Metformin Use Before Assisted Reproductive Technology and Pregnancy Outcomes in Women With Polycystic Ovary Syndrome: A Nationwide Cohort Study

Abstract

Metformin has been hypothesized to improve ovulation and pregnancy rates in women with polycystic ovary syndrome (PCOS) undergoing assisted reproductive technology (ART). This population‐based cohort study assessed pregnancy outcomes associated with metformin use prior to ART in women with PCOS. We included 10,178 women with PCOS undergoing their first fresh embryo transfer cycle under the national health insurance program of South Korea between 2017 and 2021. Women who received metformin for ovulation induction within 12 weeks prior to ovarian stimulation were compared with those who did not receive metformin. ART outcomes included successful oocyte retrieval, embryo transfer, and clinical pregnancy, while pregnancy outcomes included spontaneous abortion, preterm birth, and live birth. Overlap‐weighted risk ratios (RR) were calculated for each outcome. Among the study population, 2,149 (21.1%) had used metformin prior to ovarian stimulation, and 8,029 (78.9%) had not. Metformin use did not improve or worsen ART outcomes, with comparable rates of oocyte retrieval (RR: 1.01, 95% confidence interval 1.00–1.02) and clinical pregnancy (RR: 1.05, 1.00–1.11), as well as no increased rates of spontaneous abortion (RR: 1.01, 0.84–1.21), preterm birth (RR: 0.99, 0.84–1.18), or live birth (RR: 1.05, 0.99–1.12). Subgroup analyses suggested modest benefits, including higher clinical pregnancy or live birth rates, among women aged ≥ 35 years, those undergoing a GnRH antagonist protocol, and those with ≥ 30 days of metformin use. In conclusion, metformin pretreatment before ART was not associated with any clinically meaningful benefits or harms in pregnancy outcomes among women with PCOS.

Study Highlights.

• WHAT IS THE CURRENT KNOWLEDGE ON THE TOPIC?

Metformin has been hypothesized to improve ovulation and pregnancy rates in women with polycystic ovary syndrome (PCOS) undergoing assisted reproductive technology (ART); however, prior clinical studies have reported inconsistent findings and were constrained by limited sample sizes.

• WHAT QUESTION DID THIS STUDY ADDRESS?

Does metformin use prior to ART improve pregnancy outcomes in women with PCOS?

• WHAT DOES THE STUDY ADD TO OUR KNOWLEDGE?

In this large nationwide cohort, metformin use prior to ART was not associated with clinically meaningful benefits or harms in pregnancy outcomes among women with PCOS. However, findings from the exploratory subgroup analyses indicate potential benefits in selected populations, underscoring the need for further targeted investigations.

• HOW MIGHT THIS CHANGE CLINICAL PHARMACOLOGY OR TRANSLATIONAL SCIENCE?

These findings inform the evidence‐based use of metformin in reproductive medicine by suggesting limited overall benefit of routine pre‐ART use. They also highlight the importance of treatment timing and patient selection in optimizing therapeutic effectiveness, supporting more individualized pharmacologic strategies and the need for targeted future studies.

Polycystic ovary syndrome (PCOS) is a common endocrine disorder affecting approximately 6–18% of women of reproductive age. ¹ , ² It is characterized by a triad of hyperandrogenism, menstrual irregularities, and polycystic ovaries, with clinical manifestations including chronic anovulation and infertility. ³ In fact, PCOS is the leading cause of anovulatory infertility, accounting for about 80% of such cases. ⁴ , ⁵ Even when ovulation occurs, subfertility may persist due to reduced oocyte quality and endometrial receptivity. ⁵ , ⁶

Infertility in women with PCOS is typically managed through a stepwise approach. First‐line therapy involves ovulation induction using agents including metformin or clomiphene citrate. If ovulation or pregnancy is not achieved, gonadotropins are used as second‐line therapy, and assisted reproductive technology (ART) such as in vitro fertilization (IVF) or intracytoplasmic sperm injection (ICSI) is considered the third‐line option. ⁵ A key pathophysiological feature of PCOS is insulin resistance with compensatory hyperinsulinemia, which contributes to excess ovarian androgen and impaired follicular development. ⁷ Metformin, an insulin‐sensitizing biguanide, targets this mechanism by improving insulin sensitivity and lowering circulating insulin levels. ⁸ Consequently, it reduces ovarian androgen output and increases sex hormone‐binding globulin, thereby ameliorating hyperandrogenism and improving ovulatory function. ⁹

