Wen Yao Mak1,2, Muyesaier Alifu1, Aole Zheng1, Xiaoqiang Xiang1
1Department of Clinical Pharmacy and Pharmacy Administration, School of Pharmacy, Fudan University, 2Clinical Research Centre (Penang General Hospital), Institute for Clinical Research, National Institute of Health
BACKGROUND: Pregnancy loss (PL) after FET imposes significant physiological and mental burden on the patients, and reduced success rates of subsequent attempt. However, comprehensive analysis of cycle-dependent risk factors in frozen-thawed embryo transfer (FET) remains limited. Traditional analysis of PL hazards typically depended on non- or semi-parametric approach to identify risks factors[1, 2], but these tools are inadequate to capture the complex relationships of real-world data (RWD), particularly between nonlinear factors that are often encountered in clinical practice. Conversely, parametric time-to-event (TTE) analysis is favored as it considers the event time itself as a variable and could handle censored data[3, 4]. The parametric TTE approach could quantify repeated events (if any), rendering it well suited to uncover novel predictors of PL over the entire FET timespan. OBJECTIVE: The study aimed to identify novel predictors of PL across FET cycles, and to improve personalized risk prediction after FET treatment. METHODS: Patient data This is a retrospective cohort study conducted at Department of Assisted Reproduction, Shanghai Ninth People’s Hospital between January 9, 2004 (the date of the first FET case at the Ninth Hospital), and December 31, 2023 (censor date). A total of 24,980 FET cycles were included in the analysis. Baseline demographics and key covariates, including female age, weight, BMI(Body Mass Index), endometrial thickness, hormonal levels (progesterone and estradiol) on FET day, ovarian reserve status, endometrium preparation protocols were extracted for analysis, among other relevant variables. Model development Different baseline hazard parameterizations were tested, including exponential, Weibull, Gompertz, log-normal and log-logistic distributions. Base hazard model selection was based on AIC. A stepwise covariate model (SCM) building procedure was performed for covariates searching (forward selection: p<0.05, backward elimination: p<0.001). For missing covariates, last observation carry forward was applied; for baseline covariates, the median was imputed. The final model was evaluated using Kaplan-Meier visual predictive checks (VPCs) for time-to-first events. The model was externally validated with three different cohorts: an earlier and a later (±1 year) cohort, and ethnic minority (non-Han Chinese). RESULTS The mean (SD) age of each cycle patient was 32.22 (4.50) years. Pregnancy loss rate of per cycle was 14.96% and the median (interquartile range) follow-up period of per cycle 8.2 (3-8.57) months. The Gompertz distribution best described the PL hazard, where the hazard increased over time until a plateau was reached. Model parameters (values, [%RSE]) were precisely estimated, (scale, ?=0.0774 [3%]; shape, ß=-0.378 [2%]), where PL hazard was aggravated by increasing female age (2.3 [6%]) and previous pregnancy count (parity=1: 0.0462 [93%], 2: 0.0419 [127%], 3: 0.298 [22%]), while female weight (-0.67 [14%]), endometrial thickness on FET day (-0.503 [15%]), progesterone concentration on FET day (-0.0355 [41%]), and FET protocols – mild stimulation (-0.217 [16%]), late stimulation (-0.144 [30%]), modified late stimulation (-0.237 [17%]), HRT (-0.102 [66%]), downregulation-HRT (-0.134 [61%]) – reduced the hazard. External validation suggested temporal drift of increasing hazard over time, possible due to the COVID-19 lock-down in the “later” cohort that aggravated PL. Validation in the ethnic minority cohort indicated good model performance through KM-VPC plots. These findings align with previous studies that have reported similar risk factors, such as advanced maternal age[1, 2, 5, 6], history of miscarriage[7] and endometrium preparation protocols[5, 8] as significant predictors of pregnancy loss in ART populations. CONCLUSION: In this study, female weight, female age, diminished ovarian reserve, endometrial thickness on FET day, progesterone concentration on FET day, previous pregnancy count number and endometrium preparation protocols were identified as important contributors to pregnancy loss risks over FET cycle. The parametric TTE method complemented existing analytical toolbox in understanding complex real-world clinical data, and could provide much added value in guiding clinical practice.
1.Hourvitz, A., et al., Role of embryo quality in predicting early pregnancy loss following assisted reproductive technology. Reprod Biomed Online, 2006. 13(4): p. 504-9. 2.Hu, L., et al., Influencing factors of pregnancy loss and survival probability of clinical pregnancies conceived through assisted reproductive technology. Reprod Biol Endocrinol, 2018. 16(1): p. 74. 3.Tran, Q.T., et al., A simple time-to-event model with NONMEM featuring right-censoring. Transl Clin Pharmacol, 2022. 30(2): p. 75-82. 4.Holford, N., A time to event tutorial for pharmacometricians. CPT Pharmacometrics Syst Pharmacol, 2013. 2(5): p. e43. 5.Ozer, G., et al., Prediction of risk factors for first trimester pregnancy loss in frozen-thawed good-quality embryo transfer cycles using machine learning algorithms. J Assist Reprod Genet, 2023. 40(2): p. 279-288.
Reference: PAGE 33 (2025) Abstr 11321 [www.page-meeting.org/?abstract=11321]
Poster: Drug/Disease Modelling - Endocrine