RESUMEN
OBJECTIVE: To derive and internally validate a clinical prediction model for live birth (LB) in women with polycystic ovary syndrome (PCOS) undergoing in vitro fertilization (IVF). DESIGN: Retrospective cohort study. SETTING: Four academic reproductive endocrinology clinics. PATIENTS: A total of 207 women with PCOS confirmed using Rotterdam criteria undergoing their first fresh IVF cycle. INTERVENTIONS: Not applicable. MAIN OUTCOME MEASURE: The primary outcome was cumulative LB per IVF cycle start. This included any LB that resulted from either fresh embryo transfer or any subsequent frozen embryo transfer from embryos obtained at the index oocyte retrieval. A prediction model was derived using multivariable logistic regression. Covariates considered for inclusion in the prediction model included demographic characteristics, medical history, and prior fertility treatment. Predicted probabilities for LB were calculated using the prediction model which included the 90% shrinkage factor for each adjusted odds ratio. RESULTS: The final model, on the basis of maximization of the area under the receiver operating characteristic curve, included age < 35 years, White race, presence of polycystic ovaries on ultrasound (polycystic ovary morphology), normal body mass index (<25 kg/m2), being metabolically healthy (no metabolic risk factors), and being a nonresponder to ovulation induction agents including letrozole and clomiphene citrate. The area under the receiver operating characteristic curve score for the model was 0.68 (95% confidence interval [CI]: 0.60, 0.77). Predicted probabilities of LB ranged from 8.1% (95% CI: 2.8, 21.5) for a woman who had no favorable predictors to 74.2% (95% CI: 59.5, 84.9) for a woman who had all favorable predictors. CONCLUSION: Our study demonstrated that, in addition to anovulation, the underlying pathophysiology and associated comorbidities alter the likelihood of a successful pregnancy in women with PCOS undergoing IVF. Further validation of this model is needed before it can serve as a tool to personalize prediction estimates for the probability of LB in women with PCOS.