The research team, led by Chia-Ming Lin, Pei-Yun Li, Yu-Nan Huang, and Pen-Hua Su of Chung Shan Medical University Hospital in Taiwan, together with collaborators including Gideon Meyerowitz-Katz at the University of Wollongong in Australia, used the TriNetX US Collaborative Network, a massive repository of de-identified electronic health records spanning 2017 to 2025. Rather than running a prospective trial, which would take years and cost millions, they employed a technique known as target trial emulation. This approach borrows the design logic of a randomized controlled trial — defining eligibility criteria, treatment strategies, and follow-up windows as if a trial were being conducted — and then applies that blueprint to observational data, using statistical tools to approximate the balance that randomization would have achieved.

Specifically, the investigators emulated three separate active-comparator target trials in adults with type 2 diabetes undergoing thyroidectomy. Each trial compared preoperative continuation of a GLP-1 receptor agonist against a different alternative: dipeptidyl peptidase-4 inhibitors, sodium-glucose cotransporter-2 inhibitors, or usual care. To reduce the chance that sicker patients were systematically channeled into one drug class, the cohorts were 1:1 propensity score matched, a method that pairs patients across treatment groups based on their baseline characteristics. Cox proportional hazards models then estimated hazard ratios for each outcome, the standard statistical measure of how quickly events accumulate in one group relative to another over time.

The safety question at the heart of the study concerned aspiration, the feared complication in which stomach contents enter the airway during anesthesia. Because GLP-1 receptor agonists delay gastric emptying, anesthesiology societies have debated whether patients should hold these medications before surgery. In this analysis, however, aspiration-specific events fell below the minimum reportable cell size in every comparison and every postoperative window — meaning so few events occurred that reporting a hazard ratio would risk identifying individual patients. A broader postoperative respiratory composite outcome, which was prespecified precisely to capture this signal indirectly, showed no excess in the GLP-1 receptor agonist groups. For a surgical community hungry for reassurance, that absence of harm is itself a meaningful finding.

The thyroidectomy-specific complications told a similarly reassuring story. Recurrent laryngeal nerve injury, one of the most dreaded complications of thyroid surgery because it can permanently alter the voice, showed no excess in patients who continued GLP-1 receptor agonists. Hypocalcemia, a drop in blood calcium that follows disturbance of the parathyroid glands during thyroidectomy, was more frequent with GLP-1 receptor agonists than with SGLT2 inhibitors in the primary analysis, with a hazard ratio of 1.43 and a 95 percent confidence interval of 0.96 to 2.12 — a signal that grazed but did not cross the threshold of statistical significance. Notably, that signal did reach nominal significance in sensitivity analyses restricted to a two-year exposure window and to new users of the drugs, a pattern the authors treated with caution rather than as proof of a causal link.

The metabolic findings were more surprising. Patients who continued GLP-1 receptor agonists before surgery had a lower risk of diabetic ketoacidosis over 24 months compared with those on DPP-4 inhibitors, with a hazard ratio of 0.56 and a 95 percent confidence interval of 0.34 to 0.91. Diabetic ketoacidosis is a life-threatening emergency in which the body, starved of effective insulin action, produces dangerous levels of blood acids, and it has been a particular concern with SGLT2 inhibitors. The lower DKA risk was reproduced in both the exposure-window and new-user designs, lending it more internal consistency than many observational findings enjoy.

Mortality results generated the study’s most eye-catching numbers. GLP-1 receptor agonist continuation was associated with lower 24-month all-cause mortality compared with DPP-4 inhibitors, at a hazard ratio of 0.49 (95 percent confidence interval 0.26 to 0.92), and compared with SGLT2 inhibitors, at a hazard ratio of 0.39 (95 percent confidence interval 0.19 to 0.78). In plain terms, patients on GLP-1 drugs appeared roughly half as likely to die within two years as matched peers on the comparator drugs. Yet the authors themselves immediately tempered the excitement: the mortality findings were not reproduced in the exposure-window and new-user designs, mortality over the immediate postoperative periods of days 1 to 30 and 1 to 90 fell below the minimum reportable cell size in every comparison, and none of the significant results survived correction for multiple comparisons.

That last point deserves emphasis, because it captures the epistemological honesty that distinguishes careful pharmacoepidemiology from headline-chasing. When researchers test many outcomes across many comparisons, some associations will appear statistically significant by chance alone — the multiple-comparisons problem. Corrections such as false discovery rate control adjust for this inflation, and in this study every nominally significant finding dissolved under that scrutiny. The team also deployed negative-control outcomes, outcomes with no plausible biological connection to the drugs under study, and these were null, probing for residual confounding and finding no obvious trace of it. Still, the authors concluded that the lower DKA and mortality risks should be regarded as hypothesis-generating rather than as evidence for clinical action.

The practical takeaway for clinicians is nonetheless concrete. In adults with type 2 diabetes undergoing thyroidectomy, preoperative GLP-1 receptor agonist continuation was not associated with excess perioperative harm — no excess aspiration-related signal, no excess respiratory composite events, no excess recurrent laryngeal nerve injury. For surgeons and anesthesiologists who have faced the dilemma of whether to interrupt a patient’s diabetes therapy before an operation, the study offers the most procedure-specific evidence to date that continuation is compatible with perioperative safety. The findings align with a broader shift in perioperative medicine, where blanket drug-holding rules are increasingly replaced by procedure- and patient-specific risk assessments grounded in real-world data.

The study’s design also matters for how its results should be used. Because it emulated target trials in retrospectively collected, de-identified records, it was not a registered prospective clinical trial, and its matched cohorts — 722, 736, and 611 patients per arm across the three comparisons — were large but finite. Rare events like aspiration can escape detection even in cohorts of this size, and residual confounding can never be fully excluded in non-randomized data. The work was funded by Taiwan’s National Science and Technology Council and institutional grants from Chung Shan Medical University Hospital, with the funding sources having no role in the analysis or manuscript. The authors declare no competing interests. What the study delivers is a rigorously constructed, transparently reported foundation — one that flags no perioperative danger from continuing GLP-1 receptor agonists before thyroidectomy, and that sketches intriguing hypotheses about metabolic and survival benefits that only properly randomized trials can confirm.

Subject of Research: Perioperative safety of preoperative GLP-1 receptor agonist continuation in adults with type 2 diabetes undergoing thyroidectomy

Article Title: Preoperative GLP-1 receptor agonist continuation and perioperative safety after thyroidectomy in type 2 diabetes: a target trial emulation using a retrospective multicenter cohort

Article References: Lin, C.-M., Li, P.-Y., Chen, J.-C., Li, P.-H., Yang, H.-W., Meyerowitz-Katz, G., Huang, Y.-N., & Su, P.-H. (2026). Preoperative GLP-1 receptor agonist continuation and perioperative safety after thyroidectomy in type 2 diabetes: a target trial emulation using a retrospective multicenter cohort. BMC Endocrine Disorders. https://doi.org/10.1186/s12902-026-02537-y

Image Credits: AI Generated

DOI: 10.1186/s12902-026-02537-y

Keywords: GLP-1 receptor agonist, thyroidectomy, type 2 diabetes, perioperative safety, aspiration, diabetic ketoacidosis, hypocalcemia, recurrent laryngeal nerve injury, target trial emulation, propensity score matching, pharmacoepidemiology, TriNetX