Predicting Cumulative Live Birth Before The First And Second Complete Cycle Of Ivf: A Population-Based Study Of Linked Cycle Data From 79,512 Women.

To develop prediction models that can estimate the cumulative probability of live birth over three complete cycles of IVF. A retrospective cohort study. Linked fresh and frozen cycle data from women who underwent IVF (including intracytoplasmic sperm injection) in 2014-2016 were extracted from the Society for Assisted Reproductive Technology Clinic Outcome Reporting System database. Discrete time logistic regression was used to develop two models to predict cumulative live birth over three complete cycles of IVF. A complete cycle was defined as all fresh and frozen-thawed embryo transfers originating from one episode of ovarian stimulation. The pre-treatment model estimates the chance of live birth before starting treatment using couple characteristics, while the post-treatment model revises predictions using updated information before the second complete cycle. Potential predictors were female age, body mass index (BMI), serum anti-Mullerian hormone (AMH), previous full-term birth, male infertility, tubal factor, diminished ovarian reserve, polycystic ovaries, endometriosis, uterine factor and unexplained infertility. The post-treatment model also included number of eggs. Age, BMI and AMH were modelled as restricted cubic splines due to their non-linear relationship with probability of live birth. To assess optimism in model discrimination and calibration, internal validation using 300 bootstrap samples was done. Of 79,512 women who underwent 103,270 complete cycles, 36,850 (46.3%) had a live birth from the first complete cycle, and cumulatively, 44,172 (55.6%) had a live birth over three complete cycles. Factors that were predictive of live birth in the pre-treatment model included female age (35 vs 25 years, adjusted odds ratio 0.62, 95% confidence interval 0.59 to 0.66), BMI (35 vs 25, 0.81, 0.77 to 0.84), previous full-term birth (1.13, 1.09 to 1.17), male infertility (1.21, 1.18 to 1.25), uterine factor infertility (0.85, 0.80 to 0.91), unexplained infertility (1.20, 1.15 to 1.25) and serum AMH (5 vs 2.4ng/mL, 1.23, 1.18 to 1.27). The post-treatment model included these factors as well as egg number (15 vs 9, 1.34, 1.25 to 1.44). The C-statistic for all models was between 0.70 and 0.72. Internal validation showed no over-optimism in discrimination or calibration performance. In a 36-year-old woman with unexplained infertility, no previous live births, BMI of 25 and AMH of 2ng/L, our pre-treatment model predicts a 51% chance of live birth over her first complete cycle (82% over three complete cycles). If the first treatment is unsuccessful and the same woman starts a second cycle at 38 years old with AMH = 1.5ng/mL, our post treatment model estimates her chance of live birth at 21% after the second complete cycle and 40% over the second and third complete cycles. These novel prediction models show accurate performance and are available to patients and clinics free of charge at sart.org. We anticipate that by providing individualised chances of success at two critical stages of IVF, they will help couples prepare emotionally and financially for IVF.