Statistical analyses

Sample size tables for logistic regression³¹ were used to determine the required sample number. The outcome prevalence was estimated to be 10%. For simple logistic regression with α = 5%, a sample size of 2236 would result in a power of 80% to detect an OR 1.20 at one standard deviation above the mean of the exposure. Normal distribution of VF, SCF, and early pregnancy BMI measures were assessed in the cohort. Pearson’s correlation coefficients were calculated to evaluate the relations between VF, SCF, early pregnancy BMI, and maternal age. T-tests were used to compare VF, SCF, and early pregnancy BMI between groups defined by clinical and demographic parameters. Chi square tests were used to assess relations between categorical covariates and outcomes. When few observations were expected (< 5), Fisher’s exact test was performed.

Simple and multiple logistic regression analyses were performed to separately examine the association between VF (in 5 mm intervals), SCF (in 5 mm intervals), and BMI (kg/m²) and the likelihood of neonatal hypoglycemia. Subsequently, simple and multiple logistic regression models were performed to evaluate the likelihood of the composite outcome, and the components of the composite outcome (Apgar < 7 at 5 min of age, umbilical artery pH ≤ 7.0, and admission to NICU). In the models with VF and SCF as exposures, adjustments were made for maternal age, early pregnancy BMI, smoking status at first antenatal visit, maternal country of birth, and parity. In the model with early pregnancy BMI as exposure, the same adjustments were made except for BMI. Directed acyclic graphs were used to select covariates. Variables included in the directed acyclic graphs were either known to be associated with the exposures and outcomes, or considered clinical relevant.

We imputated information on smoking status at first antenatal visit, since this information was missing in 51% of the women. The imputation was performed by the random hot deck method³². By using information on maternal age, BMI, and country of birth, controls for women with known smoking status were matched among women with unknown smoking status. Thereafter, a random control was drawn for every woman with known smoking status, and given the same smoking status as its match. After the imputation, the prevalence of smoking at the first antenatal visit in the cohort was 3.6%.

Data were missing on Apgar score and umbilical artery pH on some of the infants (Supplementary Table ¹). We analyzed the cases that had available data, hence, the number of mother–child dyads differed between the different outcomes studied. We did not perform any data imputation besides that on smoking status at first antenatal visit.

IBM SPSS Statistics version 27 was used for all statistical analyses. Statistical significance was considered to be indicated by a nominal two-side P-value < 0.05.