Temporal Trends and Clinical Characteristics of Incretin‐Based Therapy Use in Women With Polycystic Ovary Syndrome: A Real‐World Cohort Study From a Polish Private Healthcare Network

ABSTRACT

Background

Glucagon‐like peptide‐1 receptor agonists (GLP‐1 RAs) and the dual GIP/GLP‐1 receptor agonist tirzepatide are increasingly used for weight and cardiometabolic management, but real‐world prescribing patterns in women with polycystic ovary syndrome (PCOS) remain poorly characterised.

Methods

We performed a retrospective cohort study of 38 263 adult women with ICD‐coded PCOS in the LUX MED network, Poland, using electronic health records data from July 2006 to April 2026. The primary analysis included women diagnosed with PCOS in 2018–2024 and evaluated prescription‐recorded initiation of GLP‐1 RA/GIP–GLP‐1 RA therapy within 365 days after the index diagnosis. Temporal trends were assessed using logistic regression with unadjusted and multivariable‐adjusted models.

Results

Overall, 4557 women (11.9%) had a recorded GLP‐1 RA/GIP–GLP‐1 RA prescription and 16 381 (42.8%) received metformin. Among incretin‐based therapy users, 75.5% had metformin co‐exposure, whereas only 11.5% had coded type 2 diabetes mellitus. Recorded BMI was higher among GLP‐1 RA/GIP–GLP‐1 RA users than among metformin‐only or untreated women (median 31.2 vs. 26.3 and 22.1 kg/m²) and greater comorbidity burden (all p < 0.001). Initiation within 365 days rose from 0.14% in 2018 to 5.97% in 2024. Calendar year was strongly associated with initiation (unadjusted OR 1.52, 95% CI, 1.44–1.59; adjusted OR 1.64, 95% CI, 1.56–1.73; both p < 0.001), with consistent results after excluding women with diabetes.

Conclusions

Incretin‐based therapy use in PCOS increased markedly over time, suggesting expansion beyond classical glycaemic indications toward weight‐focused and cardiometabolic management. Studies using dispensing, persistence, safety and outcome data are needed.

1. Introduction

Polycystic ovary syndrome (PCOS) is one of the most common endocrine disorders in women of reproductive age, affecting an estimated 10%–13% of this population and is associated with a broad spectrum of reproductive, metabolic and psychological manifestations [1, 2, 3]. Beyond menstrual irregularity, hyperandrogenism and infertility, women with PCOS frequently present with increased adiposity, insulin resistance, dyslipidaemia, hypertension, obstructive sleep apnoea and depressive symptoms, and carry an elevated long‐term risk of type 2 diabetes mellitus and cardiovascular disease [3, 4, 5, 6]. The recent 2025 Lancet Diabetes and Endocrinology Commission has reframed PCOS explicitly as a clinical complication of obesity, reinforcing weight and cardiometabolic management as central components of PCOS care [7, 8].

Metformin has long been the principal pharmacological option for the metabolic manifestations of PCOS, supported by international guidelines as an adjunct to lifestyle modification, but its effect on body weight is generally modest, and many women remain inadequately controlled [4, 9]. Over the past decade, glucagon‐like peptide‐1 receptor agonists (GLP‐1 RAs) and more recently the dual glucose‐dependent insulinotropic polypeptide (GIP)/GLP‐1 receptor agonist tirzepatide, have transformed the pharmacological treatment of obesity and type 2 diabetes mellitus, producing clinically meaningful and sustained weight loss with concurrent improvements in glycaemic and cardiometabolic risk profiles [10, 11, 12, 13]. Randomized and observational studies in PCOS suggest favourable effects on body weight, insulin resistance, androgen levels and menstrual regularity, although evidence remains limited by small trial size, short follow‐up and incomplete reproductive‐outcome data [3, 14, 15, 16, 17].

Despite this clinical promise, real‐world prescribing patterns of incretin‐based therapy in women with PCOS remain poorly characterised. It is unclear how frequently these agents are used, whether prescribing is concentrated among women with coexisting type 2 diabetes mellitus or obesity, and how patterns have evolved with the adoption of more potent agents. Rapidly expanding use may reflect a clinically important shift from diabetes‐centred prescribing towards broader cardiometabolic and weight‐focused management in PCOS, with implications for clinical guidelines, reimbursement policy, drug supply and the design of future trials. We therefore characterized the real‐world use of GLP‐1 RA/GIP–GLP‐1 RA therapy in a large cohort of adult women with ICD‐coded PCOS from a major Polish private healthcare network, focusing on temporal trends, the cardiometabolic profile of treated women and the changing clinical phenotype of incretin‐based therapy initiators.

2. Materials and Methods

This was a retrospective observational cohort study based on anonymized electronic health record data from LUX MED, a large private healthcare provider in Poland that operates approximately 300 outpatient clinics nationwide [18, 19, 20, 21, 22]. The dataset comprised routinely collected outpatient information generated between July 2006 and April 2026. We included adult women with an International Classification of Diseases, Tenth Revision (ICD‐10) code for PCOS (E28.2); the index date was the first recorded PCOS diagnosis. The case definition was not validated against full clinical diagnostic criteria, and the cohort should therefore be interpreted as women with ICD‐coded rather than clinically adjudicated PCOS. No restrictions were applied based on prior therapy. Each patient is represented by a single record in the LUX MED electronic health record platform, identified by a unique patient identifier assigned at registration and maintained across all network facilities; encounters from patients attending more than one clinic are therefore merged into one longitudinal record rather than duplicated.

