Introduction

The rapid expansion of incretin-based therapies has reshaped the pharmacological management of obesity and type 2 diabetes mellitus [1]. Glucagon-like peptide-1 recep-tor agonists (GLP-1 RAs) are now central to the treatment of both conditions, providing clinically meaningful improve-ments in glycemic control and body weight reduction [1–4]. The therapeutic landscape has evolved quickly with the introduction of more potent agents, including semaglutide and the dual glucose-dependent insulinotropic polypeptide (GIP)/GLP-1 receptor agonist tirzepatide, which have dem-onstrated superior efficacy in randomized controlled tri-als [1, 5, 6]. Despite robust efficacy in controlled settings, real-world use of GLP-1 RAs is frequently limited by poor long-term persistence [7–10]. In Poland, available popula-tion-level data suggest low treatment continuation, which may attenuate expected clinical benefits [6]. Moreover, uptake for weight management among individuals with-out diabetes remains relatively low, reflecting a complex interplay of access barriers, reimbursement limitations, and patient preferences [3, 11–14]. These observations under-score that treatment trajectories in routine clinical practice often diverge substantially from those observed in trials [1,3, 7, 10, 12]. As the number of available agents and for-mulations continues to grow, questions regarding treatment sequencing, persistence, and real-world patterns of use have become increasingly relevant [1].

    Within this context, therapeutic switching between GLP-1 RAs represents an important but understudied phe-nomenon [15]. Switching is conceptually distinct from dis-continuation: it reflects a modification of therapy within the incretin class and may arise in relation to treatment experi-ence, formulation preferences, or external prescribing con-straints [10, 15, 16]. Prior real-world evidence indicates that switching is common and may serve as a pragmatic alterna-tive to discontinuation or treatment intensification [12, 15]. However, local switching patterns - particularly in settings where prescribing is strongly shaped by formulation avail-ability and market dynamics - remain poorly characterized [17]. In Poland, GLP-1 RA selection may be influenced by the availability of specific formulations (including oral vs. subcutaneous preparations), evolving reimbursement policies, and supply fluctuations [11, 13]. Recent work has described distinct preference patterns for oral versus subcu-taneous semaglutide that appear sensitive to reimbursement status and market conditions [13]. Nevertheless, large-scale analyses characterizing GLP-1 RA switching and its tem-poral dynamics in a private, predominantly nonreimbursed healthcare setting are lacking.

    We therefore aimed to characterize GLP-1 RA switch-ing trajectories in a large cohort from the LUX MED net-work, a major private healthcare provider in Poland. Using prescription sequences spanning 2018 through 2025, we quantified switching frequency, described therapy-to-ther-apy transitions, and evaluated associations between cur-rent therapy, patient characteristics, and switching while explicitly accounting for switching opportunity and calen-dar time. This approach provides a real-world description of how incretin-based therapies are used in clinical practice and how switching patterns vary over time, including after the market entry of newer agents. The study was designed to characterize switching structure and temporal patterns rather than to determine causal drivers of switching or opti-mal therapeutic sequencing.

