Triglyceride-glucose index and assisted reproductive outcomes in non-PCOS women: evidence from a propensity score-weighted cohort of 9,903 patients.

A total of 9,903 patients were included, with a mean age of 34.2 years and a mean BMI of 22.4 kg/m². The average infertility duration was 4.7 years, and the mean AMH level was 3.1 ng/mL. Most patients had secondary infertility (69.2%), and tubal factor was the leading infertility diagnosis (61.3%). IVF was the predominant fertilization method (67.7%), and the agonist regimen was the most commonly used ovarian stimulation protocol (56.4%). Table  1 and eTable 1 present the baseline characteristics of patients in the oocyte/embryo analysis cohort stratified by quartiles of the TyG index (Q1 ≤ 7.95, Q2 = 7.96–8.28, Q3 = 8.29–8.67, Q4 ≥ 8.68). These characteristics reflect the PS-IPW–weighted analytic population; therefore, the effective sample size represents the sum of weights rather than the number of unique patients. Before PS-IPW, patients in the higher TyG quartiles, particularly Q4, were older and had higher BMI and total cholesterol levels, with additional differences observed in the distribution of infertility type, infertility diagnoses, and ovarian stimulation protocols. The most pronounced imbalances were noted for BMI (SMD = 0.53), total cholesterol (SMD = 0.43), and age (SMD = 0.20). After PS-IPW, a total of 10,077.4 weighted samples were included, with approximately balanced distribution across quartiles. Most variables achieved satisfactory balance (SMD < 0.1), although BMI remained slightly imbalanced (SMD = 0.11). Table 1 Baseline characteristic by quartiles of the Triglyceride–Glucose (TyG) index after Propensity Score Inverse Probability Weighting (PS-IPW) Characteristic Overall, n  = 10,077.4 Q1 = 2,564.0 Q2 = 2,540.9 Q3 = 2,486.6 Q4 = 2,485.9 SMD Age 34.20 (5.19) 34.11 (5.19) 34.32 (5.19) 34.23 (5.08) 34.13 (5.29) 0.0004 BMI (kg/m 2 ) 22.63 (3.82) 22.92 (4.46) 22.83 (4.28) 22.34 (3.24) 22.43 (3.01) 0.1102 AMH (ng/mL) 3.12 (2.63) 3.12 (2.56) 3.14 (2.57) 3.10 (2.65) 3.14 (2.74) 0.0019 Infertility duration (y) 4.65 (3.89) 4.62 (3.71) 4.63 (3.96) 4.65 (3.97) 4.69 (3.94) 0.0168 Gravidity 1.62 (1.58) 1.59 (1.58) 1.65 (1.59) 1.61 (1.53) 1.65 (1.63) 0.0121 Parity 0.40 (0.61) 0.40 (0.59) 0.40 (0.61) 0.41 (0.62) 0.41 (0.63) 0.0097 Basal E2 (pg/mL) 51.59 (133.02) 50.88 (107.65) 51.27 (148.47) 53.02 (145.39) 51.20 (126.99) 0.0031 Basal FSH (mIU/mL) 6.24 (3.43) 6.32 (2.90) 6.24 (3.44) 6.24 (4.11) 6.16 (3.17) 0.0230 Basal LH (mIU/mL) 3.74 (3.28) 3.78 (3.17) 3.72 (3.29) 3.74 (3.28) 3.72 (3.36) 0.0081 Total Gn 1722.51 (840.15) 1734.36 (844.81) 1719.41 (838.87) 1713.36 (846.96) 1722.62 (830.13) 0.0081 TC (mmol/L) 4.88 (0.93) 4.91 (0.97) 4.92 (0.94) 4.86 (0.88) 4.82 (0.94) 0.0640 Fertilization method (%) 0.0137 IVF 6703.7 (66.5) 1679.5 (65.5) 1692.5 (66.6) 1679.1 (67.5) 1652.6 (66.5) ICSI 3373.7 (33.5) 884.5 (34.5) 848.4 (33.4) 807.5 (32.5) 833.3 (33.5) Type of infertility (%) b 0.0121 Primary 3121.6 (31.0) 828.3 (32.3) 756.9 (29.8) 771.0 (31.0) 765.3 (30.8) Secondary 6955.8 (69.0) 1735.6 (67.7) 1784.0 (70.2) 1715.5 (69.0) 1720.7 (69.2) Infertility diagnosis (%) c 0.0170 Tubal factor 6078.2 (60.3) 1499.3 (58.5) 1554.8 (61.2) 1524.9 (61.3) 1499.1 (60.3) Male factor 1308.7 (13.0) 365.8 (14.3) 320.0 (12.6) 307.6 (12.4) 315.3 (12.7) Ovulatory 899.1 (8.9) 252.0 (9.8) 218.4 (8.6) 212.5 (8.5) 216.2 (8.7) Genetic factor 634.7 (6.3) 153.2 (6.0) 150.3 (5.9) 155.8 (6.3) 175.5 (7.1) Endometriosis 205.7 (2.0) 49.4 (1.9) 54.6 (2.1) 55.0 (2.2) 46.8 ( 1.9) RPL 294.5 (2.9) 86.8 (3.4) 70.1 (2.8) 70.5 (2.8) 67.0 (2.7) Other 656.6 (6.5) 157.6 (6.1) 172.7 (6.8) 160.2 (6.4) 166.0 (6.7) Ovarian stimulation protocol, No. (%) d 0.0152 Agonist 5711.7 (56.7) 1493.4 (58.2) 1426.4 (56.1) 1404.6 (56.5) 1387.3 (55.8) Antagonist 3012.2 (29.9) 743.4 (29.0) 777.8 (30.6) 735.7 (29.6) 755.3 (30.4) Luteal-phase stimulation 113.8 (1.1) 24.3 (0.9) 23.9 (0.9) 26.1 (1.0) 39.5 (1.6) PPOS 785.4 (7.8) 192.1 (7.5) 194.2 (7.6) 203.5 (8.2) 195.6 (7.9) Mild Stimulation 398.5 (4.0) 98.9 (3.9) 103.8 (4.1) 102.8 (4.1) 93.0 (3.7) Natural cycles 55.8 (0.6) 11.9 (0.5) 14.8 (0.6) 13.9 (0.6) 15.2 (0.6) Data are mean ± SD or n (%). Values are PS-IPW–weighted; sample sizes may be non-integers. SMD indicates standardized mean difference; |SMD| ≥ 0.10 suggests meaningful imbalance. TyG quartile cutoffs: Q1 ≤ 7.95, Q2 7.96–8.28, Q3 8.29–8.67, Q4 ≥ 8.68 Abbreviations : BMI body mass index (calculated as weight in kilograms divided by height in meters squared), AMH anti-Müllerian hormone, E2 estradiol, FSH follicle-stimulating hormone, LH luteinizing hormone, Gn gonadotropin, TC total cholesterol, IVF in vitro fertilization, ICSI intracytoplasmic sperm injection, PPOS progestin-primed ovarian stimulation, RPL recurrent pregnancy loss, SMD standardized mean difference a Percentages may not total 100% owing to rounding b Infertility type was categorized as primary (patients