LAUSR

Despite the vast improvements of assisted reproduction, IVF live birth rates have only slowly improved. Delayed childbearing and the well-known negative effect of age on reproduction is one likely explanation. Another often cited explanation is that we still cannot choose with high accuracy the embryo with the highest implantation potential for transfer. Various technologies (metabolomics, proteomics, timelapse algorithm-based selection (TLM), preimplantation genetic testing for aneuploidy (PGT-A)) have been evaluated as tools to improve embryo selection [1]. TLM has received much attention in recent years. While only a few randomized controlled trials have been published assessing TLM’s benefits (11 clinical trials listed on Pubmed from the past 8 years), far more reviews, meta-analyses and commentaries (4 times as many) have been published on it. Reviews, expert opinions, and meta-analyses often come up with conflicting conclusions leaving clinicians in a difficult position when couples need to be counseled [2, 3]. There are multiple reasons for conflicting opinions. It is always said that a review or meta-analysis is only as good as the studies it is based on. The TLM studies in the literature are the perfect example why it is almost impossible to come up with a good meta-analysis. The various RCTs (randomized controlled trials) enrolled rather heterogeneous patient populations. Culture conditions have not been standardized (culture medium, O2 concentration) and even within an RCT, the conditions under which experimental and control embryos are cultured could differ. Due to technical differences in the equipment, it is also hard to compare the various time-lapse systems. The day of embryo transfer and the number of embryos transferred in many of the studies are not similar either. Zaninovic et al. in their retrospective analysis tried to control as many confounding variables as possible (culture conditions similar though different culture media used; Embryoscope was used by both centers; two sets of analysis, one with day 3 development and single embryo transfer, one with day 5 development) but the oocyte source was still heterogeneous (autologous vs. donated oocytes) that may limit the generalizability of the results [4]. Time-lapse monitoring is claimed to have at least two advantages over Bstandard^ embryo culture. It provides undisturbed culture conditions and significantly more embryo observation points. These multiple observations allow us to build algorithms that may be predictive of clinical outcome after the transfer of the selected embryo. However, the different studies use different algorithms. Some use only early kinetic markers. Others use both early and late kinetic markers and yet others add morphology to the kinetic markers as well. Finally, the different studies use algorithms to predict various clinical outcomes. Therefore, it is not surprising that meta-analyses published within one year of each other, which base their findings on only slightly different RCTs after including-excluding certain studies for various issues, draw opposing conclusions [2, 3]. Zaninovic et al. studied the correlation between a few early kinetic markers and implantation/blastocyst formation using two datasets. They measured receiver operator characteristic curve (ROC) and area under the curve (AUC) to measure how good the correlation was between these markers and endpoints. They obtained similar distribution of the data in the two clinics and the kinetics of early events tended to fall within similar time ranges. However, the predictive ability of the studied parameters was relatively poor as the AUCs fell in the 0.5– 0.6 range. The combination of various parameters to provide an algorithm to improve the predictive value was not studied. As mentioned above, multiple algorithms have been proposed by the different research groups. However, external validation has not been able to reproduce their predictive power [5]. Petersen et al. have proposed a universally applicable algorithm, but their results have not been reproduced by others [6]. It has been suggested that local, clinic-specific algorithms * Peter Kovacs [email protected]