Antenatal steroids and risk of bronchopulmonary dysplasia : a lack of effect or a case of over-adjustment?

Although antenatal steroids reduce risk factors for bronchopulmonary dysplasia (BPD) in preterm infants, their effect on BPD is conflicting. We hypothesised that the lack of protective effect found in some studies could derive from over-adjustment during analysis, caused by controlling for factors intermediate in the causal pathway between treatment and outcome. We prospectively studied a cohort of infants 23-32 weeks gestation <1500 g, admitted to 10 tertiary-level neonatal units in Lombardy (Italy) in 1999-2002; 1118 neonates out of 1314 survived to 36 weeks; 15.9% developed BPD (oxygen requirement at 36 weeks); 82% were treated with steroids. In univariable analysis, steroids were not significantly protective against BPD; some intermediate factors (mechanical ventilation, greater severity of illness as measured by Clinical Risk Index for Babies score, patent ductus arteriosus) were significantly positively associated with (i.e. were risk factors for) BPD (OR = 11.0, 1.55, 4.42, respectively, all P < 0.001), and negatively associated with (i.e. prevented by) steroids (OR = 0.58, 0.92, and 0.58, respectively, all P < 0.01). In multiple logistic regression models using propensity scores, without the above-mentioned intermediate risk factors, steroid-treated infants had a lower risk of BPD (OR 0.59 [95% CI 0.36, 0.97], P = 0.036); male sex (OR = 2.08), late-onset sepsis (OR = 4.26), and birthweight (OR = 0.63 for 100 g increase) were also associated with BPD, all P < 0.001. When intermediate risk factors for BPD were also added to the model, the effect of steroids disappeared; ventilation (OR = 3.03), increased illness severity (OR = 1.11), and patent ductus arteriosus (OR = 1.90) were significant risk factors. This study suggests that including variables that are potential mediators in the causal chain can obscure the ability to detect a protective effect of treatment. We observed such a phenomenon in our analyses of the relationship between antenatal steroids and BPD, suggesting that steroid effect is partly mediated through a reduction in the classical risk factors.