Methods

This retrospective cohort study used anonymized elec-tronic health record (EHR) data from LUX MED, a large private healthcare provider in Poland, spanning January 1, 2018, through December 31, 2025 [8, 11, 13, 18]. Data were extracted from the enterprise EHR system, which cov-ers approximately 300 outpatient clinics nationwide. We assembled a utilization cohort of adults (aged≥18 years) with at least two recorded GLP-1 RA prescriptions, enabling observation of changes between consecutive prescriptions. Switching was assessed from prescription sequences recon-structed using drug-name fields and their corresponding ser-vice dates. Because linked pharmacy dispensing data were unavailable and continuous treatment exposure could not be directly observed, the prescription-to-prescription inter-val was treated as the primary observable decision unit for switching analyses. GLP-1 RA exposure included all formu-lations available in Poland during the study period: inject-able agents (semaglutide [Ozempic/Wegovy], liraglutide [Saxenda/Victoza], dulaglutide [Trulicity], and tirzepatide [Mounjaro]) and oral semaglutide (Rybelsus). Demographic characteristics (age and sex) were determined at the index date (first recorded GLP-1 RA prescription). Comorbidities defined by International Classification of Diseases, Tenth Revision (ICD-10) codes were treated as binary indicators and included type 1 diabetes mellitus (E10), type 2 diabetes mellitus (E11), arterial hypertension (I10), coronary artery disease (I25), atrial fibrillation or flutter (I48), chronic heart failure (I50), obstructive sleep apnea (G47.3), dyslipidemia (E78), chronic kidney disease (N18), and acute pancreatitis (K85); these variables serve as clinical proxies and do not capture disease severity, duration, or treatment indication. Because routine EHR coding may involve overlap between E10 and E11, these diagnoses were not treated as definitive phenotypic classifications but as recorded comorbidity indi-cators within the source database. The study was designed as a utilization and switching analysis rather than an effec-tiveness study. Longitudinal efficacy endpoints (e.g., weight change, glycated hemoglobin trajectories, or symptom measures) were not extracted, because outpatient diagnos-tic reporting is incompletely structured and sporadic, and adherence metrics such as the proportion of days covered could not be calculated without linked pharmacy dispens-ing data. In addition, the spacing between prescriptions is irregular in routine care and may partly reflect the obser-vation process itself; therefore, switching analyses were interpreted as conditional on recorded follow-up within the network rather than as fully observed continuous-time treat-ment trajectories.

    The study protocol (No. 62/2024) was approved by the Bioethics Committee at Jan Kochanowski University in Kielce, Poland. The study was conducted in accordance with the Declaration of Helsinki, and the requirement for informed consent was waived owing to the retrospective design and use of anonymized data.

Statistical methods

Continuous variables were assessed for distributional char-acteristics and are presented as mean (SD) or median (IQR), as appropriate. Between-group comparisons used the Welch t test for approximately normally distributed variables and the Mann–Whitney U test otherwise. Categorical variables are reported as n (%) and compared using the chi-square test or Fisher exact test when expected cell counts were less than 5. Because treatment exposure in this database was observed through prescription sequences rather than through dispensing-confirmed continuous use, the primary inferential analysis used a transition-level discrete-time haz-ard framework in which each prescription-to-prescription interval (Rx_t→Rx_t+1) contributed one observation, and the dependent variable was switching (yes/no) at that inter-val. This time scale was selected because the next recorded prescription represents the most reliable observable clinical decision point at which switching can be identified. Current therapy (the “from” agent) was modeled as a categorical exposure, with subcutaneous semaglutide as the reference. The primary model adjusted for switching opportunity using the transition number and for temporal market dynamics using calendar time, modeled with month-of-year fixed effects and a linear time trend. Baseline covariates included age, sex, body mass index (BMI) at the first prescrip-tion, type 2 diabetes mellitus, hypertension, dyslipidemia, chronic kidney disease, heart failure, sleep apnea, coronary artery disease, and atrial fibrillation or flutter. Inference used patient-clustered robust standard errors to account for within-patient correlation across multiple observed transi-tions. Because prescription intervals are irregularly spaced in routine care, the transition-level model was interpreted as estimating switching propensity at the next observed prescription, conditional on continued follow-up within the network, rather than a fully observed continuous-time treat-ment hazard. Several sensitivity analyses were performed to assess robustness. First, the elapsed time since the index prescription (in days) was added to the transition-level model. Second, calendar time was alternatively modeled as a restricted cubic spline over calendar month to allow for nonlinear temporal dynamics. Third, an additional transi-tion-level sensitivity analysis incorporated both the elapsed time since index therapy and the interval length between consecutive prescriptions to further assess the influence of irregular spacing. Fourth, a Cox proportional-hazards model was fitted for the time to the first observed switch, with censoring at the last recorded GLP-1 RA prescription date. Because the transition-level model is anchored on current therapy and evaluates switching at each observed prescrip-tion interval, whereas the Cox model is anchored on index therapy and evaluates the time to the first observed switch, direct numerical concordance between these estimates was not expected — particularly for late-entry therapies with shorter observation windows. All results are reported as odds ratios (OR) or hazard ratios (HR) with 95% confidence intervals (CI). All tests were two-sided, and statistical sig-nificance was set at an alpha level of 0.05. Analyses were performed in Python version 3.11.2 (pandas 2.2.3, NumPy 1.24.0, SciPy 1.14.1, statsmodels 0.14.3).

