect on body weight in people with type 2 diabetes. Clin Endocrinol (Oxf). 2023;99(4):401–8. [DOI] [PubMed] [Google Scholar]
- 24.Sencar ME, Sakiz D, Calapkulu M, Hepsen S, Kizilgul M, Ozturk IU, et al. The effect of Exenatide on thyroid-stimulating hormone and thyroid volume. Eur Thyroid J. 2019;8(6):307–11. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 25.Azran C, Hanhan-Shamshoum N, Irshied T, Ben-Shushan T, Dicker D, Dahan A, et al. Hypothyroidism and levothyroxine therapy following bariatric surgery: a systematic review, meta-analysis, network meta-analysis, and meta-regression. Surg Obes Relat Dis. 2021;17(6):1206–17. [DOI] [PubMed] [Google Scholar]
- 26.Azran C, Porat D, Fine-Shamir N, Hanhan N, Dahan A. Oral levothyroxine therapy postbariatric surgery: biopharmaceutical aspects and clinical effects. Surg Obes Relat Dis. 2019;15(2):333–41. [DOI] [PubMed] [Google Scholar]
- 27.Tian Z, Nie Y, Li Z, Wang P, Zhang N, Hei X, et al. Total weight loss induces the alteration in thyroid function after bariatric surgery. Front Endocrinol (Lausanne). 2024;15:1333033. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 28.Hauge C, Breitschaft A, Hartoft-Nielsen ML, Jensen S, Bækdal TA. Effect of oral semaglutide on the pharmacokinetics of thyroxine after dosing of levothyroxine and the influence of co-administered tablets on the pharmacokinetics of oral semaglutide in healthy subjects: an open-label, one-sequence crossover, single-center, multiple-dose, two-part trial. Expert Opin Drug Metab Toxicol. 2021;17(9):1139–48. [DOI] [PubMed] [Google Scholar]
- 29.Little D DJ, Bartelt K, Longo J. Patients Who Lose Weight After Starting GLP-1s More Likely to See Thyroid Level Improvement. Epic Research. [https://epicresearch.org/articles/patients-who-lose-weight-after-starting-glp-1s-more-likely-to-see-thyroid-level-improvement]
- 30.Alijla F, Buttia C, Reichlin T, Razvi S, Minder B, Wilhelm M, et al. Association of diabetes with atrial fibrillation types: a systematic review and meta-analysis. Cardiovasc Diabetol. 2021;20(1):230. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 31.Sha R, Baines O, Hayes A, Tompkins K, Kalla M, Holmes AP, et al. Impact of obesity on atrial fibrillation pathogenesis and treatment options. J Am Heart Assoc. 2024;13(1):e032277. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 32.Homan EA, Reyes MV, Hickey KT, Morrow JP. Clinical overview of obesity and diabetes mellitus as risk factors for atrial fibrillation and sudden cardiac death. Front Physiol. 2018;9:1847. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 33.Xu Y, Boyle TA, Lyu B, Ballew SH, Selvin E, Chang AR, et al. Glucagon-like peptide-1 receptor agonists and the risk of atrial fibrillation in adults with diabetes: a real-world study. J Gen Intern Med. 2024;39(7):1112–21. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 34.Adamou A, Barkas F, Milionis H, Ntaios G. Glucagon-like peptide-1 receptor agonists and stroke: a systematic review and meta-analysis of cardiovascular outcome trials. Int J Stroke. 2024;19(8):876–87. [DOI] [PubMed] [Google Scholar]
- 35.Tribulova N, Knezl V, Shainberg A, Seki S, Soukup T. Thyroid hormones and cardiac arrhythmias. Vascul Pharmacol. 2010;52(3–4):102–12. [DOI] [PubMed] [Google Scholar]
- 36.Gong IY, Atzema CL, Lega IC, Austin PC, Na Y, Rochon PA, et al. Levothyroxine dose and risk of atrial fibrillation: a nested case-control study. Am Heart J. 2021;232:47–56. [DOI] [PubMed] [Google Scholar]
- 37.Wilcox L, Van Dril E. Suppressed thyroid stimulating hormone levels after initiation of a subcutaneous glucagon-like peptide-1 receptor agonist in a post-thyroidectomy patient managed with levothyroxine case report. J Am Pharm Assoc (2003). 2024;64(6):102185. [DOI] [PubMed] [Google Scholar]
