This retrospective cohort study included patients who underwent IVF or intracytoplasmic sperm injection (ICSI) treatment at the Center for Reproductive Medicine, Nanfang Hospital, Southern Medical University, China, between January 2015 and April 2024. This study was reviewed and approved by the Institutional Review Board of Nanfang Hospital (Ethical approval number: NFEC-2025-452) and conducted in accordance with the Declaration of Helsinki. Informed consent was waived because all data used here were fully anonymized.

We analyzed two cohorts. Cohort 1 (Blastulation analyses) included all embryos cultured to the blastocyst stage with complete parameter records. After excluding embryos with missing annotations or day-3 counts outside 4–16-cell or day-2 counts outside 2–6-cell, 64, 853 embryos from 12, 474 IVF/ICSI cycles were included in the analysis. Cohort 2 (Blastocyst transfer analyses) included all single day-5 or day-6 blastocyst transfers in frozen-thawed embryo transfer (FET) cycles. After the same exclusions, 2,669 embryos were included in the analysis. These were further stratified by maternal age at oocyte pick-up and blastocyst quality: (1) young maternal age (<35 years), top-quality blastocyst transfer (n = 1,098); (2) young maternal age (<35 years), non-top-quality blastocyst transfer (n = 830); (3) advanced maternal age (≥35 years), top-quality blastocyst transfer (n = 327); (4) advanced maternal age (≥35 years), non-top-quality blastocyst transfer (n = 414). Day-3 cell number were categorized into four groups: 4–7-cell, 8-cell(reference), 9–10-cell, and 11–16-cell.

Ovarian stimulation was performed using standard GnRH-agonist or GnRH-antagonist protocols. A subset of patients received alternative stimulation approaches, including minimal stimulation, progestin-primed ovarian stimulation and luteal phase ovarian stimulation. Final oocyte maturation was triggered when at least one follicle reached 18 mm, using 2000–10000 IU hCG (Livzen, China) or 250 μg Ovidrel (Merck-Serono, Switzerland), with or without 0.2 mg Triptorelin (Decapeptyl, Ferring, Switzerland). Oocyte retrieval was performed 34–36 hours later under transvaginal ultrasound guidance. Fertilization was achieved by conventional IVF or ICSI; in severe male-factor cases, sperm retrieval was performed via percutaneous epididymal sperm aspiration (PESA) or testicular sperm aspiration (TESA). After insemination, embryos were cultured in pre-equilibrated cleavage medium under mineral oil at 37 °C, 6% CO₂, and 5% O₂. At 68 ± 2 hours post insemination (day-3), blastomere count, fragmentation, and symmetry were assessed according to Istanbul consensus guidelines (27). Selected embryos were cultured further in blastocyst medium to day-5 or day-6 and were then graded according to Gardner criteria (expansion graded 1-6; the inner cell mass and trophectoderm graded A-C).

Endometrial preparation was conducted following three protocols: 1) Natural cycle: ultrasonographic monitoring was initiated on menstrual days 10–12. When the leading follicle reached 14 mm, urinary LH was assessed three times daily using test strips. Upon observation of a qualitative color change (from weak to positive), serum LH testing was performed via electrochemiluminescence immunoassay. After detection of the LH surge (by serum testing) and an endometrial thickness ≥8 mm, luteal support was initiated with either vaginal or oral progesterone on the following day (designated as progesterone day-0). Embryo transfer was performed on progesterone day-5. 2) Hormone replacement therapy (HRT) cycle: oral estradiol valerate was initiated on menstrual days 2–4 or after full pituitary suppression was achieved with a GnRH agonist, either as a fixed dose (6 mg/day) or in a step-up regimen (4→6→8 mg/day) for 10–12 days. If endometrial thickness did not reach ≥8 mm, the dose was increased to 10 mg/day until this threshold was met. Then luteal support was initiated (designated as progesterone day-0). Embryo transfer was performed on progesterone day-5. 3) Ovulation induction (OI): Letrozole (2.5–5.0 mg/day for 5 days) was initiated on menstrual cycle days 2–5, either alone or supplemented with human menopausal gonadotropin (HMG; 37.5–75 IU/day). Luteal support was started on the day after the detection of the LH surge (designated progesterone day 0). Embryo transfers were performed on progesterone day 5.