Results

The study population comprised 54,260 patients with at least one recorded GLP-1 RA prescription. Among the 42,423 patients with more than one recorded prescription (switch-observable cohort), 29.7% experienced at least one switch and 14.3% had two or more switches. The distribution of index therapies is shown in Table 1. The median time from the first prescription to the first observed switch was 307 days (148–574). First-switch transitions were dominated by moves toward semaglutide and tirzepatide, particularly from liraglutide and dulaglutide (Fig. 1, Supplementary Fig-ure S1). The first-switch matrix provides structural context for the transition-level model by illustrating the dominant therapy-to-therapy pathways within the observed switching

    Baseline characteristics of the switch-observable cohort are summarized in Table 1. Compared with the no-switch group (n=29,803), patients who switched (n=12,620) were slightly older and more frequently female. Cardiometabolic comorbidities—including obesity, type 2 diabetes mel-litus, hypertension, dyslipidemia, chronic kidney disease, and obstructive sleep apnea—were more prevalent among patients who switched. BMI at the first prescription was modestly higher in switchers (33.9 [5.7] vs. 33.4 [5.8] kg/ m²; P<0.001). Switchers also had substantially greater treat-ment exposure, reflected by a higher total prescription count (11 [7–19] vs. 4 [3–8]; P<0.001) and an earlier index year (2022 [2022–2023] vs. 2024 [2023–2025]; P<0.001), con-sistent with longer available observation time among earlier initiators. The distribution of index therapy also differed between groups (P<0.001): switching was most frequent among initiators of liraglutide or dulaglutide and least fre-quent among tirzepatide initiators. Multiple switching (≥2 switches) was observed in 6074 of 42,423 patients (14.3%), representing 48.1% of those with at least one switch. Patients with multiple switches had substantially higher prescription

counts and earlier index years, with modest differences in age and body mass index (Supplementary Table S1).

    In the transition-level analysis, 12,620 patients contrib-uted 27,095 prescription-to-prescription transitions. After adjustment for switching opportunity (transition number) and calendar time (month-of-year fixed effects and a linear time trend), with patient-clustered robust standard errors and subcutaneous semaglutide as the reference, current therapy remained strongly associated with switching at the subsequent transition (Fig. 2, Supplementary Table S2). Compared with subcutaneous semaglutide, liraglutide was associated with substantially lower odds of switching (OR, 0.02; 95% CI, 0.01–0.03; P<0.001). Dulaglutide showed a nonsignificant trend toward higher switching odds (OR, 1.70; 95% CI, 0.95–3.04; P=0.07), whereas oral semaglu-tide did not differ significantly (OR, 1.30; 95% CI, 0.78–2.17; P=0.31). Calendar time was independently associated with switching (OR, 1.04 per month; 95% CI, 1.04–1.05; P<0.001), indicating substantial temporal variation beyond patient-level characteristics. Among baseline covariates, older age was associated with higher switching odds (OR, 1.17 per 10 years; 95% CI, 1.06–1.29; P=0.002), whereas coronary artery disease was associated with lower odds (OR, 0.50; 95% CI, 0.29–0.87; P=0.01). Sex (OR, 0.84; 95% CI, 0.67–1.04; P=0.11), BMI (OR, 1.00 per kg/m²; 95% CI, 0.99–1.02; P=0.70), type 2 diabetes mellitus (OR, 1.19; 95% CI, 0.96–1.47; P=0.11), and other comorbidities were not significantly associated with switching. Estimates for tirzepatide as current therapy were not estimable owing to quasi-complete separation (very few or no nonswitch-ing transitions in this subgroup). This non-estimability precluded comparative interpretation of current tirzepatide within the transition-level model; therefore, this exposure was retained for descriptive completeness but was not used to support inferences regarding relative switching stability.

