Pre-pregnancy LDL/HDL and total Cholesterol/HDL ratios are strong predictors of gestational diabetes mellitus in women undergoing assisted reproductive technologies.

Data for this retrospective study was collected in a time period of two years from January 2017 to December 2018 at the Reproductive and Genetic Hospital CITIC-Xiangya in Changsha, China. The primary study focused on the effect of vitamin D levels on IVF/ICSI outcomes [ 10 ]. The study was conducted according to the Declaration of Helsinki for Medical Research involving Human Subjects and was approved by the hospital’s ethics committee (approval number: LL-SC-2018-014). All patients gave their informed consent prior to study participation. From a total of 2569 patients included in the initial study, 767 patients were included in this analysis, as patients who did not achieve a successful pregnancy and those with incomplete data regarding lipids were excluded. All participants underwent their first IVF/ICSI treatment, were all primigravida, and were aged between 18 and 40 years. Patients were excluded, as previously described [ 10 ], in any of the following cases: [ 1 ] if they had received oocyte donation, 2) uterine malformation, 3) endometriosis, 4) uterine adhesions, 5) untreated hydrosalpinx, 6) uterine myoma, 7) Cushing syndrome, 8) adult-onset adrenogenital syndrome, 9) hypothalamic or pituitary disease causing infertility, 10) diabetes mellitus type 1 or 2 prior to pregnancy, or 11) hypertension prior to pregnancy. Further details on patient selection can be seen in Supplementary Fig.  1 . The lipid parameters (LDL, HDL, TC, and TG) were measured at baseline in a fasting state – before pregnancy and any IVF/ICSI procedures – in the central hospital laboratory as part of the routine evaluation of the patients. All clinical chemical analyses in this laboratory are part of strict quality controls for clinical chemistry according to the laws of the People’s Republic of China. In addition, we also calculated lipid ratios, which are not routinely assessed in a clinical context but may be better indicators of lipid status than measured lipid parameters [ 11 ]. During IVF, patients receive ovarian hyperstimulation, after which oocytes are retrieved, cultured, and inseminated (if necessary, using ICSI). The created embryos are cultured and then transferred into the uterus. If a successful pregnancy was achieved, the patients were followed up during and after their pregnancy through self-reports and phone calls. All patients underwent a GDM screening (oral glucose tolerance test) during their pregnancy. GDM was diagnosed according to the IADPSG (International Association of Diabetes in Pregnancy Study Groups) 2010 guidelines [ 12 ]: if the measured plasma glucose was either 1) ≥ 5.1mmol/L (92 mg/dL) before, 2) ≥ 10.0mmol/L (180 mg/dL) one hour after, or 3) ≥ 8.5mmol/L (153 mg/dL) two hours after taking a 75 g glucose solution. To test for normality, we created histograms and used the Kolmogorov-Smirnov as well as the Shapiro-Wilk test. Patient characteristics are represented as frequency (%) or median (interquartile range, IQR). To test for a significant difference in population characteristics between GDM and non-GDM patients, we used the Chi-square (χ 2 ) test for categorical variables and the Mann-Whitney U test for continuous variables. Multivariate logistic regression considered all significant factors from Table  1 (age, BMI, fasting plasma glucose, LH, and estrogen) in Model A and risk factors for GDM from the literature in Model B (age, BMI, fasting plasma glucose measured before embryo transfer, multiparity, total testosterone, antral follicle count (AFC), anti-Müllarian hormone (AMH), and vitamin D). Since we did not have documentation of the presence of polycystic ovary syndrome (PCOS), which are patients with an elevated risk of GDM [ 13 ], we included three surrogate parameters in the multivariate regression: total testosterone, which is a biochemical