Pulmonary adverse events associated with GLP-1 receptor agonists: a systematic review of respiratory safety signals Authors: Aarushi Ahuja, Sai Prasad, Sahana Manoharan, Arya Sen, Richa Rana, Mansi Mody, FNU Paras, Archita Archita, Sana Shaik, Kartik Mehta, Rachel Manoj, Prakashini Mishra, Subhi Mehta, Pulkit Kapoor Published: 2026 ABSTRACT BackgroundGlucagon-like peptide-1 receptor agonists (GLP-1 RAs) are among the most rapidly growing drug classes in contemporary medicine, approved for type 2 diabetes mellitus (T2DM) and obesity management. Although gastrointestinal adverse effects are well characterised, the pulmonary safety profile of GLP-1 RAs remains incompletely defined, representing an important evidence gap given the scale of global prescribing.ObjectiveThis systematic review aims to synthesise published evidence on respiratory adverse events temporally associated with GLP-1 RA therapy in adults, characterise their patterns and severity, appraise the risk of bias using validated design-appropriate tools, describe proposed pathophysiological mechanisms, and grade the certainty of evidence using the Grading of Recommendations Assessment, Development, and Evaluation (GRADE) framework. Given the heterogeneity of the included evidence—spanning case reports, pharmacovigilance analyses, and observational cohorts—this review is framed primarily as a signal-detection and evidence-mapping exercise rather than a causal-inference analysis.Eligibility criteriaAdults (≥ 18 years) receiving any approved GLP-1 RA for any indication. Eligible study designs included randomised controlled trials (RCTs), retrospective and prospective observational cohorts, pharmacovigilance disproportionality analyses, and case reports or series. Studies were excluded if they enrolled paediatric populations, involved animal or in vitro experiments, reported only metabolic or cardiovascular outcomes without respiratory adverse-event data, or were narrative reviews, editorials, or commentaries without primary data.MethodsA systematic review was conducted following PRISMA 2020 guidelines (PROSPERO: CRD420261305875). PubMed, Embase, and Scopus were searched from inception to January 2026. Risk of bias was assessed using the Cochrane Risk of Bias 2 (RoB2) tool for RCTs, the Newcastle-Ottawa Scale (NOS) for cohort studies, and Joanna Briggs Institute (JBI) checklists for case reports, case series, and pharmacovigilance studies. GRADE was applied at the outcome-domain level; evidence from uncontrolled designs (case reports, case series, pharmacovigilance analyses) was initially rated very low certainty, as these designs are inherently uncontrolled and start below the lowest GRADE tier. Given substantial heterogeneity, meta-analytic pooling was not performed; a pre-specified narrative synthesis following the Synthesis Without Meta-Analysis (SWiM) reporting guideline was conducted.ResultsNineteen studies met inclusion criteria: four RCTs (n = 1,884 participants), five retrospective cohort studies (n = 1,122,653 participants), three pharmacovigilance analyses (n = 498,892 spontaneous adverse-event reports), and seven case reports or case series (n = 13 patients). Spontaneous reports do not represent unique exposed individuals and were not pooled with participant counts. Risk of bias: all four RCTs were rated low risk by RoB2 (with some concerns for open-label outcome ascertainment in three trials, a domain-specific issue that does not alter the overall RoB2 category from low); three of five cohort studies were rated moderate and one high risk by NOS (one cohort study rating not reported, classified as not reported/unable to rate); all three pharmacovigilance studies were rated high risk by JBI; and case reports were rated as methodologically adequate by JBI, though these instruments cannot overcome the inherent anecdotal nature and very limited generalisability of single-case designs. Upper respiratory tract infections (URTIs) were the most frequently reported adverse event (8 of 12 reporting studies, 66.7%), occurring at rates comparable to controls across RCTs (GRADE: MODERATE certainty). Pharmacovigilance analyses identified disproportionate reporting signals for dyspnoea and asthma-like events, particularly with exenatide, using the reporting odds ratio (ROR 2.14, 95% CI 1.88–2.43; Cazzola 2024); these signals are hypothesis-generating only and do not establish causality or incidence. Serious adverse events reported in temporal association with GLP-1 RA exposure included anaphylaxis with bronchospasm (n = 4 cases), acute eosinophilic pneumonia (n = 1), perioperative aspiration pneumonitis or pneumonia (multiple perioperative cases across two cohort studies and one case series), acute respiratory distress syndrome (ARDS; n = 2, one fatal requiring extracorporeal membrane oxygenation [ECMO]), and spontaneous pneumomediastinum (n = 1). A large global retrospective cohort study (n = 331,863 matched patients; Henney 2024) demonstrated a potentially meaningful reduction in incident pneumonia with GLP-1 RAs compared with dipeptidyl peptidase-4 (DPP-4) inhibitors (hazard ratio [HR] 0.60, 95% confidence interval [CI] 0.58–0.62; GRADE: LOW certainty); this finding should be interpreted cautiously given the retrospective design, active and non-inert comparator, and likelihood of residual confounding. Evidence certainty for serious respiratory events was very low, driven by sparse case reports and high-risk pharmacovigilance data.ConclusionGLP-1 RA therapy has a respiratory safety profile that is neither uniformly benign nor hazardous, stratified by mechanism, agent subclass, and clinical context. Common upper airway symptoms are mild and comparable to controls. Rare but clinically consequential events—including anaphylaxis, eosinophilic pneumonia, perioperative aspiration, and ARDS—have been reported in temporal association with GLP-1 RA exposure, predominantly in case reports and pharmacovigilance data of very low certainty; these signals are hypothesis-generating and should not be interpreted as confirmed causal risks. Exendin-4-based agents (exenatide and lixisenatide) appear to carry the highest reported risk of hypersensitivity reactions, attributable to their non-human structural origin. Clinicians should maintain heightened awareness for aspiration risk in perioperative settings; individualised, extended pre-procedural fasting intervals calibrated to the specific agent’s pharmacokinetic profile should be considered. Prospective, standardised, and adequately powered studies with pre-specified respiratory endpoints are required to move from signal detection to causal inference. Introduction Glucagon-like peptide-1 receptor agonists (GLP-1 RAs) are one of the fastest-growing pharmacological classes in contemporary medicine. Initially approved for glycaemic control in T2DM, their indications have expanded to include chronic weight management and cardiovascular risk reduction [1, 2]. With semaglutide, liraglutide, dulaglutide, and the dual glucose-dependent insulinotropic polypeptide (GIP)/GLP-1 agonist tirzepatide achieving global blockbuster status, more than 50 million patients worldwide currently receive GLP-1 RA therapy [1]. The pharmacodynamics of GLP-1 RAs extend beyond the pancreas and the gut. GLP-1 receptors (GLP-1R) are expressed in pulmonary epithelial cells, airway smooth muscle, alveolar macrophages, and pulmonary vasculature [3, 4]. This distribution provides a mechanistic substrate for both beneficial and adverse respiratory pharmacodynamic effects. The well-documented class effect of delayed gastric emptying additionally generates a biologically plausible aspiration risk, particularly in perioperative and sedated patients, although the clinical magnitude of this risk remains incompletely characterised [18, 19]. Despite this biological plausibility, pulmonary adverse events associated with GLP-1 RAs have not been systematically characterised. Existing safety summaries are confined to regulatory labels, narrative reviews, and individual pharmacovigilance reports, whereas gastrointestinal adverse effects dominate the published safety discourse. The rapid scale-up of GLP-1 RA prescriptions, including in populations with pre-existing pulmonary diseases, renders this evidence gap increasingly consequential. This systematic review was designed primarily as a signal-detection and evidence-mapping exercise, given the predominance of non-randomised and non-comparative data in this area. Its specific aims were to: (1) identify and synthesise published evidence on pulmonary adverse events reported in temporal association with GLP-1 RA exposure; (2) characterise the pattern, severity, and clinical outcomes of reported respiratory events; (3) appraise the risk of bias using validated, design-appropriate tools; (4) describe proposed pathophysiological mechanisms; and (5) grade the certainty of evidence using the GRADE framework, applied at the outcome-domain level, to inform clinical guidance. Subgroup and sensitivity analyses were pre-specified but were not feasible given the heterogeneity of available data and the predominantly non-comparative design of included studies; this is documented explicitly in the protocol. Methods Protocol and reporting standards This systematic review was conducted and reported in accordance with the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) 2020 guidelines [20]. The review protocol was prospectively registered with PROSPERO (CRD420261305875). Search strategy and information sources A systematic literature search was performed in PubMed, Embase, and Scopus from database inception to 31 January 2026 (exact last search date). No language restrictions were applied. Trial registries (ClinicalTrials.gov, WHO ICTRP), grey literature sources, preprint servers (medRxiv), conference abstract databases, and regulatory pharmacovigilance databases (FAERS, EudraVigilance) were additionally searched; no additional eligible studies beyond those