Drug exposure was ascertained from first‐prescription dates; throughout the manuscript, ‘use’ refers to prescription‐recorded use and does not imply dispensing, purchase, adherence, persistence, dosing or medication administration. GLP‐1 RA and dual GIP/GLP‐1 receptor agonist exposure included dulaglutide, liraglutide, semaglutide, tirzepatide and albiglutide. Metformin and other glucose‐lowering therapies—sodium‐glucose cotransporter‐2 (SGLT2) inhibitors, dipeptidyl peptidase‐4 (DPP‐4) inhibitors, sulfonylureas, thiazolidinediones and α‐glucosidase inhibitors—were identified analogously. For descriptive analyses, women were classified into four mutually exclusive treatment groups based on therapy ever recorded during follow‐up: (1) no listed therapy, (2) metformin only, (3) GLP‐1 RA/GIP–GLP‐1 RA therapy with or without other therapies and (4) other non–GLP‐1 therapy, defined as any glucose‐lowering therapy other than metformin, with or without metformin, but without GLP‐1 RA/GIP–GLP‐1 RA exposure. These groups described treatment‐associated clinical profiles rather than baseline exposure groups, because treatment classification incorporated prescriptions recorded after the PCOS index date. Comorbidities were ascertained from binary ICD‐10 indicators: type 2 diabetes mellitus (E11), hypertension (I10), dyslipidaemia (E78), depression (F32 or F33) and obstructive sleep apnoea (G47.3). A cardiovascular disease composite was defined as the presence of any of the following codes: G45 (transient cerebral ischemic attack), I21 (acute myocardial infarction), I25 (chronic ischemic heart disease), I26 (pulmonary embolism), I35.0 (aortic valve stenosis), I48 (atrial fibrillation/flutter), I50 (heart failure), I63 (cerebral infarction) and I64 (stroke, not specified as haemorrhage or infarction). Consistent with our prior LUX MED analyses [18, 19, 20, 21, 22], these variables serve as clinical proxies and do not capture disease severity, duration, age at onset or treatment indication. Body mass index (BMI) was computed from recorded weight and height in kg/m²; analyses involving BMI used available‐case data, and the proportion of women with a recorded BMI is reported alongside summary statistics. Obesity was defined as BMI ≥ 30 kg/m².

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

2.1. Statistical Analysis

Continuous variables are summarized as median (interquartile range) and categorical variables as counts (percentages). Recorded clinical characteristics were descriptively compared across ever‐recorded treatment‐pattern groups using the Kruskal–Wallis test and the χ ² test or Fisher exact test, as appropriate. Because these groups were defined using therapy recorded at any time during follow‐up, these comparisons describe treatment‐trajectory contrasts rather than causal baseline differences. For the primary temporal analysis, we assessed whether a woman started GLP‐1 RA/GIP–GLP‐1 RA therapy within the first 365 days after her first recorded PCOS diagnosis. Women who had a GLP‐1 RA/GIP–GLP‐1 RA prescription before the first recorded PCOS diagnosis were kept in the cohort, but that earlier prescription was not counted as a post‐diagnosis initiation event. Women diagnosed in 2025–2026 were not included in this analysis because they did not all have a full 365 days of potential follow‐up before the end of data collection. Robustness to this choice was examined in sensitivity analyses using alternative windows of 180 and 730 days. Annual proportions are reported with 95% confidence intervals (CIs; Wilson method). Temporal trends were assessed using logistic regression with calendar year of PCOS diagnosis as the predictor of interest. Model A (unadjusted) included calendar year as the sole predictor. Multivariable models additionally adjusted for age, BMI, type 2 diabetes mellitus, hypertension, dyslipidaemia, depression, obstructive sleep apnoea and the cardiovascular disease composite. For ease of clinical interpretation, continuous predictors were modelled as linear terms and rescaled before modelling. The calendar year of PCOS diagnosis was analysed per 1‐year increment; therefore, the odds ratio represents the change in odds of treatment initiation for each later year of diagnosis. Age was analysed per 5‐year increment and BMI per 5 kg/m² increment; therefore, their odds ratios represent the change in odds associated with each 5‐year older age or each 5 kg/m² higher BMI. Because BMI was missing in 17 707 women (46.3%) with uneven distribution across treatment groups, the primary adjusted model (Model C) handled missing BMI using multiple imputation by chained equations with predictive mean matching and 20 imputed datasets, including the outcome and all covariates among auxiliary predictors; estimates were pooled using Rubin's rules. The imputation approach assumed that BMI was missing at random conditional on the observed variables included in the imputation model. To assess robustness to violations of this assumption, we conducted prespecified missing‐not‐at‐random sensitivity analyses using global delta‐adjustment and treatment‐group–specific pattern‐mixture offsets applied to imputed BMI values. Subgroup analyses of the primary endpoint were performed within strata defined by age band (< 30, 30–39, ≥ 40 years) and recorded BMI category (< 25, 25–29.9, 30–34.9, ≥ 35 kg/m²), estimating the calendar‐year association within each stratum. A complete‐case multivariable analysis (Model B) was performed as a sensitivity analysis for missing‐data handling. A further sensitivity analysis (Model D) re‐fit Model C after excluding women with coded type 2 diabetes mellitus. Departures from linearity were assessed by modelling calendar year as a restricted cubic spline with boundary knots at 2018 and 2024 and an internal knot at 2021, comparing fit with the linear specification using the Akaike information criterion and a likelihood‐ratio test. Results are reported as odds ratios (ORs) or adjusted odds ratios (aORs) with 95% CIs. Temporal changes in the clinical profile of GLP‐1 RA/GIP–GLP‐1 RA initiators were assessed according to calendar year of first recorded therapy initiation, using linear regression for continuous outcomes and logistic regression for binary outcomes. To aid interpretability, we also calculated crude absolute annual initiation proportions and directly standardized proportions. Age‐standardized proportions were standardized to the age distribution of the full 2018–2024 cohort; age/BMI‐standardized proportions were calculated among women with recorded BMI and standardized to the joint age‐ and BMI‐category distribution of the complete‐case cohort. All tests were two‐sided, with an α level of 0.05. Analyses were performed using Python 3.11 (Python Software Foundation, Wilmington, DE, USA) with the pandas, NumPy, SciPy, statsmodels, scikit‐learn and matplotlib packages.

3. Results

The study cohort comprised 38 263 adult women with PCOS identified by ICD‐10 code E28.2. The first recorded PCOS diagnosis occurred on July 5, 2006 and the most recent on April 22, 2026; the earliest GLP‐1 RA/GIP–GLP‐1 RA prescription was recorded on November 12, 2018. The median age at the index date was 33.0 years (29.0–39.0), and 1419 women (3.7%) had an ICD‐coded diagnosis of type 2 diabetes mellitus. BMI was available in 20 556 women (53.7%), with a median of 24.8 kg/m² (21.5–29.8) (Table 1).

TABLE 1.

Recorded clinical characteristics according to ever‐recorded treatment pattern.