who have never conceived) or secondary (patients with at least one prior conception) c Infertility diagnosis categories were mutually exclusive; when multiple causes were present in the same patient, only one primary diagnosis was assigned. Tubal factor infertility refers to damage or blockage of the fallopian tubes; male factor infertility indicates reduced sperm count or motility or other sperm dysfunction; ovulatory dysfunction refers to abnormal or absent ovulation; genetic factor includes known chromosomal or genetic abnormalities associated with infertility; endometriosis-related infertility refers to ectopic endometrial tissue affecting fertility; RPL (recurrent pregnancy loss) was defined as two or more consecutive pregnancy losses; “other” includes causes of infertility not classified into the above categories d Ovarian stimulation protocols included agonist, antagonist, luteal-phase stimulation, PPOS (progestin-primed ovarian stimulation), mild stimulation, and natural cycle protocols Baseline characteristic by quartiles of the Triglyceride–Glucose (TyG) index after Propensity Score Inverse Probability Weighting (PS-IPW) Data are mean ± SD or n (%). Values are PS-IPW–weighted; sample sizes may be non-integers. SMD indicates standardized mean difference; |SMD| ≥ 0.10 suggests meaningful imbalance. TyG quartile cutoffs: Q1 ≤ 7.95, Q2 7.96–8.28, Q3 8.29–8.67, Q4 ≥ 8.68 Abbreviations : BMI body mass index (calculated as weight in kilograms divided by height in meters squared), AMH anti-Müllerian hormone, E2 estradiol, FSH follicle-stimulating hormone, LH luteinizing hormone, Gn gonadotropin, TC total cholesterol, IVF in vitro fertilization, ICSI intracytoplasmic sperm injection, PPOS progestin-primed ovarian stimulation, RPL recurrent pregnancy loss, SMD standardized mean difference a Percentages may not total 100% owing to rounding b Infertility type was categorized as primary (patients who have never conceived) or secondary (patients with at least one prior conception) c Infertility diagnosis categories were mutually exclusive; when multiple causes were present in the same patient, only one primary diagnosis was assigned. Tubal factor infertility refers to damage or blockage of the fallopian tubes; male factor infertility indicates reduced sperm count or motility or other sperm dysfunction; ovulatory dysfunction refers to abnormal or absent ovulation; genetic factor includes known chromosomal or genetic abnormalities associated with infertility; endometriosis-related infertility refers to ectopic endometrial tissue affecting fertility; RPL (recurrent pregnancy loss) was defined as two or more consecutive pregnancy losses; “other” includes causes of infertility not classified into the above categories d Ovarian stimulation protocols included agonist, antagonist, luteal-phase stimulation, PPOS (progestin-primed ovarian stimulation), mild stimulation, and natural cycle protocols eTables 2 and 3 summarize the baseline characteristics of patients undergoing fresh embryo transfer. Before weighting ( n  = 5,255), higher TyG quartiles were also associated with older age, higher BMI, and higher total cholesterol, along with differences in infertility type, day of transfer, and stimulation protocols. The largest imbalances were again observed for BMI (SMD = 0.52), total cholesterol (SMD = 0.42), and age (SMD = 0.21). After PS-IPW, 5,341.4 weighted samples were included, and nearly all variables were well balanced across quartiles, although BMI showed mild residual imbalance (SMD = 0.13). Overall, application of PS-IPW substantially reduced baseline differences, ensuring comparability of groups for subsequent outcome analyses. Table  2 summarizes the cycle outcomes across TyG index quartiles after application of PS-IPW. Before weighting, significant differences were observed among the four quartiles in most cycle variables, including the numbers of oocytes retrieved, mature oocytes, fertilized oocytes, blastocysts cultured, blastocyst formation, and time from stimulation initiation to oocyte retrieval (all p < .01). After PS-IPW, these differences were attenuated, and no clear pattern of between-quartile variation was observed for cycle outcomes. Table 2 Association between TyG index quartiles and cycle characteristics after PS-IPW: relative risks or coefficients with 95% CIs Cycle Characteristics Q1 = 2,564.0 Q2 = 2,540.9 Q3 = 2,486.6 Q4 = 2,485.9 Oocytes retrieved 12.2 (7.8) 11.7 (8) 11.6 (7.6) 10.9 (8) Ref (1.0) 1.043 (0.996–1.091) 1.008 (0.973–1.045) 0.988 (0.951–1.026) Mature oocytes 10.6 (6.9) 10.2 (7.1) 10 (6.7) 9.5 (7.1) Ref (1.0) 1.039 (0.994–1.086) 1.001 (0.966–1.038) 0.982 (0.944–1.021) Mature rate (per oocyte retrieved) a 88.2 (0.3) 88.4 (0.3) 87.8 (0.3) 88.7 (0.3) Ref (1.0) -0.054 (-1.019–0.911) -0.709 (-1.643–0.226) -0.352 (-1.373–0.668) Total fertilized 7.4 (5.3) 7.1 (5.3) 7 (5) 6.6 (5.3) Ref (1.0) 1.044 (0.994–1.096) 0.998 (0.958–1.039) 