RESUMEN
OBJECTIVES: Develop an up to date prediction model using recent cycle data and key pre-treatment predictor variables to estimate a couple's individualised probability of a cumulative live birth after one cycle of ovarian stimulation and transfer of all frozen embryos, before the first embryo transfer. STUDY DESIGN: This was a retrospective cohort study. To estimate the cumulative live birth rate we only included couples who had used all embryos from their initial stimulation or achieved a live birth. We constructed a logistic regression model using live birth as a dependent variable and age group, duration of infertility, primary vs. secondary infertility, insemination method, cause of infertility, Anti-Mullerian Hormone (AMH), Follicle Stimulating Hormone (FSH) and antral follicle count (AFC) as our independent variables and used a backward elimination method to create the best fitting regression models to predict the probability of a cumulative live birth (p < 0.05 for elimination). RESULTS: There were 516 complete cycles of ovarian stimulation resulting in 357 livebirths giving a cumulative livebirth rate of 69.2 % (95 % CI 66.0-74.0). Women with a live birth had significantly lower median age (34 years [IQR 31-37] vs. 36 years [IQR 33-39], p = 0.01) and FSH (6.7 iu/L [IQR 5.8-7.9] vs. 7.4 iu/L [IQR 6.2-8.6] and a significantly higher median AMH (22.1 pmol/L [IQR 12.1-30.9] vs. 10.5 pmol/L [IQR 7.3-20.7], p = 0.01) and AFC (18 [IQR 12-26] vs. 12 [IQR 9-19], p = 0.01). The backward conditional logistic regression model retained age category, FSH category and AMH category as significant independent predictors. The area under the curve for this model was 0.68 (95 % CI 0.63 - 0.73). CONCLUSION: Our prediction model estimates a couple's individualised probability of achieving a live birth after their first complete cycle of IVF using all known pre-treatment predictors. LIMITATIONS, REASONS FOR CAUTION: The study population were only those eligible for NHS funded IVF treatment which have strict ovarian reserve criteria. Exclusion of those with very low egg reserve is likely to influence the predictive capacity of out model. Furthermore, our model was developed using cycle data from one unit and thus its predictive capacity has not been assessed on an independent cohort of women. We therefore welcome external geographical validation of our model prior to its use in clinical practice.