Characteristic	Overall	No therapy	Metformin only	GLP‐1 RA/GIP ± other	Other non–GLP‐1 therapy	p
n	38 263	20 745	12 755	4557	206	—
Age, years	33.0 [29.0–39.0]	33.0 [29.0–39.0]	33.0 [29.0–38.0]	34.0 [30.0–40.0]	39.0 [32.0–46.0]	< 0.001
BMI, kg/m2 a	24.8 [21.5–29.8]	22.1 [20.2–25.0]	26.3 [22.8–30.8]	31.2 [27.1–35.8]	29.6 [25.6–34.4]	< 0.001
BMI available	20 556 (53.7)	8889 (42.8)	8189 (64.2)	3339 (73.3)	139 (67.5)	< 0.001
BMI ≥ 30 kg/m2 a	5042 (24.5)	670 (7.5)	2348 (28.7)	1959 (58.7)	65 (46.8)	< 0.001
Type 2 diabetes	1419 (3.7)	144 (0.7)	658 (5.2)	525 (11.5)	92 (44.7)	< 0.001
Hypertension	3404 (8.9)	1002 (4.8)	1348 (10.6)	962 (21.1)	92 (44.7)	< 0.001
Dyslipidaemia	5554 (14.5)	2082 (10.0)	2160 (16.9)	1221 (26.8)	91 (44.2)	< 0.001
Depression	4085 (10.7)	1805 (8.7)	1424 (11.2)	821 (18.0)	35 (17.0)	< 0.001
Sleep apnoea	292 (0.8)	65 (0.3)	110 (0.9)	111 (2.4)	6 (2.9)	< 0.001
Cardiovascular disease	458 (1.2)	170 (0.8)	164 (1.3)	107 (2.3)	17 (8.3)	< 0.001
Metformin	16 381 (42.8)	0 (0.0)	12 755 (100.0)	3442 (75.5)	184 (89.3)	—
SGLT2 inhibitor	291 (0.8)	0 (0.0)	0 (0.0)	160 (3.5)	131 (63.6)	—
DPP‐4 inhibitor	166 (0.4)	0 (0.0)	0 (0.0)	93 (2.0)	73 (35.4)	—
Sulfonylurea	58 (0.2)	0 (0.0)	0 (0.0)	34 (0.7)	24 (11.7)	—

Note: Treatment groups were defined according to therapy ever recorded during follow‐up; this table therefore describes recorded clinical profiles by treatment trajectory rather than baseline characteristics at the index polycystic ovary syndrome diagnosis. Data are presented as median [interquartile range] for continuous variables and n (%) for categorical variables. p values reflect comparison across the four treatment groups.

Abbreviations: BMI, body mass index; DPP‐4, dipeptidyl peptidase‐4; GIP, glucose‐dependent insulinotropic polypeptide; GLP‐1 RA, glucagon‐like peptide‐1 receptor agonist; SGLT2, sodium‐glucose cotransporter‐2.

^a Computed among patients with a recorded BMI measurement (n = 20 556).

3.1. Overall Treatment Patterns

During the observation period, 4557 women (11.9%) had at least one GLP‐1 RA/GIP–GLP‐1 RA prescription and 16 381 (42.8%) had received metformin (Tables 1 and S1). Among GLP‐1 RA/GIP–GLP‐1 RA users, 3442 (75.5%) had concurrent or prior metformin exposure, whereas only 525 (11.5%) carried an ICD‐coded diagnosis of type 2 diabetes mellitus, suggesting that prescription‐recorded incretin‐based therapy use occurred predominantly outside coded type 2 diabetes. The most frequently prescribed agents in the class were semaglutide (n = 2347; 6.1%), liraglutide (n = 1625; 4.2%), tirzepatide (n = 1203; 3.1%) and dulaglutide (n = 374; 1.0%); no albiglutide prescriptions were recorded. Other glucose‐lowering agents were uncommon: SGLT2 inhibitors in 291 women (0.8%), DPP‐4 inhibitors in 166 (0.4%) and sulfonylureas in 58 (0.2%).

3.2. Recorded Clinical Profile According to Ever‐Recorded Treatment Pattern

Women were classified into four mutually exclusive treatment groups based on therapy ever recorded during follow‐up—no listed therapy (n = 20 745; 54.2%), metformin only (n = 12 755; 33.3%), GLP‐1 RA/GIP–GLP‐1 RA therapy with or without other agents (n = 4557; 11.9%), and other non–GLP‐1 therapy (n = 206; 0.5%) and Table 1 therefore describes clinical profiles associated with treatment trajectories rather than baseline characteristics at PCOS diagnosis. The proportion with recorded BMI varied across groups, being lowest among untreated women (42.8%) and highest among GLP‐1 RA/GIP–GLP‐1 RA users (73.3%). Median BMI was 31.2 kg/m² (27.1–35.8) among GLP‐1 RA/GIP–GLP‐1 RA users, 26.3 kg/m² (22.8–30.8) among metformin‐only users, and 22.1 kg/m² (20.2–25.0) among untreated women, with corresponding obesity prevalences of 58.7%, 28.7%, and 7.5%. Comorbidity burden followed the same gradient: among GLP‐1 RA/GIP–GLP‐1 RA users, type 2 diabetes mellitus, hypertension, dyslipidaemia, depression, obstructive sleep apnoea and cardiovascular disease were present in 11.5%, 21.1%, 26.8%, 18.0%, 2.4%, and 2.3%, respectively—substantially higher than in the metformin‐only or untreated groups (all p < 0.001; Table 1).

3.3. Temporal Trends in First‐Prescribed Agent

The annual number of women initiating GLP‐1 RA/GIP–GLP‐1 RA therapy rose from 6 in 2018 and 23 in 2019 to 1030 in 2024 and 1225 in 2025 (Table S2; Figure 1). The 451 initiations recorded in 2026 represent a partial calendar year (January–April 2026). The class composition of first‐prescribed agents shifted substantially over time. Initiations in 2018–2019 were limited exclusively to dulaglutide (n = 29). From 2020 onward, liraglutide became the predominant first agent, peaking at 57.9% in 2022 and declining to 30.5% in 2024 and 9.8% in 2025. In the most recent years, tirzepatide was marginally more frequent in 2025 (45.5% vs. 44.1% for semaglutide), whereas semaglutide was marginally more frequent in early 2026 (50.1% vs. 47.5% for tirzepatide). In early 2026, semaglutide and tirzepatide together represented over 97% of first‐prescribed incretin‐based therapy, while older agents had become marginal.

FIGURE 1.

