Deep phenotyping to predict live birth outcomes in IVF – the authors suggest that their mathematical model is 100x more accurate than using age to predict IVF outcomes. The authors took data from previous IVF cycles at Stanford and developed a model using complicated statistical tests. They tested the model on IVF cycles from a later time in their clinic. They did not have the data for some variables that we currently think are important such as antral follicle counts and AMH (Anti-Mullerian Hormone).
However, their model requires a primary IVF cycle and is only used for subsequent cycles. Stanford University has a patent on the model and the authors hope to commercialize this and other medical forecasting tools. Click here to read the entire abstract.
This is an important area of research for physicians in any specialty but certainly ours. We would love to have a model to provide the most accurate probabilities so that patients can make informed decisions. We still have factors that probably influence outcomes but we are not sure how to use the data and this study tries to address that problem.
I wonder though whether the model will hold up when applied to datasets from clinics with substantially better outcomes that those at Stanford. If the egg quality from stimulation and the embryo quality is better at some clinics, will that influence the model prediction accuracy?