already identified were found through these supplementary sources. The search strategy for PubMed combined Medical Subject Headings (MeSH) terms and free-text keywords in the following domains:GLP-1 RA exposure terms: semaglutide OR liraglutide OR dulaglutide OR exenatide OR tirzepatide OR lixisenatide OR mazdutide OR albiglutide OR “glucagon-like peptide-1 receptor agonist” [MeSH] OR “GLP-1 receptor agonist”;Respiratory outcome terms: “Respiratory Tract Diseases” [MeSH] OR “Pneumonia” [MeSH] OR “Aspiration Pneumonia” [MeSH] OR “Dyspnea” [MeSH] OR “Bronchospasm” [MeSH] OR “Respiratory Failure” [MeSH] OR “Pulmonary” OR “Lung” OR “Airway”;Safety terms: “Drug-Related Side Effects and Adverse Reactions” [MeSH] OR “Pharmacovigilance” [MeSH] OR “adverse event” OR “adverse effect” OR “safety”. GLP-1 RA exposure terms: semaglutide OR liraglutide OR dulaglutide OR exenatide OR tirzepatide OR lixisenatide OR mazdutide OR albiglutide OR “glucagon-like peptide-1 receptor agonist” [MeSH] OR “GLP-1 receptor agonist”; GLP-1 RA exposure terms: semaglutide OR liraglutide OR dulaglutide OR exenatide OR tirzepatide OR lixisenatide OR mazdutide OR albiglutide OR “glucagon-like peptide-1 receptor agonist” [MeSH] OR “GLP-1 receptor agonist”; Respiratory outcome terms: “Respiratory Tract Diseases” [MeSH] OR “Pneumonia” [MeSH] OR “Aspiration Pneumonia” [MeSH] OR “Dyspnea” [MeSH] OR “Bronchospasm” [MeSH] OR “Respiratory Failure” [MeSH] OR “Pulmonary” OR “Lung” OR “Airway”; Respiratory outcome terms: “Respiratory Tract Diseases” [MeSH] OR “Pneumonia” [MeSH] OR “Aspiration Pneumonia” [MeSH] OR “Dyspnea” [MeSH] OR “Bronchospasm” [MeSH] OR “Respiratory Failure” [MeSH] OR “Pulmonary” OR “Lung” OR “Airway”; Safety terms: “Drug-Related Side Effects and Adverse Reactions” [MeSH] OR “Pharmacovigilance” [MeSH] OR “adverse event” OR “adverse effect” OR “safety”. Safety terms: “Drug-Related Side Effects and Adverse Reactions” [MeSH] OR “Pharmacovigilance” [MeSH] OR “adverse event” OR “adverse effect” OR “safety”. Eligibility criteria Population Adults (≥ 18 years) receiving any GLP-1 RA for any approved indication (T2DM, obesity, or cardiovascular risk reduction). Intervention Exposure to any GLP-1 RA, including semaglutide, liraglutide, dulaglutide, exenatide, tirzepatide, lixisenatide, mazdutide, and albiglutide. Comparator Placebo, standard care, or alternative glucose-lowering or weight-loss therapy (including DPP-4 inhibitors, sodium-glucose cotransporter-2 [SGLT2] inhibitors, sulfonylureas, or insulin). Studies without a comparator group (e.g., case reports, single-arm cohorts) were eligible for inclusion as descriptive evidence. Outcomes Primary outcomes included the following pre-specified respiratory adverse events: upper respiratory tract infections (URTIs), chronic cough, upper airway symptoms (nasopharyngitis, pharyngitis, rhinitis), dyspnoea, bronchospasm, aspiration pneumonitis, aspiration pneumonia, non-aspiration community-acquired pneumonia, and acute respiratory failure (including ARDS). Secondary outcomes included hospitalisation, intensive care unit (ICU) admission, requirement for ventilatory support (invasive or non-invasive), drug discontinuation attributable to respiratory adverse events, and respiratory mortality. Study designs RCTs, retrospective and prospective observational cohorts, pharmacovigilance disproportionality analyses of spontaneous reporting databases, and case reports or series. Exclusion criteria Studies were excluded if they: enrolled paediatric populations (< 18 years); involved animal experiments or in vitro studies; evaluated non-GLP-1-based therapies exclusively without a GLP-1 RA arm; reported only metabolic or cardiovascular outcomes without any respiratory adverse-event data; or were narrative reviews, editorials, commentaries, or guidelines without original patient-level data. No publication date or country restrictions were applied. Study selection All identified records were imported into Rayyan. Three independent reviewers (A.A., S.P., S.M.) performed title and abstract screening, followed by full-text assessment by two reviewers (S.P., A.A.) for eligibility. Discrepancies at both the title/abstract and full-text stages were resolved through discussion and consensus agreement. Inter-rater agreement was calculated at both stages: κ = 0.91 at title/abstract screening and κ = 0.88 at full-text assessment. The study selection process is summarised in a PRISMA 2020 flow diagram (Fig. 1). Data extraction Data were extracted into a pre-piloted standardised extraction form by one reviewer and independently verified by a second reviewer. Extracted variables included: study design and setting; participant demographics (age, sex, sample size); GLP-1 RA agent, dose, route, and duration of exposure; comparator type; respiratory adverse events (type, severity, time to onset, management); hospitalisation and ICU admission; ventilatory support requirements; drug discontinuation; mortality; proposed mechanisms; and funding sources and conflicts of interest. For cohort studies and RCTs, we extracted hazard ratios (HRs) or odds ratios (ORs) with 95% CIs where reported. For pharmacovigilance studies, we extracted reporting odds ratios (RORs) or information component (IC) values to measure disproportionality. These pharmacovigilance effect measures are signal-detection statistics and must not be interpreted as incidence estimates or causal risk ratios. Missing data: Where outcome data were not reported or unclear, the corresponding study author was contacted by electronic mail. If no response was received within four weeks, the data were recorded as not reported (NR) in the extraction table. No assumptions were made for missing data, and no imputation was performed. Risk of bias assessment The risk of bias was appraised independently by two reviewers using the following validated, design-specific tools: Cochrane Risk of Bias 2 (RoB2) tool for RCTs, assessing five domains: randomisation process, deviations from intended interventions, missing outcome data, measurement of the outcome, and selection of the reported result.Newcastle-Ottawa Scale (NOS) for retrospective cohort studies, scoring selection, comparability, and outcome domains out of 9. NOS scores of 0–4 were classified as high risk, 5–6 as moderate risk, and 7–9 as low risk of bias.Joanna Briggs Institute (JBI) Critical Appraisal Checklist for Case Reports and Case Series, assessing patient demographics, exposure clarity, outcome description, causal reasoning, and follow-up.JBI Critical Appraisal Checklist applied to pharmacovigilance analyses, assessing data source validity, reporting completeness, confounding control, and causality assessment. It is noted that no single universally validated appraisal instrument exists specifically for pharmacovigilance disproportionality analyses; the JBI checklist was adapted and applied in this context, and its limitations are acknowledged. Cochrane Risk of Bias 2 (RoB2) tool for RCTs, assessing five domains: randomisation process, deviations from intended interventions, missing outcome data, measurement of the outcome, and selection of the reported result. Cochrane Risk of Bias 2 (RoB2) tool for RCTs, assessing five domains: randomisation process, deviations from intended interventions, missing outcome data, measurement of the outcome, and selection of the reported result. Newcastle-Ottawa Scale (NOS) for retrospective cohort studies, scoring selection, comparability, and outcome domains out of 9. NOS scores of 0–4 were classified as high risk, 5–6 as moderate risk, and 7–9 as low risk of bias. Newcastle-Ottawa Scale (NOS) for retrospective cohort studies, scoring selection, comparability, and outcome domains out of 9. NOS scores of 0–4 were classified as high risk, 5–6 as moderate risk, and 7–9 as low risk of bias. Joanna Briggs Institute (JBI) Critical Appraisal Checklist for Case Reports and Case Series, assessing patient demographics, exposure clarity, outcome description, causal reasoning, and follow-up. Joanna Briggs Institute (JBI) Critical Appraisal Checklist for Case Reports and Case Series, assessing patient demographics, exposure clarity, outcome description, causal reasoning, and follow-up. JBI Critical Appraisal Checklist applied to pharmacovigilance analyses, assessing data source validity, reporting completeness, confounding control, and causality assessment. It is noted that no single universally validated appraisal instrument exists specifically for pharmacovigilance disproportionality analyses; the JBI checklist was adapted and applied in this context, and its limitations are acknowledged. JBI Critical Appraisal Checklist applied to pharmacovigilance analyses, assessing data source validity, reporting completeness, confounding control, and causality assessment. It is noted that no single universally validated appraisal instrument exists specifically for pharmacovigilance disproportionality analyses; the JBI checklist was adapted and applied in this context, and its limitations are acknowledged. Discrepancies were resolved through consensus discussion. Where reporting-bias assessment was not feasible because of sparse, heterogeneous, or uncontrolled evidence, this was recorded explicitly and considered narratively within the certainty assessment. The overall risk of bias was rated as low, moderate, or high for each study. The ROBINS-I tool (Risk Of Bias In Non-randomised Studies of Interventions) was considered for non-randomised studies; however, given that several included cohort studies were descriptive rather than interventional in design and that the review objective was signal-detection rather than comparative effectiveness estimation, the NOS was retained as the primary instrument for cohort studies, consistent with the pre-specified protocol. Effect measures Pre-specified expected effect measures by outcome and study type were as follows: risk ratio or odds ratio (RCTs); hazard ratio or incidence rate ratio (cohort