Annual number and relative distribution of first‐prescribed GLP‐1 RA and GIP–GLP‐1 RA agents in women with polycystic ovary syndrome, 2018–2026. (A) Annual number of patients initiating incretin‐based therapy, stratified by the first recorded agent. (B) Relative distribution (%) of first‐prescribed agents by calendar year of initiation. Data for 2026 cover a partial calendar year (January–April 2026) and should be interpreted with caution. Albiglutide was eligible but not prescribed in this cohort and is therefore omitted from the table. GIP, glucose‐dependent insulinotropic polypeptide; GLP‐1 RA, glucagon‐like peptide‐1 receptor agonist.

3.4. Primary Temporal Endpoint

The primary analysis included 21 577 women diagnosed with PCOS in 2018–2024, of whom 760 (3.5%) initiated GLP‐1 RA/GIP–GLP‐1 RA therapy within 365 days. The annual proportion meeting the endpoint rose more than 40‐fold across the study period, from 0.14% in 2018 to 5.97% in 2024, with year‐specific proportions of 0.99% (2020), 3.02% (2021), 5.03% (2022) and 5.54% (2023) (Figure S1). Directly standardized proportions showed the same pattern, increasing from 0.08% to 6.39% after age standardization and from 0.11% to 7.11% after age/BMI standardization (Table S3), indicating that the temporal increase was not explained by shifts in the age or BMI composition of the cohort.

In the unadjusted model (Model A), each additional calendar year of PCOS diagnosis was associated with a 52% increase in the odds of incretin‐based therapy initiation (OR 1.52; 95% CI, 1.44–1.59; p < 0.001; Table 2). The temporal trend was robust to multivariable adjustment and to alternative handling of missing BMI. In the primary BMI‐imputed model (Model C), each additional calendar year was associated with an aOR of 1.64 (95% CI, 1.56–1.73; p < 0.001). BMI was the strongest individual correlate (aOR 1.99 per 5 kg/m²; 95% CI, 1.88–2.11), followed by depression (aOR 1.58; 95% CI, 1.28–1.94) and age (aOR 1.20 per 5 years; 95% CI, 1.12–1.28; all p < 0.001). Type 2 diabetes mellitus showed a borderline non‐significant association with initiation (aOR 1.32; 95% CI, 0.96–1.81; p = 0.09), and dyslipidaemia was modestly associated (aOR 1.25; 95% CI, 1.03–1.53; p = 0.027). Hypertension, obstructive sleep apnoea, and the cardiovascular disease composite were not independently associated with initiation in any specification. In sensitivity analyses addressing potential departures from the missing‐at‐random assumption for BMI, delta‐adjustment (up to +3 kg/m² added to all imputed values) and pattern‐mixture scenarios assigning the largest upward BMI shifts to untreated women left the calendar‐year association essentially unchanged (aOR range, 1.64–1.68), and BMI remained a strong independent correlate even under the most extreme adversarial assumption (aOR 1.81 per 5 kg/m²; Table S4). In sensitivity analyses using alternative post‐diagnosis initiation windows, the temporal trend remained consistent. In BMI‐imputed adjusted models, each later calendar year of PCOS diagnosis was associated with higher odds of GLP‐1 RA/GIP–GLP‐1 RA initiation within 180 days (aOR 1.67; 95% CI, 1.56–1.78) and within the primary 365‐day window (aOR 1.64; 95% CI, 1.56–1.73), both among women diagnosed in 2018–2024, and within 730 days (aOR 1.74; 95% CI, 1.64–1.85) among women diagnosed in 2018–2023, restricted to ensure complete 2‐year follow‐up (Table S5). The complete‐case analysis (Model B) yielded a consistent year coefficient (aOR 1.52; 95% CI, 1.43–1.62), confirming robustness to missing‐data handling. After excluding women with coded type 2 diabetes mellitus (Model D), the temporal trend was unchanged (aOR 1.65; 95% CI, 1.56–1.74), and BMI remained the dominant predictor (aOR 2.04 per 5 kg/m²; 95% CI, 1.92–2.17), indicating that the rising probability of incretin‐based therapy initiation following PCOS diagnosis was not driven by glycaemic indication. In subgroup analyses, the temporal increase in GLP‐1 RA/GIP–GLP‐1 RA initiation within 365 days was present across all age and BMI strata (Table S6). The odds of initiation rose significantly with each later calendar year among women aged < 30 years (OR 1.49; 95% CI, 1.34–1.65), 30–39 years (OR 1.63; 95% CI, 1.52–1.74) and ≥ 40 years (OR 1.62; 95% CI, 1.46–1.80). Among women with recorded BMI, absolute initiation rates increased monotonically with BMI category (from 0.92% at < 25 kg/m² to 15.23% at ≥ 35 kg/m²), yet a significant temporal increase was evident in every category, including women with BMI < 25 kg/m² (OR 2.26; 95% CI, 1.73–2.97) and 25–29.9 kg/m² (OR 1.41; 95% CI, 1.26–1.59), consistent with a broadening of prescribing beyond the most obese phenotype.

TABLE 2.

Temporal trend in GLP‐1 RA/GIP–GLP‐1 RA initiation within 365 days of polycystic ovary syndrome diagnosis: Unadjusted, complete‐case, BMI‐imputed and diabetes‐excluded models, 2018–2024.

Predictor	Model A: Unadjusted		Model B: Complete‐case		Model C: BMI‐imputed		Model D: T2DM excluded
	OR (95% CI)	p	aOR (95% CI)	p	aOR (95% CI)	p	aOR (95% CI)	p
PCOS diagnosis year (per 1‐year increment)	1.52 (1.44–1.59)	< 0.001	1.52 (1.43–1.62)	< 0.001	1.64 (1.56–1.73)	< 0.001	1.65 (1.56–1.74)	< 0.001
Age (per 5‐year increment)	—	—	1.17 (1.09–1.26)	< 0.001	1.20 (1.12–1.28)	< 0.001	1.20 (1.12–1.29)	< 0.001
BMI (per 5 kg/m2 increment)	—	—	1.95 (1.84–2.07)	< 0.001	1.99 (1.88–2.11)	< 0.001	2.04 (1.92–2.17)	< 0.001
Type 2 diabetes mellitus	—	—	1.23 (0.87–1.73)	0.24	1.32 (0.96–1.81)	0.09	—	—
Hypertension	—	—	1.06 (0.83–1.36)	0.64	1.03 (0.81–1.31)	0.81	1.03 (0.80–1.32)	0.84
Dyslipidaemia	—	—	1.21 (0.98–1.49)	0.08	1.25 (1.03–1.53)	0.027	1.32 (1.08–1.63)	0.008
Depression	—	—	1.44 (1.15–1.80)	0.001	1.58 (1.28–1.94)	< 0.001	1.60 (1.28–1.99)	< 0.001
Obstructive sleep apnoea	—	—	1.01 (0.57–1.81)	0.96	1.03 (0.58–1.84)	0.91	1.07 (0.58–2.00)	0.82
Cardiovascular disease composite	—	—	1.38 (0.75–2.56)	0.30	1.34 (0.74–2.43)	0.34	1.50 (0.81–2.79)	0.20

