This study used the publicly available National Vital Statistics System (NVSS) dataset (2019.01.01–2023.12.31) from the United States National Center for Health Statistics (NHCS), which contains detailed information on all births registered in the U.S. (23). The dataset is de-identified and publicly available to the public, and NCHS has assumed responsibility for obtaining ethical approval for data collection and publication. Therefore, no further ethical approval was required for this study. All analyses conform to the Statement for Reporting of Observational Studies in Epidemiology (STROBE) and follow the Vital Statistics User Agreement.

The birth dataset from 2019 to 2023 (n = 18,328,446) was used in this study, and the following exclusion criteria were set: (i) women who conceived without ART (n = 18,045,994). ART includes in vitro fertilization (IVF) and gamete intrafallopian transfer (GIFT); (ii) multiple pregnancies (n = 42,015); (iii) non-live births (n = 1,173); (iv) age at delivery less than 18 years old (n = 8); (v) pre-pregnancy BMI information was missing (n = 3,439); (vi) gestational age <24 weeks or ≥42 weeks or missing (n = 758); (vii) missing data (n = 37,822).

The exposure variable in this study was pre-pregnancy BMI. Mother's pre-pregnancy BMI (kg/m²) is calculated as: (mother's pre-pregnancy weight (lb)/[mother's height (in)]²) × 703. Mother's height and pre-pregnancy weight information was collected through Mother's Worksheet. Pre-pregnancy BMI was used in continuous form in nonlinear relationship analyses. For descriptive and logistic regression analyses, pre-pregnancy BMI was categorized into 6 groups according to World Health Organization (WHO) criteria (24–26), including underweight (<18.50 kg/m²); normal weight (18.5–24.9 kg/m²); overweight (25–29.9 kg/m²) and obesity (Class I: 30–34.9 kg/m², Class II: 35–39.9 kg/m² and Class III: ≥ 40 kg/m²), of which 18.5–24.9 kg/m² was the reference group.

The primary outcome was PTB (<37 weeks), and the secondary outcomes were the categories of PTB. The World Health Organization (WHO) defines preterm birth as delivery occurring before 37 weeks of gestation. PTB is divided into 3 subtypes according to the number of weeks of gestation: extremely preterm (<28 weeks), very preterm (28 to <32 weeks), and moderate to late preterm (32 to <37 weeks) (1). Gestational weeks data integrated obstetric estimates (OE) using data from ultrasound, last menstrual period (LMP), and physical examination. The specific methods of assessment were described in detail in our previous study (27). Researchers have shown that the method of estimating PTB rates based on OE has been shown to have excellent specificity and negative and positive predictive value (28).

Collection of information on relevant potential confounders: age of the pregnancy (year) (age at delivery minus gestational age), race (White, Black, Asian, or other), marital status (married or unmarried), education level (high school and below or higher than high school), smoking before pregnancy (yes or no), insurance (Medicaid, private, self-pay, or other), parity before the current pregnancy (0, or ≥ 1), obstetrical history (previous termination or fetal loss, neonatal death, PTB, or cesarean section), and pre-pregnancy comorbidities (pre-pregnancy diabetes or hypertension).

Except for marital status (14.72%), the incidence of missing data for other covariates was low (<4%; Supplementary Table 1). This was mainly due to California's systematic cessation of providing information on mothers' marital status since 2017 due to state law restrictions (29–31). The absence of marital status information is directly related to the privacy of marital status records themselves (such as “married/unmarried”). Little's missing completely at random (MCAR) test rejected the assumption of MCAR (P < 0.001, Supplementary Table 2), suggesting that data missingness may be either missing at random (MAR) or missing not at random (MNAR). Based on background information and relevance analysis (Supplementary material 2), the missingness may depend on observed variables (such as maternal age, race, and education level) and marital status itself (i.e., marital status information involves privacy concerns, and California law prohibits its reporting). Given the complexity and uncertainty surrounding missing data mechanisms, we selected complete case analysis (CCA) as a conservative approach to minimize potential bias. CCA has been demonstrated to yield valid results when missing data are MAR or close to MNAR, particularly when the proportion of missing data is relatively low (32, 33).

Continuous variables are presented as mean ± standard deviation (SD), and categorical variables are presented as frequencies or percentages. For analysis of baseline characteristics, statistical differences between pre-pregnancy BMI groups were tested using t-tests or one-way analysis of variance (ANOVA) for continuous variables and chi-squared or Fisher tests for categorical variables.

We examined the relationship between pre-pregnancy BMI and PTB in individuals who conceived after ART treatment using logistic regression models. The following covariates were used to adjust the multivariable model: age of the pregnancy, race, marital status, education level, smoking before pregnancy, insurance, parity before the current pregnancy, previous termination or fetal loss, previous neonatal death, previous PTB, previous cesarean section, pre-pregnancy diabetes, and pre-pregnancy hypertension to calculate adjusted odds ratios (ORs) and 95% confidence intervals (95% CIs). The selection criteria for the inclusion of covariates were determined by an integrated approach that considered univariate regression analysis (Supplementary Table 3), reports in the relevant literature, and a combination of Directed Acyclic Graphs (DAGs) guidance (Supplementary Figure 1).

