RESUMO
PURPOSE: Epoprostenol (also called prostaglandin, PGI) is used for pulmonary hypertension in newborns with congenital diaphragmatic hernia (CDH) in some centers. The effects of PGI on survival in newborns with CDH were examined. METHODS: A retrospective analysis of the Congenital Diaphragmatic Hernia Study Group registry between 2007and 2011 was performed. Patient-level logistic regression was applied in a subset of 29 hospitals with a history of PGI use to relate the probability of death to the use of PGI within 7 days of surgery after controlling for ethnicity, prenatal diagnosis, prenatal steroids, CDH defect, chromosomal abnormalities, liver location, complex cardiac anomalies, 5-minute Apgar score, and operative day of life. This analysis was repeated after excluding 50% of the patients with the lowest probabilities (< 0.042 mean propensity score) of receiving treatment. To reduce confounding by indication, a separate mixed effects logistic regression analysis was performed in 58 hospitals to relate the hospital-level mortality to the proportion of patients administered PGI after controlling for hospital-level covariates. RESULTS: Epoprostenol was administered within 7 days of surgery for 80 (7.3%) of these subjects. Epoprostenol use was associated with higher mortality (odds ratio [OR] 4.39, 95% confidence interval [CI] 2.04-9.48) in the patient-level analyses without covariate adjustment. The direct association of epoprostenol use with mortality was partially reduced after covariate adjustment (adjusted OR 2.24, 95% CI 0.95-5.29, p = 0.07), and further attenuated after both covariate adjustment and restriction of the analysis to patients with propensity scores > 0.042 (adjusted OR 1.71, 95% CI 0.68-4.29, p = 0.26). A total of 182 of the 1,639 patients included in the center-level dataset died after 7 days of operation. There was no statistically significant association of mortality with the proportion of patients administered epoprostenol in hospital-level analysis (adjusted OR 0.63, 95% CI 0.34-1.17 per 25% increase, p = 0.15). CONCLUSION: The discrepancy of results between the hospital and patient-level analyses suggests that the association of mortality and PGI in the patient-level analyses resulted from bias by indication. Hospital-level results provided no evidence of a benefit of PGI use on survival, but may have failed to detect a true benefit due to limited statistical power. Further use of PGI in this population should only be recommended after rigorous evaluation, such as a randomized controlled trial.