Owing to these metabolic and reproductive effects, metformin has been empirically used as an adjunct therapy in women undergoing ART to improve pregnancy outcomes and reduce treatment‐related risks. ¹⁰ , ¹¹ However, clinical evidence on its efficacy remains inconsistent. Several small randomized trials have investigated peri‐ART metformin use in women with PCOS, yielding mixed results. ⁸ , ¹² , ¹³ These studies were limited by modest sample sizes (n = 40–160) and heterogeneity in patient populations and treatment protocols. Moreover, the reproductive effects of metformin may differ by factors such as age, insulin resistance, and ovarian stimulation protocol. ⁹ , ¹² , ¹³ Additional analyses are therefore warranted to explore these potential sources of heterogeneity. To address these uncertainties, the present study aimed to address this gap by leveraging nationwide real‐world data to assess whether pre‐ART metformin use improves the success rate of ART and pregnancy outcomes in women with PCOS.

MATERIALS AND METHODS

Study design and data source

This study was conducted using the target trial emulation framework. We first defined a hypothetical pragmatic trial and then emulated it using real‐world data (see Table S1 for details on the target trial protocol and emulation strategy). ¹⁴ We utilized data from South Korea's Health Insurance Review and Assessment (HIRA) database, which spans from October 2016 to September 2022. This nationwide claims database covers the entire Korean population of over 52 million people, making it highly representative. It provides longitudinal data on International Classification of Diseases, 10th Revision (ICD‐10) diagnosis codes, medical procedures, and prescribed medications across inpatient and outpatient settings, along with socioeconomic and demographic information, up to the point of death or emigration. The study was approved by HIRA (No. M20230424005) and the Institutional Review Board of Sungkyunkwan University, South Korea (No. 2025‐07‐109). The requirement for informed consent was waived due to the use of deidentified data.

Eligibility criteria and exposure

We included women who underwent an ovarian stimulation protocol for female infertility (ICD‐10 code N97) between October 1, 2017 and September 30, 2021. Ovarian stimulation was defined as a prescription for a gonadotropin in combination with either a GnRH agonist or antagonist within a 2‐week period. The date of the first gonadotropin prescription was considered the cohort entry date. We excluded those who were undergoing a frozen embryo transfer protocol to focus on those undergoing their first ART cycle. The cohort was then limited to those diagnosed with polycystic ovary syndrome (ICD‐10 code E28.2). Women with any diagnosis of diabetes mellitus (ICD‐10 codes E10–E14) were excluded to avoid confounding, as metformin could be prescribed for glycemic control in this population. In the Korean National Health Insurance system, metformin is reimbursed only for type 2 diabetes mellitus or for ovulation induction in PCOS; therefore, this exclusion criterion ensures that metformin use reflects ovulation induction.

From this final cohort, the exposed group was defined as women who were prescribed metformin from 12 weeks before the cohort entry date up to the day prior, whereas the unexposed group included women with no metformin prescriptions during this period. The 12‐week window was selected to capture metformin use during the period immediately preceding ovarian stimulation, consistent with clinical practice and the duration of metformin treatment evaluated in prior peri‐ART studies. ¹³ This window was intended to identify women with recent pre‐ART metformin exposure while minimizing inclusion of remote metformin use that was unlikely to reflect treatment during the ART cycle.