0.992 (0.949–1.037) Fertilization rate (per oocyte retrieved) a 61.6 (0.5) 62.7 (0.5) 61.5 (0.5) 62.2 (0.5) Ref (1.0) 1.246 (-0.277–2.768) -0.016 (-1.518–1.487) 0.791 (-0.796–2.379) Blastocyst cultured 9.1 (5.2) 9 (5.2) 8.6 (5) 8.6 (5.4) Ref (1.0) 1.034 (0.982–1.089) 0.979 (0.939–1.021) 0.974 (0.927–1.023) Blastocyst formation 5.9 (4.1) 5.6 (4) 5.4 (3.8) 5.3 (4.2) Ref (1.0) 1.002 (0.940–1.067) 0.955 (0.906–1.006) 0.953 (0.896–1.014) Blastocyst formation rate a 62.6 (1.1) 61 (0.7) 61.3 (0.6) 60.1 (0.9) Ref (1.0) -1.722 (-3.969–0.526) -1.219 (-3.314–0.876) -2.371 (-4.775–0.033) Time from stimulation initiation to oocyte retrieval (days) a 11.6 (3.1) 11.3 (2.9) 11.4 (2.7) 11.1 (2.9) Ref (1.0) -0.069 (-0.247–0.109) -0.056 (-0.219–0.107) -0.079 (-0.253–0.096) Data are presented as mean ± SD for count outcomes and mean ± SE for proportion outcomes (expressed as percentages). Q1 (lowest TyG quartile) served as the reference group. Relative risks (RRs, 95% CIs) were estimated using PS-IPW Poisson regression with a log link and robust standard errors for count outcomes a Coefficients (β, 95% CIs) from linear regression were reported for proportion outcomes (percentage points) and for the continuous outcome of time from stimulation initiation to oocyte retrieval (days). Models were adjusted for age, BMI, AMH, infertility duration, gravidity, parity, basal E2, FSH, LH, and total gonadotropin dose, TC, fertilization method (IVF vs. ICSI), infertility type (primary vs. secondary), infertility diagnosis, and ovarian stimulation protocol Association between TyG index quartiles and cycle characteristics after PS-IPW: relative risks or coefficients with 95% CIs Data are presented as mean ± SD for count outcomes and mean ± SE for proportion outcomes (expressed as percentages). Q1 (lowest TyG quartile) served as the reference group. Relative risks (RRs, 95% CIs) were estimated using PS-IPW Poisson regression with a log link and robust standard errors for count outcomes a Coefficients (β, 95% CIs) from linear regression were reported for proportion outcomes (percentage points) and for the continuous outcome of time from stimulation initiation to oocyte retrieval (days). Models were adjusted for age, BMI, AMH, infertility duration, gravidity, parity, basal E2, FSH, LH, and total gonadotropin dose, TC, fertilization method (IVF vs. ICSI), infertility type (primary vs. secondary), infertility diagnosis, and ovarian stimulation protocol In the adjusted PS-IPW regression analyses, the mean number of oocytes retrieved ranged from 12.2 in Q1 to 10.9 in Q4, but relative risks across quartiles were not statistically significant. Similar null findings were observed for mature oocytes, fertilization outcomes, and blastocyst numbers, with effect estimates close to unity and narrow confidence intervals. Proportion outcomes, including mature rate and fertilization rate per oocyte retrieved, were also comparable across quartiles. Of note, the blastocyst formation rate exhibited a decreasing trend with higher TyG quartiles; in Q4 compared with Q1, the adjusted difference was − 2.37% points (β = − 2.37; 95% CI, − 4.78 to 0.03), which did not reach statistical significance, and other quartile contrasts were likewise non-significant. Time to oocyte retrieval was similar across groups. In a secondary analysis comparing the lower (Q1/Q2) and higher (Q3/Q4) TyG groups (eTable 5), similar trends were observed. Patients in the higher TyG half had modestly fewer mature oocytes (9.8 vs. 10.4; RR = 0.97, 95% CI 0.95–0.99), fewer fertilized oocytes (6.8 vs. 7.3; RR = 0.97, 95% CI 0.94–0.99), and reduced blastocyst numbers (5.3 vs. 5.7; RR = 0.95, 95% CI 0.92–0.99) compared with those in the lower TyG half. Other cycle outcomes, including fertilization rate, blastocyst formation rate, and time to oocyte retrieval, did not significantly differ between groups. An RR was estimated for live birth, clinical pregnancy, and miscarriage using sequential models, including unadjusted, multivariable adjusted, PS-IPW, and the combination of PS-IPW with multivariable adjustment (Table  3 ). Before weighting, quartile groups differed modestly in live birth ( p = .04), but these differences were no longer significant after PS-IPW. Across the fully adjusted models, live birth and clinical pregnancy rates were generally comparable among TyG quartiles. Notably, patients in Q3 demonstrated a modestly higher likelihood of live birth (RR = 1.12; 95% CI, 1.01–1.23) and clinical pregnancy (RR = 1.09; 95% CI, 1.01–1.19) compared with Q1, whereas miscarriage rates did not differ significantly across quartiles. Table 3 Live birth, clinical pregnancy, and miscarriage across quartiles of the TyG index Outcomes Events, n / N (%) Relative risk (95% CI) Unadjusted Multivariable adjusted a PS-IPW b PS-IPW + multivariable adjusted a, b Live birth Q1 564 (42.9%) Reference Reference Reference Reference Q2 558 (42.5%) 0.989 (0.905–1.081) 