Do glucagon-like peptide-1 receptor agonists influence cervical fusion outcomes? A systematic review and meta-analysis

Abstract

Background

Glucagon-like peptide-1 receptor agonists (GLP-1RAs) are increasingly used in patients with obesity and type 2 diabetes, but their procedure-specific effects in cervical fusion remain unclear.

Methods

A systematic review and meta-analysis following PRISMA 2020 and MOOSE, prospectively registered in PROSPERO, searched PubMed, Embase, and Web of Science through April 28, 2026. Comparative studies evaluating perioperative GLP-1RA use in adults undergoing cervical fusion were included. Two reviewers independently screened studies, extracted data, and assessed risk of bias. Random-effects meta-analyses were performed in Review Manager Web (RevMan Web). Certainty of evidence was assessed using the GRADE approach for clinically important outcomes.

Results

Six propensity score-matched retrospective cohort studies involving 17,310 patients were included. In the primary analyses, GLP-1RA use was not associated with significant differences in pseudarthrosis (odds ratio [OR] 0.98, 95% confidence interval [CI] 0.33–2.89; p = .96), dysphagia (OR 0.89, 95% CI 0.38–2.05; p = .68), hospital readmission (OR 0.67, 95% CI 0.39–1.17; p = .12), or device failure (OR 0.83, 95% CI 0.44–1.56; p = .41). Secondary outcomes were also not significantly different between groups.

Conclusions

Perioperative GLP-1RA use was not associated with a consistent pattern of worse postoperative outcomes after cervical fusion. Prospective, procedure-specific studies with standardized exposure and outcome definitions are needed.

Background

Glucagon-like peptide-1 receptor agonists (GLP-1RAs) have rapidly become major pharmacologic interventions for type 2 diabetes, obesity, and cardiometabolic risk reduction. In 2022, 2.5 billion adults were overweight and 890 million were living with obesity worldwide [1]. Large randomized trials have shown clinically meaningful body-weight reduction with semaglutide and tirzepatide, and semaglutide reduced major adverse cardiovascular events in patients with overweight or obesity and established cardiovascular disease [[2], [3], [4]]. As use of these agents expands among patients considered for spine surgery, their perioperative safety and potential association with spine-specific outcomes remain clinically important questions.

Beyond their cardiometabolic benefits, GLP-1RAs may influence postoperative recovery and fusion biology. Improvements in glycemic control and reductions in adiposity may attenuate metabolic and inflammatory conditions unfavorable to surgical recovery, whereas appetite suppression and weight loss may affect lean mass and nutritional adequacy; experimental and clinical data also suggest possible effects on osteoblastic activity and bone remodeling [[2], [3], [4], [5], [6], [7]]. These mechanisms provide a biologic rationale for evaluating both perioperative complications and fusion-related outcomes, although their net clinical effect after cervical fusion remains uncertain.

Cervical fusion is among the most commonly performed spine procedures for degenerative cervical pathology. In a national database analysis from the United States, Saifi et al. [8] identified 1,059,403 anterior cervical discectomy and fusion procedures performed between 2006 and 2013. Although cervical fusion is generally effective, postoperative morbidity remains clinically relevant. A comprehensive systematic review and meta-analysis of complications after anterior cervical discectomy and fusion included 222 studies and 50,584 patients and reported an overall postoperative complication rate of 16% [9]. These data highlight the importance of evaluating perioperative exposures that may be associated with surgical recovery and fusion-related outcomes in metabolically complex patients.

Current evidence regarding GLP-1RA use in spine surgery is heterogeneous and predominantly observational. Previous systematic reviews and meta-analyses have pooled anatomically and surgically diverse cohorts, including cervical, lumbar, and mixed spinal procedures, limiting procedure specific inference for cervical fusion [[10], [11], [12]]. Moreover, recent cervical fusion studies have reported conflicting findings, with lower pseudarthrosis risk after anterior cervical discectomy and fusion in some analyses and higher pseudarthrosis and dysphagia rates among semaglutide users after posterior cervical fusion in another [[13], [14], [15]]. Therefore, the objective of this systematic review and meta-analysis was to evaluate the association between perioperative GLP-1RA use and surgical and postoperative outcomes in adults undergoing cervical fusion.

