“…Potential confounders were selected based on their significance in the univariable analysis ( P < 0.1) and because they have been associated with increased risk for PTB. The potential confounders included in the multivariable model were maternal age (continuous), prenatal care initiation (during first 5 months, 6 months or later/no initiation/unknown), maternal education (some high school or less, high school graduate, some college, college graduate or more), race/ethnicity (Non‐Hispanic White, Non‐Hispanic Black, Asian, Hispanic, Other), parity (nulliparous, multiparous), smoking during pregnancy (yes, no), presence of eating disorder (yes/no), anaemia (yes/no), pre‐existing diabetes (yes/no), pre‐existing hypertension (yes/no), gestational diabetes (yes/no), gestational hypertension (yes/no), placental abruption (yes/no), and maternal height (continuous) to reduce further potential residual confounding associated with the BMI algorithm . Because risk‐adjusted PTB could result from medical intervention, we compared the relation between severity of underweight and PTB in births that were spontaneous and in births that resulted from medical intervention.…”