    Sensitivity analyses yielded directionally consistent results. In the extended transition-level model, the time since the index prescription was independently associated with switching (OR, 1.04 per 30 days; 95% CI, 1.03–1.05; P<0.001), whereas the principal therapy associations remained qualitatively unchanged. In a sensitivity analy-sis in which the linear calendar-time trend was replaced with a restricted cubic spline over calendar month, the principal therapy associations again remained qualita tively unchanged. In an additional sensitivity analysis incorporating both the elapsed time since index therapy and the interval length between consecutive prescriptions, longer prescription gaps were independently associated with switching, whereas the principal therapy associations remained directionally unchanged (Supplementary Table S3). As a complementary survival-analytic sensitivity analy-sis, we fitted a Cox proportional-hazards model for the time to the first observed switch (censoring at the last recorded GLP-1 RA prescription date; 12,264 events among 12,620 patients). Relative to subcutaneous semaglutide, initiation with liraglutide (HR, 0.71; 95% CI, 0.67–0.74; P<0.001)

and dulaglutide (HR, 0.72; 95% CI, 0.68–0.76; P<0.001) was associated with a lower hazard of first switch, whereas oral semaglutide was associated with a higher hazard (HR, 1.31; 95% CI, 1.24–1.39; P<0.001). Initiation with tirzepa-tide was also associated with a higher hazard of first switch (HR, 2.38; 95% CI, 2.11–2.68; P<0.001), consistent with strong time- and market-entry effects on observed switch-ing. In this model, older age (HR, 0.94 per 10 years; 95% CI, 0.93–0.96; P<0.001) and type 2 diabetes mellitus (HR, 0.84; 95% CI, 0.81–0.88; P<0.001) were associated with a lower hazard of first switch, whereas sex and BMI were not significantly associated (Supplementary Table S4). Direc-tionally similar therapy contrasts were also observed in stratified analyses of patients with and without type 2 diabe-tes, as well as after exclusion of patients with any E10 code.

    Switching patterns differed according to type 2 diabetes status and BMI. Among patients without type 2 diabetes, the most frequent first switches were liraglutide to subcuta-neous semaglutide (n=1,549) and liraglutide to tirzepatide (n=891), followed by subcutaneous semaglutide to oral semaglutide (n=697). Among patients with type 2 diabetes, the most common transitions were dulaglutide to subcuta-neous semaglutide (n=888) and subcutaneous semaglutide to oral semaglutide (n=611), followed by subcutaneous semaglutide to dulaglutide (n=529) and oral semaglutide to subcutaneous semaglutide (n=514); class-level first-switch distributions differed between type 2 diabetes strata (P<0.001). When stratified by BMI, liraglutide to sub-cutaneous semaglutide was the most frequent first switch in both the BMI less than 30 kg/m² and BMI 30 kg/m² or greater groups (n=436 and n=1,242, respectively), whereas

the BMI 30 kg/m² or greater group showed relatively more liraglutide-to-tirzepatide and dulaglutide-to-subcutaneous-semaglutide transitions; class-level distributions differed modestly between BMI strata (P=0.014). Among patients with two or more switches, motif analysis demonstrated a predominance of oscillatory, bidirectional trajectories. At the class level (Fig. 3A), the most frequent motif was sema-glutide to liraglutide/dulaglutide to semaglutide (16.4%), followed by liraglutide/dulaglutide to semaglutide to lira-glutide/dulaglutide (12.9%) and a four-state extension of the same pattern (9.5%), indicating frequent cycling rather than progressive escalation. At the formulation-aware level (Fig. 3B), subcutaneous semaglutide acted as a central hub, with common motifs including subcutaneous semaglutide to dulaglutide to subcutaneous semaglutide (4.5%) and subcutaneous semaglutide to tirzepatide to subcutaneous semaglutide (4.0%), alongside recurrent oral semaglutide intermediates, consistent with periodic trials of alternative formulations between injectable agents. The motif analysis

provides structural context for the regression framework by showing that repeated switching was frequently oscillatory rather than unidirectional.