- 38.Hernán MA, Robins JM. Using big data to emulate a target trial when a randomized trial is not available. Am J Epidemiol. 2016;183(8):758–64. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 39.Hernán MA, Alonso A, Logan R, Grodstein F, Michels KB, Willett WC, et al. Observational studies analyzed like randomized experiments: an application to postmenopausal hormone therapy and coronary heart disease. Epidemiology. 2008;19(6):766–79. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 40.Hernán MA, Wang W, Leaf DE. Target trial emulation: a framework for causal inference from observational data. JAMA. 2022;328(24):2446–7. [DOI] [PubMed] [Google Scholar]
- 41.Zhao JZ, Ruzieh M, Du F, Lian Y, Foy AJ, Platt RW, et al. Association between use of WATCHMAN device and 1-year mortality using high-dimensional propensity scores to reduce confounding. Circ Cardiovasc Qual Outcomes. 2025; Apr;18(4):e011188. [DOI] [PubMed]
- 42.Brown E, Heerspink HJL, Cuthbertson DJ, Wilding JPH. SGLT2 inhibitors and GLP-1 receptor agonists: established and emerging indications. Lancet. 2021;398(10296):262–76. [DOI] [PubMed] [Google Scholar]
- 43.Zhang HD, Ding L, Mi LJ, Zhang AK, Zhang K, Jiang ZH, et al. Sodium-glucose co-transporter-2 inhibitors for the prevention of atrial fibrillation: a systemic review and meta-analysis. Eur J Prev Cardiol. 2024;31(7):770–9. [DOI] [PubMed] [Google Scholar]
- 44.Frost L, Vestergaard P, Mosekilde L. Hyperthyroidism and risk of atrial fibrillation or flutter: a population-based study. Arch Intern Med. 2004;164(15):1675–8. [DOI] [PubMed] [Google Scholar]
- 45.Lanham TN, Morgan FH. Management of atrial flutter in thyroid storm. J Endocr Soc. 2021;5(Suppl 1):A941. [Google Scholar]
- 46.Venkatesh AK, Mei H, Kocher KE, Granovsky M, Obermeyer Z, Spatz ES, et al. Identification of emergency department visits in Medicare administrative claims: approaches and implications. Acad Emerg Med. 2017;24(4):422–31. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 47.Bode W, Ptaszek LM. Management of atrial fibrillation in the emergency department. Curr Cardiol Rep. 2021;23(12):179. [DOI] [PubMed] [Google Scholar]
- 48.Lee SE, Kim HJ, Ro YS. Epidemiology of stroke in emergency departments: a report from the National Emergency Department Information System (NEDIS) of Korea, 2018–2022. Clin Exp Emerg Med. 2023;10(S):S48-s54. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 49.Jiao T, Platt RW, Douros A, Filion KB. Use of a statistical adaptive treatment strategy approach for emulating randomized controlled trials using observational data: the example of blood-pressure control strategies for the prevention of cardiovascular events among individuals with hypertension at high cardiovascular risk. Am J Epidemiol. 2023;192(9):1576–91. [DOI] [PubMed] [Google Scholar]
- 50.Dischinger PC, Ho SM, Kufera JA. Medical conditions and car crashes. Annu Proc Assoc Adv Automot Med. 2000;44:335–46. [PMC free article] [PubMed] [Google Scholar]
- 51.Zhu S, Layde PM, Guse CE, Laud PW, Pintar F, Nirula R, et al. Obesity and risk for death due to motor vehicle crashes. Am J Public Health. 2006;96(4):734–9. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 52.Scirica BM, Bhatt DL, Braunwald E, Steg PG, Davidson J, Hirshberg B, et al. Saxagliptin and cardiovascular outcomes in patients with type 2 diabetes mellitus. N Engl J Med. 2013;369(14):1317–26. [DOI] [PubMed] [Google Scholar]
- 53.White WB, Cannon CP, Heller SR, Nissen SE, Bergenstal RM, Bakris GL, et al. Alogliptin after acute coronary syndrome in patients with type 2 diabetes. N Engl J Med. 2013;369(14):1327–35. [DOI] [PubMed] [Google Scholar]