Methods

This systematic review and meta-analysis was conducted in accordance with the Cochrane Handbook for Systematic Reviews of Interventions [16] and reported according to the PRISMA 2020 statement and the MOOSE reporting recommendations [17,18]. The protocol was prospectively registered in PROSPERO (CRD420261410956). Because only previously published aggregate data were analyzed, institutional review board approval was not required.

Search strategy

A comprehensive literature search was performed in PubMed (MEDLINE), Embase, and Web of Science from inception through April 28, 2026. No language, date, or study-design filters were applied. Search terms included free-text terms related to spine surgery, spinal fusion, arthrodesis, decompression and fusion, discectomy and fusion, instrumented fusion, glucagon-like peptide-1 receptor agonists (GLP-1RAs), and individual GLP-1-based agents, including semaglutide, liraglutide, dulaglutide, exenatide, lixisenatide, and tirzepatide. The search strategy was intentionally broad to maximize sensitivity across spine surgery populations. Reference lists of included studies and relevant reviews were manually screened to identify additional eligible records. Grey literature sources were also searched. The full electronic search strategy for all databases is provided in Supplementary Material S1.

Eligibility criteria

Studies were considered eligible if they met all of the following criteria: (1) randomized controlled trials or analytical observational studies, including cohort and case-control designs; (2) inclusion of adult patients undergoing cervical fusion, including anterior cervical discectomy and fusion, anterior cervical fusion, posterior cervical fusion, or mixed anterior/posterior cervical fusion cohorts with extractable cervical fusion data; (3) comparison between patients receiving preoperative or perioperative GLP-1RAs and patients not receiving GLP-1RAs; and (4) reporting of at least one prespecified postoperative outcome of interest, including pseudarthrosis/nonunion, dysphagia, postoperative respiratory complications, postoperative infection or surgical site infection, hospital readmission, venous thromboembolism, acute kidney injury, implant or hardware failure, or revision/reoperation.

Studies were excluded from the systematic review if they met any of the following criteria: (1) noncomparative studies, case reports, case series without a control group, reviews, editorials, letters, commentaries, or conference abstracts without a full-text publication; (2) animal or in vitro studies; (3) studies not involving cervical fusion, including studies limited to nonfusion cervical procedures, lumbar fusion, thoracic fusion, or general spine surgery without extractable cervical fusion-specific data; (4) studies evaluating weight-loss interventions other than GLP-1RAs without a separable GLP-1RA group; or (5) absence of any prespecified extractable outcome.

For the primary quantitative synthesis, potential population overlap was reassessed study by study and outcome by outcome. A plausible direct overlap concern required use of the same database or related data network, including differently labeled TriNetX networks, together with substantial concordance in surgical population, procedure and approach, available study period, GLP-1RA exposure definition, matching strategy, follow-up window, and outcome reporting. When such a concern was identified, one report was retained for the relevant pooled analysis, prioritizing procedure specificity, exposure clarity, completeness of outcome reporting, relevant follow-up, and/or matched sample size. Thus, no reports judged to have plausible direct overlap were intentionally included together in the same forest plot; nonselected reports were retained for narrative synthesis when relevant. Although individual-level linkage was unavailable, the included reports differed across these key design and cohort features, making clinically meaningful residual overlap unlikely.

Mixed-spine or mixed-procedure cohorts were included only when cervical fusion-specific data were separately extractable; otherwise, they were excluded from the quantitative synthesis.

Study selection and data extraction

All records retrieved from the databases were imported into Rayyan QCRI for management, duplicate identification, and screening [19]. Screening was conducted independently by 2 reviewers, who assessed titles and abstracts, followed by full-text review to determine study eligibility according to the predefined inclusion and exclusion criteria. Disagreements at any stage were resolved by consensus or, when necessary, by a third reviewer.