Study outcomes

We evaluated four ART outcomes: successful oocyte retrieval, 11 or more oocytes retrieved, embryo transfer, and clinical pregnancy. We also assessed three pregnancy outcomes: spontaneous abortion, preterm birth, and live birth. Successful oocyte retrieval was assessed within 4 weeks of the cohort entry date, and embryo transfer within 5 weeks. Clinical pregnancy was defined by the presence of a fetal ultrasound code within 8 weeks after embryo transfer. Spontaneous abortion was assessed within a prespecified outcome window from the operationally defined last menstrual period (LMP, 17 days before embryo transfer) through 20 weeks of gestation. Preterm birth was defined as a delivery occurring between 21 and 36 weeks after the LMP, while live birth was defined as any delivery occurring between 21 and 44 weeks after the LMP. Cases with diagnostic codes for fetal death were excluded from both live birth and preterm birth outcomes. Pregnancy episodes and outcomes were identified using relevant diagnosis and procedure codes using a hierarchical algorithm adapted from previously developed algorithms for identifying pregnancy outcomes in Korean claims databases. ¹⁵ , ¹⁶ The specific codes used to define all study variables are provided in Table S2 .

Covariates

We identified a wide range of demographic and clinical covariates based on their potential association with both the exposure and outcomes. These included demographics (e.g., maternal age, receipt of medical aid), indication‐related characteristics (e.g., fertilization method, other ovarian induction medication use), maternal comorbidities (e.g., pelvic inflammatory disease, leiomyoma, endometriosis), concurrent maternal medication use (e.g., acetaminophen, corticosteroids, antibiotics), and healthcare utilization patterns (e.g., outpatient visits, emergency room visits, inpatient visits). The assessment windows for each covariate are illustrated in Figure S1 .

Statistical analysis

To emulate randomization conditional on observed covariates, we applied a propensity score (PS) overlap weighting approach. PS was estimated using logistic regression models incorporating all identified covariates to predict the likelihood of metformin exposure. We assessed covariate balance between the exposed and unexposed groups both before and after weighting using absolute standardized mean differences (aSDs), with an aSD greater than 0.1 considered indicative of significant imbalance.

Because individuals who fail at an earlier stage of ART are no longer at risk for subsequent outcomes, we assessed stage‐specific risks conditional on successful completion of the preceding stage. ¹⁷ Specifically, the probability of success at each stage was estimated conditional on successful completion of the preceding stage: oocyte retrieval among all eligible ART cycles, embryo transfer among cycles with successful oocyte retrieval, clinical pregnancy among cycles with successful embryo transfer, and live birth, preterm birth, or spontaneous abortion among cycles resulting in clinical pregnancy. Thus, the probability of progressing through the ART pathway can be represented as the product of the corresponding stage‐specific conditional probabilities. Because conditioning on successful completion of an intermediate ART stage can result in collider bias, these conditional analyses should be interpreted as associations at each sequential stage of the ART process rather than as the overall effect of metformin on downstream outcomes.

For each outcome, we estimated the absolute risk per 100 cycles using binomial distributions. In addition, we calculated crude and PS‐weighted risk ratios (RRs), along with their corresponding 95% confidence intervals (CIs). All weighted analyses were conducted using robust variance estimators to account for the application of propensity score weights. PS were re‐calculated for each analysis. All data management and analysis were performed using SAS version 9.4 (SAS Institute Inc., Cary, NC, USA).

Subgroup and sensitivity analyses

We performed three prespecified subgroup analyses to explore potential effect modification. First, the study population was stratified by maternal age (< 35 vs. ≥ 35 years), given its strong association with fertility outcomes. We also examined subgroups based on ovarian stimulation protocol (GnRH agonist vs. GnRH antagonist). Lastly, because metformin prescriptions within the exposure assessment window could represent different treatment durations, we evaluated duration of metformin use before ovarian stimulation (< 30 vs. ≥ 30 days) to explore potential duration‐response relationships.

To assess the robustness of our main findings, we performed a series of sensitivity analyses. First, to address potential exposure misclassification, we redefined metformin exposure as having at least two prescriptions on distinct dates during the exposure assessment window. ¹⁸ Second, we narrowed the exposure assessment window to the 4 weeks prior to ovarian stimulation to examine the effect of more immediate exposure. Third, we limited the inclusion period to January 2019 through September 2021, coinciding with the expansion of ART reimbursement in Korea in 2019. Fourth, to increase the validity of identifying women with infertility due to PCOS and minimize confounding by indication, we restricted the analysis to those who also received clomiphene, a commonly used ovulation induction agent. Finally, we conducted external adjustment for maternal body mass index (BMI), impaired fasting glucose, pre‐pregnancy physical activity, tobacco use, and alcohol consumption using an external validation set derived from a subset of the study population. Although these variables were not available in the primary dataset, they are considered important potential confounders (see eAppendix 1 ).