parameter for hyperandrogenism, one of the main diagnostic criteria of PCOS [ 14 , 15 ], AFC, which corresponds to the polycystic ovarian morphology (PCOM), another criterion of PCOS, and AMH, which is a biochemical parameter that is often increased in women with PCOS as it correlates with PCOM [ 16 ]. We also included vitamin D in the analysis, as some studies have also found a correlation between low vitamin levels and GDM [ 17 , 18 ]; however, the association is uncertain as there have been contrasting findings [ 19 , 20 ]. Receiver operating characteristic (ROC) analysis was used to determine optimal cut-off values for binary analysis. Furthermore, we created a forest plot to represent the results of subgroup analyses and 3D plots to show interactions between lipids and other risk factors for GDM. All data was analyzed using SPSS (Statistical Package for Social Sciences), version 29.0 (IBM Corporation, Armonk, New York, USA). We created Fig.  1 using GraphPad Prism 6 (GraphPad Software Inc., San Diego, California, USA) and Figs.  2 a-b and 3 using SPSS. The statistical significance level was set to p  < 0.05. Fig. 1 LDL/HDL levels in GDM versus non-GDM patients. Plots are presented as median and IQR (interquartile range). p -values were calculated with the Mann-Whitney U test. Abbrevations GDM = gestational diabetes mellitus; LDL = low-density lipoproteins; HDL = high-density lipoproteins. Plots for other lipid parameters can be seen in Supplementary Fig.  4 a-f LDL/HDL levels in GDM versus non-GDM patients. Plots are presented as median and IQR (interquartile range). p -values were calculated with the Mann-Whitney U test. Abbrevations GDM = gestational diabetes mellitus; LDL = low-density lipoproteins; HDL = high-density lipoproteins. Plots for other lipid parameters can be seen in Supplementary Fig.  4 a-f Fig. 2 3D Plots – Effects of LDL/HDL ratio and other GDM risk factors on GDM incidence in women undergoing IVF/ICSI. LDL/HDL ratio is displayed with fasting plasma glucose ( a ) and LH ( b ). Comparing the group’s lowest vs. highest tertile of LDL/HDL ratio and blood glucose, 7.9% developed GDM in the lowest and 25.5% in the highest group ( p  < 0.001). Women in the lowest LDL/HDL tertile and highest LH tertile had a GDM rate of 6.6%, whilst women in the highest LDL/HDL tertile and lowest LH tertile had a GDM rate of 28.2% ( p  < 0.001). Abbreviations GDM = gestational diabetes mellitus; E2 = estradiol; ICSI = intracytoplasmic sperm injection; IVF = in-vitro fertilization; LDL = low-density lipoproteins; LH = luteinizing hormone; HDL = high-density lipoproteins. 3D Plots for age, BMI, and estradiol can be seen in Supplementary Fig.  2 a-c 3D Plots – Effects of LDL/HDL ratio and other GDM risk factors on GDM incidence in women undergoing IVF/ICSI. LDL/HDL ratio is displayed with fasting plasma glucose ( a ) and LH ( b ). Comparing the group’s lowest vs. highest tertile of LDL/HDL ratio and blood glucose, 7.9% developed GDM in the lowest and 25.5% in the highest group ( p  < 0.001). Women in the lowest LDL/HDL tertile and highest LH tertile had a GDM rate of 6.6%, whilst women in the highest LDL/HDL tertile and lowest LH tertile had a GDM rate of 28.2% ( p  < 0.001). Abbreviations GDM = gestational diabetes mellitus; E2 = estradiol; ICSI = intracytoplasmic sperm injection; IVF = in-vitro fertilization; LDL = low-density lipoproteins; LH = luteinizing hormone; HDL = high-density lipoproteins. 