Construction of a Prediction Model for Sleep Quality in Embryo Repeated Implantation Failure Patients Undergoing Assisted Reproductive Technology Based on Machine Learning: A Single-Center Retrospective Study.

Objective: Constructing a predictive model for sleep quality in embryo Repeated Implantation Failure(RIF) patients using multiple machine learning algorithms, verifying its performance, and selecting the optimal model.

Methods: Retrospective collection of clinical data from RIF patients who underwent assisted reproductive technology at the Reproductive Medicine Center of Tongji University Affiliated Obstetrics and Gynecology Hospital from January 2022 to June 2022, divided into a training set and a validation set in an 8:2 ratio. Use Lasso regression to screen variables and construct a risk prediction model using six machine learning algorithms. Evaluate the validity of the model using the area under the curve (AUC), and comprehensively evaluate the performance of the model based on F1 score, accuracy, sensitivity, and specificity. Use SHAP method to explain the contribution of each variable in the optimal model to the occurrence of sleep disorders.

Results: A total of 404 RIF patients were included in the study. The incidence of sleep disturbances was 48.76%. After LASSO regression analysis, nine variables were selected for inclusion in the model. The RF model has an AUC of 0.941, Accuracy of 0.938, Specification of 0.950, and F1 score of 0.938 in the validation set, making it the optimal model for this study. According to the SHAP feature importance ranking of the RF model, the factors influencing sleep quality in RIF patients were E2, SDS, Fertiqol, FSH, daily exercise time, weekly shift work hours, coffee consumption, sunbathing, and SAS.

Conclusion: The RF model is the optimal model for predicting the sleep quality of RIF patients. Its sleep quality is not only affected by physiological factors, but also by psychological and lifestyle factors. Medical personnel should implement intervention strategies as early as possible based on relevant risk factors to improve the sleep quality of this population.