studies); reporting odds ratio or information component (pharmacovigilance analyses). All effect estimates are presented with 95% CIs. For pharmacovigilance analyses, ROR and IC values are presented as disproportionality statistics only; they do not represent incidence estimates, relative risks, or causal effect measures. Evidence synthesis and GRADE Given the substantial clinical, methodological, and statistical heterogeneity across the included studies—spanning four different study designs, multiple GLP-1 RA agents, diverse respiratory outcomes, and markedly different sample sizes—formal meta-analytic pooling was not considered appropriate for most outcomes. This decision was pre-specified in the study protocol. For outcomes for which two or more cohort studies reported comparable effect estimates (e.g., pneumonia incidence), individual effect estimates with 95% CIs are presented in a forest plot (Fig. 2) to allow visual comparison without inappropriate statistical pooling. A structured narrative synthesis was conducted, organised by adverse event category, and informed by the Synthesis Without Meta-Analysis (SWiM) reporting guideline [21]. Pre-specified subgroup categories included GLP-1 RA agent subclass (exendin-4-based vs. human GLP-1 analog), clinical indication (T2DM vs. obesity vs. cardiovascular risk reduction), and event category (aspiration vs. non-aspiration respiratory events). These subgroup analyses were not feasible in quantitative form owing to the heterogeneity and non-comparative nature of available data; findings were addressed narratively. The certainty of evidence for each outcome domain was appraised using the GRADE framework, applied at the outcome-domain level rather than as a single overall certainty rating for the review [22]. Evidence from uncontrolled designs—case reports, case series, and spontaneous pharmacovigilance analyses—was initially rated as very low certainty because of serious risk of bias, absence of denominators or comparators, imprecision, and inability to establish causality. RCT evidence started at high certainty and was downgraded according to standard GRADE domains. Upgrading considerations (large effect size, dose-response gradient, plausible confounding) were applied where applicable. The possibility of publication bias was noted where assessable (e.g., for pharmacovigilance analyses and case reports); formal funnel plot analysis was not feasible given the non-comparative and heterogeneous evidence base. Results Study selection A total of 1,271 records were identified through database searches. After the removal of 483 duplicates, 788 records were screened by title and abstract, of which 742 were excluded. Forty-six articles underwent full-text review; of these, 27 were excluded for the following reasons: no primary respiratory adverse-event data reported (n = 14), non-GLP-1 therapy only (n = 6), paediatric population (n = 3), and narrative review or editorial without primary data (n = 4). Nineteen studies met all pre-specified inclusion criteria and were included in the final synthesis. The complete PRISMA 2020 flow diagram, including all flow elements (records identified, duplicates removed, records screened, full texts assessed, exclusions with reasons, and final included studies), is presented in Fig. 1. Study characteristics The 19 included studies comprised four RCTs (n = 1,884 participants), five retrospective cohort studies (n = 1,122,653 participants), three pharmacovigilance analyses (n = 498,892 spontaneous adverse-event reports), and seven case reports or case series (n = 13 patients). Because spontaneous pharmacovigilance reports do not represent unique exposed individuals or denominator-based populations, they are not pooled with participant counts; the clinical participant total across RCTs and cohort studies was 1,124,537 participants, of whom an estimated 913,179 were GLP-1 RA-exposed. The publication years ranged from 2012 to 2025. Studies originated from the United States (n = 8), China (n = 4), Israel (n = 2), the United Kingdom (n = 1), Japan (n = 1), Switzerland (n = 1), and multinational settings (n = 2). GLP-1 RA agents studied included semaglutide (n = 8 studies), exenatide (n = 6), liraglutide (n = 5), dulaglutide (n = 4), tirzepatide (n = 3), lixisenatide (n = 2), mazdutide (n = 1), and albiglutide (n = 1); several studies examined multiple agents concurrently. Full study characteristics are detailed in Table 1. Table 1Characteristics of included studies (N = 19)#Author (Year) [Ref]CountryStudy DesignTotal NGLP-1 Agent(s)IndicationDurationComparatorPrimary Resp. AE ReportedOverall RoB1Cazzola (2024) [1]NRPharmacovigilance (FAERS retrospective)251,145 reportsAlbiglutide, Dulaglutide, Exenatide, Liraglutide, Lixisenatide, Semaglutide, TirzepatideNRNRAll other FAERS drugsDyspnoea, asthma-like events, bronchial hyperreactivity, allergic respiratory symptomsHIGH (JBI Pharmacovigilance Checklist)2Zhang (2025) [2]ChinaPharmacovigilance (FAERS retrospective)20,350 reportsTirzepatideT2DMNR (2 phases)Other FAERS drugsUpper respiratory infections, dyspnoea, obstructive sleep apnoeaHIGH (JBI Pharmacovigilance Checklist)3Li Liu (2025) [3]ChinaRCT Phase 3 (open-label)440LY05008 (Dulaglutide biosimilar)T2DM24 weeksDulaglutideURTILOW (RoB2; some concerns D2/D4)4Amin (2024) [4]USACase Report1LiraglutideT2DM2 weeksN/AAngioedema, acute hypoxaemic respiratory failure, ARDSLOW* (JBI Case Report Checklist)5Grell (2024) [5]USACase Report (conference abstract)1SemaglutideObesity6 monthsN/ASpontaneous pneumomediastinumHIGH (incomplete abstract data)6Kadowaki (2022) [6]JapanRCT Phase 3 (open-label)442TirzepatideT2DM52 weeksNo active comparatorNasopharyngitis, URTILOW (RoB2; some concerns D2/D4)7Ji (2023) [7]ChinaRCT Phase 2 (double-blind, placebo-controlled)180MazdutideObesity24 weeksPlaceboUpper respiratory eventsLOW (RoB2; all domains low)8Foer (2023) [8]USARetrospective observational cohort1,642GLP-1 RAs (multiple)T2DM + Asthma/COPD6 monthsDPP-4i, SGLT2i, sulfonylureasRespiratory exacerbationsMODERATE (NOS 5/9)9T. Liu (2019) [9]USACase Report1DulaglutideT2DM1 doseN/AAnaphylaxisLOW* (JBI Case Report Checklist)10Shamriz (2019) [10]IsraelCase Report1Lixisenatide / ExenatideT2DMFirst doseN/AAnaphylaxis, bronchospasmLOW* (JBI Case Report Checklist)11Henney (2024) [11]Multinational (TriNetX)Retrospective Cohort (EHR)331,863GLP-1 RAs (multiple)T2DM12 monthsDPP-4 inhibitorsIncident pneumonia (HR 0.60, 95% CI 0.58–0.62)MODERATE (NOS 6/9)12Fishman (2025) [12]USARetrospective cohort (no control)58Semaglutide, Dulaglutide, Liraglutide, Exenatide, TirzepatideT2DM/Obesity (post-lung transplant)1.4 years (median)NonePneumonia, hospitalisationHIGH (NOS 3/9)13Askin (2016) [13]USACase Report1Exenatide ERT2DM3 weeksN/AAcute eosinophilic pneumoniaLOW* (JBI Case Report Checklist)14Russell-Jones (2012) [14]MultinationalRCT Phase 3 (open-label)820Exenatide (once weekly)T2DM26 weeksMetformin / Pioglitazone / Sitagliptin / InsulinNasopharyngitis, coughLOW (RoB2; some concerns D2/D4)15Steveling (2014) [15]SwitzerlandCase Report1ExenatideT2DM/Obesity1 monthN/AEosinophilic pneumonia, bronchospasmLOW* (JBI Case Report Checklist)16Nakhla (2025) [16]USAPharmacovigilance (multi-database: FAERS, EudraVigilance, DAEN, VigiBase)Multiple databasesSemaglutide, Liraglutide, Dulaglutide, Exenatide, TirzepatideT2DM/ObesityNRSGLT2i, sulfonylureas, DPP-4iAspiration events (no positive GLP-1 RA signal identified)HIGH (JBI Pharmacovigilance Checklist)17Rehan (2024) [17]USACase Report (conference abstract)1GLP-1 RA (not otherwise specified)ObesityNRN/AARDS requiring ECMO (fatal)HIGH (incomplete abstract)18Avraham (2024) [18]IsraelCase Series (n = 2)2SemaglutideT2DM / ObesityLast dose 4–6 days pre-opN/AAspiration pneumonitis (perioperative)MODERATE (JBI Case Series Checklist)19Rashid (2025) [19]USARetrospective cohort (IBM MarketScan)138,980Semaglutide, Albiglutide, Liraglutide, Dulaglutide, Lixisenatide, ExenatideT2DM/Obesity (surgical)Up to 1 year pre-surgeryStandard careAspiration pneumonia (adjusted HR 1.28, 95% CI 0.98–1.67)MODERATE (NOS 5/9)Abbreviations: ARDS, acute respiratory distress syndrome; CI, confidence interval; DB, double-blind; DPP-4i, dipeptidyl peptidase-4 inhibitor; ECMO, extracorporeal membrane oxygenation; EHR, electronic health records; ER, extended release; FAERS, FDA Adverse Event Reporting System; GLP-1 RA, glucagon-like peptide-1 receptor agonist; HR, hazard ratio; JBI, Joanna Briggs Institute; NOS, Newcastle-Ottawa Scale; NR, not reported; PC, placebo-controlled; RCT, randomised controlled trial; RoB, risk of bias; RoB2, Cochrane Risk of Bias 2 tool; ROR, reporting odds ratio; SGLT2i, sodium-glucose cotransporter-2 inhibitor; T2DM, type 2 diabetes mellitus; URTI, upper respiratory tract infection. *LOW* = methodologically adequate by JBI checklist with acknowledged very limited generalisability inherent to single-case designs Table 1Characteristics of included studies (N = 19)#Author (Year) [Ref]CountryStudy DesignTotal NGLP-1 Agent(s)IndicationDurationComparatorPrimary Resp. AE ReportedOverall RoB1Cazzola (2024) [1]NRPharmacovigilance (FAERS retrospective)251,145 reportsAlbiglutide, Dulaglutide, Exenatide, Liraglutide, Lixisenatide, Semaglutide, TirzepatideNRNRAll other FAERS drugsDyspnoea, asthma-like events, bronchial hyperreactivity, allergic respiratory symptomsHIGH (JBI Pharmacovigilance Checklist)2Zhang (2025) [2]ChinaPharmacovigilance (FAERS retrospective)20,350 reportsTirzepatideT2DMNR (2 phases)Other FAERS drugsUpper respiratory infections, dyspnoea, obstructive sleep apnoeaHIGH (JBI Pharmacovigilance Checklist)3Li Liu (2025) [3]ChinaRCT Phase 3 (open-label)440LY05008 (Dulaglutide biosimilar)T2DM24 weeksDulaglutideURTILOW (RoB2; some concerns D2/D4)4Amin (2024) [4]USACase Report1LiraglutideT2DM2 weeksN/AAngioedema, acute hypoxaemic respiratory failure, ARDSLOW* (JBI Case Report Checklist)5Grell (2024) [5]USACase Report (conference abstract)1SemaglutideObesity6 monthsN/ASpontaneous pneumomediastinumHIGH (incomplete abstract data)6Kadowaki (2022) [6]JapanRCT Phase 3 (open-label)442TirzepatideT2DM52 weeksNo active comparatorNasopharyngitis, URTILOW (RoB2; some concerns D2/D4)7Ji (2023) [7]ChinaRCT Phase 2 (double-blind, placebo-controlled)180MazdutideObesity24 weeksPlaceboUpper respiratory eventsLOW (RoB2; all domains low)8Foer (2023) [8]USARetrospective observational cohort1,642GLP-1 RAs (multiple)T2DM + Asthma/COPD6 monthsDPP-4i, SGLT2i, sulfonylureasRespiratory exacerbationsMODERATE (NOS 5/9)9T. Liu (2019) [9]USACase Report1DulaglutideT2DM1 doseN/AAnaphylaxisLOW* (JBI Case Report Checklist)10Shamriz (2019) [10]IsraelCase Report1Lixisenatide / ExenatideT2DMFirst doseN/AAnaphylaxis, bronchospasmLOW* (JBI Case Report Checklist)11Henney (2024) [11]Multinational (TriNetX)Retrospective Cohort (EHR)331,863GLP-1 RAs (multiple)T2DM12 monthsDPP-4 inhibitorsIncident pneumonia (HR 0.60, 95% CI 0.58–0.62)MODERATE (NOS 6/9)12Fishman (2025) [12]USARetrospective cohort (no control)58Semaglutide, Dulaglutide, Liraglutide, Exenatide, TirzepatideT2DM/Obesity (post-lung transplant)1.4 years (median)NonePneumonia, hospitalisationHIGH (NOS 3/9)13Askin (2016) [13]USACase Report1Exenatide ERT2DM3 weeksN/AAcute eosinophilic pneumoniaLOW* (JBI Case Report Checklist)14Russell-Jones (2012) [14]MultinationalRCT Phase 3 (open-label)820Exenatide (once weekly)T2DM26 weeksMetformin / Pioglitazone / Sitagliptin / InsulinNasopharyngitis, coughLOW (RoB2; some concerns D2/D4)15Steveling (2014) [15]SwitzerlandCase Report1ExenatideT2DM/Obesity1 monthN/AEosinophilic pneumonia, bronchospasmLOW* (JBI Case Report Checklist)16Nakhla (2025) [16]USAPharmacovigilance (multi-database: FAERS, EudraVigilance, DAEN, VigiBase)Multiple databasesSemaglutide, Liraglutide, Dulaglutide, Exenatide, TirzepatideT2DM/ObesityNRSGLT2i, sulfonylureas, DPP-4iAspiration events (no positive GLP-1 RA signal identified)HIGH (JBI Pharmacovigilance Checklist)17Rehan (2024) [17]USACase Report (conference abstract)1GLP-1 RA (not otherwise specified)ObesityNRN/AARDS requiring ECMO (fatal)HIGH (incomplete abstract)18Avraham (2024) [18]IsraelCase Series (n = 2)2SemaglutideT2DM / ObesityLast dose 4–6 days pre-opN/AAspiration pneumonitis (perioperative)MODERATE (JBI Case Series Checklist)19Rashid (2025) [19]USARetrospective cohort (IBM MarketScan)138,980Semaglutide, Albiglutide, Liraglutide, Dulaglutide, Lixisenatide, ExenatideT2DM/Obesity (surgical)Up to 1 year pre-surgeryStandard careAspiration pneumonia (adjusted HR 1.28, 95% CI 0.98–1.67)MODERATE (NOS 5/9)Abbreviations: ARDS, acute respiratory distress syndrome; CI, confidence interval; DB, double-blind; DPP-4i, dipeptidyl peptidase-4 inhibitor; ECMO, extracorporeal membrane oxygenation; EHR, electronic health records; ER, extended release; FAERS, FDA Adverse Event Reporting System; GLP-1 RA, glucagon-like peptide-1 receptor agonist; HR, hazard ratio; JBI, Joanna Briggs Institute; NOS, Newcastle-Ottawa Scale; NR, not reported; PC, placebo-controlled; RCT, randomised controlled trial; RoB, risk of bias; RoB2, Cochrane Risk of Bias 2 tool; ROR, reporting odds ratio; SGLT2i, sodium-glucose cotransporter-2 inhibitor; T2DM, type 2 diabetes mellitus; URTI, upper respiratory tract infection. *LOW* = methodologically adequate by JBI checklist with acknowledged very limited generalisability inherent to single-case designs Characteristics of included studies (N = 19) Abbreviations: ARDS, acute respiratory distress syndrome; CI, confidence interval; DB, double-blind; DPP-4i, dipeptidyl peptidase-4 inhibitor; ECMO, extracorporeal membrane oxygenation; EHR, electronic health records; ER, extended release; FAERS, FDA Adverse Event Reporting System; GLP-1 RA, glucagon-like peptide-1 receptor agonist; HR, hazard ratio; JBI, Joanna Briggs Institute; NOS, Newcastle-Ottawa Scale; NR, not reported; PC, placebo-controlled; RCT, randomised controlled trial; RoB, risk of bias; RoB2, Cochrane Risk of Bias 2 tool; ROR, reporting odds ratio; SGLT2i, sodium-glucose cotransporter-2 inhibitor; T2DM, type 2 diabetes mellitus; URTI, upper respiratory tract infection. *LOW* = methodologically adequate by JBI checklist with acknowledged very limited generalisability inherent to single-case designs Risk of bias in included studies All four RCTs were rated as having a low overall risk of bias by RoB2. The main concern across three of the four RCTs was the open-label design, which introduced potential detection bias for subjectively ascertained adverse events such as dyspnoea and cough (Domain 2 and Domain 4: some concerns). It is emphasised that under RoB2 conventions, ‘some concerns’ at the domain level is distinct from the overall study-level rating; a study may receive ‘some concerns’ on individual domains while retaining a ‘low’ overall rating when the overall judgement considers that the domain concerns are unlikely to materially affect the outcome. The double-blind, placebo-controlled phase 2 mazdutide trial (Ji 2023) received low ratings across all five RoB2 domains. Among the five retrospective cohort studies, three were rated moderate (NOS scores 5–6/9) and one had a high risk of bias (Fishman 2025; NOS 3/9, absence of comparator group). The rating for the fifth cohort study (Henney 2024) was moderate (NOS 6/9). All three pharmacovigilance studies were rated as having high risk of bias, reflecting the inherent structural limitations of spontaneous reporting databases: absence of denominator data precluding true incidence estimation, underreporting and stimulated-reporting biases, inability to establish causality, frequent missing covariate data, and potential for duplicate reports. Of the seven case reports and series, five were rated as methodologically adequate by JBI (with acknowledged very limited generalisability inherent to single-case designs), and two were rated as having high risk of bias due to incomplete data in conference abstracts. Overall, 9/19 studies (47%) were rated as low risk of bias, 4/19 (21%) moderate risk, and 6/19 (32%) high risk. The risk of bias summary is presented in Table 2. Table 2Risk of bias assessment summary (N = 19 studies)#Author (Year)Study DesignRoB ToolD1/SelectionD2/Comp.D3/OutcomeD4D5Key ConcernsOverall RoB1Cazzola (2024)Pharmacovigilance (FAERS)JBI Pharmacovigilance ChecklistN/AN/AN/ALHNo denominator data; underreporting and stimulated-reporting bias; causality not establishable; duplicate reports possibleHIGH2Zhang (2025)Pharmacovigilance (FAERS)JBI Pharmacovigilance ChecklistN/AN/AN/ALHSame limitations as above; 99.2% US-sourced reports; 93.7% consumer-submitted; limited generalisabilityHIGH3Li Liu (2025)RCT Phase 3 (open-label)RoB2LMLMLOpen-label design: potential detection bias for subjective adverse events; industry fundingLOW4Amin (2024)Case ReportJBI Case Report ChecklistN/AN/ALLN/AAdequate case description; prior ACE inhibitor reaction may confound attributionLOW*5Grell (2024)Case Report (conference abstract)JBI Case Report ChecklistN/AN/AUCLMIncomplete data; conference abstract only; exposure details unverifiableHIGH6Kadowaki (2022)RCT Phase 3 (open-label)RoB2LMLMLOpen-label; no placebo arm for adverse event comparison; single-country (Japan)LOW7Ji (2023)RCT Phase 2 (DB, PC)RoB2LLLLLLow concerns across all domains; double-blind, placebo-controlled designLOW8Foer (2023)Retrospective CohortNOSMMM——Active comparator bias; residual confounding by indication; retrospective adverse event ascertainment; NOS 5/9MODERATE9T. Liu (2019)Case ReportJBI Case Report ChecklistN/AN/ALLN/AAdequate case description; temporal association clearly documentedLOW*10Shamriz (2019)Case ReportJBI Case Report ChecklistN/AN/ALLN/APrior drug tolerance documented; strengthens drug specificityLOW*11Henney (2024)Retrospective Cohort (EHR)NOSLMM——Non-inert comparator (DPP-4 inhibitors); residual confounding; NOS 6/9; EHR coding accuracyMODERATE12Fishman (2025)Retrospective Cohort (no control)NOSHHM——No comparator group; highly selected population; small n = 58; NOS 3/9HIGH13Askin (2016)Case ReportJBI Case Report ChecklistN/AN/ALLN/ABAL confirmation strengthens diagnosis; single caseLOW*14Russell-Jones (2012)RCT Phase 3 (open-label)RoB2LMLMLOpen-label; multiple active comparators limits isolation of respiratory signalLOW15Steveling (2014)Case ReportJBI Case Report ChecklistN/AN/ALLN/AHistological confirmation; rechallenge not performedLOW*16Nakhla (2025)Pharmacovigilance (FAERS)JBI Pharmacovigilance ChecklistN/AN/AN/ALHDisproportionality measure only; same FAERS structural limitations; comparator classes not inactiveHIGH17Rehan (2024)Case Report (conference abstract)JBI Case Report ChecklistN/AN/AUCLUCIncomplete data; agent unspecified; abstract only; causality unverifiableHIGH18Avraham (2024)Case Series (n = 2)JBI Case Series ChecklistN/AN/ALLMSmall series; causal mechanism biologically plausible but confounders possibleMODERATE19Rashid (2025)Retrospective Cohort (MarketScan)NOSMMM——Claims-based data; coding accuracy concerns; confounding by surgical complexity; NOS 5/9MODERATERoB2 domains: D1 = Randomisation process; D2 = Deviations from intended interventions; D3 = Missing outcome data; D4 = Outcome measurement; D5 = Selection of reported result. L = Low; M = Moderate/Some Concerns; H = High; UC = Unclear; N/A = Not Applicable; — = Domain not applicable to this tool. NOS scoring: 0–4 = High RoB; 5–6 = Moderate RoB; 7–9 = Low RoB. For RoB2, ‘Some concerns’ at the domain level is consistent with an overall ‘Low’ study-level rating when domain concerns are unlikely to materially affect the outcome of interest Table 2Risk of bias assessment summary (N = 19 studies)#Author (Year)Study DesignRoB ToolD1/SelectionD2/Comp.D3/OutcomeD4D5Key ConcernsOverall RoB1Cazzola (2024)Pharmacovigilance (FAERS)JBI Pharmacovigilance ChecklistN/AN/AN/ALHNo denominator data; underreporting and stimulated-reporting bias; causality not establishable; duplicate reports possibleHIGH2Zhang (2025)Pharmacovigilance (FAERS)JBI Pharmacovigilance ChecklistN/AN/AN/ALHSame limitations as above; 