RESULTS

Patient characteristics

Among 156,258 women who underwent an ovarian stimulation protocol during the study period, 10,578 had PCOS. After excluding 400 women with a history of diabetes mellitus, 10,178 women met the eligibility criteria for inclusion. Of these, 2,149 (21.1%) had used metformin prior to ovarian stimulation, and 8,029 (78.9%) had not (Figure 1 ). Before PS weighting, metformin users were less likely to be aged > 40 years (age 40–44: 5.8% vs. 9.4%, aSD: 0.138), more likely to undergo a GnRH agonist protocol (67.9% vs. 52.9%, aSD: 0.309), less likely to receive aromatase inhibitors (27.4% vs. 39.4%, aSD: 0.256), and had a lower prevalence of genital tract polyps (9.3% vs. 13.4%, aSD: 0.130). Nevertheless, the PS distributions showed good overlap between the two groups before weighting (Figure S2 ), and baseline characteristics were well balanced after PS weighting (Table 1 ).

Figure 1.

Study flow diagram. HIRA, Health Insurance Review and Assessment; PCOS, polycystic ovary syndrome.

Table 1.

Baseline maternal characteristics before and after propensity score weighting, stratified by metformin use

	Before weighting, n (%)		aSDa	After weighting, n (%)b
	Metformin users	Metformin non‐users		Metformin users	Metformin non‐users
Total, n	2,149	8,029	N/A	3,186	3,186
Demographic characteristics
Age group (years)
< 30	257 (12.0)	899 (11.2)	0.024	384 (12.1)	384 (12.1)
30–34	1,041 (48.4)	3,559 (44.3)	0.083	1,524 (47.8)	1,524 (47.8)
35–39	725 (33.7)	2,770 (34.5)	0.016	1,074 (33.7)	1,074 (33.7)
40–44	124 (5.8)	755 (9.4)	0.138	201 (6.3)	201 (6.3)
≥ 45	2 (0.1)	46 (0.6)	0.083	4 (0.1)	4 (0.1)
Cohort entry year
2017	129 (6.0)	237 (3.0)	0.148	150 (4.7)	150 (4.7)
2018	620 (28.9)	2,145 (26.7)	0.048	902 (28.3)	902 (28.3)
2019	525 (24.4)	2,111 (26.3)	0.043	799 (25.1)	799 (25.1)
2020	419 (19.5)	2,039 (25.4)	0.142	670 (21.0)	670 (21.0)
2021	456 (21.2)	1,497 (18.6)	0.064	666 (20.9)	666 (20.9)
Medical aid recipients	40 (1.9)	158 (2.0)	0.008	62 (1.9)	62 (1.9)
Nulliparity	1,891 (88.0)	6,922 (86.2)	0.053	2,792 (87.6)	2,792 (87.6)
Indication‐related characteristics
Type of ART
IVF	1,109 (51.6)	2,973 (37.0)	0.297	1,525 (47.8)	1,525 (47.8)
ICSI	743 (34.6)	3,573 (44.5)	0.204	1,186 (37.2)	1,186 (37.2)
Unknown	297 (13.8)	1,483 (18.5)	0.127	476 (14.9)	476 (14.9)
Type of protocol
GnRH agonist	1,459 (67.9)	4,251 (52.9)	0.309	2,054 (64.5)	2,054 (64.5)
GnRH antagonist	690 (32.1)	3,778 (47.1)	0.309	1,133 (35.5)	1,133 (35.5)
Other ovarian induction
Aromatase inhibitor	589 (27.4)	3,163 (39.4)	0.256	958 (30.1)	958 (30.1)