    Calendar-time analyses of transition rates revealed clear temporal variability (Fig. 4A). Overall switching intensity displayed distinct peaks, and subcutaneous-to-oral sema-glutide transitions showed intermittent increases (Fig. 4B), a pattern compatible with episodic substitution away from injectable semaglutide during discrete time windows. In parallel, transitions terminating in tirzepatide increased during specific periods after market introduction (Fig. 4C), suggesting accelerated uptake during time-limited inter vals. Denominators were displayed alongside rates, and a 3-month rolling mean was used to distinguish sustained shifts from isolated monthly spikes.

Discussion

In this large real-world analysis from a private healthcare network in Poland, switching among GLP-1 receptor ago-nists was frequent and temporally dynamic. Because only approximately one-third of patients had type 2 diabetes mellitus, the observed treatment trajectories likely reflect obesity management and weight-focused prescribing in addition to glycemic intensification. Consistent with prior real-world reports [12, 15, 16], we observed substantial het-erogeneity in switching patterns across agents, including prominent transitions toward semaglutide and tirzepatide and a sizeable subgroup of patients with repeated switch-ing. A contemporaneous US claims-based analysis by Xie et al. of 126,984 adults with overweight or obesity with-out diabetes reported a 12-month switching frequency of 20.6%, with liraglutide-to-semaglutide as the most common first transition and injectable semaglutide serving as a cen-tral switching hub [17]. Our findings extend this picture in several directions: the longer observation window captures the tirzepatide market-entry period more completely, the mixed-indication cohort reflects a broader prescribing pop-ulation, and the transition-level framework characterizesswitching propensity conditional on observed opportunities rather than treating switching as a single binary outcome within a fixed follow-up window.

    A central methodological challenge in switching research is that switching is inherently time-dependent and oppor-tunity-driven. Patient-level “ever-switched” outcomes are strongly influenced by observation intensity: patients with more recorded prescriptions and earlier index years have greater opportunity to switch. In the patient-level models, cumulative prescription count dominated the association with switching, supporting the concern that such models largely capture opportunity rather than clinical determi-nants. For this reason, we prioritized a transition-level (dis-crete-time hazard) framework that models the probability of switching at each prescription interval, conditions on observed opportunities, and adjusts for calendar time. In this time-aware approach, current therapy remained strongly associated with switching at the next interval compared with subcutaneous semaglutide, whereas most baseline comorbidities showed smaller or nonsignificant associa-tions. These findings should be interpreted as associations with switching propensity at a given prescription interval, not as causal effects of therapies or patient characteristics. The particularly large inverse contrast for liraglutide should not be interpreted as evidence of superior treatment stabil-ity or comparative effectiveness; rather, it likely reflects a combination of residual confounding, structural imbalance in observed prescription sequences, and the limitations of routine EHR-based transition data.

    The transition-level model and the Cox model capture complementary aspects of switching and therefore answer different questions. The transition-level model estimates switching propensity at the next observed prescription inter-val, conditional on current therapy and continued follow-up, whereas the Cox model estimates the time from index therapy to the first observed switch. For late-entry agents - particularly tirzepatide - these estimands are differentially affected by market-entry timing, censoring, and incomplete capture of prescriptions outside the network. We therefore interpret the Cox analysis as a robustness check for time structure rather than as an alternative primary comparison of therapy stability.