99.2% US-sourced reports; 93.7% consumer-submitted; limited generalisabilityHIGH3Li Liu (2025)RCT Phase 3 (open-label)RoB2LMLMLOpen-label design: potential detection bias for subjective adverse events; industry fundingLOW4Amin (2024)Case ReportJBI Case Report ChecklistN/AN/ALLN/AAdequate case description; prior ACE inhibitor reaction may confound attributionLOW*5Grell (2024)Case Report (conference abstract)JBI Case Report ChecklistN/AN/AUCLMIncomplete data; conference abstract only; exposure details unverifiableHIGH6Kadowaki (2022)RCT Phase 3 (open-label)RoB2LMLMLOpen-label; no placebo arm for adverse event comparison; single-country (Japan)LOW7Ji (2023)RCT Phase 2 (DB, PC)RoB2LLLLLLow concerns across all domains; double-blind, placebo-controlled designLOW8Foer (2023)Retrospective CohortNOSMMM——Active comparator bias; residual confounding by indication; retrospective adverse event ascertainment; NOS 5/9MODERATE9T. Liu (2019)Case ReportJBI Case Report ChecklistN/AN/ALLN/AAdequate case description; temporal association clearly documentedLOW*10Shamriz (2019)Case ReportJBI Case Report ChecklistN/AN/ALLN/APrior drug tolerance documented; strengthens drug specificityLOW*11Henney (2024)Retrospective Cohort (EHR)NOSLMM——Non-inert comparator (DPP-4 inhibitors); residual confounding; NOS 6/9; EHR coding accuracyMODERATE12Fishman (2025)Retrospective Cohort (no control)NOSHHM——No comparator group; highly selected population; small n = 58; NOS 3/9HIGH13Askin (2016)Case ReportJBI Case Report ChecklistN/AN/ALLN/ABAL confirmation strengthens diagnosis; single caseLOW*14Russell-Jones (2012)RCT Phase 3 (open-label)RoB2LMLMLOpen-label; multiple active comparators limits isolation of respiratory signalLOW15Steveling (2014)Case ReportJBI Case Report ChecklistN/AN/ALLN/AHistological confirmation; rechallenge not performedLOW*16Nakhla (2025)Pharmacovigilance (FAERS)JBI Pharmacovigilance ChecklistN/AN/AN/ALHDisproportionality measure only; same FAERS structural limitations; comparator classes not inactiveHIGH17Rehan (2024)Case Report (conference abstract)JBI Case Report ChecklistN/AN/AUCLUCIncomplete data; agent unspecified; abstract only; causality unverifiableHIGH18Avraham (2024)Case Series (n = 2)JBI Case Series ChecklistN/AN/ALLMSmall series; causal mechanism biologically plausible but confounders possibleMODERATE19Rashid (2025)Retrospective Cohort (MarketScan)NOSMMM——Claims-based data; coding accuracy concerns; confounding by surgical complexity; NOS 5/9MODERATERoB2 domains: D1 = Randomisation process; D2 = Deviations from intended interventions; D3 = Missing outcome data; D4 = Outcome measurement; D5 = Selection of reported result. L = Low; M = Moderate/Some Concerns; H = High; UC = Unclear; N/A = Not Applicable; — = Domain not applicable to this tool. NOS scoring: 0–4 = High RoB; 5–6 = Moderate RoB; 7–9 = Low RoB. For RoB2, ‘Some concerns’ at the domain level is consistent with an overall ‘Low’ study-level rating when domain concerns are unlikely to materially affect the outcome of interest Risk of bias assessment summary (N = 19 studies) RoB2 domains: D1 = Randomisation process; D2 = Deviations from intended interventions; D3 = Missing outcome data; D4 = Outcome measurement; D5 = Selection of reported result. L = Low; M = Moderate/Some Concerns; H = High; UC = Unclear; N/A = Not Applicable; — = Domain not applicable to this tool. NOS scoring: 0–4 = High RoB; 5–6 = Moderate RoB; 7–9 = Low RoB. For RoB2, ‘Some concerns’ at the domain level is consistent with an overall ‘Low’ study-level rating when domain concerns are unlikely to materially affect the outcome of interest Pulmonary adverse-event patterns Upper respiratory tract infections and airway symptoms Upper airway symptoms and URTIs were the most frequently reported pulmonary adverse events, identified in eight of the 12 studies that explicitly reported this outcome (66.7%, 8 of 12 reporting studies). Across all four included RCTs, URTI rates were consistently reported but occurred at rates comparable to comparator arms without statistically significant excess risk attributable to GLP-1 RA exposure [3, 6, 7, 14]. In the tirzepatide SURPASS J-COMBO phase 3 trial (Kadowaki 2022; n = 442), nasopharyngitis occurred in 17.0% of participants and did not differ significantly from comparator rates [6]. In the mazdutide phase 2 trial (Ji 2023; n = 180), upper respiratory events were symmetrically distributed across the three dose groups and placebo (p > 0.05 for all comparisons) [7]. The dulaglutide biosimilar RCT (Li Liu 2025; n = 440) reported URTIs in both arms with no significant between-group difference [3]. The retrospective cohort study by Foer 2023 (n = 1,642 patients with comorbid asthma or COPD receiving diabetes pharmacotherapy) demonstrated no significant excess respiratory exacerbation risk with GLP-1 RAs compared with DPP-4 inhibitors, SGLT2 inhibitors, or sulfonylureas [8]. GRADE certainty for this outcome: MODERATE (evidence from RCTs; downgraded one level for open-label ascertainment concerns in three trials). Dyspnoea and bronchospasm: pharmacovigilance signals Dyspnoea was the most frequently flagged serious respiratory signal in pharmacovigilance datasets, reported in 10 of 13 studies addressing this outcome. The largest pharmacovigilance analysis, Cazzola 2024 (251,145 FAERS reports across all approved GLP-1 RAs), identified disproportionate reporting signals for asthma-like events (ROR 1.63, 95% CI 1.41–1.88), bronchial hyperreactivity, and allergic respiratory symptoms [1]. Exenatide demonstrated the strongest disproportionality signal for dyspnoea (ROR 2.14, 95% CI 1.88–2.43), and was associated with 32 respiratory-related deaths in the FAERS database [1]. The authors attributed this signal to exenatide’s exendin-4 structural origin, which shares only 50% amino acid homology with endogenous human GLP-1, promoting anti-drug antibody formation and IgE-mediated hypersensitivity. It is critically important to contextualise these pharmacovigilance findings. Disproportionality metrics such as RORs are signal-detection statistics; they indicate that a given adverse event is reported with a higher frequency for a drug relative to all other drugs in the database. They do not constitute incidence estimates, relative risks, or causal evidence. These findings should be considered exploratory and hypothesis-generating rather than confirmatory. All three pharmacovigilance analyses were rated high risk of bias, and their findings should be framed accordingly. Zhang 2025 (FAERS; n = 20,350 tirzepatide reports) identified a reporting signal for dyspnoea with tirzepatide (ROR 1.18, 95% CI 0.95–1.47), which did not reach statistical significance [2]. Nakhla 2025 (multi-database pharmacovigilance analysis including FAERS, EudraVigilance, DAEN, and WHO VigiBase) found no positive disproportionality signals for aspiration events with GLP-1 RAs across any of the four databases searched; GLP-1 RAs ranked lower than DPP-4 inhibitors for aspiration reporting across databases [16]. Hypersensitivity reactions: anaphylaxis, angioedema, and eosinophilic pneumonia Four case reports documented severe immune-mediated pulmonary hypersensitivity events reported in temporal association with GLP-1 RA exposure [4, 9, 10, 15]. Amin 2024 described a 58-year-old woman who developed acute angioedema with tongue swelling, severe dyspnoea, and acute hypoxaemic respiratory failure within six hours of liraglutide administration, requiring ICU admission, intubation, and mechanical ventilation before subsequent recovery [4]. Shamriz 2019 reported anaphylaxis following a first dose of lixisenatide in a patient who had previously tolerated exenatide and liraglutide, demonstrating drug-specific rather than universal class-level hypersensitivity [10]. T. Liu 2019 documented anaphylaxis following a single dose of dulaglutide [9]. Steveling 2014 reported pulmonary infiltrates and peripheral eosinophilia in a patient receiving exenatide, consistent with a drug hypersensitivity reaction; radiological and laboratory resolution occurred following drug discontinuation [15]. Askin 2016 reported biopsy-confirmed acute eosinophilic pneumonia with BAL eosinophilia in a patient receiving exenatide extended-release for three weeks, which resolved completely after drug withdrawal and corticosteroid therapy [13]. The structural distinction between exendin-4-based agents (exenatide, lixisenatide; 50% structural homology with human GLP-1) and human GLP-1 analogues (liraglutide, semaglutide, dulaglutide; ≥90% homology) appears clinically relevant: immune-mediated serious events cluster predominantly with exendin-4-based agents across the published literature, consistent with their higher immunogenic potential [1, 9, 10, 15]. As these findings are derived exclusively from case reports, they represent very-low-certainty evidence and cannot establish class-level incidence. Aspiration events: perioperative and post-procedural risk Aspiration pneumonitis or pneumonia was reported in four studies with biologically coherent perioperative clustering [18, 19]. Avraham 2024 presented a case series of two patients who developed aspiration pneumonitis following elective surgery conducted under general anaesthesia; both patients had received semaglutide within four to six days of their procedure and had followed standard nil-by-mouth protocols [18]. Nasogastric aspiration at induction revealed unexpectedly large gastric residual volumes, supporting delayed gastric emptying as a plausible mechanistic explanation. Rashid 2025 (retrospective surgical cohort; n = 138,980 patients, of whom n = 2,944 were GLP-1 RA-exposed) found a trend toward increased postoperative aspiration events in GLP-1 RA-exposed patients (adjusted HR 1.28, 95% CI 0.98–1.67), which did not reach conventional statistical significance but is directionally consistent with biological plausibility [19]. The pharmacological basis for this risk is well established: GLP-1 RAs delay gastric emptying via