Clomiphene citrate	1,331 (61.9)	5,224 (65.1)	0.065	2,012 (63.1)	2,012 (63.1)
Hospital visits for PCOS
< 4	1,495 (69.6)	5,639 (70.2)	0.015	2,195 (68.9)	2,195 (68.9)
4–8	436 (20.3)	1,503 (18.7)	0.040	649 (20.4)	649 (20.4)
≥ 9	218 (10.1)	887 (11.0)	0.029	342 (10.7)	342 (10.7)
Maternal comorbidities
Depression/mood disorder	34 (1.6)	156 (1.9)	0.027	51 (1.6)	51 (1.6)
Endometriosis	63 (2.9)	321 (4.0)	0.058	100 (3.1)	100 (3.1)
Epilepsy	5 (0.2)	22 (0.3)	0.008	8 (0.2)	8 (0.2)
Autoimmune disorder	541 (25.2)	2,125 (26.5)	0.030	806 (25.3)	806 (25.3)
Pelvic inflammatory disease	105 (4.9)	492 (6.1)	0.054	168 (5.3)	168 (5.3)
Polyp of genital tract	200 (9.3)	1,079 (13.4)	0.130	325 (10.2)	325 (10.2)
Leiomyoma	107 (5.0)	572 (7.1)	0.090	170 (5.3)	170 (5.3)
Pre‐gestational hypertension	51 (2.4)	143 (1.8)	0.042	72 (2.3)	72 (2.3)
Psychotic disorder	0 (0.0)	7 (0.1)	0.042	0 (0.0)	0 (0.0)
Renal disease	8 (0.4)	35 (0.4)	0.010	13 (0.4)	13 (0.4)
Serious infection	41 (1.9)	160 (2.0)	0.006	62 (2.0)	62 (2.0)
Thyroid disorder	329 (15.3)	1,143 (14.2)	0.030	476 (14.9)	476 (14.9)
TORCH infection	17 (0.8)	85 (1.1)	0.028	27 (0.8)	27 (0.8)
Substance dependence	4 (0.2)	5 (0.1)	0.035	4 (0.1)	4 (0.1)
Other medication exposure
Antibiotics	1,486 (69.1)	5,824 (72.5)	0.075	2,228 (69.9)	2,228 (69.9)
Antihypertensives	93 (4.3)	327 (4.1)	0.013	135 (4.3)	135 (4.3)
Opioid analgesics	1,114 (51.8)	4,335 (54.0)	0.043	1,667 (52.3)	1,667 (52.3)
NSAIDs	1,528 (71.1)	5,915 (73.7)	0.057	2,289 (71.9)	2,289 (71.9)
Known teratogens	48 (2.2)	301 (3.7)	0.089	80 (2.5)	80 (2.5)
Benzodiazepines	161 (7.5)	652 (8.1)	0.023	242 (7.6)	242 (7.6)
Anticonvulsants	33 (1.5)	145 (1.8)	0.021	53 (1.7)	53 (1.7)
Antipsychotics	14 (0.7)	58 (0.7)	0.009	20 (0.6)	20 (0.6)
Antidepressants	21 (1.0)	86 (1.1)	0.009	31 (1.0)	31 (1.0)
Acetaminophen	307 (14.3)	943 (11.7)	0.076	432 (13.6)	432 (13.6)
Spironolactone	0 (0.0)	2 (0.0)	0.022	0 (0.0)	0 (0.0)
Systemic corticosteroids	1,115 (51.9)	4,333 (54.0)	0.042	1,674 (52.6)	1,674 (52.6)
Healthcare utilization
No. outpatient visits
< 3	25 (1.2)	45 (0.6)	0.065	30 (0.9)	30 (0.9)
3–5	79 (3.7)	172 (2.1)	0.091	98 (3.1)	98 (3.1)
6–8	150 (7.0)	391 (4.9)	0.089	200 (6.3)	200 (6.3)
≥ 9	1,895 (88.2)	7,421 (92.4)	0.144	2,859 (89.7)	2,859 (89.7)
No. inpatient visits
0	1,906 (88.7)	6,976 (86.9)	0.055	2,814 (88.3)	2,814 (88.3)
1	175 (8.1)	847 (10.5)	0.083	275 (8.6)	275 (8.6)
≥ 2	68 (3.2)	206 (2.6)	0.036	97 (3.0)	97 (3.0)
No. emergency room visits
0	1,844 (85.8)	6,659 (82.9)	0.079	2,714 (85.2)	2,714 (85.2)
1	251 (11.7)	1,137 (14.2)	0.074	389 (12.2)	389 (12.2)
≥2	54 (2.5)	233 (2.9)	0.024	83 (2.6)	83 (2.6)

