Cosmos Study

GLP-1s Lead to Greater Weight Loss and A1C Improvement Than Metformin in Patients with PCOS

November 13, 2025

Team A:Kersten Bartelt, RNDan GreerEmily Higgs

Team B:Louis Kazaglis, MDTed Stamp

Key Findings

Among patients with polycystic ovary syndrome (PCOS), the median one-year weight change was an 11.5% loss for those treated with a GLP-1, compared to a 1.9% loss among those treated with metformin.
Among patients with PCOS, 55.7% of those prescribed a GLP-1 lost at least 10% of their body weight, compared to just 13.7% of those prescribed metformin.
Among patients with PCOS who were prescribed a GLP-1, 83.5% experienced an HbA1c reduction, compared to only 55.9% of those prescribed metformin.

Polycystic ovary syndrome (PCOS) affects 10 to 13% of reproductive-age women and is closely linked to obesity, insulin resistance, and increased risk of type 2 diabetes.¹ Metformin has historically been the first-line therapy for metabolic management in PCOS.¹ Recently, GLP-1s have emerged as alternatives due to their effects on weight reduction and glycemic control.² With rising off-label use of GLP-1s in PCOS, comparative evidence with metformin is needed to inform clinical practice and health policy.

We studied 36,674 women aged 18 to 50 with a diagnosis of PCOS who filled a new prescription for either a GLP-1 or metformin between January 2021 and November 2024. Patients were required to have at least one baseline measurement of weight or HbA1c within a year before drug initiation and one follow-up measurement between 9 and 15 months after initiation.

At one year, patients with PCOS on GLP-1s experienced significantly greater weight loss than those on metformin. The median weight change among GLP-1 users was an 11.5% reduction, compared to a 1.9% reduction for metformin users. Many GLP-1 patients (55.7%) lost over 10% of their body weight, while only 13.7% of metformin patients reached that threshold.

Figure 1

One-Year Percent Change in Body Weight by Treatment

Figure 1. The distribution of one-year weight change among patients with PCOS by treatment type.

For glycemic outcomes, the distribution of absolute HbA1c changes showed greater reductions for GLP-1 users compared to metformin users, as seen in Figure 2. A lower HbA1c is a marker of better blood glucose regulation, with a value below 5.7% representing normal glucose levels, a value between 5.7% and 6.4% representing prediabetes, and a value over 6.4% representing diabetes.³ Patients on GLP-1s had a median HbA1c reduction of 0.5 points, while those on metformin had a median HbA1c reduction of 0.1 points. 83.5% of patients who received GLP-1s had a reduction of their HbA1c, while only 55.9% of those who received metformin did.

Figure 2

HbA1c Change by Treatment

Figure 2. The distribution of HbA1c change among patients with PCOS by treatment type.

A sensitivity analysis accounting for factors such as demographics, starting BMI classification, residence in a socially vulnerable area, comorbidities, and baseline HbA1c level also found that patients on GLP-1s were more likely to experience weight loss and to reduce their HbA1c levels.

These data come from Cosmos, a dataset created in collaboration with a community of Epic health systems representing more than 300 million patient records from 1,800 hospitals and more than 41,000 clinics from all 50 U.S. states, Canada, Lebanon, and Saudi Arabia. This study was completed by two teams that worked independently, each composed of a clinician and research scientists. The two teams came to similar conclusions. Graphics by Brian Olson.

References

Teede HJ, Tay CT, Laven JJE, et al. Recommendations From the 2023 International Evidence-based Guideline for the Assessment and Management of Polycystic Ovary Syndrome. J Clin Endocrinol Metab. 2023;108(10):2447-2469. doi:10.1210/clinem/dgad463
Lin S, Deng Y, Huang J, et al. Efficacy and safety of GLP-1 receptor agonists on weight management and metabolic parameters in PCOS women: a meta-analysis of randomized controlled trials. Sci Rep. 2025;15(1):16512. Published 2025 May 13. doi:10.1038/s41598-025-99622-4
Ellis RR. Hemoglobin A1c (HbA1c): What to know if you have diabetes or prediabetes or are at risk for these conditions. Harvard Health. June 30, 2025. https://www.health.harvard.edu/diseases-and-conditions/hemoglobin-a1c-hba1c-what-to-know-if-you-have-diabetes-or-prediabetes-or-are-at-risk-for-these-conditions. Accessed November 5, 2025.

