Analyzing the statistical pitfalls of controlling for later weight and height in the association between small-for-gestational-age and later obesity in a cohort of early preterm infants
Introduction: Numerous studies indicated that infants born small-for-gestational age (SGA) have higher risks of adult chronic diseases including obesity and hypertension. However, recent publications have suggested that the association between SGA and later chronic diseases may be due to overcontrolling for later body size measures. This study aims to analyze the effect of controlling for weight and height in the association between SGA and overweight and obesity at 3 years corrected age (CA) in a cohort of early preterm infants (< 32 weeks gestational age)
Methods: Data were obtained from the Preterm Infant Multicentre Growth Study (2001-2014). The association between SGA and overweight and obesity was analyzed using multiple logistic regression models: 1) crude 2) controlling for baseline covariates including maternal and paternal age, education and maternal smoking 3) controlling for baseline covariates with additional adjustments for weight and height z-score at 21 months CA, separately. Marginal Structural Models were used with stabilized inverse probability weights to estimate the direct effect of SGA on overweight and obesity at 3 years CA, with weight as an intermediate variable
Results: The crude model and that only controlling for baseline covariates yielded no association between SGA and later overweight and obesity (OR 0.88 p-value 0.8, OR 0.94 p-value 0.9, respectively). Controlling for later height reversed the direction of the effect (OR 2.15 p-value 0.3) and controlling for weight reversed the direction of the effect and provided a strong significant association (OR 6.6 p-value 0.03). The marginal structural model provided a similar effect as the statistical approach controlling for only baseline covariates (OR 0.71 p-value 0.6)
Conclusion: Controlling for later body size measures, including weight and height, can attenuate and reverse the effect of the association between SGA and later overweight and obesity thus providing a biased estimate.