Abstract
<title>Abstract</title> <p> Background We sought to determine whether endometrial preparation protocols influence early pregnancy loss risk after ART via a corpus luteum (CL)-dependent mechanism, and to evaluate CL status as an independent predictor in a multivariable risk model. Methods Single-center retrospective cohort study. We screened 6485 IVF/ICSI or FET cycles achieving clinical pregnancy at our center (2012–2026). After excluding ongoing pregnancies and losses to follow-up, 5872 cycles with known outcomes (live birth or spontaneous abortion) were analyzed, with 98 ectopic pregnancies excluded separately. Cycles were classified by endogenous CL presence: CL-present (fresh embryo transfer + FET with natural cycle) versus CL-absent (FET-HRT + FET-GnRHa + HRT). The training set comprised 2012–2022 data (n = 3589); the independent temporal validation set comprised 2022–2026 data (n = 2283). Results After excluding ectopic pregnancies, loss rates showed a clear CL-dependent gradient. The two CL-present groups had the lowest and comparable rates (fresh ET 15.0%, FET-NC 13.9%, <italic>P</italic> > 0.8). The two CL-absent FET groups were significantly higher (FET-GnRHa + HRT 18.9%, FET-HRT 19.9%; four-group χ² <italic>P</italic> = 0.0001). The pooled CL+ group (15.0%) and CL− group (19.5%) differed by 4.5 percentage points ( <italic>P</italic> < 0.00001). After multivariable adjustment, CL absence remained an independent risk factor (aOR = 1.121, 95% CI 1.029–1.223, <italic>P</italic> = 0.009). The independent cryopreservation effect could not be fully isolated because the FET-NC subgroup was underpowered (n = 144). The 7-variable model achieved an AUC of 0.6463 in independent temporal validation, significantly outperforming age alone (AUC = 0.6011, ΔAUC = 0.0452, DeLong <italic>P</italic> = 0.0001), with a risk gradient from 3.3% to 37.8% (11-fold). Conclusions Fresh ET and FET-NC (both CL-present) had the lowest and comparable loss rates. CL absence was an independent risk factor (aOR = 1.121). The independent cryopreservation effect could not be fully isolated because the FET-NC subgroup was underpowered. The 7-variable model effectively discriminated four risk tiers and significantly outperformed age-alone assessment. </p>