3D Plots for age, BMI, and estradiol can be seen in Supplementary Fig.  2 a-c Fig. 3 Forest plot – subgroup analysis of GDM prevalence and the association between LDL/HDL ratio and GDM. The first part of the table shows the prevalence of GDM in each of the subgroups. p* = represents the significance of a difference in GDM rate between the subgroups, calculated using the Chi-squared test. OR and p** show the effect size of a high LDL/HDL ratio (cut-off calculated using the Youden-Index) on the GDM rate in each of the subgroups. Abbreviations BMI = body mass index; CI = confidence interval; DiaRR = diastolic blood pressure; E2 = estradiol; GDM = gestational diabetes mellitus; LDL = low-density lipoproteins; LH = luteinizing hormone; HDL = high-density lipoproteins; N = no GDM; OR = odds ratio; SysRR = systolic blood pressure Forest plot – subgroup analysis of GDM prevalence and the association between LDL/HDL ratio and GDM. The first part of the table shows the prevalence of GDM in each of the subgroups. p* = represents the significance of a difference in GDM rate between the subgroups, calculated using the Chi-squared test. OR and p** show the effect size of a high LDL/HDL ratio (cut-off calculated using the Youden-Index) on the GDM rate in each of the subgroups. Abbreviations BMI = body mass index; CI = confidence interval; DiaRR = diastolic blood pressure; E2 = estradiol; GDM = gestational diabetes mellitus; LDL = low-density lipoproteins; LH = luteinizing hormone; HDL = high-density lipoproteins; N = no GDM; OR = odds ratio; SysRR = systolic blood pressure

Of the 767 patients included in this study, 119 (15.5%) developed GDM during their pregnancy. All measured pre-pregnancy lipid parameters (LDL, HDL, TC, and TG) were not normally distributed (Supplementary Fig.  2 a-d), which was also confirmed by the statistical Kolmogorov-Smirnov and Shapiro-Wilk tests for normality (all p  < 0.001). 15.0% of the patients had elevated LDL levels, 9.0% had low HDL, 8.2% had elevated TC, and 15.5% had elevated TG levels according to the AACE (American Association of Clinical Endocrinologists and American College of Endocrinology) 2017 guidelines [ 21 ] and Chinese guidelines [ 22 ] (Supplementary Fig.  3 ). In total, 233 women (30.4%) had at least one pathological lipid measurement to be diagnosed with dyslipidemia. Detailed characteristics of the study cohort are displayed in Table  1 . Overall, the median age was 29.0 years (IQR 27.0–31.0), and the median BMI was 21.33 kg/m 2 (IQR 19.53–23.01). For the lipid parameters, median LDL was 2.23 mmol/L (IQR 2.66–3.10), median HDL was 1.43 mmol/L (IQR 1.22–1.67), median TC was 4.13 mmol/L (IQR 3.67–4.60), and median TG was 1.00 mmol/L (IQR 0.73–1.14). Table  1 also compares these characteristics between patients with and without the diagnosis of GDM. Women who developed GDM were statistically older (30.0 versus 29.0 years, p  < 0.001), had a higher BMI (21.91 versus 21.23 kg/m 2 , p  = 0.005), fasting glucose (5.21 versus 5.10 mmol/L, p  = 0.002), LDL (2.90 versus 2.62 mmol/L, p  = 0.004), TG (1.12 versus 0.98 mmol/L, p  = 0.010), and lower HDL (1.32 versus 1.44 mmol/L, p  < 0.001), LH (luteinizing hormone) (3.26 versus 3.73 mIU/mL, p  = 0.003), and estradiol (32.00 versus 33.00 pg/mL, p  = 0.046). Table 1 Clinical characteristics of study participants by GDM incidence Parameters All ( n  = 767) GDM ( n  = 119) Non-GDM ( n  = 648) p -Value Age, years 29.0 (27.0–31.0) 30.0 (28.0–33.0) 29.0 (27.0–31.0) < 0.001 Weight, kg 53.00 (48.00–58.00) 54.50 (49.00–60.00) 53.00 (48.00–58.00) 0.142 BMI, kg/m 2 21.33 (19.53–23.01) 21.91 (20.20-23.74) 21.23 (19.44–22.86) 0.005 TC, mmol/L 4.13 (3.67–4.60) 4.25 (3.74–4.76) 4.10 (3.65–4.57) 0.059 LDL, mmol/L 2.23 (2.66–3.10) 2.90 (2.31–3.30) 2.62 (2.21–3.09) 0.004 HDL, mmol/L 1.43 (1.22–1.67) 1.32 (1.12.1.55) 1.44 (1.23–1.68) < 0.001 TG, mmol/L 1.00 (0.73–1.41) 1.12 (0.78–1.60) 0.98 (0.72–1.38) 0.010 Fasting glucose, mmol/L 5.12 (4.87–5.37) 5.21 (4.99–5.49) 5.10 (4.86–5.35) 0.002 Gestational hypertension, n (%) 26 (3.4%) 5 (4.2%) 21 (3.2%) 0.594 Systolic RR, mmHg 115 (108–121) 117 (109–124) 115 (108–121) 0.200 Diastolic RR, mmHg 76 (70–81) 77 (70–81) 76 (70–81) 0.308 AMH, ng/mL 6.23 (4.06–10.28) 5.24 (3.73–10.62) 6.49 (4.19–10.28) 0.096 FSH, mIU/mL 5.64 (4.81–6.64) 5.51 (4.58–6.42) 5.69 (4.85–6.66) 0.191 LH, mIU/mL 3.65 (2.59–5.09) 3.26 (2.29–4.46) 3.73 (2.67–5.13) 0.003 Estradiol, pg/mL 33.00 (27.00–43.00) 32.00 (24.00–40.00) 33.00 (27.00–43.00) 0.046 Testosterone, ng/mL 0.27 (0.21–0.34) 0.27 (0.20–0.34) 0.27 (0.21–0.34) 0.789 AFC 15.00 (12.00–21.00) 14.00 (11.00–20.00) 15.00 (12.00–22.00) 0.109 Infertility type 0.787  1° 449 (58.5%) 71 (59.7%) 378 (58.3%)  2° 318 (41.5%) 48 (40.3%) 270 (41.7%) Fertilization method 0.459  IVF 509 (66.4%) 84 (70.6%) 425 (65.6%)  ICSI 132 (17.2%) 16 (13.4%) 116 (17.9%)  IVF + ICSI 126 (16.4%) 19 (16.0%) 107 (16.5%) Pregnancy type 0.332  Singleton 496 (64.7%) 412 (63.6%) 84 (70.6%)  Multiple 262 (34.2%) 227 (35.0%) 35 (29.4%) Delivery method 0.337  Normal 213 (27.8%) 36 (30.3%) 177 (27.3%)  Cesarean 544 (70.9%) 83 (69.7%) 461 (71.1%) Data is presented as frequency, n (%) or median (IQR, interquartile range). p -values were calculated with the Chi-square (χ 2 ) test for categorical variables and the Mann-Whitney U test for continuous variables. Abbreviations: AFC = antral follicle count; AMH = anti-Müllarian hormone; BMI = body mass index; GDM = gestational diabetes mellitus; FSH = follicle-stimulating hormone; HDL = high-density lipoproteins; ICSI = intracytoplasmic sperm injection; IVF = in-vitro fertilization; LDL = low-density lipoproteins; LH = luteinizing hormone; RR = blood pressure; TC = total cholesterol; TG = triglycerides Clinical characteristics of study participants by GDM incidence Data is presented as frequency, n (%) or median (IQR, interquartile range). p -values were calculated with the Chi-square (χ 2 ) test for categorical variables and the Mann-Whitney U test for continuous variables. Abbreviations: AFC = antral follicle count; AMH = anti-Müllarian hormone; BMI = body mass index; GDM = gestational diabetes mellitus; FSH = follicle-stimulating hormone; HDL = high-density lipoproteins; ICSI = intracytoplasmic sperm injection; IVF = in-vitro fertilization; LDL = low-density lipoproteins; LH = luteinizing hormone; RR = blood pressure; TC = total cholesterol; TG = triglycerides Figure  1 graphically illustrates the comparison of LDL/HDL between GDM and non-GDM patients and Supplementary Fig.  4 a-f further show this for measured lipid parameters and other lipid ratios, including TC/HDL and TG/HDL. Along with LDL, HDL, and TG, all three lipid ratios show a significant difference between the two groups ( p  < 0.05), whilst TC was not statistically different ( p  = 0.059). In further analysis, we performed multivariate logistic regression, creating Model A (considering confounders that were statistically significant in Table  1 : age, BMI, fasting plasma glucose, LH, and estradiol) and Model B (considering confounders previously described in literature [ 8 ]: age, BMI, fasting plasma glucose, multiparity, total testosterone, AFC, AMH, and total vitamin D). Both models showed that continuous LDL, HDL, LDL/HDL ratio, and TC/HDL ratio were still significantly different in GDM versus non-GDM patients (Table  2 ). The ratios LDL/HDL and TC/HDL displayed higher odds ratios (OR) (OR 1.628 for LDL/HDL and 1.452 for TC/HDL in Model A), as well as more significant p -values ( p  < 0.001) than continuous LDL (OR 1.351 and p  = 0.035 in Model A). HDL, which is inversely proportional to GDM incidence, also still showed a significant difference between GDM and non-GDM patients (OR 0.431, p  = 0.013 in Model A). However, the other lipid parameters, TC, TG, and TG/HDL ratio, were not significantly associated with GDM after adjustment for confounders ( p  = 0.171, 0.341, and 0.253 in Model A, respectively). Full tables of the multivariate regression can be seen in Supplementary Tables 1 – 5 . Table 2 Multivariate logistic regression for lipid parameters and GDM incidence B OR 95% CI Standard Error p -Value LDL   Continuous LDL  Model A 0.301 1.351 [1.021, 1.788] 0.143 0.035  Model B 0.404 1.499 [1.074, 2.091] 0.170 0.017   Binary LDL (cut-off 2.95mmol/L)  Model A 0.594 1.812 [1.196, 2.745] 0.212 0.005  Model B 0.710 2.033 [1.255, 3.293] 0.246 0.004 HDL   Continuous HDL  Model A -0.842 0.431 [0.221, 0.839] 0.340 0.013  Model B -0.991 0.371 [0.168, 0.821] 0.405 0.014   Binary HDL (cut-off 