1.040 (0.956–1.133) 0.964 (0.831–1.119) 1.008 (0.908–1.118) Q3 603 (45.9%) 1.070 (0.982–1.166) 1.124 (1.033–1.223) 1.075 (0.932–1.241) 1.116 (1.014–1.229) Q4 531 (40.4%) 0.941 (0.860–1.031) 1.047 (0.953–1.151) 1.021 (0.876–1.190) 1.058 (0.951–1.178) Clinical pregnancy Q1 688 (52.4%) Reference Reference Reference Reference Q2 678 (51.6%) 0.985 (0.916–1.061) 1.019 (0.949–1.094) 0.970 (0.863–1.090) 1.002 (0.919–1.093) Q3 722 (55%) 1.050 (0.978–1.128) 1.079 (1.006–1.157) 1.064 (0.954–1.188) 1.094 (1.011–1.185) Q4 682 (51.9%) 0.991 (0.921–1.067) 1.059 (0.981–1.144) 1.048 (0.932–1.179) 1.079 (0.989–1.177) Miscarriage Q1 111 (8.4%) Reference Reference Reference Reference Q2 104 (7.9%) 0.937 (0.725–1.211) 0.896 (0.693–1.157) 0.998 (0.733–1.358) 0.986 (0.727–1.336) Q3 105 (8%) 0.947 (0.733–1.222) 0.871 (0.673–1.129) 1.018 (0.751–1.380) 1.009 (0.749–1.358) Q4 136 (10.4%) 1.225 (0.965–1.556) 1.071 (0.824–1.391) 1.227 (0.901–1.671) 1.226 (0.910–1.650) a Adjusted for age, BMI, AMH, gravidity, parity, infertility duration (years), E2, FSH, LH, endometrial thickness, number of embryos transferred, total gonadotropin dose, total cholesterol (TC), fertilization method, day of transfer, type of infertility, infertility diagnosis, and ovarian stimulation protocol b The propensity score was estimated using the same set of covariates as in model a, and IPW were applied to estimate the average treatment effect Live birth, clinical pregnancy, and miscarriage across quartiles of the TyG index a Adjusted for age, BMI, AMH, gravidity, parity, infertility duration (years), E2, FSH, LH, endometrial thickness, number of embryos transferred, total gonadotropin dose, total cholesterol (TC), fertilization method, day of transfer, type of infertility, infertility diagnosis, and ovarian stimulation protocol b The propensity score was estimated using the same set of covariates as in model a, and IPW were applied to estimate the average treatment effect Given these findings, interaction analyses were performed to assess whether the association between Q3 and outcomes varied across baseline factors (eTable 4). Significant modifiers included BMI, gravidity, parity, endometrial thickness, number of embryos transferred, transfer day, infertility type, and stimulation protocol. For instance, the positive association of Q3 with live birth was more evident among women with lower endometrial thickness and those undergoing antagonist stimulation, whereas higher BMI was linked to an increased risk of miscarriage. In a simplified two-group analysis comparing the lower (Q1/Q2) and higher (Q3/Q4) TyG halves (eTable 6), patients in the higher TyG group showed slightly increased live birth (RR = 1.09; 95% CI, 1.01–1.16) and clinical pregnancy rates (RR = 1.08; 95% CI, 1.02–1.14), while miscarriage remained similar. Interaction testing in this two-group framework (eTable 7) yielded consistent modifiers, including BMI, endometrial thickness, stimulation protocol, and male-factor infertility, supporting the robustness of the quartile-specific findings. Consistent with these analyses, Fig.  1 illustrates that mature and fertilization ratios remained largely stable across TyG bins, blastocyst formation declined modestly at higher TyG levels, and clinical outcomes followed a non-linear pattern, with live birth and clinical pregnancy rates peaking at intermediate TyG values while miscarriage showed no consistent trend. In a sensitivity analysis excluding all patients with endometriosis, the associations between TyG quartiles and clinical outcomes remained unchanged (eTable 8), further supporting the robustness of the main findings. Fig. 1 Relationship between TyG index and oocyte, embryo, and pregnancy outcomes in ART cycles. A Oocyte mature ratio, B Fertilization ratio, C Blastocyst formation ratio, D Clinical pregnancy rate, E Live birth rate, and F Miscarriage rate. The TyG index was divided into bins of 0.5 units. For each bin, the average outcome ratio or rate was calculated and plotted (red line with dots). Numbers above points indicate the sample size within each bin Relationship between TyG index and oocyte, embryo, and pregnancy outcomes in ART cycles. A Oocyte mature ratio, B Fertilization ratio, C Blastocyst formation ratio, D Clinical pregnancy rate, E Live birth rate, and F Miscarriage rate. The TyG index was divided into bins of 0.5 units. For each bin, the average outcome ratio or rate was calculated and plotted (red line with dots). Numbers above points indicate the sample size within each bin