Note: The endpoint was initiation of GLP‐1 RA/GIP–GLP‐1 RA therapy within 365 days after the first recorded PCOS diagnosis. The analysis was limited to women diagnosed in 2018–2024 to ensure that each patient had the opportunity for a complete 365‐day observation period. Continuous predictors were rescaled so that each odds ratio reflects a clinically meaningful increment: calendar year per 1‐year increment (the year‐on‐year change in odds of initiation), age per 5‐year increment and BMI per 5 kg/m² increment. Values are odds ratios (ORs) or adjusted odds ratios (aORs) with 95% confidence intervals (CIs). Model A was an unadjusted logistic regression model including calendar year only. Model B was a complete‐case multivariable model. Model C was the primary BMI‐imputed multivariable model using 20 imputed datasets. Model D repeated Model C after excluding women with coded type 2 diabetes mellitus. Model A/C: N = 21 577, events = 760; Model B: N = 13 578, events = 636; Model D: N = 20 896, events = 701.

Abbreviations: aOR, adjusted odds ratio; BMI, body mass index; CI, confidence interval; GIP, glucose‐dependent insulinotropic polypeptide; GLP‐1 RA, glucagon‐like peptide‐1 receptor agonist; ICD‐10, International Classification of Diseases, Tenth Revision; OR, odds ratio; PCOS, polycystic ovary syndrome; T2DM, type 2 diabetes mellitus.

3.5. Functional Form of the Temporal Trend

Year‐specific proportions suggested a non‐linear pattern, with very low uptake in 2018–2020, rapid expansion in 2021–2022 and relative stabilization in 2023–2024. A restricted cubic spline (boundary knots at 2018 and 2024, internal knot at 2021) provided a better fit than the linear specification (p < 0.001; Figure 2), confirming non‐linearity and indicating that uptake was concentrated mainly after 2020–2021.

FIGURE 2.

Non‐linear temporal pattern of GLP‐1 RA/GIP–GLP‐1 RA initiation within 365 days of polycystic ovary syndrome diagnosis, 2018–2024. Annual observed proportions are shown as points with Wilson 95% confidence intervals. The fitted curve represents a restricted cubic spline logistic regression model for calendar year, with boundary knots at 2018 and 2024 and an internal knot at 2021; the shaded area indicates the 95% confidence interval of the fitted spline. The dashed line represents the corresponding linear calendar‐year model. CI, confidence interval; GIP, glucose‐dependent insulinotropic polypeptide; GLP‐1 RA, glucagon‐like peptide‐1 receptor agonist; PCOS, polycystic ovary syndrome.

3.6. Temporal Evolution of the Initiator Phenotype

The clinical profile of women initiating incretin‐based therapy shifted progressively toward a younger and less metabolically burdened population (Table 3). Median age at initiation declined from 46.0 years in 2018 to 33.0 years in 2025, and median BMI from 33.9 kg/m² in 2019 to 29.1 kg/m² in 2025. The proportion of initiators meeting the obesity threshold fell from 73.1% in 2022 to 44.5% in 2025. Type 2 diabetes mellitus prevalence among initiators decreased from 29.0% in 2020 to 9.3% in 2024 and 6.4% in 2025, while concomitant metformin use declined from over 90% before 2024 to approximately 63% in 2025 and 61% in 2026. Tests for trend confirmed that calendar year of initiation was associated with significantly lower age, BMI, obesity prevalence, type 2 diabetes mellitus prevalence, metformin co‐use, hypertension, dyslipidaemia and depression (all p for trend < 0.001). BMI availability among initiators was not stable over the study period: it showed no significant trend during 2018–2023 (OR per calendar year 1.09; 95% CI, 0.97–1.23; p = 0.13) but declined thereafter, with the odds of BMI availability decreasing across 2018–2026 (OR 0.82; 95% CI, 0.78–0.86; p < 0.001); the decline in recorded BMI and obesity prevalence among recent initiators should therefore be interpreted partly in light of this changing measurement completeness.

TABLE 3.

Phenotype of GLP‐1 RA/GIP–GLP‐1 RA initiators by year of first prescription.

Year	n	Age, years (median)	BMI, kg/m2 (median)	BMI available, n	BMI available, %	BMI ≥ 30 kg/m2, % a	Type 2 diabetes, %	Metformin, %	Hypertension, %	Dyslipidaemia, %	Depression, %
2018	6	46.0	29.1	5	83.3	40.0	16.7	100.0	50.0	100.0	16.7
2019	23	41.0	33.9	18	78.3	66.7	39.1	100.0	47.8	60.9	13.0
2020	100	39.0	33.4	74	74.0	63.5	29.0	96.0	45.0	47.0	25.0
2021	282	38.0	33.6	230	81.6	72.2	23.8	94.3	39.0	42.9	23.0
2022	672	36.0	33.1	539	80.2	73.1	18.2	90.2	29.3	35.6	24.4
2023	768	35.0	32.6	633	82.4	67.9	12.1	89.3	23.3	28.5	21.1
2024	1030	34.0	31.0	722	70.1	57.3	9.3	69.5	18.0	22.2	16.6
2025	1225	33.0	29.1	839	68.5	44.5	6.4	62.9	13.6	20.4	14.4
2026 b	451	34.0	28.7	279	61.9	43.4	6.4	60.5	14.4	21.3	12.0
p for trend		< 0.001	< 0.001			< 0.001	< 0.001	< 0.001	< 0.001	< 0.001	< 0.001

Note: Values are medians for continuous variables and percentages for categorical variables.

Abbreviations: BMI, body mass index; GIP, glucose‐dependent insulinotropic polypeptide; GLP‐1 RA, glucagon‐like peptide‐1 receptor agonist.