1.32mmol/L)  Model A -0.593 0.553 [0.355, 0.860] 0.226 0.009  Model B -0.619 0.539 [0.321, 0.902] 0.263 0.019 LDL/HDL   Continuous LDL/HDL  Model A 0.487 1.628 [1.244, 2.130] 0.137 < 0.001  Model B 0.560 1.750 [1.286, 2.383] 0.157 < 0.001   Binary LDL/HDL (cut-off 2.02)  Model A 0.671 1.957 [1.258, 3.044] 0.225 0.003  Model B 0.782 2.186 [1.308, 3.656] 0.262 0.003 TC/HDL   Continuous TC/HDL  Model A 0.373 1.452 [1.164, 1.811] 0.113 < 0.001  Model B 0.429 1.536 [1.196, 1.973] 0.128 < 0.001   Binary TC/HDL (cut-off 3.15)  Model A 0.664 1.942 [1.243, 3.034] 0.228 0.004  Model B 0.817 2.264 [1.346, 3.807] 0.265 0.002 TG/HDL   Continuous TG/HDL  Model A 0.119 1.126 [0.919, 1.381] 0.104 0.253  Model B 0.148 1.160 [0.927, 1.451] 0.114 0.194   Binary TG/HDL (cut-off 0.68)  Model A 0.537 1.710 [1.104, 2.649] 0.223 0.016  Model B 0.731 2.076 [1.240, 3.475] 0.263 0.005 p -values were calculated using multivariate logistic regression. Binary cut-offs were determined with ROC analysis. Model A: adjusted to all significant factors in Table  1 (age, BMI, pre-pregnancy fasting plasma glucose, LH, and E2); Model B: adjusted to risk factors from literature (age, BMI, pre-pregnancy fasting plasma glucose, multiparity, total testosterone, antral follicle count, anti-Müllarian hormone, and total vitamin D). Abbreviations: B = Regression coefficient; BMI = body mass index; CI = Confidence Interval for OR; GDM = gestational diabetes mellitus; HDL = high-density lipoproteins; LDL = low-density lipoproteins; LH = luteinizing hormone; OR = odds ratio; ROC = receiver operating characteristic; TC = total cholesterol; TG = triglycerides. The full tables can be found in Supplementary Tables 1 – 5 Multivariate logistic regression for lipid parameters and GDM incidence p -values were calculated using multivariate logistic regression. Binary cut-offs were determined with ROC analysis. Model A: adjusted to all significant factors in Table  1 (age, BMI, pre-pregnancy fasting plasma glucose, LH, and E2); Model B: adjusted to risk factors from literature (age, BMI, pre-pregnancy fasting plasma glucose, multiparity, total testosterone, antral follicle count, anti-Müllarian hormone, and total vitamin D). Abbreviations: B = Regression coefficient; BMI = body mass index; CI = Confidence Interval for OR; GDM = gestational diabetes mellitus; HDL = high-density lipoproteins; LDL = low-density lipoproteins; LH = luteinizing hormone; OR = odds ratio; ROC = receiver operating characteristic; TC = total cholesterol; TG = triglycerides. The full tables can be found in Supplementary Tables 1 – 5 To find an optimal cut-off value, we performed ROC analyses and calculated the Youden-Index. The cut-off values were as follows: LDL 2.95 mmol/L, HDL 1.32 mmol/L, LDL/HDL 2.02, and TC/HDL 3.15. In binary multivariate logistic regression, these cut-offs were shown to have better odds ratios – for LDL (OR 1.812 in Model A) and the lipid ratios (OR LDL/HDL 1.957; TC/HDL 1.942; TG/HDL 1.710), but not for HDL (OR 0.553) (Table  2 ). The 3D plots (Fig.  2 a-b and Supplementary Fig.  5 a-c) further show that the pre-pregnancy LDL/HDL ratio is strongly related to GDM, but also show interactions with other risk factors for GDM identified in Table  1 . The data shows that 7.9% of the women developed GDM if they were in the lowest tertiles for LDL/HDL ratio and fasting glucose, whilst 25.5% developed GDM if they were in the highest tertiles for both categories. 6.6% of women in the lowest tertile for LDL/HDL ratio and the highest tertile of LH received the diagnosis of GDM, compared to 28.2% of women in the highest LDL/HDL ratio tertile and lowest LH tertile. Statistically, this was also confirmed using the Mann-Whitney U test, showing a significant difference in GDM-rates ( p  < 0.001) in both cases. This shows that the interaction between these factors significantly affects GDM risk. Analysis of subgroups (displayed in Fig.  3 ) showed that there was a higher prevalence of GDM in women above the age of 29 years, with a high BMI, high fasting plasma glucose, and blood pressure. The rate of GDM was 9.8% in women aged < 29 years compared to 20.0% in women aged ≥ 29 years. However, the odds ratios show that an elevated LDL/HDL ratio also had a relevant effect on GDM occurrence in women who were younger, had a lower BMI, and had lower blood pressure.