Data extraction was performed independently by the same 2 reviewers using a structured Microsoft Excel spreadsheet developed for this review. Each reviewer independently verified the extracted data to ensure accuracy and minimize transcription errors. Any discrepancies were resolved by discussion and, when necessary, adjudicated by a third reviewer.

The following variables were extracted: study design and year of publication; country or region; data source; study period; sample size in the GLP-1RA and control groups; baseline demographic and clinical characteristics, including smoking or nicotine use, body mass index, and diabetes or glycemic-control variables when reported; type of cervical fusion procedure, surgical approach, and number of fused levels when available; graft and instrumentation characteristics when reported; GLP-1RA agent or class; exposure definition and timing relative to surgery; reported treatment duration and perioperative discontinuation, when available; control group definition; matching or adjustment methods; prespecified postoperative outcomes; and follow-up duration. For each reported outcome, we additionally extracted the study-specific outcome definition, source and method of ascertainment, and postoperative assessment window.

Outcomes

The primary outcome was pseudarthrosis after cervical fusion.

Secondary outcomes included dysphagia, postoperative respiratory complications, postoperative infection or surgical site infection, hospital readmission, venous thromboembolism, acute kidney injury, and implant or hardware failure. Postoperative respiratory complications included pulmonary aspiration, aspiration pneumonia, or postoperative pneumonia, according to the definition reported by each study. Venous thromboembolism included deep vein thrombosis, pulmonary embolism, or a combined venous thromboembolism outcome when reported. Implant or hardware failure included implant failure, hardware failure, device failure, or material failure.

For pseudarthrosis and implant or hardware failure, the longest available follow-up was prioritized for the primary analysis. For perioperative complications, including dysphagia, postoperative respiratory complications, postoperative infection or surgical site infection, hospital readmission, venous thromboembolism, and acute kidney injury, the earliest available postoperative follow-up was prioritized, as these outcomes were considered short-term postoperative events. When quantitative pooling was not appropriate because of incompatible outcome reporting, findings were summarized narratively.

For all outcomes, the original study-specific definition and ascertainment method were retained. Pseudarthrosis or nonunion was eligible for synthesis when reported as a postoperative diagnosis after cervical fusion; no unreported clinical or radiographic confirmation was inferred. Revision or reoperation, when reported, was treated as a distinct postindex operative outcome and was not considered interchangeable with hospital readmission, emergency department utilization, or implant or device failure; no unreported indication for revision or surgical approach was inferred. Because revision or reoperation was reported using nonuniform procedural definitions and postoperative assessment windows, these findings were summarized narratively rather than pooled.

Definitions

In this review, “perioperative GLP-1RA use” was an umbrella term for study-specific medication exposure definitions in relation to surgery. It did not imply uniform treatment duration, confirmed day-of-surgery use, or uninterrupted therapy. Definitions included active preoperative use, minimum preoperative duration criteria, and prescription- or fill-based windows spanning preoperative and postoperative periods. Individual agents were considered within the GLP-1RA or GLP-1-based therapy class as reported by the source study.

Cervical fusion was defined as any anterior or posterior cervical arthrodesis procedure, including anterior cervical discectomy and fusion, anterior cervical fusion, posterior cervical fusion, or mixed anterior/posterior cervical fusion cohorts when cervical fusion-specific data were extractable.

Risk of bias assessment

Risk of bias was assessed using the risk of bias in nonrandomized studies of interventions tool (ROBINS-I) [20], because all included studies were observational comparative studies. The review team prespecified the main confounding domains considered most relevant to this clinical question, including age, sex, body mass index, obesity, diabetes status, smoking status, comorbidity burden, indication for surgery, anterior versus posterior surgical approach, type of cervical fusion procedure, number of fused levels when available, and baseline metabolic risk factors.