^a Computed among initiators with a recorded BMI measurement.

^b 2026 data are partial, covering January through April 2026.

4. Discussion

In this large real‐world cohort of 38 263 adult women with PCOS from a major Polish private healthcare network, GLP‐1 RA/GIP–GLP‐1 RA therapy was prescribed to approximately one in eight women (11.9%), with the proportion of women initiating incretin‐based therapy within 365 days of PCOS diagnosis rising more than 40‐fold between 2018 and 2024—from 0.14% to 5.97%. Each additional calendar year of PCOS diagnosis was associated with 64% higher adjusted odds of initiation in the multiply imputed primary model (aOR 1.64). Although GLP‐1 RA/GIP–GLP‐1 RA users carried a more adverse cardiometabolic profile than untreated or metformin‐only women, most users did not have ICD‐coded type 2 diabetes mellitus and the temporal trend persisted unchanged after excluding women with coded diabetes (aOR 1.65). Over the study period, initiators became progressively younger and less metabolically burdened, while the first‐prescribed agent shifted from dulaglutide and liraglutide toward semaglutide and tirzepatide. Together, these findings indicate a substantial expansion of incretin‐based therapy use in PCOS beyond classical glycaemic indications toward broader cardiometabolic and weight‐focused management.

PCOS is a heterogeneous disorder in which reproductive manifestations frequently coexist with obesity, insulin resistance, dyslipidaemia, hypertension, obstructive sleep apnoea, depressive symptoms and elevated long‐term cardiometabolic and type 2 diabetes risk [3, 4, 5, 6]. The recent 2025 Lancet Diabetes and Endocrinology Commission has reframed PCOS explicitly as a clinical complication of obesity, providing a conceptual rationale for adiposity‐targeted pharmacotherapy in this population [7, 8]. Against this background, the preferential use of GLP‐1 RA/GIP–GLP‐1 RA therapy among women with higher BMI and a greater burden of cardiometabolic comorbidities is clinically coherent. BMI was the strongest individual correlate of treatment initiation across all multivariable specifications, with an approximately twofold higher adjusted odds of initiation per 5 kg/m² increment (aOR 1.99). Depression was also independently associated with initiation (aOR 1.58), plausibly reflecting the bidirectional association between obesity and depressive symptoms in PCOS [23, 24] and the role of weight‐related psychosocial burden in healthcare‐seeking behaviour. By contrast, type 2 diabetes mellitus showed only a borderline association (aOR 1.32), and traditional vascular comorbidities (hypertension, sleep apnoea, the cardiovascular disease composite) showed no independent association—a pattern that is consistent with real‐world prescribing being more strongly associated with adiposity‐related considerations than with coded glycaemic indication, although the specific treatment indication could not be directly ascertained from prescription data [21, 25].

The temporal increase was non‐linear, with very low uptake in 2018–2020, rapid expansion in 2021–2022 and relative stabilization in 2023–2024. The linear year coefficient should therefore be interpreted as a parsimonious summary of the average annual increase rather than a constant year‐on‐year multiplicative change. The observed trajectory is consistent with the regulatory and market evolution of incretin‐based therapies—notably the EU authorization of semaglutide 2.4 mg for weight management in 2022 and the subsequent introduction of tirzepatide—and with country‐specific factors including evolving private‐market availability, shifting reimbursement and out‐of‐pocket prescribing dynamics, intermittent supply constraints and growing patient demand [21, 22]. In contrast to several Western European systems, incretin‐based therapy for weight management is not reimbursed by the Polish public payer (Narodowy Fundusz Zdrowia): agents licensed for obesity (e.g., the 2.4‐mg semaglutide formulation and tirzepatide) are dispensed entirely out of pocket, and partial reimbursement of semaglutide is restricted to patients with type 2 diabetes mellitus meeting stringent glycaemic and cardiovascular‐risk criteria. The marked post‐2021 acceleration in uptake therefore occurred in a predominantly self‐pay environment, suggesting that patient demand and prescriber awareness—rather than reimbursement incentives—were the principal drivers of adoption. The relative stabilization observed in 2023–2024 also coincided with EU‐wide supply shortages of semaglutide and liraglutide, which began in 2022 and deteriorated in late 2023, prompting European Medicines Agency recommendations to limit new initiations. Because the present analysis is based on prescription records without linkage to dispensing, reimbursement or supply data, we frame these factors as plausible contextual drivers consistent with the observed time structure rather than as established mechanisms [21, 22]. Our findings are directionally consistent with a recent US analysis of 94 365 women with PCOS in the TriNetX network, in which the age‐standardized incidence of new GLP‐1 RA use rose from 0.07% in 2011 to 17.1% in 2025, with a parallel decline in the proportion of type 2 diabetes mellitus‐driven prescribing (58% in 2019–34% in 2024) and a corresponding increase in obesity‐driven prescribing without diabetes mellitus (35%–58%) [25]. Direct numerical comparison is limited by differences in case definition, healthcare system, reimbursement and outcome construction—most notably, our primary endpoint captured prescription‐recorded initiation within 365 days after the index PCOS diagnosis, whereas the US analysis reported any new GLP‐1 RA use after PCOS diagnosis without a defined time window. Nevertheless, the convergent direction, magnitude and phenotype/agent shifts across two markedly different healthcare systems strengthen the inference that the broadening of incretin‐based therapy use beyond classical glycaemic indications is a generalizable rather than country‐specific phenomenon.

The shift in first‐prescribed agents and in initiator phenotype is itself clinically informative. Earlier initiation was dominated by dulaglutide and liraglutide; in the most recent years, semaglutide and tirzepatide together represented over 97% of new prescriptions, mirroring the broader displacement of earlier‐generation GLP‐1 RAs by agents with greater weight‐loss efficacy [10, 11, 12, 13]. Concurrently, initiators became younger, less obese and less frequently diabetic, and the proportion on concomitant metformin therapy declined from over 90% before 2024 to approximately 60% in 2025–2026 (all p for trend < 0.001). These observations are consistent with a progressive broadening of the indication for incretin‐based therapy in PCOS—from a relatively narrow, high‐cardiometabolic‐risk phenotype toward a wider population in whom adiposity and metabolic risk are present but the classical diabetic phenotype is absent. The high proportion of metformin co‐exposure among GLP‐1 RA/GIP–GLP‐1 RA users (75.5%) suggests that incretin‐based therapy is most often added to, rather than substituted for, established metabolic treatment, although treatment sequence and clinical rationale could not be reconstructed from the available data [9, 15, 16]. This downward trend may further signal an early shift away from metformin as the default first‐line metabolic agent in obesity‐driven PCOS, although co‐use remained the majority pattern throughout.