To further explore the potential nonlinear dose-response relationship between pre-pregnancy BMI and PTB, we developed a restricted cubic spline (RCS) model with 4 nodes (5^th, 35^th, 65^th, and 95^th). The likelihood ratio test was used for non-linearity hypothesis testing (P for non-linearity). The bootstrap resampling method and likelihood ratio test were employed to determine the minimum point and its confidence interval. In this model, pre-pregnancy BMI was used as a continuous variable, and subjects within the mean ± 3SD range (n = 195,180) were included. Using the median as a reference point, the corresponding odds ratio and 95% confidence interval (CI) were calculated after adjusting for all covariates. The results were presented graphically to intuitively show the relationship between changes in BMI levels and the risk of PTB.

Subgroup analyses were performed based on with or without previous termination or fetal loss, and interactions were assessed using likelihood ratio tests.

Pre-pregnancy diabetes and pre-pregnancy hypertension may have causal mechanisms (mediating pathways) that influence PTB through BMI, rather than simply being confounding factors (34–37). These two conditions have dual attributes in the analysis: they may be confounding factors that need to be adjusted for, or they may be mediating variables in the association between BMI and PTB. Therefore, sensitivity analysis excluded not only women with a history of PTB but also participants with pre-pregnancy diabetes and pre-pregnancy hypertension in order to assess confounding effects and the robustness of the association. In addition, we adjusted the analysis range for pre-pregnancy BMI.

All analyses were performed using R Statistical Software (Version 4.2.2, http://www.R-project.org, The R Foundation) and the Free Statistics analysis platform (Version 2.1.1, Beijing, China). Statistical testing was two-sided. Statistical significance was defined as P < 0.05.

A total of 18,328,446 records were included in the NVSS 2019–2023 birth dataset. Based on the exclusion criteria, 197,237 cases were ultimately included in the analysis (Figure 1). The mean age of the mothers participating in the study was 35 ± 5 years; 79.68% were White; the majority (n = 89,603,45.43%) had a pre-pregnancy BMI of 18.5–24.9 kg/m², 3,643 (1.85%) were <18.50 kg/m², 53,248 (27.00%) were 25–29.9 kg/m², 28,686 (14.54%) were 30–34.9 kg/m², 14,195 (7.20%) were 35–39.9 kg/m², and 7,862 (3.99%) were ≥40 kg/m²; 22,592 (11.45%) were PTB, and the group of 18.5–24.9 kg/m² had the lowest rate of PTB (9.09%). Detailed baseline characteristics are shown in Table 1.

The association between pre-pregnancy BMI and PTB is shown in Table 2, including crude models without adjustment for covariates and adjusted models after adjustment for all covariates. Compared with the reference group, pre-pregnancy BMI <18.5 and ≥ 25 kg/m² were both associated with an increased risk of PTB. In the adjusted model OR (95% CI) for BMI <18.5 kg/m²: 1.14 (1.02–1.27); 25–29.9 kg/m²: 1.26 (1.21–1.30); 30–34.9 kg/m²: 1.50 (1.44–1.56); 35–39.9 kg/m²: 1.64 (1.55–1.72); ≥40 kg/m²: 1.75 (1.64–1.87). Subsequently, in the RCS model analysis adjusted for all covariates, participants with pre-pregnancy BMI within the mean ± 3SD range (n = 195,180) were included. A nonlinear relationship was observed between pre-pregnancy BMI and PTB (P for non-linearity: <0.001, Figure 2), with the lowest risk of PTB at approximately 21.6 (21.3, 21.9), followed by a gradual increase in risk, which then plateaued after reaching 35.7 (35.5, 35.9).

In the multinomial logistic regression model for different subgroups of PTB (Supplementary Table 4), similar results were observed in the moderate to late PTB group, while no statistically significant differences were found in the very PTB and extremely PTB groups when maternal BMI was <18.5 kg/m² before pregnancy [0.97 (0.66–1.42); 1.34 (0.82–2.18)]. The trend with increasing BMI appears to be slightly stronger for very PTB and extremely PTB than in the moderate to late PTB group.

An interaction was observed in the subgroup analysis based on with or without previous termination or fetal loss (Supplementary Table 5; P for interaction = 0.002). In multivariate-adjusted logistic regression, pre-pregnancy BMI <18.5 kg/m² was associated with an increased risk of PTB in the subgroup without previous termination or fetal loss [OR (95% CI): 1.18 (1.03–1.36), P = 0.015], while there was no statistically significant difference in PTB risk in the subgroup with previous termination or fetal loss [OR (95% CI): 1.06 (0.88–1.28), P = 0.549]. In the groups with pre-pregnancy BMI ≥ 25 kg/m², it was observed that the risk of PTB increased with increasing BMI. This was more evident in subgroups without previous termination or fetal loss in the 25.0–29.9 kg/m² and 30.0–34.9 kg/m² groups and more evident in subgroups with previous termination or fetal loss in the 35.0–39.9 kg/m² and ≥ 40.0 kg/m² groups. In the multivariate-adjusted RCS model (Supplementary Figure 2), the lowest point of PTB risk in the subgroup with previous termination or fetal loss occurred at approximately 22.6 (22.3, 22.8), while in the subgroup without previous termination or fetal loss, it occurred at approximately 21.2 (20.7, 21.6). The two CI groups did not overlap, further supporting the modifying effect of previous termination or fetal loss on the association between BMI and PTB.

The results of the sensitivity analysis (Figure 3, Table 3, Supplementary Table 6) were similar to the overall results, with no evidence of a change in the nonlinear association between pre-pregnancy BMI and PTB in women who conceived with ART treatment.