Search

Infertility is a huge problem nowadays, not only from the medical but also from the social point of view. A key step to improve treatment outcomes is the possibility of effective prediction of treatment result. In a situation when a phenomenon with more than 2 states needs to be explained, e.g. pregnancy, miscarriage, non-pregnancy, the use of multinomial logistic regression is a good solution. The aim of this paper is to select those features that have a significant impact on achieving clinical pregnancy as well as those that determine the occurrence of spontaneous miscarriage (non-pregnancy was set as the reference category). Two multi-factor models were obtained, used in predicting infertility treatment outcomes. One of the models enabled to conclude that the number of follicles and the percentage of retrieved mature oocytes have a significant impact when prediction of treatment outcome is made on the basis of information about oocytes. The other model, built on the basis of information about embryos, showed the significance of the number of fertilized oocytes, the percentage of at least 7-cell embryos on day 3, the percentage of blasts on day 5, and the day of transfer.

Infertility is currently a common problem with causes that are often unexplained, which complicates treatment. In many cases, the use of ART methods provides the only possibility of getting pregnant. Analysis of this type of data is very complex. More and more often, data mining methods or artificial intelligence techniques are appropriate for solving such problems. In this study, classification trees were used for analysis. This resulted in obtaining a group of patients characterized most likely to get pregnant while using in vitro fertilization.

Many factors play an important role in prediction of infertility treatment outcome (for example, female age and quality of oocytes or embryos are the most important prognostic factors concerning positive IVF outcome). The purpose of this study was to identify a set of variables that could fulfill criteria for prediction of pregnancy in IVF patients through the application of data mining – using the discriminant analysis method. The principle of this method is to establish a set of rules that allows one to place multi-dimensional objects into one of two analyzed groups (pregnant or not pregnant). Six hundred and ten IVF cycles were included in the analysis and the following variables were taken into consideration: female age, number and quality of retrieved oocytes, number and quality of embryos, number of transferred embryos, and outcome of treatment. Discriminant analysis allowed for the creation of a model with a 51.22% correctness of prediction to achieve pregnancy during IVF treatment and with 74.07% correctly predicted failure of pregnancy. Therefore, the created model is more suitable for the prediction of a negative outcome (lack of pregnancy) during IVF treatment and offers an option for adjustments to be made during infertility treatment.

Principal Component Analysis is one of the data mining methods that can be used to analyze multidimensional datasets. The main objective of this method is a reduction of the number of studied variables with the mainte- nance of as much information as possible, uncovering the structure of the data, its visualization as well as classification of the objects within the space defined by the newly created components. PCA is very often used as a preliminary step in data preparation through the creation of independent components for further analysis. We used the PCA method as a first step in analyzing data from IVF (in vitro fertilization). The next step and main purpose of the analysis was to create models that predict pregnancy. Therefore, 805 different types of IVF cy- cles were analyzed and pregnancy was correctly classified in 61-80% of cases for different analyzed groups in obtained models.