For the blastulation analyses (cohort 1), primary predictors comprised day-3 blastomere counts (4–16). Baseline and other embryo characteristics were collected, including female age, infertility type (primary or secondary), stimulation protocol (agonist, antagonist, or unconventional), the number of oocytes retrieved, fertilization method (IVF or ICSI), pronuclear (PN) pattern (2PN or non-2PN), day-2 cell number, day-3 fragmentation rate (<10% or ≥10%), symmetry (symmetric or asymmetric), day-3 observation time recorded as hours post insemination (hpi), and patient identifier. Two primary blastulation outcomes were evaluated: usable blastocyst (blastulation) was defined as ≥3BC on day-5 or day-6; top-quality blastocyst was defined as ≥4BB on day-5.

For the blastocyst transfer analysis (cohort 2), primary predictors comprised day-3 blastomere counts (4–7, 8, 9–10 or 11–16). Potential confounders included blastocyst morphology (expansion: 3–6; inner cell mass: A/B; trophectoderm: A/B/C), day of blastocyst formation (5 or 6), and cycle factors (female age, endometrial thickness on transfer day and endometrial preparation protocol [natural cycle, hormone replacement, hormone replacement + gonadotropin-releasing hormone agonist, or ovulation induction]). Three primary pregnancy outcomes were evaluated: the clinical pregnancy rate (CPR) was defined as the number of transfer cycles resulting in detection of both a gestational sac and fetal heartbeat at 4 weeks post–blastocyst transfer, divided by the total number of embryo transfer cycles. The miscarriage rate was calculated as the number of miscarriages divided by the number of clinical pregnancies. The live birth rate (LBR) was calculated as the number of live births divided by the total number of embryo transfer cycles.

Continuous variables are presented as mean ± standard deviation (SD), while categorical variables are presented as percentages. Differences in baseline characteristics (including potential confounders) and outcomes between the cell number groups were assessed using t-tests for continuous variables and χ² tests for categorical variables. Statistical significance for these unadjusted group comparisons was defined as a two-tailed p-value < 0.05.

Multivariate logistic regression analyses were conducted to assess the independent association between day−3 cell number and each outcome (two blastulation and three pregnancy outcomes). Confounders for each model were selected via a best−subset approach by comparing all possible variable combinations and retaining the model with the minimum Akaike Information Criterion (AIC) value. For blastulation outcomes (involving multiple embryos per cycle), generalized linear mixed−effects models (GLMMs) were employed, incorporating random intercepts for patient identifier and day-3 observation time. For blastocyst transfer outcomes, generalized estimating equations (GEE) logistic regression was used. All models were adjusted for their corresponding confounders, and results are reported as adjusted odds ratios (aORs) with 95% confidence intervals (CIs).

A total of 64,853 day-3 embryos from Cohort 1 underwent extended culture to day-5/6. Among these, 21,749 (33.5%) developed into usable blastocysts, and 7,221 (11.1%) were top-quality blastocysts (≥4BB on day 5). Baseline cycle characteristics and other embryo parameters are detailed in Table 1. Both blastulation rates and top-quality blastocyst rates, as well as their corresponding aORs across day-3 cell number groups, are presented in Figure 1 and Table 2.

Compared to 8-cell embryos (rate: 56.3%), usable blastocyst rates were significantly lower in the 4-, 5-, 6-, 7-, 9- and 10-cell groups (rates: 11.7–46.2%; aORs: 0.11–0.84, all 95% CIs excluded 1), higher in the 13-, 14-, 15- and 16-cell groups (rates: 62.4–78.8%; aORs: 1.42–3.19, all 95% CIs excluded 1), and comparable in the 11- and 12-cell groups (rates: 49.8% and 56.3%; aORs: 0.96 [0.85–1.09] and 1.08 [0.93–1.24]). A similar pattern was observed for top-quality blastocysts, but the positive shift was from an earlier cell-number threshold: relative to the 8-cell (rate: 24.0%), the top-quality blastocyst rate was significantly lower in the 4-, 5-, 6-, 7- and 9-cell groups group (rates: 0.4%–15.0%; aORs: 0.02–0.86, all 95% CIs excluded 1), but higher from the 10-cell group onward (rates:18.6%–60.6%; aORs:1.13–6.97, all 95% CIs excluded 1).