The institutional review board of Guangzhou Women and Children’s Medical Center, Liuzhou Hospital, approved this study (protocol number 2025 − 147). Informed consent was not required from participants because this was a retrospective study in which no personally identifying information was collected. The study population included women who underwent their first autologous IVF/ICSI cycle at the Reproductive Medicine Center of Liuzhou Hospital, Guangzhou Women and Children’s Medical Center, between January 1, 2020, and December 31, 2023. Eligibility required available fasting triglyceride (TG) and fasting plasma glucose (FPG) measurements before ovarian stimulation. Cycles using donor oocytes, donor sperm, or thawed oocytes, as well as those with no oocytes or no mature oocytes retrieved, were excluded. Patients with polycystic ovary syndrome (PCOS), type 1 or type 2 diabetes, or use of medications affecting glucose and lipid metabolism within 3 months before oocyte retrieval (e.g., systemic glucocorticoids, statins/fibrates, SGLT2 inhibitors, GLP-1 receptor agonists, or metformin) were also excluded. Pregnancy outcome analyses were restricted to fresh embryo transfer cycles, whereas embryologic outcome analyses included all cycles with fertilization and embryo culture, including PGT-A cycles. Patient age at oocyte retrieval was obtained from identification records; infertility duration was self-reported; and infertility diagnosis and stimulation protocol were documented by physicians. Height and weight were measured at the initial visit to calculate BMI. Embryologic data (fertilization method, blastocyst culture, embryo transfer) were recorded by the embryology laboratory, and hormonal and biochemical parameters were obtained from institutional laboratory databases. Patients underwent individualized ovarian stimulation at the discretion of the treating physician. Stimulation protocols included long or short agonist, antagonist, luteal-phase stimulation, progestin-primed ovarian stimulation (PPOS), mild stimulation, and natural cycle regimens. Subcutaneous gonadotropins were administered, with dosage and combinations adjusted according to baseline characteristics and ovarian response. Follicular development was monitored using transvaginal ultrasound in conjunction with serum estradiol, luteinizing hormone, and progesterone measurements. Final oocyte maturation was triggered with human chorionic gonadotropin (hCG) and/or gonadotropin-releasing hormone agonist (GnRH-a) when appropriate follicular criteria were met. Oocytes were retrieved by transvaginal aspiration, and embryos were either transferred fresh or cryopreserved at the blastocyst stage according to clinical indications. The detailed procedures have been reported previously [ 29 , 30 ]. Conventional insemination (IVF) or intracytoplasmic sperm injection (ICSI) was performed according to clinical indications. For conventional IVF, cumulus–oocyte complexes were inseminated with prepared motile sperm at an approximate concentration of 50,000 per oocyte, typically 4–6 h after oocyte retrieval. Following insemination, cumulus cells were removed after 4–5 h, and oocytes were transferred into fresh culture medium for continued incubation. For ICSI, morphologically normal and motile sperm were immobilized and individually injected into the cytoplasm of metaphase II (MII) oocytes. All embryos were subsequently cultured in bench-top incubators under controlled conditions of temperature, humidity, and gas composition. Fertilization was assessed 16–18 h after insemination or injection. Fresh embryo transfer was performed at the discretion of the treating physician. Embryos were transferred either on day 3 after oocyte retrieval or, in selected cases, at the blastocyst stage on day 5 or 6, depending on embryo development and quality. Embryo assessment before transfer followed standard morphological criteria [ 31 ], and the number of embryos transferred complied with institutional clinical practice guidelines. The TyG index was calculated as Ln [TG (mg/dL) × FPG (mg/dL)/2]. At our center, fasting triglycerides and fasting plasma glucose are routinely measured as part of the standard metabolic assessment before IVF/ICSI treatment; thus, TyG in this study represents a baseline, cross-sectional proxy of IR rather than long-term IR history. For statistical analysis, TyG was treated both as a continuous variable and as a categorical variable. To examine potential non-linear associations, patients were divided into quartiles based on the observed distribution of TyG values in the study population. The quartile cut-points (Q1 ≤ 7.95, Q2 7.96–8.28, Q3 8.29–8.67, Q4 ≥ 8.68) were determined empirically from the dataset and were not part of the a priori study design, rather than based on prior clinical thresholds. The primary outcomes were clinical pregnancy, live birth, and miscarriage. Clinical pregnancy was defined as an intrauterine gestational sac on ultrasound; when ultrasound data were unavailable, clinical pregnancy status was inferred from documented pregnancy outcomes (live birth, miscarriage, or ectopic pregnancy). Live birth was defined as the delivery of one or more live-born infants. Miscarriage was defined as the loss of an intrauterine gestation before 20 weeks of gestation. Secondary outcomes included oocyte