The persistence of the temporal increase after excluding women with coded type 2 diabetes mellitus is one of the most policy‐relevant findings of this study. It indicates that the observed expansion is not a consequence of growing diabetes treatment but reflects broader use for obesity, insulin resistance and overall cardiometabolic risk management. This is particularly relevant in PCOS because the syndrome is typically recognized in young women, in whom early cardiometabolic risk modification may carry long‐term implications for the development of type 2 diabetes mellitus, atherosclerotic cardiovascular disease and other obesity‐related sequelae [2]. The clinical promise of pharmacological weight management in PCOS is supported by randomized and observational studies showing that GLP‐1 RA therapy consistently reduces body weight and central adiposity, with emerging but lower‐certainty evidence for improvements in insulin resistance, androgen profiles, menstrual cyclicity and fertility‐related outcomes [3, 8]. Whether the rapid real‐world expansion documented here translates into durable improvements in weight, metabolic health, fertility, mental health, quality of life or cardiovascular outcomes cannot be determined from prescription data alone and requires dedicated prospective evaluation. The rapid uptake of incretin‐based therapies in a reproductive‐aged PCOS population also has important safety implications. Recent large cohort studies have not consistently shown an increased risk of major congenital malformations after periconceptional GLP‐1 RA exposure, but data remain limited and particularly sparse for continued use during pregnancy and lactation [26, 27]. Treatment initiation should therefore be accompanied by counselling on pregnancy planning, timing of discontinuation before planned conception, and effective contraception when pregnancy is not desired—considerations particularly relevant for tirzepatide, for which delayed gastric emptying may reduce oral contraceptive exposure during initiation and dose escalation [8, 28, 29, 30]. Beyond reproductive considerations, the broadening phenotype of initiators highlights the need to define the benefit–risk balance in PCOS populations not yet well represented in randomized trials, and the dominant uptake of semaglutide and tirzepatide raises questions about affordability, drug supply and long‐term treatment sustainability. The absence of dispensing, persistence and outcome data also underscores the need for linked prescribing–dispensing–outcome registries: our group has previously documented low long‐term persistence with GLP‐1 RA therapy in Polish real‐world cohorts [20] and substantial within‐class switching activity, both of which complicate the interpretation of ‘real’ treatment exposure in PCOS care. Such exposure misclassification could bias estimates of uptake and the recorded clinical profile of women categorized as incretin‐based therapy users.

This study has several strengths. It provides the first detailed real‐world characterization of GLP‐1 RA/GIP–GLP‐1 RA prescribing in women with PCOS in a Central European healthcare setting and, to our knowledge, the largest such cohort outside the United States, with comprehensive prescription‐level capture across an eight‐year period of rapid therapeutic class evolution. The cohort size (n = 38 263) and breadth of the observation window enabled granular description of year‐on‐year prescribing dynamics, including the transition from earlier‐generation GLP‐1 RAs to semaglutide and tirzepatide and the parallel shift in initiator clinical phenotype. The analytic approach combined descriptive characterization with formal regression modelling of temporal trends and pre‐specified sensitivity analyses—multiple imputation of missing BMI, complete‐case modelling, and exclusion of women with coded type 2 diabetes mellitus—that collectively support the robustness of the headline temporal finding. Nevertheless, several limitations should be acknowledged. First, this was a retrospective analysis of routinely collected electronic health record data from a single private healthcare network in Poland and is subject to the inherent limitations of observational pharmacoepidemiology, including residual confounding, incomplete data capture and potential misclassification. PCOS was identified using ICD‐10 code E28.2 and was not validated against the Rotterdam criteria, transvaginal ultrasound findings or biochemical evidence of hyperandrogenism; the cohort should therefore be interpreted as women with ICD‐coded PCOS rather than a phenotypically homogeneous PCOS population. Nonetheless, ICD‐based case definitions are the standard approach for identifying PCOS in large administrative and electronic health record–based studies, and a recent systematic review reported a pooled positive predictive value of 88% for such definitions, albeit with substantial between‐study heterogeneity [31]. More importantly, any residual misclassification is unlikely to be correlated with calendar year of diagnosis; being effectively non‐differential with respect to time, it is therefore unlikely to account for the directional temporal trends we observed, which reflect within‐cohort prescribing dynamics rather than between‐cohort prevalence comparisons. Comorbidities were ascertained analogously from ICD‐10 codes and treated as binary indicators that do not capture disease severity, duration, age at onset or treatment adequacy [18, 19, 20, 21, 22]; obesity is not consistently coded in this dataset and was therefore identified only through measured BMI, where available. Second, the dataset did not include several clinically relevant PCOS‐related variables, including menstrual phenotype, androgen and sex hormone–binding globulin concentrations, ovarian ultrasound findings, fertility status and pregnancy intention, HOMA‐IR, glycated haemoglobin, fasting glucose, detailed lipid profile, lifestyle interventions and weight‐trajectory history. Consequently, we could not determine the specific clinical indication for GLP‐1 RA/GIP–GLP‐1 RA therapy in individual women—whether prescribing was directed at weight management, glycaemic control, insulin resistance, ovulatory dysfunction or fertility‐related indications or broader cardiometabolic risk reduction. The dominant role of BMI and the absence of coded type 2 diabetes mellitus in most initiators are most consistent with weight‐focused prescribing, but this interpretation is inferential rather than directly demonstrated. Third, drug exposure was based on outpatient prescription dates rather than confirmed dispensing, reimbursement, purchase or actual medication use. Prescriptions should therefore be interpreted as treatment initiation or intended use rather than confirmed pharmacological exposure, and we were unable to assess dose, persistence, adherence, discontinuation, within‐class switching or clinical response. Fourth, BMI was missing in 17 707 women (46.3%) and was unevenly distributed across treatment groups, being most frequently recorded in GLP‐1 RA/GIP–GLP‐1 RA users (73.3%) and least frequently in women with no listed therapy (42.8%), a pattern unlikely to satisfy the missing‐completely‐at‐random assumption. Multiple imputation reduced complete‐case bias and yielded a temporal trend coefficient that was, if anything, stronger than the complete‐case estimate; delta‐adjustment and pattern‐mixture sensitivity analyses spanning a range of missing‐not‐at‐random assumptions left this trend essentially unchanged (Table S4), indicating that residual missing‐not‐at‐random bias is unlikely to account for our findings. Separately, BMI availability among initiators declined in the most recent years; the observed decrease in recorded BMI and obesity prevalence among recent initiators should therefore be interpreted with caution, as it may partly reflect changing measurement completeness rather than a purely phenotypic shift, and because the characteristics of initiators with unrecorded BMI are unknown, the direction of any resulting bias cannot be determined. In addition, because patients may attend multiple offices within the network and individual physicians may practice across sites, patients could not be unambiguously assigned to a single clinic or provider, and formal multilevel modelling was not performed; in the absence of cluster‐robust standard errors, the precision of our estimates may be modestly overstated should residual centre‐level heterogeneity exist, although the relative homogeneity of the network limits this possibility. Fifth, the temporal relationship between PCOS diagnosis, comorbidity development and treatment initiation could not always be fully established. A small proportion of women (n = 588; 1.5%) had at least one GLP‐1 RA/GIP–GLP‐1 RA prescription before the first recorded PCOS diagnosis date, plausibly reflecting prior incretin‐based therapy for obesity or diabetes, delayed PCOS recognition or incomplete capture of historical diagnoses preceding the patient's entry into the LUX MED network. Although pre‐index prescriptions were not counted as events in the primary 365‐day post‐index endpoint, this pattern illustrates the inherent limitation of using first recorded diagnosis as a proxy for true disease onset. Sixth, the marked temporal increase and the shift in initiator phenotype documented here are descriptive, and this study was not designed to identify the underlying causal drivers. The observed trends are temporally consistent with the introduction of more potent agents (semaglutide for chronic weight management, tirzepatide), evolving reimbursement and out‐of‐pocket prescribing dynamics and broader changes in obesity treatment paradigms, but the relative contribution of regulatory, reimbursement, prescriber and patient‐level factors cannot be quantified. We therefore frame these as plausible drivers consistent with the observed time structure rather than as established mechanisms. Seventh, the primary temporal endpoint analysis was restricted to PCOS diagnosis years 2018–2024 to ensure complete 365‐day post‐index follow‐up. Data from 2025 to 2026 were used only for descriptive analyses, and 2026 figures cover a partial calendar year (January–April 2026); estimates for the most recent years should therefore be interpreted with caution. Finally, the cohort was assembled from a single private healthcare network in Poland and may differ from the broader population of women with PCOS in socioeconomic status, health‐seeking behaviour, access to specialist care, and willingness to pay for non‐reimbursed therapies—characteristics that disproportionately affect uptake of incretin‐based agents prescribed predominantly for weight management. Care episodes and prescriptions occurring outside the network—including incretin‐based therapy initiated by non–LUX MED providers—are not captured, which would underestimate true treatment prevalence; because such out‐of‐network prescribing is unlikely to be related to calendar year, the resulting exposure misclassification is most plausibly non‐differential and would tend to attenuate rather than exaggerate the observed temporal trend. Absolute estimates of treatment prevalence and post‐diagnosis initiation rates may therefore not be generalizable to publicly insured, rural or socioeconomically disadvantaged populations, although the relative temporal patterns and between‐group contrasts are likely to remain informative.