From the measured lipid parameters, LDL and HDL were significantly different in GDM vs. non-GDM patients even after adjustment for confounders. The LDL/HDL and TC/HDL ratios were most significantly associated with the incidence of GDM in women undergoing IVF/ICSI. 119 (15.5%) of the 767 patients were diagnosed with GDM. This corresponds to the results of a large-scale meta-analysis, which estimated China’s overall GDM rate to be at 14.8% [ 23 ]. However, there are significant regional differences: several similar studies with pregnant women in different cities of China found the prevalence of GDM to be at 15.2% [ 24 ] and 19.7% in Beijing [ 25 ], and 10.5% in Shanghai [ 26 ]. Concerning lipid parameters, the prevalence of dyslipidemia amongst adults in China has been estimated to be 34.0%, though it is significantly higher in men and older people [ 1 ]. In this cross-sectional analysis, the average age of the participants was 50.24 years, and the rates of dyslipidemia were significantly higher in men (41.9%) than in women (32.5%). In our cohort of women undergoing IVF/ICSI, 30.4% of the study participants had at least one lipid parameter outside the recommended values of the AACE 2017 [ 21 ] and Chinese guidelines from 2016 [ 22 ], which is comparable to observed dyslipidemia rates in China. Recorded factors that were significantly associated with GDM were age, BMI, fasting glucose, LH, and E2 (all p  < 0.05). Age and BMI are known from the literature as risk factors for GDM [ 8 ], as well as hyperglycemia before pregnancy, which is represented by the pre-pregnancy fasting glucose measurement. One study suggested that the FSH/LH ratio could be an early predictor of GDM for women undergoing IVF [ 27 ]. A possible explanation could be the relationship between LH and estrogen, whereby the latter is known to be a protective factor against diabetes and modify insulin resistance [ 28 ]. Low levels of SHBG (sex hormone binding globulin), which are also closely interrelated with LH levels, are also associated with increased insulin resistance and therefore with the occurrence of GDM [ 29 ]. However, the exact mechanisms remain to be further explored. From the initial univariate analysis, three of the four measured lipid parameters, LDL, HDL, and TG, were significantly different in GDM versus non-GDM patients. Of these, HDL showed the strongest association ( p  < 0.001), whilst TC was not significantly different ( p  = 0.059). After adjusting for confounding factors, TG also did not show a significant association with GDM (in both Models A and B). The lipid ratios LDL/HDL and TC/HDL showed strong associations with GDM incidence, even after adjustment for confounders. Looking at binary LDL/HDL for example, it showed a significant odds ratio (OR 1.957 [95%CI 1.258, 3.044], even higher than for BMI with GDM (Supplementary Tables  3 a-d)). Our findings suggest that analyzing lipid ratios, along with other known risk factors such as age, baseline fasting glucose, and LH, could be a cost-efficient method to identify women undergoing IVF/ICSI who have a high risk for developing GDM. Previous studies that also measured the same four lipid parameters as in our study, but in a normally conceived pregnancy, have found that TG and HDL were significantly associated with GDM, but not TC and LDL [ 2 , 30 ]. Another study in China by Shen et al. [ 26 ] found that whilst TG was associated with GDM throughout pregnancy, TC and LDL were only higher in the first trimester in women who developed GDM. This indicates how lipid parameters also significantly change throughout the course of pregnancy. Even though