| Objective: With the increasing burden of global infertility diseases and the liberalization of China’s fertility policy,the clinical application of assisted reproduction will be further increased,so it is urgent to evaluate the characteristics and clinical effects of patients with a set of models that can accurately predict pregnancy outcomes.This study aimed to construct a prediction model based on live birth and cumulative live birth probability in infertile patients treated with in vitro fertilization-embryo transfer/cytoplasmic sperm injection(IVF-ET/ICSI)before and after treatment,and to validate and evaluate the model,in order to obtain a set of clinically usable prediction tools.By comparing and discussing the advantages and disadvantages of multiple adaptive regression splines(MARS)to construct nonlinear models and traditional logistic regression models in predicting patients.Methods: General data,follicle monitoring records,embryo culture records and neonatal records were collected from January 2015 to December 2020 in a reproductive medicine center in Jiangxi Province.Restrictive cubic splines were used to test the nonlinear effects of continuous independent variables and live birth and cumulative live birth.According to the time group,the patient cycle from 2015 to 2018 was used as the modeling group,and live birth and cumulative live birth were used as dichotomous outcomes,and the MARS method was used to construct a live birth prediction model before treatment(Model 1),a live birth prediction model after treatment(Model 2),a cumulative live birth prediction model before treatment(Model 3),and a cumulative live birth prediction model after treatment(Model 4)according to whether treatment information was included.Taking the 2019-2020 cycle as the verification group,the area under the curve(AUC),net weight classification index(NRI),and comprehensive discriminant improvement index(IDI)were used to verify the accuracy of the predictive model differentiation,and the calibration curve and brier score were used to evaluate the calibration degree of the model.The traditional stepwise logistic regression method was used to reconstruct the prediction model,and the validation group was used to compare the accuracy of its prediction with the MARS model by the same index,and the applicability of the model was evaluated by decision curve analysis(DCA).Results: A total of 24,672 cycles that met the criteria were included in the study,of which 13,616 were born live in the first embryo transfer,with an average live birth rate of 55.03%;16,745 cumulative live births were obtained after the end of the complete cycle,and the average cumulative live birth rate reached 67.50%.The median age of women receiving treatment was 30 years,and the type of infertility was mainly secondary infertility(59.87%),and the median age of men was 32 years.The RCS model showed that there was a significant nonlinear correlation between female age,AFC,h CG intimal thickness,egg acquisition and logarithmic probability of live birth,and the nonlinear relationship between female age,AFC,basal FSH,basal LH,basal E2,h CG intimal thickness,progesterone,egg harvest,number of high-quality embryos,embryonic live birth score and logarithmic probability of cumulative live birth was also statistically significant.Four prediction models were successfully constructed by MARS method for 17101 cycles(69.3%)in the modeling group,among which model1 retained female age,antral follicle count(AFC),previous ovulation induction times,and scarred uterus as effective predictors.Model 2 retained a total of 8 valid variables including female age,number of eggs obtained,number of embryos transferred,cycle type,h CG intimal thickness,ovulation induction protocol,embryo type,and previous ovulation induction times.Model 3 retained a total of five valid variables: female age,AFC,previous ovulation induction times,basal follicle-stimulating hormone(FSH),and basal luteinizing hormone(LH).Model 4 included seven valid variables: female age,number of high-quality embryos,number of eggs obtained,progesterone(P),score of live embryo,number of previous ovulation induction and thickness of the h CG intima.The results of five-fold cross-validation showed that the average AUCs of the four models were 0.658(95% CI: 0.649～0.667),0.681(95% CI: 0.672～0.690),0.725(95%CI: 0.716～0.734)and 0.797(95% CI: 0.789～0.805),and the average calibration slope was close to 1.The sensitivity analysis of this model by using the median & mode imputation method and the complete data set can prove that the prediction performance of the fitted model is stable and reliable.The prediction ability of the model was verified by using the information of 7571 cases(30.7%)in the verification group,and the AUC of the centralized model was 0.641,0.657,0.712 and 0.802,and the calibration Brier scores were 0.231,0.225,0.182 and 0.160,respectively.Comparing the difference in prediction accuracy between post-treatment and pre-treatment models in the outcomes of live birth and cumulative live birth,the prediction accuracy of the post-treatment models of the two groups was significantly improved compared with that before treatment,among which the NRI of the live birth model increased by 4.24%(95% CI:2.50%～5.99%),and the IDI increased by 1.73%(95% CI: 1.38%～2.09%).The NRI of the cumulative live birth model increased by 13.69%(95% CI: 11.43%～15.95%),and the IDI increased by 9.99%(95% CI: 9.29%～10.68%),and the model prediction performance after treatment was good.Compared with the traditional logistic model,the AIC value of the MARS model is smaller,and in terms of model prediction,the AUC of the MARS model is 0.802 and the BRIER score is 0.160,which are better than the traditional model;after using the MARS model,the NRI increases by 3.38%(95%CI: 1.63%～5.13%),and the IDI increases by 2.63%(95% CI: 2.13%～3.14%),and the difference is statistically significant.The DCA results showed that when the threshold was in the range of 30%～80%,the MARS model had a higher net benefit than the traditional model,and the clinical applicability of the model was better.Conclusions:(1)In this study,the prediction model of live birth and cumulative live birth before and after treatment was successfully constructed by using the clinical information of infertility patients,and the prediction model using cumulative live birth as the outcome was better.(2)The prediction model after treatment has significantly improved the prediction effect compared with the prediction before treatment.(3)In this study,the MARS method was used to establish the prediction model,and the prediction performance of the MARS model was better than that of the traditional Logistic model. |
PDF Full Text Request