We thank Lorthe et al for the interest in our study and their excellent questions. First, our primary outcome included both early- and late-onset neonatal sepsis because the parent trial did not differentiate between these conditions. While the inclusion of both early- and late-onset neonatal sepsis may have made it more difficult to detect a significant difference in early-onset sepsis, the low incidence of the combined diagnosis was reassuring. Second, cases of stillbirth and neonatal deaths were excluded because perinatal mortality was not the primary outcome evaluated. While it is possible that stillbirth could be due to sepsis, other etiologies such as cord accidents and placental abruption may be more likely in the setting of regular fetal testing. There were 8 cases of stillbirth; all occurred in the preterm premature rupture of membranes (PPROM) latency <4 weeks group. We ran a linear regression analysis using latency time as a continuous variable and did not find an association between latency time and stillbirth. Neonatal mortality was not included because it is known to be closely inversely related to gestational age at delivery, which is highly associated with latency. If there was additive risk from sepsis from prolonged PPROM, addition of neonatal death would bias our model away from showing increased risk. There were 90 (6.5%) neonatal deaths in the PPROM <4 weeks group and 8 (3.9%) neonatal deaths in the prolonged latency group; this difference was not statistically significant ( P = .15). The unadjusted odds ratio for neonatal death in the prolonged latency group was 0.58 (95% confidence interval, 0.28–1.22). While evaluation of overall perinatal mortality in the context of PPROM would be interesting, it was not the focus of our analysis. Third, in response to the concern of using multiple potentially time-dependent variables such as latency and multiple courses of steroids, we believed that given the biological plausibility of a second course of steroids being related to the outcome it was important to include this variable in the model. We agree with the authors that models should be as parsimonious as possible; however, this goal should not be achieved by excluding potentially important covariates and many management interventions associated with PPROM are related to latency duration. Furthermore, we confirmed the lack of multicollinearity using variance inflation factors. Finally, regarding the statistical presentation of the adjusted model, while adding global P values would provide additional modeling information, we elected to present our results in the most accessible means possible.
WordPress theme by UFO themes