ART, assisted reproductive technology; aSD, absolute standardized mean difference; ICSI, intracytoplasmic sperm injection; IVF, in vitro fertilization; NSAIDs, non‐steroidal anti‐inflammatory drugs; PCOS, polycystic ovary syndrome; SD, standard deviation.

^a A maximum pairwise aSD of over 0.1 indicates a significant imbalance between groups.

^b The numbers presented represent weighted pseudo‐population counts, calculated as the sum of the overlap weights. Overlap weighting targets the population with the greatest covariate overlap and achieves exact balance in the weighted marginal distributions of the measured baseline covariates; therefore, the weighted distributions and corresponding aSDs are identical (aSD = 0) between exposure groups.

Primary analysis

In the primary analysis, there were no significant differences in ART outcomes between metformin users and non‐users (Figure 2 ). Clinical pregnancy was achieved in 49.7% of metformin users and 43.8% of non‐users. Although marginal, this difference remained after adjustment (RR: 1.05, 1.00–1.11). The RR for oocyte retrieval was 1.01 (95% CI: 1.00–1.02), for retrieval of ≥ 11 oocytes was 1.03 (0.97–1.09), and for embryo transfer was 1.00 (0.98–1.02).

Figure 2.

Relative risks for study outcomes following metformin use prior to ovarian stimulation in women with polycystic ovary syndrome. CI, confidence interval; PS, propensity score.

Similarly, live birth was achieved in 40.1% of metformin users and 34.8% of non‐users. After adjustment, although not statistically significant, metformin pretreatment was associated with a trend toward higher rates of live birth (RR: 1.05, 0.99–1.12). No significant differences were observed in adverse pregnancy outcomes, with RRs of 1.01 (0.84–1.21) for spontaneous abortion and 0.99 (0.84–1.18) for preterm birth. The denominators for each risk set used for risk estimation are provided in Table S3 .

Subgroup analysis

Figure 3 presents the results of the subgroup analysis. Among women of higher maternal age (≥ 35 years), metformin use was associated with a greater number of oocytes retrieved (RR: 1.13, 95% CI: 1.03–1.25) and a higher clinical pregnancy rate (RR: 1.10, 95% CI: 1.01–1.19) but was not associated with any significant adverse pregnancy outcomes. Among women undergoing a GnRH antagonist protocol, metformin use was associated with a modest increase in live birth rate (RR: 1.11, 95% CI: 1.01–1.22). In addition, metformin use of 30 or more days prior to ovarian stimulation was associated with a higher number of oocytes retrieved (RR: 1.12, 1.02–1.24) and a higher clinical pregnancy rate (RR: 1.11, 1.02–1.21).

Figure 3.

Results of the subgroup analysis stratified by maternal age, ovarian stimulation protocol, and metformin use duration. CI, confidence interval; GnRH, gonadotropin‐releasing hormone; RR, weighted risk ratio.

Sensitivity analysis

Results from the sensitivity analyses were consistent with the primary findings, with overlapping 95% CIs (Figure 4 ). Details of the results of the secondary analyses can be found in: Tables S4–S11 . Proximity of metformin exposure to ART procedure did not affect the risks of study outcomes. External adjustment using data from a subset of the study population for maternal BMI, impaired fasting glucose, physical activity, and tobacco and alcohol use yielded similar estimates, supporting the validity of the findings against the influence of these unmeasured confounders (Table S12 ).

Figure 4.

Results of the sensitivity analysis. CI, confidence interval; RR, weighted risk ratio. ^aMaternal body mass index, impaired fasting glucose, physical activity, smoking, and drinking.