- 54.Green JB, Bethel MA, Armstrong PW, Buse JB, Engel SS, Garg J, et al. Effect of sitagliptin on cardiovascular outcomes in type 2 diabetes. N Engl J Med. 2015;373(3):232–42. [DOI] [PubMed] [Google Scholar]
- 55.Karakus KE, Shah VN, Akturk HK. Tirzepatide-induced rapid weight loss-related thyrotoxicosis. JAMA Intern Med. 2024;184(10):1246–7. [DOI] [PubMed] [Google Scholar]
- 56.Pharmacologic Approaches to Glycemic Treatment. Standards of Care in Diabetes-2025. Diabetes Care. 2025;48(1 Suppl 1):S181-s206. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 57.Marso SP, Daniels GH, Brown-Frandsen K, Kristensen P, Mann JF, Nauck MA, et al. Liraglutide and cardiovascular outcomes in type 2 diabetes. N Engl J Med. 2016;375(4):311–22. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 58.Gerstein HC, Colhoun HM, Dagenais GR, Diaz R, Lakshmanan M, Pais P, et al. Dulaglutide and cardiovascular outcomes in type 2 diabetes (REWIND): a double-blind, randomised placebo-controlled trial. Lancet. 2019;394(10193):121–30. [DOI] [PubMed] [Google Scholar]
- 59.Cesaro A, Pastori D, Acerbo V, Biccirè FG, Golino M, Panico D, Prati F, Abbate A, Lip GYH, Calabrò P: Reduction of New Onset of Atrial Fibrillation in Patients Treated with Semaglutide: An updated systematic review and meta regression analysis of randomized controlled trials. Eur J Prev Cardiol 2025; Apr 28:zwaf257. [DOI] [PubMed]
- 60.Rivera FB, Cruz LLA, Magalong JV, Ruyeras J, Aparece JP, Bantayan NRB, et al. Cardiovascular and renal outcomes of glucagon-like peptide 1 receptor agonists among patients with and without type 2 diabetes mellitus: a meta-analysis of randomized placebo-controlled trials. Am J Prev Cardiol. 2024;18:100679. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 61.Badve SV, Bilal A, Lee MMY, Sattar N, Gerstein HC, Ruff CT, et al. Effects of GLP-1 receptor agonists on kidney and cardiovascular disease outcomes: a meta-analysis of randomised controlled trials. Lancet Diabetes Endocrinol. 2025;13(1):15–28. [DOI] [PubMed] [Google Scholar]
- 62.Zinman B, Wanner C, Lachin JM, Fitchett D, Bluhmki E, Hantel S, et al. Empagliflozin, Cardiovascular Outcomes, and Mortality in Type 2 Diabetes. N Engl J Med. 2015;373(22):2117–28. [DOI] [PubMed] [Google Scholar]
- 63.Neal B, Perkovic V, Mahaffey KW, de Zeeuw D, Fulcher G, Erondu N, et al. Canagliflozin and cardiovascular and renal events in type 2 diabetes. N Engl J Med. 2017;377(7):644–57. [DOI] [PubMed] [Google Scholar]
- 64.Zelniker TA, Bonaca MP, Furtado RHM, Mosenzon O, Kuder JF, Murphy SA, et al. Effect of dapagliflozin on atrial fibrillation in patients with type 2 diabetes mellitus: insights from the DECLARE-TIMI 58 trial. Circulation. 2020;141(15):1227–34. [DOI] [PubMed] [Google Scholar]
- 65.Jaiswal V, Ang SP, Kumar D, Deb N, Jaiswal A, Joshi A, et al. Sodium-glucose cotransporter-2 inhibitors and arrhythmias: a meta-analysis of 38 randomized controlled trials. JACC Adv. 2025;4(3):101615. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 66.Mousavi A, Shojaei S, Soleimani H, Semirani-Nezhad D, Ebrahimi P, Zafari A, et al. Safety, efficacy, and cardiovascular benefits of combination therapy with SGLT-2 inhibitors and GLP-1 receptor agonists in patients with diabetes mellitus: a systematic review and meta-analysis of randomized controlled trials. Diabetol Metab Syndr. 2025;17(1):68. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 67.Guan B, Chen Y, Yang J, Yang W, Wang C. Effect of bariatric surgery on thyroid function in obese patients: a systematic review and meta-analysis. Obes Surg. 2017;27(12):3292–305. [DOI] [PubMed] [Google Scholar]