5. Conclusions

In this large Polish real‐world cohort, GLP‐1 RA/GIP–GLP‐1 RA therapy in women with PCOS expanded markedly between 2018 and 2024, with initiation within 365 days of PCOS diagnosis rising from 0.14% to 5.97%. The temporal trend persisted after excluding women with coded type 2 diabetes mellitus, indicating expansion beyond classical glycaemic indications toward broader cardiometabolic and weight‐focused management. Because these estimates reflect prescription‐recorded initiation rather than confirmed pharmacological exposure, studies using dispensing, persistence and long‐term reproductive, metabolic and cardiovascular outcome data are needed to evaluate the appropriateness and effectiveness of incretin‐based therapy in this population.

Author Contributions

Conceptualization: A.D. and Z.S. Methodology: A.D. and Z.S. Formal analysis: A.D. and N.K. Data curation: J.D.‐K. and A.R. Resources: J.D.‐K. and A.R. Investigation: A.D. and N.K. Writing – original draft: A.D. Writing – review and editing: N.K., K.S., A.A., J.D.‐K., A.R. and Z.S. Supervision: J.D.‐K. and Z.S. All authors read and approved the final manuscript.

Funding

The authors have nothing to report.

Disclosure

The authors declare that generative artificial intelligence (Claude Opus 4.7, 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.

Conflicts of Interest

The authors declare no conflicts of interest.

Supporting information

Table S1: Ever use of glucose‐lowering and weight‐management medications in the polycystic ovary syndrome cohort (N = 38 263).

Table S2: First‐prescribed GLP‐1 RA or GIP–GLP‐1 RA agent by calendar year of initiation in women with polycystic ovary syndrome (N = 4557).

Table S3: Crude and directly standardized annual proportions of GLP‐1 RA/GIP–GLP‐1 RA initiation within 365 days after PCOS diagnosis, 2018–2024.

Table S4: Sensitivity of the temporal trend in GLP‐1 RA/GIP–GLP‐1 RA initiation to departures from the missing‐at‐random assumption for body mass index, 2018–2024.

Table S5: Temporal trend in GLP‐1 RA/GIP–GLP‐1 RA initiation under alternative post‐diagnosis initiation windows.

Table S6: Stratified temporal trends in GLP‐1 RA/GIP–GLP‐1 RA initiation within 365 days after PCOS diagnosis, by age and BMI category, 2018–2024.

Figure S1: Temporal trends in GLP‐1/GIP therapy within 365 days after first recorded diagnosis of polycystic ovary syndrome.

Dziewierz A., Kulicka N., Stolarska K., et al., “Temporal Trends and Clinical Characteristics of Incretin‐Based Therapy Use in Women With Polycystic Ovary Syndrome: A Real‐World Cohort Study From a Polish Private Healthcare Network,” Diabetes, Obesity and Metabolism 28, no. 9 (2026): 8045–8056, 10.1111/dom.71017.

Data Availability Statement

The data that support the findings of this study are available on request from the corresponding author. The data are not publicly available due to privacy and ethical restrictions, as they derive from the LUX MED healthcare network and are subject to institutional data‐governance requirements.