Discussion
A convincing correlation between endometrial thickness development and patients' BMI could not be substantiated. However, FSH and LH levels exhibited a surprising decreasing trend with increasing BMI, supporting the evolutionary selective role of nutritional status. Our SVM model outperforms previous models; however, to confidently predict the outcome of embryo transfer, further optimization is necessary.
Methods
537 participants undergoing IVF/ICSI cycles with successful oocyte retrieval were enrolled. Subjects were divided into four BMI based groups: underweight (UW; n=32), normal weight (NW; n=324), overweight (OW; n= 115), obesity (OB; n=66). Anthropometric and anamnestic parameters, characteristics of stimulation, endometrial thickness on the day of hCG injection, at puncture, at embryo transfer, FSH, LH, AMH, partner's age and the semen analysis indicators, embryo quality, clinical pregnancy, were recorded and analyzed. Support Vector Machine (SVM) was built to predict potential pregnancies based on medical data using 22 dimensions.
Results
In accordance with BMI categories, when examining pregnant/non-pregnant division, the average age of pregnant women was significantly lower in the UW (30.9 ± 4.48 vs. 35.3 ± 5.49 years, p=0.022), NW (34.2 ± 4.25 vs. 36.3 ± 4.84 years, p<0.001), and OW (33.8 ± 4.89 vs. 36.3 ± 5.31 years, p=0.009) groups. Considering FSH, LH, and AMH levels in each BMI category, a statistically significant difference was observed only in the NW category FSH was significantly lower (7.8 ± 2.99 vs. 8.6 ± 3.50 IU/L, p=0.032) and AMH (2.87 ± 2.40 vs. 2.28 ± 2.01 pmol/L, p=0.021) was higher in pregnant women. There were no further statistically significant differences observed between the pregnant and non-pregnant groups across any BMI categories, especially concerning endometrial development. Surprisingly, BMI and weight correlated negatively with FSH (r=-0.252, p<0.001; r=-0.206, p<0.001, respectively) and LH (r= -0.213, p<0.001; r= -0.195, p<0.001) in the whole population. SVM model average accuracy on predictions was 61.71%.