both central and peripheral GLP-1R-mediated mechanisms, increasing intragastric residual volume. For long-acting weekly formulations, such as subcutaneous semaglutide (half-life approximately 7 days), standard preoperative fasting intervals may be insufficient to normalise gastric emptying. Point-of-care gastric ultrasound assessment immediately before induction has been proposed as a practical risk mitigation strategy. GRADE certainty for aspiration outcomes: VERY LOW. Pneumonia incidence: a paradoxical protective signal In contrast to the adverse signals described above, Henney et al. 2024—the largest study in this review—demonstrated a statistically robust reduction in the risk of incident pneumonia among GLP-1 RA users compared with DPP-4 inhibitor users in a global federated electronic health records cohort using the TriNetX platform (331,863 matched patients with T2DM; adjusted HR 0.60, 95% CI 0.58–0.62) [11]. This 40% relative risk reduction was consistent across pre-specified subgroup analyses, including age, sex, body mass index, and baseline comorbidities. However, important methodological limitations substantially constrain causal interpretation of this finding. The observational design introduces the potential for residual confounding. DPP-4 inhibitors, the active comparator employed, are not biologically inert: by inhibiting dipeptidyl peptidase-4, they increase endogenous GLP-1 bioavailability, which may attenuate or confound the apparent advantage conferred by exogenous GLP-1 RA therapy. Furthermore, the direction of confounding is uncertain, as healthier and more adherent patients may be selectively prescribed GLP-1 RAs (confounding by indication). These limitations are emphasised here as primary considerations rather than footnotes; the pneumonia risk reduction signal is an interesting hypothesis warranting prospective investigation, not an established causal effect. GRADE certainty: LOW. Acute respiratory failure and life-threatening events Acute hypoxaemic or hypercapnic respiratory failure was documented in six studies (50% of those reporting this outcome) [4, 13, 15, 17]. The most severe events were documented in case reports. Rehan 2024 (conference abstract) described a 33-year-old man with obesity who developed bilateral pulmonary infiltrates and ARDS following GLP-1 RA exposure, requiring veno-venous ECMO; the patient died, representing the only reported fatal case judged by the case authors as temporally associated with a pulmonary adverse event in this review [17]. A second case of ARDS was reported in a patient receiving liraglutide, who survived following critical care support [4]. Grell 2024 described spontaneous pneumomediastinum in a patient receiving semaglutide for weight loss, hypothesised to result from vomiting-related barotrauma secondary to gastroparesis-induced emesis [5]. Given that the Rehan 2024 case was reported in a conference abstract with incomplete data, the agent unspecified, and causality unverifiable by independent assessment, this event is characterised as a reported temporal association rather than a confirmed causal relationship. Severity outcomes and secondary endpoints Hospitalisation due to respiratory causes was documented in 11 of 14 reporting studies (78.6%), although case reports heavily influenced this proportion and aggregate absolute event numbers were small. ICU admission was confirmed in three studies. Mechanical ventilation was required in three studies, including the ECMO case. Non-invasive ventilation was not confirmed in any included study. Drug discontinuation due to respiratory adverse events was documented in six studies, most frequently in case reports of anaphylaxis or severe hypersensitivity reactions. Respiratory mortality was reported in three studies: Cazzola 2024 identified 32 respiratory-related deaths in the FAERS database associated with GLP-1 RAs (noting that FAERS deaths are not definitively attributable to GLP-1 RA exposure); Zhang 2025 identified 55 total deaths in tirzepatide FAERS reports (not all confirmed respiratory); and one fatality was documented in the ARDS case report by Rehan 2024 [1, 2, 17]. A summary of all pulmonary adverse events across included studies is presented in Table 3. Table 3Summary of pulmonary adverse events across included studiesAdverse Event CategoryStudies (Y)Studies (N)NR% Reporting (Y/[Y + N])Evidence BaseClinical SignificanceGRADE CertaintyKey Studies [Ref]Chronic Cough28920.0%Very low (2 case reports)Low–ModerateVERY LOW[4,13]Upper Airway Symptoms / URTI84766.7% (8/12)Moderate (RCTs + cohorts)Mild; comparable to controlsMODERATE[3,6,7,8,14]Dyspnoea103676.9% (10/13)Low–Moderate (pharmacovigilance + cohorts)Moderate–High; exenatide highest disproportionalityLOW[1,2,4,8,16]Bronchospasm28920.0%Very low (case reports)Low; predominantly exendin-4 agentsVERY LOW[9,10,15]Aspiration Pneumonitis (perioperative)48733.3%Very low (case series + one cohort)High in surgical contextVERY LOW[18,19]Aspiration Pneumonia29818.2%Very low–LowModerateVERY LOW[16,19]Non-Procedural Pneumonia (protective signal)48733.3%Low (large retrospective cohort)Potentially high benefit; HR 0.60LOW[8,11,12]Acute Hypoxaemic Respiratory Failure66750.0%Very low (case reports)High; life-threateningVERY LOW[4,13,17]Hypercapnic Respiratory Failure38827.3%Very low (case reports)HighVERY LOW[4,17]Anaphylaxis / Angioedema48733.3%Very low (case reports)High; predominantly exendin-4 agentsVERY LOW[4,9,10,15]Eosinophilic Pneumonia11089.1%Very low (single case)High; immune-mediatedVERY LOW[15]Spontaneous Pneumomediastinum11089.1%Very low (single conference abstract)Moderate (barotrauma/vomiting)VERY LOW[5]Y = studies explicitly reporting the event; N = studies explicitly reporting absence of the event; NR = outcome not reported. Percentage calculated as Y/(Y + N). Evidence base and GRADE certainty apply at the outcome-domain level. Pharmacovigilance studies (rated high risk of bias) contribute to Y counts but provide disproportionality signals only, not incidence data Table 3Summary of pulmonary adverse events across included studiesAdverse Event CategoryStudies (Y)Studies (N)NR% Reporting (Y/[Y + N])Evidence BaseClinical SignificanceGRADE CertaintyKey Studies [Ref]Chronic Cough28920.0%Very low (2 case reports)Low–ModerateVERY LOW[4,13]Upper Airway Symptoms / URTI84766.7% (8/12)Moderate (RCTs + cohorts)Mild; comparable to controlsMODERATE[3,6,7,8,14]Dyspnoea103676.9% (10/13)Low–Moderate (pharmacovigilance + cohorts)Moderate–High; exenatide highest disproportionalityLOW[1,2,4,8,16]Bronchospasm28920.0%Very low (case reports)Low; predominantly exendin-4 agentsVERY LOW[9,10,15]Aspiration Pneumonitis (perioperative)48733.3%Very low (case series + one cohort)High in surgical contextVERY LOW[18,19]Aspiration Pneumonia29818.2%Very low–LowModerateVERY LOW[16,19]Non-Procedural Pneumonia (protective signal)48733.3%Low (large retrospective cohort)Potentially high benefit; HR 0.60LOW[8,11,12]Acute Hypoxaemic Respiratory Failure66750.0%Very low (case reports)High; life-threateningVERY LOW[4,13,17]Hypercapnic Respiratory Failure38827.3%Very low (case reports)HighVERY LOW[4,17]Anaphylaxis / Angioedema48733.3%Very low (case reports)High; predominantly exendin-4 agentsVERY LOW[4,9,10,15]Eosinophilic Pneumonia11089.1%Very low (single case)High; immune-mediatedVERY LOW[15]Spontaneous Pneumomediastinum11089.1%Very low (single conference abstract)Moderate (barotrauma/vomiting)VERY LOW[5]Y = studies explicitly reporting the event; N = studies explicitly reporting absence of the event; NR = outcome not reported. Percentage calculated as Y/(Y + N). Evidence base and GRADE certainty apply at the outcome-domain level. Pharmacovigilance studies (rated high risk of bias) contribute to Y counts but provide disproportionality signals only, not incidence data Summary of pulmonary adverse events across included studies Y = studies explicitly reporting the event; N = studies explicitly reporting absence of the event; NR = outcome not reported. Percentage calculated as Y/(Y + N). Evidence base and GRADE certainty apply at the outcome-domain level. Pharmacovigilance studies (rated high risk of bias) contribute to Y counts but provide disproportionality signals only, not incidence data Proposed mechanisms The following section describes mechanistic hypotheses proposed in the included studies and broader literature. These mechanisms represent biologically plausible explanations, not established causal pathways, and should be interpreted as context for clinical vigilance rather than confirmed pharmacological certainty. Four principal, non-mutually exclusive mechanisms have been proposed. First, IgE-mediated hypersensitivity is operative primarily for exendin-4-based agents (exenatide, lixisenatide) due to their non-human structural origin; immunogenic anti-drug antibody formation may drive anaphylaxis, bronchospasm, angioedema, and eosinophilic pneumonitis [1, 9, 10, 15]. Second, gastroparesis-mediated aspiration arises from GLP-1 RA-induced delayed gastric emptying, elevating intragastric residual volume and aspiration risk in anaesthetised or sedated patients; this effect is particularly pronounced with long-acting weekly formulations [18, 19]. Third, direct pulmonary GLP-1 receptor activation may produce pharmacodynamic effects on airway smooth muscle contractility, pulmonary epithelial permeability, and alveolar macrophage function, potentially contributing to dyspnoea and bronchial hyperreactivity signals observed in pharmacovigilance data [1, 3]. Fourth, anti-inflammatory and weight-loss-mediated pleiotropic effects—including reduction of systemic and pulmonary inflammation, improved respiratory mechanics with adiposity reduction, and microbiome-mediated changes in mucosal immunity—may explain the protective pneumonia signal in large real-world cohorts, though this mechanism requires prospective validation [11]. GRADE certainty of evidence GRADE was applied at the outcome-domain level rather than as