Studies were assessed across the 7 ROBINS-I domains: bias due to confounding, selection of participants, classification of interventions, deviations from intended interventions, missing data, measurement of outcomes, and selection of the reported result. Overall risk-of-bias judgments followed the recommended ROBINS-I algorithm.

Risk-of-bias assessments were performed independently by 2 reviewers. Disagreements were resolved by consensus and, when necessary, by adjudication from a third reviewer.

Statistical analysis and synthesis methods

Meta-analyses were performed using Review Manager Web (RevMan Web) [21]. All studies included in the quantitative synthesis used propensity score matching; therefore, outcome data were extracted from the matched cohorts rather than from the unmatched source populations. For dichotomous outcomes, odds ratios (ORs) with 95% confidence intervals (CIs) were used as the summary effect measure.

Whenever available, postmatching event counts and total sample sizes were extracted and used to calculate ORs. This approach was prioritized because adjusted ORs were not consistently reported across all studies and outcomes. When event counts were not clearly extractable, particularly because of suppressed cells reported as “<10” or “<11,” adjusted or precomputed ORs with their corresponding standard errors were used when available through the generic inverse-variance method.

Because clinical and methodological heterogeneity was expected across observational studies, random-effects models were used. Between-study variance was estimated using the restricted maximum likelihood (REML) method. Confidence intervals for pooled effects were calculated using the Hartung–Knapp–Sidik–Jonkman (HKSJ) method when tau² was greater than zero and more than 2 studies contributed to the analysis [22]. When tau² was equal to zero, or when only 2 studies contributed to the analysis, Wald-type confidence intervals were used.

Statistical heterogeneity was assessed using Cochran's Q test, the I² statistic, and tau² [23]. A p value <.10 for Cochran's Q test or I² > 25% was considered suggestive of relevant between-study heterogeneity. For the pooled effect estimate, a 2-sided p value <.05 was considered statistically significant.

For outcomes that could not be quantitatively pooled because of incomplete reporting, incompatible outcome definitions, or insufficient extractable data, findings were summarized narratively. Formal assessment of small-study effects or publication bias, including funnel plot inspection or regression-based testing, was not performed because fewer than 10 studies contributed to each quantitative synthesis.

When an included study reported more than 1 separately extractable and mutually exclusive cervical fusion cohort, each cohort was entered as a separate comparison in the corresponding meta-analysis, provided that this did not result in double counting of participants within the same pooled estimate.

Exploratory sensitivity analyses

Exploratory sensitivity analyses excluding Ng et al. [15] were conducted for pseudarthrosis and dysphagia to assess the influence of a clinically and methodologically distinct cohort on the primary pooled estimates. Ng et al. [15] evaluated active semaglutide use exclusively in patients undergoing posterior cervical fusion and used administrative coding for outcome ascertainment. These analyses were not intended to estimate a protective, procedurespecific, or medication-specific effect of GLP-1RA therapy.

Certainty of evidence

Certainty of evidence was assessed using the Grading of Recommendations Assessment, Development and Evaluation (GRADE) approach for 4 clinically important outcomes selected for the Summary of Findings table: pseudarthrosis, dysphagia, hospital readmission, and surgical site infection [24]. The assessment considered risk of bias, inconsistency, indirectness, imprecision, and publication bias, with the ROBINS-I assessment informing judgments regarding risk of bias. Anticipated absolute effects were calculated by applying pooled odds ratios to pooled control risks. Publication bias was not formally assessed because fewer than 10 studies contributed to each quantitative synthesis. Device failure was not included in the Summary of Findings table because event counts were suppressed in some source cohorts, precluding reliable estimation of absolute effects. Detailed certainty judgments and explanations are presented in Supplementary Table S1.

Results

Study selection

As shown in Fig. 1, a total of 160 records were identified through database searching, including 44 from PubMed, 77 from Embase, and 39 from Web of Science. After the removal of 61 duplicate records, 99 records remained for title and abstract screening. Of these, 77 records were excluded for not meeting the eligibility criteria. A total of 22 reports were sought for retrieval, and all were successfully obtained.