our analysis showed differing results, whereby TG was only significantly associated with GDM in univariate analysis but not after adjustment for confounders, we also found that there was a strong, inverse association with HDL, as well as a positive association with LDL. In addition, many studies did not assess lipid ratios, which, in our study, had the strongest statistical association with GDM. For further studies, it would be meaningful to include calculated lipid ratios, as they may be a better indicator of lipid status [ 11 ]. The subgroup analysis in our cohort confirmed that older women, women with a higher BMI, higher pre-pregnancy fasting plasma glucose, blood pressure, as well as low LH showed a higher prevalence of GDM. As seen in Fig.  3 , the LDL/HDL ratio also had a significant effect on GDM risk in women who were younger, with a lower BMI, and lower blood pressure. Contrastingly, O’Malley et al. found that the association between dyslipidemia and GDM was only present in patients with obesity [ 3 ]. Our study indicates that there might be a high significance of lipid levels in our study cohort of Chinese women undergoing IVF/ICSI and in women with a low risk of GDM. While the median BMI in our study cohort was well within the normal range at 21.33 kg/m 2 , O’Malley’s study focused on women with at least one risk factor for GDM and the cohort had an overall obesity rate of 57.4%. The 3D plots (Fig.  2 a-b and Supplementary Fig.  5 a-c), which display the relationship between LDL/HDL with other GDM risk factors on the incidence of GDM, show that in our cohort, the LDL/HDL ratio was an even stronger predictor for GDM incidence than BMI. However, the analyzed factors also interact with each other, meaning that there are complex underlying mechanisms related to the development of GDM. For example, women in our cohort with a high LDL/HDL ratio and a high pre-pregnancy fasting glucose level had a much higher risk of developing GDM, compared to women with low LDL/HDL and fasting glucose (Fig.  2 a). Concerning the association between lipids and GDM development, there are several explanations and hypotheses, but the exact mechanisms remain unclear. Eppel et al. found that high levels of TG were associated with insulin resistance as well as β-cell dysfunction [ 31 ], which are both factors that play a significant role in the pathogenesis of GDM. In combination, impaired β-cells cannot compensate for the increased insulin resistance, thus leading to higher plasma glucose levels and the development of diabetes. Another mechanism is an alteration in the lipid metabolism during pregnancy, whereby lipid levels increase at the beginning of pregnancy as a result of the anabolic metabolism and then are broken down again [ 32 ]; increased fatty acids and glycerol are transferred from the breakdown of adipose tissue to the placenta as well as to the liver for gluconeogenesis [ 33 ] – thus, producing more glucose and leading to GDM. Another case-control study found that women who developed GDM had lower pre-pregnancy LDL peak diameter size as well as lower HDL levels [ 34 ]. Smaller LDL particles are more prone to oxidation, leading to β-cell dysfunction, which could also contribute to increased GDM risk. Further research into the exact pathomechanisms is required to fully understand how lipids contribute to the development of GDM, especially in women undergoing IVF/ICSI, a subgroup of women with a very high GDM risk. There are several study limitations, including that the family history of diabetes, the diagnosis of PCOS, and the use of medication, which are important risk factors for GDM, as well as pre-pregnancy HbA1c, were not recorded because this was a retrospective analysis of a previously prospective study focusing on IVF/ICSI outcomes. However, we have corrected for the biochemical diagnostic criteria for PCOS in the form of total testosterone concentration, as well as AFC and AMH in Model B of the multivariate logistic