DISCUSSION

In this large, population‐based, real‐world cohort study, we found that metformin use before ART did not confer a significant improvement in fertility outcomes for women with PCOS, nor was it associated with an increased risk of adverse pregnancy outcomes. In the PS‐weighted analysis, metformin users and non‐users had non‐differential probabilities of oocyte retrieval, embryo transfer, clinical pregnancy, and live birth. These findings provide meaningful evidence that helps clarify the longstanding uncertainty in this field, which has largely resulted from previous studies being constrained by low number of events and low quality of evidence. ⁸ , ¹² , ¹³ By leveraging a nationwide dataset and robust trial emulation methods, our study strengthens the current evidence base and suggests that routine metformin pretreatment may not meaningfully improve ART outcomes in all women with PCOS.

Importantly, our findings regarding ART‐related pregnancy outcomes in women with PCOS are consistent with recent evidence syntheses. The most recent Cochrane review, which included 13 randomized controlled trials involving around 1,100 women with PCOS combined, concluded that there is no conclusive evidence that metformin use before or during ART improves live birth rates or reduces miscarriage risk. ¹³ Also consistent with our findings, a meta‐analysis demonstrated that metformin use was not associated with significant improvements in clinical pregnancy or live birth rates compared with controls. ¹² In line with this evidence, specialty societies such as the Royal College of Obstetricians and Gynaecologists and the European Society of Human Reproduction and Embryology have expressed that metformin has a limited role in enhancing reproductive outcomes in women with PCOS. ¹⁹ , ²⁰ Our findings align with this perspective, suggesting that adjunctive metformin therapy provides no meaningful benefit for pregnancy outcomes in women with PCOS undergoing ART.

Metformin has long been used as an adjunct to ovarian stimulation because of its potential to enhance reproductive outcomes through metabolic and endocrine mechanisms. The drug improves insulin sensitivity by suppressing hepatic glucose production and increasing glucose uptake and utilization in peripheral tissues. ²¹ In addition, metformin may directly inhibit androgen synthesis in ovarian thecal cells, thereby mitigating the hyperandrogenic milieu characteristic of polycystic ovary syndrome. ²² However, the inconsistency between its mechanistic rationale and observed clinical benefits may be partly explained by differences in the duration of pretreatment. In our study, the broad definition of pre‐ART metformin exposure was intended to capture recent use during the clinically relevant period preceding ovarian stimulation, but encompassed women with heterogeneous treatment durations. Notably, longer metformin use (30 days or more) was associated with a higher probability of clinical pregnancy compared with shorter use, suggesting that treatment duration may be relevant to its potential reproductive effects. Consistently, a clinical trial requiring 2 months of metformin pretreatment demonstrated significant improvement in pregnancy outcomes, ²³ whereas another with only 16 days of pretreatment showed contrary results. ²⁴ Similarly, another study found that continuation of preconceptional metformin therapy throughout the first trimester improves pregnancy outcomes in women with PCOS. ²⁵ Currently, there is no consensus on the optimal duration of metformin use before ART, but our findings suggest that sufficient exposure duration may be an important consideration when evaluating its potential benefits.

Our exploratory subgroup analyses suggested that two additional patient subsets might experience modest benefits from metformin use. First, higher maternal age was associated with improved oocyte retrieval and clinical pregnancy. It is postulated that metformin might have stronger metabolic and ovulatory benefits, especially in women with higher age and related insulin resistance. ²⁶ Second, we found that in women undergoing a GnRH antagonist protocol for ovarian stimulation, metformin use was associated with a slightly higher live birth rate compared to non‐use. However, there have been inconsistent results across trials. ¹³ , ²⁴ One clinical trial found a lower live birth rate among women receiving metformin with a GnRH agonist protocol—interpreted by the authors as a chance finding—while other studies showed inconclusive results for GnRH agonist protocol (pooled RR: 1.30, 95% CI: 0.94–1.79). One possible explanation is that GnRH antagonist protocols are often chosen for PCOS patients at high risk of OHSS (ovarian hyperstimulation syndrome), ²⁷ and metformin's known ability to lower OHSS incidence ¹³ , ²⁸ might allow safer completion of such cycles, indirectly improving the chances of a live birth. Taken together, these findings raise the possibility that the clinical context may influence the association between metformin use and ART outcomes. However, because the primary analyses were largely null, these findings should be interpreted cautiously given the potential for chance findings from multiple comparisons.