Fig. 1.

PRISMA flow diagram of study selection. The diagram summarizes record identification, screening, eligibility assessment, and final inclusion in the meta-analysis. PRISMA, preferred reporting items for systematic reviews and meta-analyses.

Overall, 22 full-text reports were assessed for eligibility. Of these, 16 reports were excluded for the following reasons: wrong population (n = 4), absence of a control group (n = 3), lack of relevant outcomes (n = 3), and overlapping populations (n = 6). Among reports excluded because of potential population overlap, cohort characteristics were reassessed study by study and outcome by outcome before quantitative synthesis. Reports judged to have plausible direct overlap were not copooled, and no report was intentionally counted more than once within the same pooled estimate. Ultimately, 6 studies were included in the meta-analysis.

Baseline characteristics

Table 1 summarizes 6 included studies [14,15,[25], [26], [27], [28]] published between 2025 and 2026. All were retrospective cohort studies conducted in the United States using large administrative databases or multicenter research networks, including MarketScan, PearlDiver, and TriNetX. Overall, the meta-analysis included 17,310 patients. In addition, all studies used propensity score matching, which strengthens baseline comparability between GLP-1 receptor agonist users and nonusers.

Table 1.

Characteristics of included studies.

Study	Data source (study period)	Population/inclusion	Surgical approach/fusion level	Sample size, N (GLP-1RA:control)	Age, y	Male:female	Maximum follow-up, y	GLP-1RA agent(s) reported	Exposure definition/window
Bank et al. [14]	TriNetX-US 2013–2023	Overweight/obesity or T2DM	Single-/multilevel ACDF	3,184 (1,592:1,592)	59 ± 10.5*	1,376:1,698 (110 unknown)	2	Dula/Sema/Lira║	Rx: 12 mo pre to 1 mo postop
Heo et al. [25]	MarketScan 2015–2022	T2DM and/or BMI > 30	1–2-level ACDF	3,196 (1,598:1,598)	54.75 (32–82)/54.53 (29–91)†,‡	1,528:1,668	1	NR	Chronic use; 6 mo pre/postop¶
Kim et al. [26]	TriNetX Global NR**	T2DM	Single-/multilevel ACDF	2,494 (1,247:1,247)	58.8 ± 10.4*	1,086:1,408	1	NR	Rx in 6 mo pre/postop
Ng et al. [15]	PearlDiver 2010–2022	PCF cohort (mostly obese)	PCF (levels NR)	1,880 (340:1,540)	60–64§	991:889	2	Sema	Active pre-op Sema; window NR
Rajkovic et al. [27]	TriNetX Research Network NR**	Degenerative pathology	PCF (levels NR)	1,392 (696:696)	63.5 ± 9.4/62.7 ± 11.0*,‡	629:763	1	Sema/Exe/Tirz/Dula/Lira	≥12 wk preop
Vatsia et al. [28]	TriNetX Diamond Network 2005–2024	T2DM	Multilevel ACF/PCF	ACF: 2,408 (1,204:1,204); PCF: 2,756 (1,378:1,378)	NR	NR	2	Dula/Lira/Lixi/Sema/Tirz#	Rx within 1 y of surgery

Population/inclusion criteria, surgical approach/fusion level, and study-specific GLP-1RA exposure information are summarized as reported by each study.

ACF, anterior cervical fusion; ACDF, anterior cervical discectomy and fusion; BMI, body mass index; Dula, dulaglutide; Exe, exenatide; GLP-1RA, glucagon-like peptide-1 receptor agonist; Lira, liraglutide; Lixi, lixisenatide; NR, not reported; PCF, posterior cervical fusion; pre-op, preoperative; postop, postoperative; Rx, prescription; Sema, semaglutide; T2DM, type 2 diabetes mellitus; Tirz, tirzepatide.

^⁎Values reported as mean ± SD.

^†Values reported as mean (range).

