Optimizing GWG in Obese Pregnant Women: A Nationwide Real-World Study Using the U.S. Natality Cohort

Detailed Description This is a population-based retrospective cohort study that leverages the U.S. National Vital Statistics System (NVSS) natality files for the years 2023 through 2025. The NVSS is a comprehensive, publicly available database compiled by the National Center for Health Statistics (NCHS) at the U.S. Centers for Disease Control and Prevention (CDC), capturing nearly all births occurring in the United States. The dataset contains de-identified maternal demographic characteristics, medical history, prenatal care utilization, labor and delivery procedures, and infant health outcomes.

Study Cohort The target population consists of nulliparous and multiparous individuals aged 18 to 45 years with singleton, live-born, term pregnancies (37 to 41 completed weeks of gestation) and pre-pregnancy obesity, defined as a Body Mass Index (BMI) of 30.0 kg/m² or higher. To minimize confounding by late or absent prenatal care, only individuals who initiated prenatal care during the first trimester (months 1 through 3 of pregnancy) are included. After applying all inclusion criteria and excluding records with missing or implausible values for key variables (gestational weight gain below -10 kg or above 50 kg, unknown delivery mode, or documented prior cesarean section), the analytic cohort is anticipated to include approximately 3 million birth records.

The cohort is stratified into three World Health Organization (WHO) obesity classes for all primary analyses: Class I (BMI 30.0-34.9 kg/m²), Class II (BMI 35.0-39.9 kg/m²), and Class III (BMI ≥ 40.0 kg/m²). This stratification allows for the assessment of effect measure modification by obesity severity and the derivation of class-specific optimal gestational weight gain targets.

Exposure and Outcome Measurements The primary exposure is total gestational weight gain (GWG), calculated as the exact arithmetic difference between maternal weight at delivery and pre-pregnancy weight as recorded on the birth certificate. Continuous GWG is categorized into eight prespecified clinical intervals: (1) < 0 kg (weight loss), (2) 0 to 2.9 kg, (3) 3.0 to 4.9 kg, (4) 5.0 to 9.0 kg (the 2009 Institute of Medicine recommended range, inclusive of both endpoints, serving as the reference category), (5) 9.1 to 10.9 kg, (6) 11.0 to 12.9 kg, (7) 13.0 to 15.9 kg, and (8) ≥ 16.0 kg.

The primary endpoints are two comprehensive composite indicators. The maternal composite (M1_equal) captures any of the following: intensive care unit (ICU) admission, blood transfusion, eclampsia (Tier 1, severe); or gestational hypertension, gestational diabetes mellitus, primary cesarean delivery, or severe perineal laceration in the absence of Tier 1 events (Tier 2, moderate). The neonatal composite (M2_equal) captures: neonatal ICU admission, prolonged assisted ventilation (> 6 hours), or neonatal seizures (Tier 1, severe); or small for gestational age (birth weight < 10th percentile using the 2017 U.S. reference), large for gestational age (> 90th percentile), immediate assisted ventilation, or 5-minute Apgar score < 7 in the absence of Tier 1 events (Tier 2, moderate). These gradings align with CDC Severe Maternal Morbidity (SMM) indices, the WHO Maternal Near-Miss framework, and the WHO Neonatal Near-Miss criteria, respectively, ensuring clinical relevance and comparability across studies.

Statistical Analysis Plan All analyses are conducted independently for each obesity class to preserve stratified inference.

Multivariable Regression. Multivariable logistic regression models estimate the independent associations between the 8-level GWG categorical variable and each composite binary outcome (M1_equal and M2_equal), adjusting for a comprehensive set of confounders: maternal age (continuous), race/ethnicity (categorical), education level, marital status, primary payment source for delivery, participation in the Women, Infants, and Children (WIC) program, parity, pre-pregnancy diabetes, chronic hypertension, history of infertility treatment, and smoking during pregnancy. For the ordinal, 3-tier graded outcomes (M1_grade and M2_grade), ordinal logistic regression is the primary approach, provided the proportional odds assumption is satisfied (assessed by the Brant test at α = 0.05). If violated, estimation downgrades to multinomial logistic regression to calculate tier-specific adjusted odds ratios.

