妊娠週数別の三胎死産：2つの三次医療センターからのレトロスペクティブコホート

Objective To describe a contemporary cohort of triplet pregnancies from two tertiary centers in a middle-income country (LMIC) and evaluate the association between gestational age (GA) at birth and perinatal death; we developed and internally validated a GA-based risk model. Study Design Retrospective cohort from two tertiary hospitals (2015-2024). The unit of analysis was the fetus/neonate, with cluster-robust standard errors at the mother level. Primary outcome: perinatal death (stillbirth ≥23+0 weeks or neonatal death ≤28 days). GA was modeled with logistic regression using restricted cubic splines; Firth penalization addressed separation where applicable. Internal validation used 200 bootstrap resamples. Results The cohort included 150 neonates; perinatal death occurred in 23 (15.4%). Median GA was 33 [31-34] weeks overall, 34.0 [32.0-34.0] in survivors, and 25.0 [24.0-28.0] in perinatal deaths. In the adjusted spline model, GA was the dominant predictor (overall Wald χ² 1,473.66, df=3, p<0.001; nonlinearity χ² 424.92, df=2, p<0.001), while severe preeclampsia was not significant (χ² 0.10, p=0.750). The category-based Firth model showed markedly elevated odds of perinatal death at earlier gestations versus ≥34 weeks: <28 weeks aOR 871.15 (95% CI 81.25-124,006.80, p<0.001); 28-<32 weeks aOR 50.22 (5.45-6,682.53, p<0.001); 32-<34 weeks aOR 5.34 (0.26-804.14, p=0.278); severe preeclampsia aOR 1.02 (0.09-7.17, p=0.984). The internally validated model demonstrated excellent discrimination (optimism-corrected AUROC, 0.972) and good overall performance (Brier score, 0.047), with a calibration intercept of -0.129, a slope of 0.696, and an Emax of 0.104. Conclusion In triplet pregnancies, GA at birth is the dominant determinant of perinatal death, with a steep risk gradient at earlier gestations. A GA-based model demonstrates excellent discrimination and acceptable calibration following bootstrap internal validation, supporting its use in informing counseling and timing-of-delivery decisions in this high-risk population.