Subgroup analyses stratified by maternal age (<35 or ≥35 years) generally mirrored the overall cohort trends (Figure 1 and STable 1). Exceptions were noted in the <35 years subgroup, where 13-cell embryos exhibited comparable usable rates to 8-cell controls (aOR: 1.30 [0.96–1.75]). In the ≥35 years subgroup, 9- and 10-cell embryos had a comparable top-quality rate to 8-cell controls (aORs: 0.87 [0.73–1.04] and 1.01 [0.81–1.25]).

Cohort 2 included 2,669 single frozen-thawed blastocyst transfer cycles. The overall CPR was 55.3%, LBR was 42.1%, and miscarriage rate was 24.0%. The mean maternal age at oocyte retrieval was 32 years and 33 years at embryo transfer. The sample sizes and number of top-quality blastocysts per day-3 cell number group were as follows: 4–7-cell (n = 829; 236), 8-cell (n = 866; 573), 9–10-cell (n = 669; 389), and 11–16-cell (n = 305; 227). The proportion of top-quality blastocysts was significantly higher in the 11–16-cell group than in the 8-cell group (74.4% vs. 66.2%, P = 0.008), and significantly lower in the 4–7, and 9–10-cell groups (28.5%, 58.1% vs. 66.2%, all P < 0.001). Baseline characteristics differed significantly among day-3 cell number groups (Table 3) and were adjusted as covariates in subsequent logistic regression analyses.

In the overall cohort2, neither the 9–10-cell nor 11–16-cell groups differed significantly from the 8-cell group in CPR (57.1% and 63.3% vs. 57.9%, P = 0.808 and 0.112, aOR = 1.05[0.85-1.30] and 1.24 [0.94-1.64]) or LBR (43.6% and 46.6% vs. 44.8%, P = 0.689 and 0.644, aOR = 1.03 [0.83-1.27] and 1.09 [0.83-1.44]). The 4–7-cell group showed significantly lower CPR (48.5% vs. 57.9%, P < 0.001) and LBR (36.3% vs. 44.8%, P < 0.001); however, these differences were not significant after adjustment to confounders (see Table 4).

Subgroup analyses revealed that blastocyst quality and maternal age significantly modified the association between day-3 cell number and pregnancy outcomes (Figure 2 and Table 4). Among women <35 years receiving top-quality blastocysts, 11–16-cell embryos were associated with significantly higher CPR (76.5% vs. 63.0%; P = 0.002; aOR = 1.95 [1.30–2.96]) and LBR (61.2% vs. 51.2%; P = 0.034; aOR = 1.58 [1.10–2.30]) compared to 8-cell embryos, with comparable miscarriage rates. Both 9–10-cell and 4–7-cell embryos showed outcomes similar to 8-cell embryos across all three endpoints.

Conversely, among women ≥35 years receiving non-top-quality blastocysts, both 11–16-cell and 4–7-cell embryos demonstrated significantly lower CPR (26.5% and 36.7% vs. 51.0%; P = 0.023 and 0.027; aOR = 0.40 [0.15–0.97] and 0.58 [0.34–0.99]) and LBR (14.7% and 23.5% vs. 38.5%; P = 0.019 and 0.011; aOR = 0.32 [0.10–0.89] and 0.50 [0.28–0.87]) versus 8-cell embryos, with no difference in miscarriage rates. The 9–10-cell group showed comparable CPR and miscarriage rates but a significantly reduced LBR (23.9% vs. 38.5%; P = 0.047; aOR = 0.51 [0.26–0.98]).

In the remaining two subgroups (women <35 years with non-top-quality blastocysts and ≥35 years with top-quality blastocysts), the link between day-3 cell number and blastocyst transfer outcomes was minimal. Although younger women with non-top-quality blastocysts showed a reduction in CPR with 11–16-cell embryos (34.1% vs 54.3%; P = 0.024; aOR = 0.49 [0.24–0.98]), this did not translate into a significant difference in LBR or miscarriage outcomes. All other comparisons in these subgroups showed no difference in outcomes relative to the 8-cell reference.