a single overall certainty rating for the review. Evidence from uncontrolled case reports, case series, and spontaneous pharmacovigilance analyses was initially rated as very low certainty because of serious risk of bias, absence of denominators or comparators, imprecision, and inability to establish causality. Upgrading was not applied to these evidence streams. RCT evidence was rated starting from high and downgraded according to GRADE domains. The complete GRADE evidence profile is presented in Table 4. Table 4GRADE evidence profile — GLP-1 receptor agonists and pulmonary adverse eventsOutcomeStudy Design (# Studies; N)Downgrading DomainsRisk of BiasInconsistencyIndirectnessImprecisionGRADE CertaintyKey FindingUpper respiratory tract infections vs. comparatorRCT + Cohort (4 RCTs + 2 cohorts; n ≈ 3,526)Open-label adverse-event ascertainment (− 1)Some concernsNot seriousNot seriousNot serious⊕⊕⊕○ MODERATEConsistent, mild, comparable to controls across randomised controlled trialsAny pulmonary adverse event — overall incidenceRCT + Cohort (4 RCTs + 4 cohorts; n≈808 K)Risk of bias (− 1); Inconsistency (− 1)Moderate–SeriousSeriousNot seriousNot serious⊕⊕○○ LOWRCTs: no serious adverse events; cohorts: heterogeneous real-world signalsPneumonia incidence (GLP-1 RA vs. DPP-4i)Retrospective Cohort (2 cohorts; n≈800 K)Risk of bias (− 1); Indirectness (− 1)SeriousNot seriousSerious (non-inert comparator)Not serious⊕⊕○○ LOWHenney 2024: HR 0.60 (95% CI 0.58–0.62); consistent signal across subgroups; causal inference limitedAspiration pneumonitis / pneumoniaCase series + pharmacovigilance (3 PV + 2 case series; n≈500 K reports)Risk of bias (− 2); Indirectness (− 1)Very seriousSeriousSeriousSerious⊕○○○ VERY LOWCase series confirm perioperative risk; Nakhla 2025 found no positive aspiration disproportionality signal for GLP-1 RAs across multiple databasesAnaphylaxis / immune-mediated pulmonary adverse eventCase reports (4 case reports; n = 4)Risk of bias (− 1); Indirectness (− 1); Imprecision (− 1)SeriousNot seriousSeriousVery serious⊕○○○ VERY LOWMechanism well-described; exendin-4 agents carry highest risk; intra-class heterogeneity documentedAcute respiratory failure / ARDSCase reports (2 case reports; n = 2)Risk of bias (− 1); Indirectness (− 1); Imprecision (− 1)SeriousNot seriousSeriousVery serious⊕○○○ VERY LOWARDS confirmed in 2 cases; 1 fatal; very sparse dataRespiratory mortalityPharmacovigilance + case reports (3 PV + 1 case report)Risk of bias (− 2); Indirectness (− 2)Very seriousSeriousVery seriousVery serious⊕○○○ VERY LOWFAERS mortality signals are not definitively attributable to GLP-1 RA exposure; 1 confirmed respiratory deathDrug discontinuation due to respiratory adverse eventMixed (6 studies; n mixed)Risk of bias (− 2); Inconsistency (− 1)Very seriousSeriousSeriousSerious⊕○○○ VERY LOWLow incidence in randomised trials; documented in 6/9 reporting studies (predominantly case reports)⊕⊕⊕⊕ = HIGH; ⊕⊕⊕○ = MODERATE; ⊕⊕○○ = LOW; ⊕○○○ = VERY LOW. GRADE applied at outcome-domain level. Uncontrolled designs (case reports, case series, pharmacovigilance) start at VERY LOW certainty before upgrading considerations. Downgrading domains: risk of bias, inconsistency, indirectness, imprecision, publication bias Table 4GRADE evidence profile — GLP-1 receptor agonists and pulmonary adverse eventsOutcomeStudy Design (# Studies; N)Downgrading DomainsRisk of BiasInconsistencyIndirectnessImprecisionGRADE CertaintyKey FindingUpper respiratory tract infections vs. comparatorRCT + Cohort (4 RCTs + 2 cohorts; n ≈ 3,526)Open-label adverse-event ascertainment (− 1)Some concernsNot seriousNot seriousNot serious⊕⊕⊕○ MODERATEConsistent, mild, comparable to controls across randomised controlled trialsAny pulmonary adverse event — overall incidenceRCT + Cohort (4 RCTs + 4 cohorts; n≈808 K)Risk of bias (− 1); Inconsistency (− 1)Moderate–SeriousSeriousNot seriousNot serious⊕⊕○○ LOWRCTs: no serious adverse events; cohorts: heterogeneous real-world signalsPneumonia incidence (GLP-1 RA vs. DPP-4i)Retrospective Cohort (2 cohorts; n≈800 K)Risk of bias (− 1); Indirectness (− 1)SeriousNot seriousSerious (non-inert comparator)Not serious⊕⊕○○ LOWHenney 2024: HR 0.60 (95% CI 0.58–0.62); consistent signal across subgroups; causal inference limitedAspiration pneumonitis / pneumoniaCase series + pharmacovigilance (3 PV + 2 case series; n≈500 K reports)Risk of bias (− 2); Indirectness (− 1)Very seriousSeriousSeriousSerious⊕○○○ VERY LOWCase series confirm perioperative risk; Nakhla 2025 found no positive aspiration disproportionality signal for GLP-1 RAs across multiple databasesAnaphylaxis / immune-mediated pulmonary adverse eventCase reports (4 case reports; n = 4)Risk of bias (− 1); Indirectness (− 1); Imprecision (− 1)SeriousNot seriousSeriousVery serious⊕○○○ VERY LOWMechanism well-described; exendin-4 agents carry highest risk; intra-class heterogeneity documentedAcute respiratory failure / ARDSCase reports (2 case reports; n = 2)Risk of bias (− 1); Indirectness (− 1); Imprecision (− 1)SeriousNot seriousSeriousVery serious⊕○○○ VERY LOWARDS confirmed in 2 cases; 1 fatal; very sparse dataRespiratory mortalityPharmacovigilance + case reports (3 PV + 1 case report)Risk of bias (− 2); Indirectness (− 2)Very seriousSeriousVery seriousVery serious⊕○○○ VERY LOWFAERS mortality signals are not definitively attributable to GLP-1 RA exposure; 1 confirmed respiratory deathDrug discontinuation due to respiratory adverse eventMixed (6 studies; n mixed)Risk of bias (− 2); Inconsistency (− 1)Very seriousSeriousSeriousSerious⊕○○○ VERY LOWLow incidence in randomised trials; documented in 6/9 reporting studies (predominantly case reports)⊕⊕⊕⊕ = HIGH; ⊕⊕⊕○ = MODERATE; ⊕⊕○○ = LOW; ⊕○○○ = VERY LOW. GRADE applied at outcome-domain level. Uncontrolled designs (case reports, case series, pharmacovigilance) start at VERY LOW certainty before upgrading considerations. Downgrading domains: risk of bias, inconsistency, indirectness, imprecision, publication bias GRADE evidence profile — GLP-1 receptor agonists and pulmonary adverse events ⊕⊕⊕⊕ = HIGH; ⊕⊕⊕○ = MODERATE; ⊕⊕○○ = LOW; ⊕○○○ = VERY LOW. GRADE applied at outcome-domain level. Uncontrolled designs (case reports, case series, pharmacovigilance) start at VERY LOW certainty before upgrading considerations. Downgrading domains: risk of bias, inconsistency, indirectness, imprecision, publication bias GRADE certainty: MODERATE for URTIs (consistent across RCTs, downgraded for open-label design); LOW for overall pulmonary adverse event incidence and the pneumonia risk reduction signal; VERY LOW for aspiration pneumonia, anaphylaxis, eosinophilic pneumonia, acute respiratory failure, drug discontinuation, and respiratory mortality. Fig. 1PRISMA 2020 flow diagram — study selection process. The diagram illustrates: identification (1,271 records from PubMed, Embase, Scopus; 483 duplicates removed), screening (788 records screened; 742 excluded by title/abstract), eligibility (46 full-text articles assessed; 27 excluded: no primary respiratory AE data n = 14, non-GLP-1 therapy only n = 6, paediatric population n = 3, narrative review/editorial n = 4), and inclusion (19 studies). Inter-rater agreement: κ = 0.91 at title/abstract screening; κ = 0.88 at full-text assessment PRISMA 2020 flow diagram — study selection process. The diagram illustrates: identification (1,271 records from PubMed, Embase, Scopus; 483 duplicates removed), screening (788 records screened; 742 excluded by title/abstract), eligibility (46 full-text articles assessed; 27 excluded: no primary respiratory AE data n = 14, non-GLP-1 therapy only n = 6, paediatric population n = 3, narrative review/editorial n = 4), and inclusion (19 studies). Inter-rater agreement: κ = 0.91 at title/abstract screening; κ = 0.88 at full-text assessment Fig. 2Forest plot of pulmonary adverse-event hazard ratios — cohort studies. Individual study estimates for four cohort studies plotted on a log-scale HR axis; estimates presented individually and not pooled owing to heterogeneous comparators, populations, and outcomes. Henney 2024: HR 0.60, 95% CI 0.58–0.62 (pneumonia, vs. DPP-4i; n = 331,863). Foer 2023: HR 0.72, 95% CI 0.51–1.01 (respiratory exacerbations, vs. active comparators; n = 1,642). Rashid 2025: HR 1.28, 95% CI 0.98–1.67 (aspiration pneumonia, perioperative; n = 138,980). Fishman 2025: HR 1.15, 95% CI 0.62–2.14 (pneumonia, post-transplant, no control). GRADE: LOW for pneumonia protective signal; VERY LOW for aspiration risk Forest plot of pulmonary adverse-event hazard ratios — cohort studies. Individual study estimates for four cohort studies plotted on a log-scale HR axis; estimates presented individually and not pooled owing to heterogeneous comparators, populations, and outcomes. Henney 2024: HR 0.60, 95% CI 0.58–0.62 (pneumonia, vs. DPP-4i; n = 331,863). Foer 2023: HR 0.72, 95% CI 0.51–1.01 (respiratory exacerbations, vs. active comparators; n = 1,642). Rashid 2025: HR 1.28, 95% CI 0.98–1.67 (aspiration pneumonia, perioperative; n = 138,980). Fishman 2025: HR 1.15, 95% CI 0.62–2.14 (pneumonia, post-transplant, no control). GRADE: LOW for pneumonia protective signal; VERY LOW for aspiration risk Discussion This systematic review, the most comprehensive synthesis to date of pulmonary adverse events reported in association with GLP-1 RA therapy, identified a complex and stratified respiratory safety profile across 19 studies encompassing over 1.1 million participants and nearly 500,000 pharmacovigilance reports. Given that the majority of included evidence consists of non-comparative, observational, and spontaneous-reporting data, the primary inferential framework of this review is one of signal detection and evidence mapping rather than causal inference. Three overlapping themes emerged, each warranting different levels of clinical and scientific weight. Common upper respiratory symptoms—URTIs, nasopharyngitis, and cough—represent a consistently documented but clinically mild class effect, comparable