and embryologic parameters: number of oocytes retrieved, number of mature (MII) oocytes, mature oocyte rate (per oocyte retrieved), number of fertilized oocytes, fertilization rate (per oocyte retrieved), number of blastocysts cultured, number of blastocysts formed, blastocyst formation rate, and duration of ovarian stimulation (days). Baseline covariates were selected based on their potential to influence ovarian response, embryologic development, and pregnancy outcomes. For analyses of embryologic outcomes, female factors included age at oocyte retrieval, body mass index (BMI), antimüllerian hormone (AMH), infertility duration, gravidity, parity, basal estradiol (E2), follicle-stimulating hormone (FSH), and luteinizing hormone (LH), total gonadotropin dose (Total Gn), and total cholesterol (TC). Treatment-related factors included fertilization method (IVF vs. ICSI), type of infertility (primary vs. secondary), infertility diagnosis (tubal factor, male factor, ovulatory, genetic factor, endometriosis, recurrent pregnancy loss, or other), and ovarian stimulation protocol (agonist, antagonist, luteal-phase stimulation, progestin-primed ovarian stimulation [PPOS], mild stimulation, or natural cycle). For analyses of pregnancy outcomes after fresh embryo transfer, covariates included the same baseline female characteristics (age, BMI, AMH, infertility duration, gravidity, parity, basal E2, FSH, and LH, Total Gn, and TC), as well as treatment-related factors: fertilization method (IVF vs. ICSI), type of infertility, infertility diagnosis, and ovarian stimulation protocol (agonist, antagonist, luteal-phase stimulation, progestin-primed protocol, mild stimulation, or natural cycle). Additional transfer-related covariates included endometrial thickness, number of embryos transferred, and day of transfer (day 3 vs. day 5/6). Analyses were performed at the patient level, restricting to the first oocyte-retrieval cycle per patient. Data were analyzed using SPSS 26.0 (IBM Corp., Armonk, NY, USA) and R within RStudio. A two-sided α of 0.05 was used. Baseline characteristics were summarized as means (standard deviations) for continuous variables and counts (percentages) for categorical variables. For outcome reporting, proportion outcomes and the time from stimulation initiation to oocyte retrieval were presented as means with standard errors. Group comparisons used chi-square or Fisher’s exact tests for categorical variables, and linear regression models for proportion outcomes, and Poisson regression models (or their weighted analogs) for count outcomes. There was no missingness for outcome variables, and other covariates had < 5% missingness. Given the low proportion of missing data, we applied median imputation for continuous variables and mode imputation for categorical variables. As a robustness check, complete-case analyses produced similar results to the primary analysis. To mitigate confounding across TyG quartiles, we estimated propensity scores including all prespecified baseline covariates and applied stabilized inverse probability weights (IPW) to estimate average treatment effects. Propensity-score performance was evaluated using the c-statistic and covariate balance diagnostics; standardized mean differences (SMD) < 0.10 were considered indicative of adequate balance. Importantly, PS-IPW does not delete individuals; all eligible patients remained in the analysis, and the weighted sample size reflects the sum of weights rather than the number of unique patients. For binary clinical outcomes, we fit generalized linear models with a Poisson distribution and robust standard errors to estimate relative risks (RRs) and 95% confidence intervals (CIs), reporting both unadjusted and adjusted (covariate-adjusted and/or PS-weighted) estimates. For continuous embryologic outcomes and proportion-type embryologic outcomes (e.g., mature, fertilization, blastocyst formation ratios), we used linear regression with robust standard errors, likewise reported in crude, adjusted, and PS-weighted forms. Prespecified interaction analyses examined effect modification by key baseline factors. To visualize and analyze the association between TyG quartiles and binary outcomes (e.g., live birth rate, clinical pregnancy rate, miscarriage rate), we used binned plots to present the proportion of outcomes within each TyG group. These plots were generated using binning techniques and visualized using ggplot2, with bin widths set to 0.5 for a clear depiction of group-level differences.

In conclusion, this large real-world cohort study demonstrates that the TyG index, as a surrogate marker of insulin resistance, was not significantly associated with impaired embryologic or pregnancy outcomes in non-PCOS women undergoing their first IVF/ICSI cycle. A modest non-linear trend was noted, with intermediate TyG levels showing slightly better outcomes, but the overall effect of TyG was limited.