Data Definitions

Study period: 1/1/2021 to 10/1/2024 with follow-up possible through 7/1/2025

Study population: inclusion

Patients with:

PCOS (encounter or billing diagnosis with ICD-10-CM code E28.2) before exposure or within two years of drug trial start
Evaluated sex of female or ambiguous
Aged 18 to 50 at drug trial start
No overlap in metformin and GLP-1 use
At least one measurement within one year prior to trial start
At least one measurement 9 to 15 months after trial start
At least 70% of days covered by their exposure medication dispense

Study population: exclusion

Patients with:

Evidence of pregnancy during baseline or outcome measurements
Definitive pregnancy start date: Between 12 months before the first reading through the follow-up reading
Definitive pregnancy end date, delivery, or if it’s ambiguous what point in the pregnancy the patient would be at (e.g. an encounter or billing diagnosis with ICD-10-CM O*): Between 12 months before the first reading through 9 months after the follow-up reading
Evidence of amputation during baseline or outcome measurements for weight outcome

Censoring

First of:

Measurement closest to 12 months.
Last dispense + days supply. Default days supply = 30 for GLP1s and 90 for metformin. Day supply capped at 180 days.
One year without dispense.

Exposures: The first of:
Metformin: Dispense of simple generic “Metformin HCl” or ATC code A10BA02
GLP-1: Dispense of a simple generic of “dulaglutide,” “exenatide,” “exenatide microspheres,” “liraglutide,” “semaglutide,” or “tirzepatide” or a pharmaceutical class of “ANTIHYPERGLYCEMIC – INCRETIN MIMETICS COMBINATION,” “ANTIHYPERGLY,INSULIN,LONG ACT-GLP-1 RECEPT.AGONIST,” “ANTI-OBESITY GLUCAGON-LIKE PEPTIDE-1 RECEP AGONIST,” or “ANTIHYPERGLY,INCRETIN MIMETIC(GLP-1 RECEP.AGONIST)”

Outcomes: Percent change in weight from baseline
Change in HbA1c

Dispense

Available medication dispense history meeting the following criteria:

Covering the six months prior to baseline measurement, including the baseline measurement
Covering >= nine months of time between drug exposure start and final measurement
With no orders of the exposure medication more than two weeks prior to the drug exposure start date
With no orders of the other exposure medication between baseline reading date and the follow-up reading date

HbA1c: A lab with LOINC code 17856-6, 17855-8, 4548-4, 4549-2, 55454-3, or 41995-2 and a value between 3.2 and 20.

Race and ethnicity: Patients were classified based on self-reported race and ethnicity that were mapped to standards as Hispanic, Multiracial, White only, Black only, or Other

Amputation: An encounter or billing diagnosis with ICD-10-CM code Z89.2*, Z89.6*, S78.011A, S98.011A, S78.012A, S98.012A, S78.019A, S98.019A, S78.021A, S98.021A, S78.022A, S98.022A, S78.029A, S98.029A, S78.111A, S98.111A, S78.112A, S98.112A, S78.119A, S98.119A, S78.121A, S98.121A, S78.122A, S98.122A, S78.129A, S98.129A, S78.911A, S98.131A, S78.912A, S98.132A, S78.919A, S98.139A, S78.921A, S98.141A, S78.922A, S98.142A, S78.929A, S98.149A, S88.011A, S98.211A, S88.012A, S98.212A, S88.019A, S98.219A, S88.021A, S98.221A, S88.022A, S98.222A, S88.029A, S98.229A, S88.111A, S98.311A, S88.112A, S98.312A, S88.119A, S98.319A, S88.121A, S98.321A, S88.122A, S98.322A, S88.129A, S98.329A, S88.911A, S98.911A, S88.912A, S98.912A, S88.919A, S98.919A, S88.921A, S98.921A, S88.922A, S98.922A, S88.929A, or S98.929A or a procedure with ICD-10-PCS code 0Y6[23478CDFGHJ]* or 0X6[012389BCDF]*

Limitations: Because medication dispense information relies upon third-party queries operating in the U.S. and this was required for the study, patients are U.S. based.