Embryo Selection by Time Lapse Monitoring for Single Embryo Transfer ("SET")

Prospective RCT comparing two embryo selection methods (TL algorithm vs standard morphology based) for SET. Infertile couples scheduled to undergo IVF or intracytoplasmic sperm injection (ICSI) and met the inclusion-exclusion criteria were enrolled. Randomization to Groups 1 or 2 was performed prior to the start of stimulation. Randomization was carried out in blocks of two, by selecting TL or control assignments from closed, opaque envelopes. Patients were blinded to allocation. Standard luteal long (Suprefact, Sanofi Aventis) and flexible gonadotropin-releasing hormone (GnRH) antagonist (Cetrotide, Merck Serono) protocols and recombinant FSH(Gonal-f, Merck Serono) were used for stimulation. Embryos were placed into individual wells of a 9-well Primo Vision culture dish (Vitrolife AB, Sweden) on day 1 of culture, after checking for fertilization. Out of incubator handling was same in the groups;: day 1: fertilization check; day 3: culture medium change; day 5: transfer. On days 3 and 5, the embryos were morphologically assessed under an inverted microscope. Embryos in TL group were imaged at 10-minute intervals. The Primo Vision software was used for the analysis of TL images. The first appearance of the 2, 3, 4 and 5 cell stages (t2, t3, t4 and t5, respectively) were annotated. Timing of the kinetic events was calculated from the time of the fertilization (t0). (for TL parameter definitions see supplement) There were no adjustments of the kinetic events depending on the method of fertilization. Blastocyst morphology was assessed using the Gardner score. Fragmentation was assessed when it reached the highest grade in the sequence of the TL images. Cytoplasmic abnormalities (vacuoles) were also recorded. The TL reference ranges were defined and fixed before the start of the recruitment, Embryos in the control group were selected based on actual morphology. The primary end point was pregnancy rate (PR; rise in β human chorionic gonadotropin [hCG]) per patient based on intention to treat. Further endpoints were ongoing pregnancy rate (OPR; pregnancy that progresses beyond 12 weeks gestation), pregnancy loss (loss of pregnancy after an initial rise in βhCG up to 12 weeks gestation), live birth rate (LBR), gestational age (GA) at delivery and birth weight of the offspring.

When calculating sample size we assumed a 13% increase in pregnancy rate from an expected 42% to 55%. We planned to run the study with 1:1 ratio between the groups and do one interim monitoring with 0.1 information fraction. Using two-sided test and Pocock boundaries for estimation, 282 patients in each group was calculated (power of 80%, p of 5%). Assuming a 10% dropout rate we needed to enroll 620 patients.

Based on the results of the interim monitoring (61 cases in two arms) the pregnancy rate in the TL group was 15% higher than in controls (58.3% vs. 44.0%). Assuming 58% pregnancy rate in the TL group and 44% in controls, a sample size of 210 per group, was needed to achieve a power of 80%, at a significance level of 5%. Assuming 10% dropout rate a total sample of size of 462 patients were needed.

The groups were compared based on maximum likelihood estimation principles. We hypothesized and then showed that the patients in the treatment arms were similar at baseline, on average, with respect to variables that might influence outcome. Chi-square tests, likelihood-ratio tests and exact tests were used for the analysis of categorical variables. Continuous variables were compared using independent group t-test and. analysis of variance. Normality and homogeneity of variances were examined.

At the interim analysis of the trial the primary endpoint (pregnancy rate) between TL and control group was compared with Fisher's exact test. After closing the trial the primary endpoint (pregnancy rate) among the groups was compared with Fisher's exact test. Bonferroni correction was used for the p-value at the end of the trial. All other p-values were considered exploratory in nature. During the analyses the extent and distribution of missing data were also examined. No imputations were made for missing data.