to controls, supported by moderate-certainty evidence from RCTs. Rare but potentially life-threatening adverse events—including anaphylaxis with bronchospasm, acute eosinophilic pneumonia, perioperative aspiration pneumonitis, and ARDS—have been reported in temporal association with GLP-1 RA exposure in case reports and cohort studies, underpinned by very-low-certainty evidence, yet carrying significant clinical consequences per event. The third theme, a paradoxical large-scale protective pneumonia signal, challenges any simplistic framing of GLP-1 RAs as uniformly deleterious to pulmonary health, though its causal status remains uncertain. Perioperative aspiration risk: clinical implications The risk of perioperative aspiration requires particular clinical attention. GLP-1 RA-mediated gastroparesis is pharmacologically predictable and dose-dependent, most pronounced with long-acting subcutaneous weekly formulations; semaglutide has a half-life of approximately seven days, compared with shorter-acting daily agents [18, 19]. Standard nil-by-mouth protocols, developed for populations with normal gastric emptying physiology, cannot reliably ensure gastric decompression in GLP-1 RA users. The case series by Avraham 2024 illustrated this concretely [18]. The American Society of Anesthesiologists (ASA) 2023 consensus guidance recommends withholding GLP-1 RAs for at least one full dosing interval before elective procedures. For weekly semaglutide, this implies a minimum seven-day drug-free window—a recommendation that requires anticipatory coordination between prescribers and surgical teams. When GLP-1 RA use cannot be confirmed or fasting adequacy is uncertain, rapid sequence induction with cricoid pressure should be the default approach. Point-of-care gastric ultrasound assessment immediately before induction offers a practical means of evaluating residual gastric volume. Post-transplant recipients and critically ill patients represent particularly high-risk subgroups in which GLP-1 RA-associated aspiration warrants active vigilance. Exendin-4 versus human GLP-1 analogue structural pharmacology The clustering of immune-mediated adverse pulmonary events—anaphylaxis, angioedema, bronchospasm, eosinophilic pneumonia—predominantly with exendin-4-based agents (exenatide, lixisenatide) rather than human GLP-1 analogues (liraglutide, semaglutide, dulaglutide) is neither coincidental nor mechanistically obscure. Exendin-4, a native peptide isolated from the Gila monster Heloderma suspectum, shares only 50% amino acid sequence homology with endogenous human GLP-1 [1]. This structural divergence confers substantially greater immunogenic potential, manifesting as anti-drug antibody formation and IgE-mediated sensitisation. Human GLP-1 analogues share ≥ 90% sequence homology with endogenous GLP-1, resulting in a lower immunogenic burden. These data support a risk-stratification framework in which exendin-4-based agents are considered to carry higher reported risk for immune-mediated pulmonary events, though the absolute incidence of these events in treated populations cannot be determined from the available evidence. Clinicians initiating exenatide or lixisenatide therapy should counsel patients on the reported risk of anaphylaxis and the importance of seeking early medical attention for dyspnoea, bronchospasm, or angioedema. Patients who develop hypersensitivity to an exendin-4-based agent should not automatically be considered allergic to the entire GLP-1 RA class; a supervised transition to a human GLP-1 analogue may be clinically appropriate [10]. The protective pneumonia signal: mechanistic considerations and limitations The reported 40% relative risk reduction in incident pneumonia with GLP-1 RAs versus DPP-4 inhibitors observed by Henney et al. 2024 is biologically plausible [11]. The proposed mechanisms are multiple: obesity-related improvements in respiratory mechanics accompanying GLP-1 RA-mediated weight loss; reduction of gastroesophageal reflux-mediated microaspiration; direct anti-inflammatory GLP-1R activation in alveolar macrophages; and potential microbiome-mediated changes in mucosal immunity. However, important methodological limitations substantially constrain causal inference and deserve emphasis as primary considerations. DPP-4 inhibitors increase endogenous GLP-1 bioavailability, making them a non-inert comparator that may attenuate the apparent advantage of exogenous GLP-1 RA therapy. Residual confounding by indication persists as a legitimate concern. Differential weight trajectory over follow-up cannot be excluded as an independent confounder in a retrospective electronic health record analysis. The magnitude and statistical robustness of this signal are notable, but prospective, randomised trial-level data with pre-specified, validated respiratory endpoints are required before this pneumonia risk reduction can be accepted as a causal effect. Implications for real-world practice and future research priorities The rapidly expanding real-world prescription of GLP-1 RAs—now encompassing patients with asthma, COPD, obstructive sleep apnoea, and post-transplant immunosuppression—renders a systematic, prospective approach to respiratory surveillance no longer optional. Current evidence is heterogeneous in design, inconsistent in outcome definition, and disproportionately reliant on pharmacovigilance signal detection rather than incidence estimation. Several future research priorities are identified. First, head-to-head comparative respiratory safety trials between exendin-4-based and human GLP-1 analogue agents are needed to quantify class-level heterogeneity. Second, prospective perioperative cohort studies with standardised gastric ultrasound protocols would address the aspiration risk question with the methodological rigour that clinical stakes demand. Third, designated respiratory safety substudies embedded within ongoing phase 3 cardiovascular outcome trials represent a tractable and cost-efficient mechanism for generating high-certainty evidence. Fourth, pharmacovigilance analyses incorporating drug utilisation denominators—currently absent from all three pharmacovigilance studies included in this review—are necessary to move from disproportionality signals to actionable incidence estimates. Strengths and limitations This systematic review has several strengths. The multi-database search strategy was comprehensive, eligibility criteria and analysis plans were pre-specified, and risk of bias assessment employed validated design-appropriate tools (RoB2, NOS, JBI). Evidence certainty was appraised using the GRADE framework applied at the outcome-domain level, and adherence to both PRISMA 2020 and SWiM reporting standards was maintained throughout. With over 1.1 million participants across four study design categories and nine GLP-1 RA agents, this review represents the most comprehensive synthesis of pulmonary GLP-1 RA safety to date. The acknowledged limitations are equally important. First, the substantial methodological heterogeneity across study designs, GLP-1 RA agents, populations, and outcome definitions precluded formal meta-analytic pooling for most outcomes. Second, the three pharmacovigilance studies—which contributed the largest number of records—carried inherent structural high risk of bias: underreporting, stimulated-reporting bias, absence of drug utilisation denominator data, and fundamental inability to establish causality. Their findings were appropriately interpreted as signal generation only. Third, nine case reports and two conference abstracts contributed descriptive evidence of very low generalisability. Fourth, the absence of a universal comparator across studies limits between-study comparability at a foundational level. Fifth, publication bias favouring dramatic adverse events may meaningfully inflate the perceived incidence of rare complications; formal correction for publication bias was not feasible given the non-comparative and heterogeneous evidence base. Sixth, several studies did not provide complete data despite author contact attempts; these were recorded as not reported. Conclusion GLP-1 receptor agonist therapy has a respiratory safety profile that is neither uniformly benign nor hazardous, stratified by mechanism, agent subclass, and clinical context. Common upper airway symptoms represent a mild, class-wide effect comparable to controls, supported by moderate-certainty evidence from RCTs. Rare but clinically consequential events—including anaphylaxis, eosinophilic pneumonia, perioperative aspiration, and ARDS—have been reported in temporal association with GLP-1 RA exposure, predominantly in case reports and pharmacovigilance data of very low certainty; these signals are hypothesis-generating and should not be interpreted as confirmed causal risks or quantified incidence estimates. A 40% relative reduction in pneumonia incidence was observed in a large retrospective cohort (HR 0.60, 95% CI 0.58–0.62), constituting an intriguing signal that complicates any simple adverse-risk framing; however, low-certainty evidence, an active and non-inert comparator, and potential residual confounding preclude causal attribution at this stage. Clinically, the most actionable priorities are: subclass risk stratification for hypersensitivity, with exendin-4-based agents carrying the highest reported risk; extended pre-procedural fasting intervals calibrated to the specific GLP-1 RA agent’s pharmacokinetic half-life; and gastric ultrasound assessment in high-risk perioperative patients when fasting adequacy is uncertain. Scientifically, prospective, standardised, and adequately powered studies with pre-specified respiratory endpoints remain absent—a gap that the scale of global GLP-1 RA prescribing has made untenable to sustain. This systematic review provides a structured evidence map that may serve as a foundation for designing such studies and for informing regulatory and clinical risk-benefit assessments.