The strengths of our study make the main findings more reliable. Our study, which included over 10,000 women with PCOS undergoing ART, represents one of the largest analyses to date and provided sufficient statistical power. We also employed a robust target trial emulation design with propensity score weighting to approximate a randomized comparison, thereby minimizing confounding bias. The cohort was drawn from a nationwide healthcare database, enhancing the generalizability of results to real‐world clinical practice and providing broader applicability than studies limited to single‐center trial populations. Another strength of our analysis is the inclusion of oocyte retrieval as an intermediate ART outcome, which has not been consistently evaluated in previous meta‐analyses.

Despite these strengths, several limitations of this study should be acknowledged. First, although we attempted to control for many baseline differences, residual confounding may still be present due to the observational nature of the study. To assess residual confounding by unmeasured confounders, we used an external dataset including information on anthropometric information and lifestyle‐related characteristics to calibrate the estimates, in which we found minimal effect from these confounders. However, detailed measures of PCOS phenotype and severity, such as hyperandrogenism, menstrual characteristics, and insulin resistance, as well as clinical factors influencing ART treatment selection, including infertility duration and detailed ovarian stimulation characteristics, were not available in the primary dataset. These factors therefore could have contributed to residual confounding. Second, because our exposure definition relied on prescription records, patient adherence to metformin could not be verified, and women who discontinued metformin before ovarian stimulation may have been classified as exposed. To address potential exposure misclassification, we conducted sensitivity analyses defining exposure as 2 or more prescriptions on distinct dates and restricting the exposure assessment window to the 4 weeks preceding embryo transfer, with consistent results. These findings provide some reassurance that the observed associations were not driven solely by intermittent prescriptions or remote metformin exposure. Third, the study population—South Korean women with PCOS undergoing fresh embryo transfer cycles—may limit generalizability and warrant replications in other settings. Fourth, because spontaneous abortion was identified from administrative claims, very early or subclinical pregnancy losses that did not result in a healthcare encounter may have been missed. Finally, while our data are robust, they cannot establish causality in the way a randomized trial could.

In summary, our results suggest that adding metformin before ART neither significantly improves the chances of pregnancy or live birth nor increases the risk of adverse pregnancy outcomes in women with PCOS. Nevertheless, the exploratory findings from our subgroup analyses highlight the possibility that treatment effects of metformin may vary across clinical contexts and patient characteristics. Future studies are warranted to systematically investigate whether specific factors modify the effects of metformin on ART outcomes in women with PCOS. Meanwhile, current evidence does not support the empirical use of metformin in all women with PCOS as an adjunct therapy to a fresh embryo transfer cycle.

FUNDING

This work was supported by a grant [RS‐2026‐25475077] from the National Research Foundation of Korea, which is funded by the Korean Government (Ministry of Science and Information and Communication Technology). This research was supported by the Korea National Institute of Health research project (2024‐ER1104‐00, ‐01, ‐02, 2025‐ER1103‐00, ‐01). The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.

CONFLICTS OF INTEREST

Dr. Shin received grants from the Ministry of Food and Drug Safety, the National Research Foundation of Korea, and grants from Pfizer, Johnson & Johnson, and GSK outside the submitted work. No other relationships or activities have influenced the submitted work.

AUTHOR CONTRIBUTIONS

Y.C. and Y.S.P. wrote the manuscript; Y.C., Y.S.P., Y.L., H.L., J.‐Y.S., and S.‐H.K. designed the research; Y.C. performed the research; Y.C. analyzed the data.

USE OF ARTIFICIAL INTELLIGENCE

ChatGPT (OpenAI) was used to assist with language editing and grammar during preparation of the manuscript. The authors reviewed all AI‐assisted edits and take full responsibility for the content, accuracy, and integrity of the manuscript.

Supporting information

Data S1

DATA AVAILABILITY STATEMENT

Data generated and/or analyzed during the current study cannot be shared publicly due to the data‐sharing policy of the Health Insurance Review and Assessment Service (HIRA) of Korea.