With TyG increasingly explored in PCOS populations as a convenient surrogate of insulin resistance for IVF risk stratification, it is important to examine its relevance in non-PCOS women, while recognizing that surrogate markers of IR—including TyG, HOMA-IR, and related indices—provide indirect estimations of insulin sensitivity and show variable performance across studies. In our cohort, women in higher TyG quartiles were older, had higher BMI, and showed less favorable metabolic profiles; however, after PS-IPW adjustment, key cycle metrics—oocyte yield, maturation, fertilization, and blastocyst counts—were broadly comparable across quartiles. More importantly for clinical decision-making, higher TyG levels did not translate into significantly worse clinical outcomes. Live birth and clinical pregnancy rates were largely similar across quartiles, with only a modest, non-linear increase observed in the intermediate range (Q3). The highest TyG group showed a slight, non-significant reduction in blastocyst formation. Overall, these findings indicate that variations in TyG may not exert a major impact on ART outcomes in non-PCOS women undergoing IVF/ICSI. Our observations are generally compatible with reports in PCOS populations showing that insulin resistance assessed with validated measures (e.g., HOMA-IR, oral glucose tolerance test [OGTT]) has limited influence on early laboratory outcomes [ 32 ]. As TyG is an indirect surrogate whose performance varies across populations, particularly in non-PCOS women, our findings should be interpreted as associative rather than causal. Wu et al. reported that TyG was negatively associated with oocyte retrieval number in PCOS patients, even after adjusting for age, BMI, and lipid levels [ 22 ]. Because PCOS patients frequently exhibit obesity and insulin resistance, TyG alone may not fully capture adiposity-related metabolic burden; therefore, TyG-BMI has been used to better reflect metabolic phenotypes. In PCOS cohorts, higher TyG-BMI has been linked to reduced oocyte retrieval and fertilized embryos [ 25 ]. Other studies have evaluated TyG-BMI in relation to pregnancy outcomes. Wu et al. and Ding et al. found that elevated TyG-BMI was associated with reduced live birth rates in both fresh and frozen cycles among PCOS patients [ 22 , 23 ]. However, Li et al. observed no significant associations between TyG-BMI and pregnancy outcomes, including biochemical pregnancy, clinical pregnancy, and live birth [ 25 ]. These discrepancies likely reflect differences in TyG-BMI cut-off values, population characteristics, and laboratory assays, underscoring the heterogeneity inherent in surrogate markers of IR. More recently, a large retrospective cohort including both PCOS and non-PCOS women ( n = 17,365) reported that higher TyG-BMI was associated with increased miscarriage rates, while live birth rates were unchanged in fresh cycles but lower in frozen and cumulative analyses [ 24 ]. In our study, live birth and clinical pregnancy rates after fresh embryo transfer were generally comparable across TyG quartiles. The modest improvements observed in Q3 suggest a non-linear association, but the overall effect was small. Interaction analyses indicated that reproductive history, ovarian stimulation strategy, and endometrial characteristics may modify this association. These findings support the notion that TyG, although reflective of certain metabolic features, may have limited discriminatory ability in predicting reproductive outcomes in non-PCOS women. At present, there is no standardized TyG threshold for identifying clinically meaningful insulin resistance. Prior studies proposed cut-offs around 8.31 for PCOS patients [ 33 ] and 8.34 for infertile women [ 34 ]. In our cohort, using a comparable division (Q1/Q2 vs. Q3/Q4), women in the higher TyG half showed slightly increased live birth and clinical pregnancy rates, while miscarriage remained similar. Notably, the absolute differences across quartiles were small (e.g., live birth 45.9% in Q3 vs. 42.9% in Q1; clinical pregnancy 55.0% in Q3 vs. 52.4% in Q1), suggesting limited clinical impact despite statistical significance in some models. Interaction analyses suggested that these small differences were influenced by reproductive history and stimulation strategy, rather than by TyG itself as a consistent predictor. We additionally explored potential effect modification by BMI and age. In interaction analyses, BMI modestly modified the association between TyG and miscarriage risk, whereas no clear effect modification by age was observed. This is consistent with emerging evidence that excess body weight may independently and synergistically influence reproductive outcomes through metabolic and endocrine pathways, and should be considered when interpreting surrogate metabolic markers such as TyG in ART settings [ 35 ]. Given that TyG—like other surrogate markers of insulin resistance—captures insulin sensitivity only indirectly and shows variable, and in some studies suboptimal, diagnostic performance compared with clamp-based measurements, the clinical relevance of these modest associations should be interpreted with caution. A plausible biological explanation for the largely null association between TyG and ART outcomes in our cohort is that TyG reflects peripheral metabolic status rather than the intra-follicular microenvironment. Mechanistic evidence suggests that reproductive impairment in PCOS is closely related to follicular endocrine–metabolic alterations, including elevated AMH levels and signatures of glycolytic disruption and mitochondrial dysfunction in follicular fluid, which may compromise granulosa cell energy metabolism and oocyte competence [ 36 ]. Therefore, in non-PCOS women, a peripheral index such as TyG may not adequately reflect intra-follicular metabolic or mitochondrial function, which could partly explain the largely null associations observed in our cohort. The major strength of this study lies in its large sample size derived from routine clinical practice, which enhances the reliability and real-world relevance of our findings. By applying propensity score weighting rather than one-to-one matching, we preserved the statistical power of the cohort while improving the comparability of baseline characteristics across TyG quartiles and reducing the risk of confounding. Another methodological strength is the restriction to women without PCOS undergoing their first IVF/ICSI cycle, thereby minimizing potential bias from previous treatments and the unique metabolic profile of PCOS. Importantly, our analysis covered fresh embryo transfer cycles and incorporated both embryologic and clinical endpoints. By simultaneously assessing oocyte yield, fertilization, and blastocyst formation in addition to pregnancy outcomes, we were able to validate earlier reports linking insulin resistance to reduced oocyte retrieval, while also extending the evaluation to embryo developmental competence. This multidimensional approach provides a more comprehensive understanding compared with prior studies that predominantly focused on PCOS populations and emphasized clinical outcomes such as implantation and pregnancy rates, which are more susceptible to uterine factors and may not directly reflect embryo quality. Nevertheless, several limitations should be acknowledged. First, as a retrospective observational study, causal relationships cannot be inferred, and residual confounding by unmeasured variables cannot be excluded despite the use of propensity score weighting. For example, lifestyle-related factors (e.g., diet and physical activity) and other metabolic interventions were not consistently available in our dataset. Second, the single-center design may limit generalizability to other populations and clinical settings. Third, our assessment of metabolic status beyond the TyG index was limited. Although BMI and total cholesterol were included as covariates, other important components of metabolic syndrome—such as waist circumference, blood pressure, high-density lipoprotein (HDL) cholesterol, and fasting insulin or HOMA-IR—were not consistently available in this retrospective dataset. Moreover, TyG was measured only once prior to ovarian stimulation, so we could not capture the chronicity or temporal variation of insulin resistance during ART. Because TyG is an indirect surrogate of insulin resistance and does not reflect other dimensions of metabolic or inflammatory dysregulation, our findings should be interpreted as associative rather than causal. As BMI alone is an imperfect proxy for adiposity and metabolic health, some metabolic abnormalities may have been misclassified, potentially attenuating true associations between TyG and ART outcomes. Fourth, because PCOS was excluded solely based on documented diagnoses in the electronic medical system, the retrospective nature of the study means that unrecognized or subclinical PCOS cannot be completely ruled out. Finally, although our analytic strategy reduced baseline imbalance, unobservable differences may still exist. Future studies with prospective, multicenter designs will be necessary to confirm and expand upon these findings. In addition, a multidimensional assessment of metabolic health—including biochemical, anthropometric, and inflammatory markers—as well as stratification of endometriosis by disease stage will help clarify whether metabolic and inflammatory pathways differ across subgroups. Furthermore, longitudinal studies tracking temporal changes in insulin resistance and integrating clinical, metabolic, and molecular markers would help elucidate causal pathways and enhance the translational relevance of future findings.