    The descriptive structure of switching provides additional context. First-switch matrices and Sankey diagrams showed that transitions frequently moved from older incretin thera-pies (liraglutide and dulaglutide) toward semaglutide and, subsequently, toward tirzepatide. Among patients with multiple switches, we observed a predominance of bidi-rectional, “return-to-origin” motifs, with cycling between semaglutide and liraglutide/dulaglutide classes accounting for a large share of trajectories [18–20]. Notably, Xie et al. reported that 92.7% of switchers in their 12-month window switched only once [17], whereas in our longer observation period nearly half of patients with any switch experienced two or more switches, and oscillatory “return-to-origin” motifs were prominent. This contrast likely reflects both the extended follow-up in the present analysis and the structural features of prescription-sequence data captured across mul-tiple calendar years, and it suggests that the cyclic nature of GLP-1 RA use becomes apparent only with sufficiently long observation. Such oscillatory patterns are consistent with nonlinear decision pathways in routine care, in which therapy changes may reflect periodic reassessment, toler-ability trade-offs, formulation preferences, or intermittent external constraints rather than a unidirectional escalation strategy [1, 18–20]. These oscillatory trajectories may also reflect broader clinical uncertainty about whether incretin-based therapy should be maintained continuously or used episodically, a question that remains unresolved even as the therapeutic armamentarium expands [1]. Figures 1 and 3 are intended as structural complements to the regression framework, illustrating the dominant first-switch pathways and recurrent oscillatory motifs within which the transition-level associations arise, rather than as independent infer-ential analyses. Importantly, these trajectory analyses are descriptive and do not establish mechanisms.

    Temporal analyses further indicated that switching pro-pensity varied over calendar time, with distinct peaks in overall switching intensity and in specific transitions (e.g., subcutaneous-to-oral semaglutide and transitions termi nating in tirzepatide) [15, 20–22]. These observations are

consistent with emerging evidence that rapid therapeutic innovation is associated with dynamic real-world treatment patterns in obesity care [17], with incumbent agents fre-quently displaced as newer, more efficacious formulations become available. Importantly, these findings were robust to a more flexible nonlinear specification of calendar time, modeled as a restricted cubic spline, which suggests that the principal therapy associations were not primarily driven by misspecification of a linear temporal trend. These time-limited patterns are compatible with episodes of substitution and adoption following market entry. However, because the dataset does not capture dispensing, individual-level reim-bursement or cost exposure, supply disruptions, or patient and clinician preferences, we cannot attribute these peaks to specific system-level drivers [23]. We therefore frame cost, availability, and tolerability as plausible hypotheses consistent with the observed time structure rather than as empirical conclusions [24–26]. Notably, supply shortages driven by rapidly increasing demand for GLP-1 RAs have been recognized as a growing concern at the class level [1, 27–29], lending plausibility to the hypothesis that at least some of the observed temporal clustering reflects access-related rather than clinically motivated switching.

    Several findings also warrant cautious interpretation in light of calendar-time and market-entry effects. Tirzepatide entered the market later and had shorter observable expo-sure windows within the study period. In the transition-level model, estimates for tirzepatide as current therapy were not estimable owing to quasi-complete separation (very few or no nonswitching transitions in that subgroup), which likely reflects a combination of short observation windows and data-capture characteristics rather than intrinsic treatment stability or instability. Cox models of time to first switch provide a familiar survival framework but are similarly sen-sitive to differential market exposure and incomplete capture of medication use outside the private network; accordingly, these were treated as sensitivity analyses rather than the pri-mary inferential approach.