^‡Age is presented separately for the GLP-1RA and control groups.

^§Age reported as the most frequent categorical range.

^║Dula, Sema, and Lira were the 3 most frequently documented agents in Bank et al. [14] (34%, 31%, and 24%, respectively).

^¶Chronic use in Heo et al. [25] required ≥3 fills in both 6-month preoperative and postoperative periods or ≥1 fill with a ≥90-day supply in both periods.

^#Vatsia et al. [28] also included pramlintide in the source exposure definition; it is not listed because it is not a GLP-1 receptor agonist.

^⁎⁎For Kim et al. [26] and Rajkovic et al. [27], the patient inclusion period was not reported; only the database query date was reported in the source studies.

In general, the study populations were concentrated in patients with type 2 diabetes mellitus, overweight/obesity, or degenerative pathology. Sample sizes ranged from 1,392 to 5,164 patients, mean age remained consistently between the late 50s and early 60s, and maximum follow-up ranged from 1 to 2 years.

Regarding the surgical profile, anterior procedures predominated. Three studies evaluated anterior cervical discectomy and fusion, 1 study included separate anterior and posterior cohorts, and 2 studies focused exclusively on posterior cervical fusion. Therefore, although all studies examined cervical fusion surgery, there was meaningful variation in surgical approach and procedure type across cohorts.

Study-specific GLP-1RA exposure definitions, reported agents, and follow-up windows are summarized in Table 1. Exposure windows ranged from active preoperative semaglutide use and a minimum 12-week preoperative exposure criterion to prescription- or fill-based exposure definitions spanning preoperative and postoperative periods. Four studies reported 1 or more individual agents, whereas 2 evaluated GLP-1RA exposure at the class level. Specific minimum treatment-duration criteria were reported in only 2 cohorts; medication dose, adherence, and exact postoperative exposure were otherwise incompletely reported. No included study evaluated perioperative withholding or discontinuation as an analyzable variable.

Outcome ascertainment was based on study-specific claims or electronic health record algorithms. Pseudarthrosis was assessed at 1 to 2 years after surgery without a uniform clinical-radiographic confirmation standard reported across cohorts; revision or reoperation was summarized narratively because procedural definitions and follow-up windows were heterogeneous.

Pseudarthrosis according to GLP-1 receptor agonist use

Five study cohorts evaluated pseudarthrosis. Overall, GLP-1RA use was not associated with a significant difference in pseudarthrosis compared with no GLP-1RA use (OR 0.98, 95% CI 0.33–2.89; p = .96; I² = 98%) (Fig. 2).

Fig. 2.

Forest plot of pseudarthrosis after cervical fusion. Pooled odds ratio comparing GLP-1RA users with nonusers.

Dysphagia according to GLP-1 receptor agonist use

Four study cohorts evaluated dysphagia. Overall, GLP-1RA use was not associated with a significant difference in dysphagia compared with no GLP-1RA use (OR 0.89, 95% CI 0.38–2.05; p = .68; I² = 83%) (Fig. 3).

Fig. 3.

Forest plot of dysphagia after cervical fusion. Pooled odds ratio comparing GLP-1RA users with nonusers.

Hospital readmission according to GLP-1 receptor agonist use

Five study cohorts evaluated hospital readmission. Overall, GLP-1RA use was not associated with a significant difference in hospital readmission compared with no GLP-1RA use (OR 0.67, 95% CI 0.39–1.17; p = .12; I² = 87%) (Fig. 4).

Fig. 4.

Forest plot of hospital readmission after cervical fusion. Pooled odds ratio comparing GLP-1RA users with nonusers.

Device failure according to GLP-1 receptor agonist use

Four study cohorts evaluated device failure. Overall, GLP-1RA use was not associated with a significant difference in device failure compared with no GLP-1RA use (OR 0.83, 95% CI 0.44–1.56; p = .41; I² = 13%) (Fig. 5).

Fig. 5.