regression. Another limitation is that baseline blood tests were performed at different times before conception. As lipid levels can be significantly influenced by lifestyle factors such as diet and physical activity as well as uncontrollable factors, it could be beneficial to have several measurements from each study individual and evaluate their average. Furthermore, it is important to note that our study was a single-center study with only Chinese women, which is important because GDM has different prevalences in different regions and ethnicities. Therefore, the findings need to be studied in other cohorts to generalize our conclusions. Strengths of this study include that this is the first study focusing on the effects of pre-pregnancy lipid levels on GDM incidence specifically in women undergoing ART. Furthermore, we measured four different lipid parameters and compared them, in addition to three different lipid ratios.

From all lipid measurements and ratios, pre-pregnancy LDL/HDL and TC/HDL ratios were most significantly associated with GDM incidence in this post-hoc analysis of women undergoing IVF/ICSI. Analysis of pre-pregnancy LDL/HDL and TC/HDL might be a novel and cost-effective tool to identify women at high GDM risk in daily clinical practice in IVF centers.

With the change in diet and an increasingly sedentary lifestyle, obesity and high lipid levels are becoming more and more prevalent. Especially China, which in the past had relatively low rates of dyslipidemia, has seen an increase to more than one-third of the general adult population [ 1 ]. Dyslipidemia can be diagnosed according to different lipid parameters: LDL (low-density lipoproteins), HDL (high-density lipoproteins), total cholesterol (TC), and/or triglycerides (TG). Several studies have shown an association between dyslipidemia and gestational diabetes mellitus (GDM) [ 2 , 3 ], which is one of the most common pregnancy complications with a comparably high incidence in Southeast Asia [ 4 ]. Another population group with an increased risk of developing GDM are women undergoing ART (assisted reproductive technologies), for which there are different hypotheses, including the influence of hormonal therapies, as well as the factor that women undergoing ART commonly have certain baseline characteristics or diagnoses causing infertility and associated with GDM such as a higher prevalence of PCOS compared to women conceiving naturally [ 5 , 6 ]. In the last decades, the use of ART has significantly increased and has given more and more infertile couples the chance to have their own child [ 7 ]. From a scientific perspective, it also offers a unique opportunity to follow the course of a pregnancy before it even occurs. Because GDM is such a common pregnancy complication with many short- and long-term complications for mother and child [ 8 ], it is important to assess potential risk factors. By identifying high-risk patients, it would be possible to take early preventative measures. To date, more attention has been paid to the effects of maternal age and body mass index (BMI) on GDM [ 9 ]. There have also been studies on the effects of lipid parameters but to a much lesser extent. Additionally, these findings have not been studied in women undergoing in-vitro fertilization (IVF) or intracytoplasmic sperm injection (ICSI), which is a population especially at risk for GDM. Therefore, this study aims to and is the first to investigate the association between pre-pregnancy lipid levels and GDM in this specific cohort. Our objective is firstly, to identify whether and which lipid parameters are associated with GDM in IVF/ICSI patients and therefore which parameter is the most suitable for tracking lipid status, even before pregnancy, to identify patients with a high-risk profile for this pregnancy complication.

Below is the link to the electronic supplementary material. Supplementary Material 1 Supplementary Material 1 Supplementary Material 2 Supplementary Material 2