Limitations and clinical implications

This study has several limitations inherent to retrospec--tive analyses of outpatient EHR prescription sequences. First, we did not capture reasons for switching (e.g., tol-erability, perceived effectiveness, patient preference, cost sharing, reimbursement changes, or supply constraints). Accordingly, observed temporal clustering and therapy-specific switching patterns should be interpreted as asso-ciations compatible with changes in access or preferences rather than as evidence of causal mechanisms. Second, switching is intrinsically time-dependent and opportunity-driven. Although patient-level “ever switched” models are highly sensitive to differential observation intensity, we mitigated this concern by prioritizing transition-level dis-crete-time hazard analyses that condition on each observed switching opportunity, adjust for calendar time, and use patient-clustered robust inference. Nevertheless, residual time-related bias cannot be fully excluded, particularly because prescriptions recorded in a private network may not capture all medication use if subsequent care occurs outside the system. Third, prescription entries reflect clini-cian orders rather than confirmed dispensing; therefore, we cannot confirm medication acquisition, adherence, or con-tinuous exposure, and apparent switches may partly reflect treatment gaps, intermittent use, or reinitiation rather than deliberate within-class substitution. Relatedly, censoring at the last recorded prescription may misclassify follow-up if patients continued therapy outside the network. In addition, the recorded prescription-sequence structure may incom-pletely capture silent continuation of unchanged therapy, which may amplify apparent imbalances between switching and non-switching transitions for some agents. The obser-vation process itself may also be informative, as patients with more frequent visits or prescriptions may differ sys-tematically with respect to disease complexity, access, or treatment behavior. Fourth, comorbidities were derived from ICD-10 codes and treated as binary proxies that do not capture disease severity, duration, or treatment indication; thus, residual clinical confounding is likely. A small pro-portion of patients (5.3%) carried a type 1 diabetes mellitus diagnosis, likely reflecting off-label prescribing and/or cod-ing imprecision; these patients were retained to reflect real-world prescribing patterns, but their inclusion should be considered when interpreting comorbidity-related findings. Finally, later-market therapies (particularly tirzepatide) have shorter observable exposure windows, which can influence both descriptive switching rates and time-to-event analy-ses despite calendar-time adjustment; for this reason, we avoided efficacy-based interpretation and reported instances of nonestimability transparently when data structure limited inference. Likewise, the large inverse contrast observed for

liraglutide should be understood as a non-causal association within the recorded transition structure, not as evidence of superior treatment stability or comparative effectiveness.

    From a clinical perspective, these findings underscore that switching among GLP-1 RAs in routine care is frequent and temporally dynamic, and that treatment stability may be influenced by a combination of patient factors and the broader prescribing environment. Consistent with recent US evidence [17], the frequent and often bidirectional switching patterns observed in our cohort are compatible with switch-ing as a form of active therapy management rather than treatment disengagement. Reframing within-class switch-ing as part of long-term therapeutic optimization - rather than as an early signal of treatment failure or abandonment - may better align clinical expectations with how GLP-1 RAs are used in routine practice, particularly in settings of rapid market evolution and formulation diversification. From a practical standpoint, clinicians should anticipate potential treatment interruptions or substitutions during periods of rapid market change or fluctuating availability and should consider proactive counseling, monitoring, and follow-up planning when initiating or modifying GLP-1 RA therapy. At the same time, the present data do not allow determina-tion of whether individual switches were clinically appro-priate, access-driven, or preference-driven. Future studies linking electronic health record prescriptions with dispens-ing data, reimbursement information, and patient-reported reasons for switching would be needed to distinguish clini-cal drivers from access- and system-related factors and to quantify their relative contributions.

Supplementary Information The online version contains

supplementary material available at https://doi.org/10.1007/s00592-0 26-02709-1.

Declarations

Conflict of interest The authors declare that they have no conflict of

Ethical approval Study has been approved by the appropriate ethics committee and has therefore been performed in accordance with the ethical standards laid down in the 1964 Declaration of Helsinki and its later amendments.

Declaration of generative AI use The authors declare that generative artificial intelligence (Claude Opus 4.6, Anthropic) was used during the preparation of this manuscript for language editing and stylistic improvement. Following its use, the authors reviewed and edited the content as needed and take full responsibility for the content of the published article.

Open Access This article is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License, which permits any non-commercial use, sharing, distribution and reproduction in any medium or format, as long as you give appropri-ate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if you modified the licensed material. You do not have permission under this licence to share adapted material derived from this article or parts of it. The images or other third party material in this article are included in the article’s Cre-ative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Com-mons licence and your intended use is not permitted by statutory regu-lation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by-nc-nd/4.0/. 

 This article is excerpted from the 《Acta Diabetologica》 by Wound World.