Non-linear Dose-Response Modeling. Restricted cubic spline (RCS) functions with five knots fixed at the 5th, 27.5th, 50th, 72.5th, and 95th percentiles of the GWG distribution are incorporated into logistic models to characterize the continuous dose-response relationship without imposing a linear shape. From the resulting spline curves, the nadir (the GWG value corresponding to the lowest adjusted odds of the composite outcome) is identified. A "safe threshold" is then defined as the GWG interval where the relative increase in risk remains ≤ 10% above the nadir (i.e., adjusted odds ratio ≤ 1.10). The final recommended optimal GWG range for each obesity class is determined by the mathematical intersection of the safe thresholds derived independently for M1_equal and M2_equal. To quantify uncertainty, 1,000 bootstrap resamples are used to construct 95% confidence intervals around the nadirs and the lower and upper bounds of the recommended ranges.

Hierarchical Gatekeeping for Multiplicity Control. To control the family-wise error rate across multiple comparisons, a prespecified three-layer hierarchical gatekeeping strategy is implemented. Layer 1 tests the global GWG × obesity class interaction via an omnibus likelihood ratio test (LRT) at α = 0.05; a significant interaction justifies proceeding with stratified analyses. Layer 2 applies a fixed-sequence testing procedure within each obesity class: binary M1_equal (Logistic LRT) → binary M2_equal (Logistic LRT) → graded M1_grade (Ordinal Logistic LRT) → graded M2_grade (Ordinal Logistic LRT). Each subsequent test is only conducted if the preceding test is statistically significant at α = 0.05. Layer 3 applies a Bonferroni correction (α ≈ 0.0167) across the three obesity classes for the sequence-level tests, ensuring strict control of Type I error. For secondary individual outcome components, P-values are adjusted using the Benjamini-Hochberg false discovery rate (FDR) procedure, with a q-value threshold of < 0.05.

Handling of Missing Data. In the primary analysis, missing covariate data are retained as an independent analytical category ("Unknown"). This approach avoids listwise deletion, preserves statistical power, and mitigates selection bias inherent to complete-case analyses in large real-world datasets. In sensitivity analyses, multiple imputation by chained equations (MICE) with 20 imputations is applied exclusively to covariates with missingness rates exceeding 10% (e.g., marital status). Complete-case analysis is also performed as an additional sensitivity check to assess the robustness of findings to missing data assumptions.

Model Diagnostics and Sensitivity Analyses. Model fit is assessed using the Hosmer-Lemeshow test (10 deciles), the area under the receiver operating characteristic curve (AUC), and the Brier score. Multicollinearity is evaluated via variance inflation factors, and influential observations are identified using Cook's distance and DFBETAs. Four prespecified sensitivity analyses are conducted: (1) a pure obesity cohort restricted to women without pre-pregnancy diabetes or chronic hypertension, (2) complete-case analysis and MICE-based imputation for missing covariates, (3) E-value calculations to quantify robustness to unmeasured confounding, and (4) expansion of the cohort to include individuals with prior cesarean sections (adjusted for in the model) to confirm that the primary exclusion criterion did not introduce critical selection bias.

Subgroup Analyses. Effect modification is explored across the following prespecified subgroups: race/ethnicity, pre-pregnancy diabetes, chronic hypertension, maternal age (< 35 years versus ≥ 35 years), and parity.

Data Integrity and Reporting This study adheres to the STROBE (Strengthening the Reporting of Observational Studies in Epidemiology) and RECORD-PE (Reporting of studies Conducted using Observational Routinely-collected health Data for Pharmacoepidemiology) reporting guidelines. All analyses will be conducted using R software (version 4.3 or higher), with the rms package for restricted cubic spline modeling. The analytic code will be made available upon reasonable request to the corresponding author. As the dataset is fully de-identified and publicly available, this study is exempt from institutional review board (IRB) approval. No patient consent is required, as the research involves no direct contact with participants or collection of private, identifiable information.