Metabolic homeostasis has a profound impact on female reproductive health and assisted reproductive technology (ART) outcomes. Insulin resistance (IR), one of the most common metabolic abnormalities, plays a critical role not only in the development of diabetes and cardiovascular disease but also as a key mechanism underlying infertility [ 1 , 2 ]. Epidemiological and clinical studies have demonstrated that IR may impair oocyte quality and embryonic developmental potential, ultimately reducing pregnancy outcomes through its effects on follicular growth, steroidogenesis, and endometrial receptivity [ 3 – 5 ]. The impact of IR is particularly evident in patients with polycystic ovary syndrome (PCOS), among whom approximately 40–75% exhibit IR [ 6 – 8 ], a prevalence markedly higher than that observed in women of reproductive age without PCOS. Consequently, IR has been recognized as a central pathophysiological feature of PCOS. Recent studies further highlight that IR contributes to reduced reproductive potential in PCOS not only through anovulation but also through compromised oocyte, embryo, and endometrial competence [ 9 ]. IR is also associated with substantially higher risks of pregnancy complications, such as gestational diabetes and preeclampsia [ 10 ]. Mechanistic evidence suggests that IR disrupts oocyte competence by altering metabolic and mitochondrial function [ 11 ], and impairs endometrial receptivity through dysregulated glucose metabolism, chronic inflammation, and abnormal hormonal signaling [ 12 ]. Together, these findings underscore the broad reproductive impact of IR. Despite the undeniable clinical significance of IR, accurate and practical assessment remains a challenge. The hyperinsulinemic–euglycemic clamp test is regarded as the gold standard; however, its complexity, high cost, and time-consuming nature limit its clinical and research utility [ 13 ]. Simpler surrogate indices, such as the homeostasis model assessment (HOMA-IR), have been widely adopted, but their reliance on fasting insulin measurements reduces reproducibility and generalizability across different laboratory conditions [ 14 , 15 ]. Thus, there remains an unmet need for an index that balances accuracy, simplicity, and feasibility for large-scale clinical and epidemiological applications. The triglyceride–glucose (TyG) index, which relies only on fasting glucose and triglyceride levels, has been proposed as a simple and practical surrogate marker for assessing IR [ 16 , 17 ]. Although TyG has demonstrated associations with multiple metabolic disorders—including diabetes, metabolic syndrome, cardiovascular disease, and nonalcoholic fatty liver disease—its performance as a surrogate of IR has shown considerable variability across studies and populations, and it does not fully replicate the diagnostic accuracy of clamp-based measurements [ 18 – 20 ]. Beyond metabolic diseases, growing interest has focused on the relevance of TyG to female reproductive health. Elevated TyG levels have been linked to ovulatory dysfunction and menstrual irregularities, particularly in women with PCOS, in whom TyG may reflect underlying metabolic disturbances [ 21 ]. A recent systematic review and meta-analysis also reported higher TyG levels in PCOS patients compared with non-PCOS women and described moderate predictive value of TyG for metabolic syndrome (area under the curve [AUC] up to 0.91), although the applicability of these findings to other reproductive populations remains uncertain [ 8 ]. In PCOS populations, several studies have evaluated TyG-related indices in the context of assisted reproduction, but the findings remain inconsistent. Some reports have suggested that higher TyG or TyG-BMI levels are associated with fewer retrieved oocytes, reduced embryo availability, or lower live birth rates [ 22 – 24 ]. By contrast, other evidence has shown that although TyG-BMI may correlate with differences in laboratory parameters, it does not translate into significant differences in clinical pregnancy or live birth rates [ 25 ]. Overall, the available findings indicate heterogeneous rather than strong or uniform evidence regarding the influence of TyG-related indices on ART outcomes in PCOS. Their predictive value remains variable, and no definitive consensus has been established. Importantly, PCOS is not the only group experiencing metabolic abnormalities. Even among non-PCOS women, subclinical metabolic disturbances may influence ART outcomes through pathways involving follicular development, embryonic competence, and endometrial receptivity [ 26 – 28 ]. However, evidence in this population remains limited, and systematic investigations of TyG in relation to ART outcomes are still lacking. The objectives of this study are therefore two-fold. First, to evaluate whether women with different TyG index levels demonstrate differences in oocyte yield, maturation rate, fertilization rate, and embryonic development. Second, to investigate the association between TyG index and clinical outcomes—including clinical pregnancy, live birth, and miscarriage—in non-PCOS women undergoing ART. Clarifying these relationships may help determine whether TyG offers incremental but limited information about metabolic status relevant to reproductive outcomes.

Supplementary Material 1. Supplementary Material 1.