The high complexity of factors contributing to IVF cycle outcomes makes manual analysis difficult, and often impossible. Through the model development process, we developed a process to maximize the signal within a dataset in order to find the greatest contributing factors for positive clinical outcomes.

Extracting patterns from the data by combining advanced computational and statistical methods helps to improve the efficiency and quality of IVF treatment and, possibly, decrease the cost. It essentially transforms medical records into medical knowledge.

Since many factors in an IVF cycle are interrelated, it is near impossible to conduct a proper root-cause analysis by deploying logarithmic regression or similar analysis. Moreover, the possible correlation between factors may cause confounding, and results obtained using one predictive factor may be quite different from those obtained with multiple predictors.

Predicting IVF cycle outcomes is, strictly speaking, a binomial classification task (0 or 1, pregnant vs “not pregnant”). At the same time, machine learning algorithms will predict a qualitative response as numerical probabilities of each class by using sigmoid or similar function.

175 clinical factors and 145 morphological and kinetic parameters of embryo development were recorded for each IVF PGT cycle. From 320 original features, 6470 synthetic features were created (by Weight of Evidence for columns, Encoding of categorical levels of the feature, Cross-Validation Target Encoding, Time series, etc.), tested, and 357 features were selected for each IVF PGT cycle.

For convenience, we named the calculated probability of a positive clinical outcome as “ReproScore”. ReproScore can be used as a precise and specific measurement during embryo selection for the embryo transfer. This single numeric metric, backed by extensive machine learning calculations described above, simplifies the embryo selection process. It provides information that is accurate and easy to interpret for the patients and clinical staff. It can also be used to make treatment plans for future IVF cycles.


ReproScore for RSC patients can be found here (providers only):