Background: In order to optimize IVF strategies, particularly with the use of single embryo transfer, good predictive models are required. Here, we develop a model to allow such prediction, and the structure of the models point to more general conclusions about the mode of action of prognostic factors.
Methods: Anonymized data from consecutive embryo transfers in five IVF centres in the UK for the 2000-2005 period were extracted and the morphological grade based on common scoring criteria was included. There were 16 096 (12 487 fresh and 3609 frozen) transfers, for 8775 couples, available for analysis. Live birth data were fitted to a model with separate sub-models for embryo and recipient effects [the 'Embryo-Uterus' (EU) model]. All covariates were included, with sub-model selection using Akaike's information criterion.
Results: Age, number of embryos created, attempt number, previous history of pregnancy, duration of infertility, day of transfer and tubal diagnosis were all identified as significant prognostic factors, along with embryo grade and growth rate. Frozen transfers were substantially less likely to lead to a live birth with odds ratios of 1/3 to 1/2 compared with fresh transfers, with no evidence of differential loss for any particular patient group. Age acts predominantly through the embryo component with only a weak effect on the uterus. The embryo number, attempt number, previous pregnancies and duration of infertility act predominantly through the uterine environment. Both sub-models show significant heterogeneity between centres.
Conclusions: The EU modelling framework has generated a model for predicting outcomes of embryo-transfer procedures, subject to the limitations of routinely collected data. With this large data set, the model allows identification of factors that act specifically on embryo viability or maternal receptivity. Variability in the two components between centres with similar overall outcomes suggests scope for further optimization of IVF treatment.