- 68.Klein I, Danzi S. Thyroid disease and the heart. Circulation. 2007;116(15):1725–35. [DOI] [PubMed] [Google Scholar]
- 69.Park B, Bakbak E, Teoh H, Krishnaraj A, Dennis F, Quan A, et al. GLP-1 receptor agonists and atherosclerosis protection: the vascular endothelium takes center stage. Am J Physiol Heart Circ Physiol. 2024;326(5):H1159-h1176. [DOI] [PubMed] [Google Scholar]
- 70.Uccellatore A, Genovese S, Dicembrini I, Mannucci E, Ceriello A. Comparison review of short-acting and long-acting Glucagon-like Peptide-1 receptor agonists. Diabetes Ther. 2015;6(3):239–56. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 71.Yuan X, Gao Z, Hao Z, Ma H, Duan K, Yang C. Effect of long-acting versus short-acting glucagon-like peptide-1 receptor agonists on improving body weight and related metabolic parameters in type 2 diabetes: a head-to-head meta-analysis. Medicine (Baltimore). 2023;102(43):e35739. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 72.Madsbad S. Review of head-to-head comparisons of glucagon-like peptide-1 receptor agonists. Diabetes Obes Metab. 2016;18(4):317–32. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 73.Pinto M, Brennan L, Diehl K, Lin S, Heacock S: Real-World Comparison of Oral Versus Injectable Semaglutide for the Reduction of Hemoglobin A(1C) and Weight in Patients with Type 2 Diabetes. J Pharm Technol 2024:87551225241289959. [DOI] [PMC free article] [PubMed]
- 74.Rayner CK, Watson LE, Phillips LK, Lange K, Bound MJ, Grivell J, et al. Effects of sustained treatment with lixisenatide on gastric emptying and postprandial glucose metabolism in type 2 diabetes: a randomized controlled trial. Diabetes Care. 2020;43(8):1813–21. [DOI] [PubMed] [Google Scholar]
- 75.Bible KC, Kebebew E, Brierley J, Brito JP, Cabanillas ME, Clark TJ Jr, et al. 2021 American Thyroid Association Guidelines for Management of Patients with Anaplastic Thyroid Cancer. Thyroid. 2021;31(3):337–86. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 76.Khurshid S, Ashburner JM, Ellinor PT, McManus DD, Atlas SJ, Singer DE, et al. Prevalence and incidence of atrial fibrillation among older primary care patients. JAMA Netw Open. 2023;6(2):e2255838. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 77.Columbo JA, Daya N, Colantonio LD, Wang Z, Foti K, Hyacinth HI, et al. Derivation and validation of ICD-10 codes for identifying incident stroke. JAMA Neurol. 2024;81(8):875–81. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 78.Jensen PN, Johnson K, Floyd J, Heckbert SR, Carnahan R, Dublin S. A systematic review of validated methods for identifying atrial fibrillation using administrative data. Pharmacoepidemiol Drug Saf. 2012;21(Suppl 1(0 1)):141–7. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 79.Ohlmeier C, Mikolajczyk R, Haverkamp W, Garbe E. Incidence, prevalence, and antithrombotic management of atrial fibrillation in elderly Germans. Europace. 2013;15(10):1436–44. [DOI] [PubMed] [Google Scholar]
- 80.Kelly-Hayes M. Influence of age and health behaviors on stroke risk: lessons from longitudinal studies. J Am Geriatr Soc. 2010;58(2):S325-328. [DOI] [PMC free article] [PubMed] [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.
Supplementary Materials
Data Availability Statement
The data that support the findings of this study are available from Centers for Medicare & Medicaid Services (CMS), but restrictions apply to the availability of these data, which were used under license for the current study, and so are not publicly available. Data are, however, available from the authors upon reasonable request and with permission of CMS.