Development and validation of a visualized prediction model for early miscarriage risk in patients undergoing IVF/ICSI procedures: a real-world multi-center study

Abstract

Background: This study focuses on the risk of early miscarriage in patients undergoing in vitro fertilization (IVF) or intracytoplasmic sperm injection (ICSI). These patients commonly experience heightened stress levels and may discontinue treatment due to emotional burdens associated with repeated failures. Despite the identification of numerous potential factors contributing to early miscarriage, there exists a research gap in integrating these factors into predictive models specifically for IVF/ICSI patients. The objective of this study is to develop a user-friendly nomogram that incorporates relevant risk factors to predict early miscarriage in IVF/ICSI patients. Through internal and external validation, the nomogram facilitates early identification of high-risk patients, supporting clinicians in making informed decisions.

Methods: A retrospective analysis was conducted on 20,322 first cycles out of 31,307 for IVF/ICSI treatment at Sun Yat-sen Memorial Hospital between January 2011 and December 2020. After excluding ineligible cycles, 6,724 first fresh cycles were included and randomly divided into a training dataset (n = 4,516) and an internal validation dataset (n = 2,208). An external dataset (n = 1,179) from another hospital was used for validation. Logistic and LASSO regression models identified risk factors, and a multivariable logistic regression constructed the nomogram. Model performance was evaluated using AUC, calibration curves, and decision curve analysis (DCA).

Results: Significant risk factors for early miscarriage were identified, including female age, BMI, number of spontaneous abortions, number of induced abortions and medical abortions, basal FSH levels, endometrial thickness on hCG day, and number of good quality embryos. The predictive nomogram demonstrated good fit and discriminatory power, with AUC values of 0.660, 0.640, and 0.615 for the training, internal validation, and external validation datasets, respectively. Calibration curves showed good consistency with actual outcomes, and DCA confirmed the clinical usefulness. Subgroup analysis revealed variations; for the elder subgroup (age ≥35 years), female age, basal FSH levels, and number of available embryos were significant risk factors, while for the younger subgroup (age <35 years), female age, BMI, number of spontaneous abortions, and number of good quality embryos were significant.

Conclusions: Our study provides valuable insights into the impact factors of early miscarriage in both the general study population and specific age subgroups, offering practical recommendations for clinical practitioners. We have taken into account the significance of population differences and regional variations, ensuring the adaptability and relevance of our model across diverse populations. The user-friendly visualization of results and subgroup analysis further enhance the applicability and value of our research. These findings have significant implications for informed decision-making, allowing for individualized treatment strategies and the optimization of outcomes in IVF/ICSI patients.

Keywords: IVF-ET; early miscarriage; individualized prediction; nomogram; pregnancy outcome.

Figure 1

General flowchart of modeling algorithm.

Figure 2

Nomogram for predicting the probability of early miscarriage following IVF/ICSI treatment was developed and validated in three patient groups: (A) Total patients, (B) Patients aged ≥35 years, and (C) Patients aged <35 years.

Figure 3

ROC curve of the Training set, Internal validation model and External validation model for total patients.

Figure 4

Calibration of the model to predict miscarriage probability in (A) Training set, (B) Internal validation model and (C) External validation model for total patients.

Figure 5

Decision curve analysis of miscarriage nomogram for the total patients.