Predicting the outcome of IVF treatments using forward selection regression and linear discriminant analysis

Abstract: As more and more In Vitro Fertilizaiton (IVF) treatments are performed each year, there is a need to better predict the outcomes of different stages of the treatment and hence get a better understanding of which hormonal and physical parameters affect the treatment outcome and in what way. In this study, the effect of interaction between baseline AMH (anti-mullerian hormone) and DFI (DNA fragmentation index) on the chance of obtaining at least one good quality embryo was investigated, but no significance was found. Then, a statistical approach was used to find predictive models for each stage of the treatment. Linear regressions were fitted to predict continuous target variables and linear discriminant analysis (LDA) was performed to predict the binary ones. It was found that baseline AMH (anti-mullerian hormone), baseline FSH (follicle stimulating hormone), and female age significantly affect the number of retrieved oocytes (commonly referred to as eggs). Further, high BMI (body-mass index) was shown to have a significant negative impact on fertilization rate and the chance of receiving at least one good quality embryo. Finally, it was shown that the number of collected oocytes has a significant impact on fertilization rate, and that fertilization rate has a significant impact on the chance of receiving at least one good quality embryo. The best models found for predicting pregnancy and live birth did not significantly outperform the naive models which they were compared to. Hence, no significant conclusions were drawn from these models. All patients in the data set went through their first IVF treatment, defined by a first and single egg retrieval. All proven significance is on a 95\% level.