Forest plot of device failure after cervical fusion. Pooled odds ratio comparing GLP-1RA users with non-users. Abbreviations: GLP-1RA, glucagon-like peptide-1 receptor agonist; PRISMA, preferred reporting items for systematic reviews and meta-analyses.

Secondary outcomes

No significant differences were observed between GLP-1RA users and nonusers for the other secondary outcomes analyzed. Specifically, postoperative respiratory complications (OR 0.60, 95% CI 0.27–1.31; p = .13; I² = 21%), surgical site infection (OR 0.97, 95% CI 0.76–1.23; p = .78; I² = 0%), venous thromboembolism (OR 0.82, 95% CI 0.33–2.03; p = .53; I² = 8%), and acute kidney injury (OR 0.77, 95% CI 0.55–1.09; p = .14; I² = 0%) were not significantly different between groups. Individual forest plots for these secondary outcomes are provided in the Supplementary Material (Supplementary Figs. S1–S4).

Exploratory sensitivity analyses excluding Ng et al.

Exploratory sensitivity analyses excluding Ng et al. [15] were conducted for pseudarthrosis and dysphagia because this cohort was clinically and methodologically distinct from the remaining studies. For pseudarthrosis, exclusion of Ng et al. [15] yielded a lower pooled estimate (OR 0.66, 95% CI 0.46–0.95; p = .04; I² = 71%), although substantial heterogeneity persisted. For dysphagia, exclusion of Ng et al. [15] did not materially change the interpretation of the primary analysis (OR 0.73, 95% CI 0.36–1.46; p = .19; I² = 48%). These analyses did not alter the interpretation of the primary pooled estimates and should be considered exploratory and hypothesis-generating. The corresponding forest plots are provided in Supplementary Figs. S5 and S6.

Risk of bias

Risk of bias in nonrandomized studies was assessed using ROBINS-I. Of the 6 included studies, 1 was judged to have low risk of bias overall, 4 were judged to have moderate risk of bias, and 1 was judged to have serious risk of bias. Vatsia et al. [28] was judged to have low overall risk of bias; Bank et al. [14], Heo et al. [25], Kim et al. [26], and Rajkovic et al. [27] were judged to have moderate overall risk of bias; and Ng et al. [15] was judged to have serious overall risk of bias. The main concerns were related to bias due to confounding, which was rated as moderate in 5 studies and serious in 1 study, and bias due to missing data, which was rated as moderate across all included studies. Concerns regarding selection of the reported result were also frequent, with 5 studies rated as having moderate risk in this domain. Full ROBINS-I judgments are presented in Supplementary Figs. S7 and S8.

Certainty of evidence

Certainty of evidence was very low for pseudarthrosis, dysphagia, and hospital readmission, and low for surgical site infection. Detailed GRADE assessments, including relative and absolute effect estimates and outcome-specific explanations, are presented in Supplementary Table S1.

Discussion

In this systematic review and meta-analysis of 6 propensity score matched retrospective cohort studies, perioperative GLP-1RA use was not associated with statistically significant differences in the primary analyses of pseudarthrosis, dysphagia, hospital readmission, device failure, or the other secondary outcomes assessed. Taken together, the primary analyses did not demonstrate a consistent pattern of worse postoperative outcomes associated with perioperative GLP-1RA use after cervical fusion. Within the limits of the available observational evidence, perioperative GLP-1RA exposure alone should not be interpreted as a consistent marker of increased postoperative risk after cervical fusion.

