The Influence of Defining Desired Outcomes on Prediction Algorithms: Mass Spectral Analysis of Spent Blastocyst Media in Ai/ML Prediction of IVF Embryo Implantation or Viability, or Both?

The current goal of much clinical IVF research is to predict which blastocyst has the greatest probability of success. The criteria of such evaluation is that the test must be non-invasive, rapid and cost effective. One of the promising approaches is metabolomic analysis of spent blastocyst culture media (SBM). Techniques such as mass spectrometry yield enormous analytical data sets and Ai/ML analysis has to be deployed. When analysing the metabolomic profile, or any data sets, to generate predictive algorithms by Ai/ML software; outcome grouping have to be accurately identified. Furthermore, the outcome group has to be clearly inputted with precision. Often the definition of success for such test develop is implantation. However, the actual goal is live births, and there is a considerable number of loss events between implantation and live birth. Implantation itself is often defined as a positive pregnancy test, but this can be transient and superficial attachment, such as seen in biochemical pregnancy; clear and prolonged, as in blighted ovum/anembryonic pregnancy or first trimester miscarriage, and indeed highly elevated and progressive, as in later trimester spontaneous abortions. Blastocyst competence prediction Ai/ML algorithms were generated against two linked but distinct outcomes implantation (a positive maternal hCG test) and viable pregnancy (defined as heart beat and survival after 16 weeks’ gestation). The two optimum complementary Ai/ML learning algorithms generated overlapped in many of the biochemical marker peaks that were used in the performance of the algorithms. However, some were unique to the “Viability” algorithm or utilised different ranges of the metabolites as cut-offs in decision tree classifications. Although statistically there was a significant correlation, this was a non-linear regression. The combined algorithm interpretation system of high implantation and high viability scoring, demonstrated an 83.5% positive predictive value for selectingviable implanting blastocysts. Since the ultimate goal is to increase live birth success rates in IVF, the application of profiling of the blastocyst must consider features of trophoblast cell metabolic biochemistry, which largely determines implantation; and early metabolic markers of the inner cell mass that reflect embryo viability.