The substantial heterogeneity observed for pseudarthrosis, dysphagia, and hospital readmission should be interpreted in the context of clinical and methodological diversity across the included cohorts. The included studies encompassed single and multilevel anterior cervical discectomy and fusion, anterior cervical fusion, and posterior cervical fusion. GLP-1RA exposure varied by reported agent and study-defined exposure window, ranging from active preoperative semaglutide use to prescription- or fill-based exposure definitions spanning preoperative and postoperative periods. Reported maximum follow-up ranged from 1 to 2 years, and outcome ascertainment varied across claims databases and electronic health record networks; pseudarthrosis was assessed using study-specific algorithms without a uniform clinical or radiographic definition. These differences may have contributed to variation in baseline risk and effect estimates and do not permit attribution of heterogeneity to any single cohort or characteristic. Although Ng et al. [15] was clinically distinct because it evaluated semaglutide exclusively in posterior cervical fusion, heterogeneity remained substantial after its exclusion. Accordingly, the pooled random-effects estimates should be interpreted as summary associations across heterogeneous cervical fusion settings rather than as approach-, level-, or medication-specific effects. Formal subgroup analyses were not performed because only 4 to 5 independent comparisons contributed to these outcomes and relevant effect modifiers were incompletely reported.

The present findings are broadly consistent with previous systematic reviews that did not demonstrate a consistent increase in perioperative complications among GLP-1RA users undergoing spine surgery [10,11]. However, some broader syntheses have also reported lower pseudarthrosis risk, particularly in pooled spinal fusion populations [11,12]. These reviews combined cervical, lumbar, and mixed surgical cohorts with different indications, exposure definitions, follow-up intervals, and outcome ascertainment methods, limiting procedure specific inference for cervical fusion. By restricting the analysis exclusively to cervical fusion, the present review extends and refines the available evidence by showing that GLP-1RA use was not associated with significant differences in pseudarthrosis or the other postoperative outcomes assessed. Thus, the favorable signal observed in more heterogeneous fusion populations was not reproduced consistently in the cervical fusion setting. The biologic plausibility of metabolic, inflammatory, and bone-related effects remains relevant, but available observational evidence does not establish a direct clinical benefit [[2], [3], [4], [5], [6], [7]].

This review has important strengths, including prospective protocol registration, independent study selection and data extraction, an exclusive focus on cervical fusion, and use of propensity score matched cohorts. ROBINS-I and GRADE assessments contextualized risk of bias and certainty. Limitations include the retrospective design of all included studies and reliance on administrative databases or electronic health record networks, leaving residual confounding possible despite matching. Claims and electronic health record-based algorithms may also have introduced outcome misclassification: no uniform clinical-radiographic standard for pseudarthrosis was reported, and coded dysphagia and implant or device failure may have varied in diagnostic specificity and clinical detail. Revision or reoperation was therefore kept distinct from pseudarthrosis and summarized narratively because its indication and surgical approach could not be consistently determined. Although Table 1 details study-specific exposure windows, medication dose, adherence, exact treatment duration, and postoperative exposure were incompletely reported; no included study evaluated perioperative withholding or discontinuation as an analyzable variable, and graft type and instrumentation were also inconsistently reported. Certainty of evidence was very low for pseudarthrosis, dysphagia, and hospital readmission, and low for surgical site infection, as presented in Supplementary Table S1.

Conclusion

Perioperative GLP-1RA use was not associated with a consistent pattern of worse postoperative outcomes after cervical fusion. Prospective, procedure specific studies with standardized exposure and outcome definitions are needed.

Credit authorship statement

Matheus Loiola Magalhães Alves: Conceptualization, Methodology, Investigation, Data curation, Formal analysis, Writing – original draft, Writing – review & editing. Antônio Olímpio da Silva Moura Costa: Methodology, Validation, Supervision, Writing – review & editing. Julia do Vale Moura Costa: Investigation, Data curation, Validation, Writing – review & editing. Gustavo Azevedo Garrido: Investigation, Data curation, Writing – review & editing. João Victor Pereira Gonzalez: Investigation, Data curation, Writing – review & editing. Julia Sader Neves Ferreira: Investigation, Data curation, Validation, Writing – review & editing. Bernardo de Andrada Pereira: Conceptualization, Methodology, Supervision, Project administration, Writing – review & editing.

Funding

This research received no specific grant from any funding agency in the public, commercial, or not-for-profit sectors.

Declaration of Competing Interests

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.