Table 5.

Sensitivity Analysis for Controlled Direct Effects Under Unmeasured Moderate and Substantial Confounding, US Singleton Births, 1995–2002^a

Perinatal Mortality	Observed Analysis (No Unmeasured Confounding)				Moderate Unmeasured Confounding (RRU = 1.5)				Severe Unmeasured Confounding (RRU = 6.0)
	Preterm = No		Preterm = Yes		Preterm = Nob		Preterm = Yes		Preterm = Nob		Preterm = Yes
	RRCDE	95% CI	RRCDE	95% CI	RRCDE	95% CI	RRCDE	95% CI	RRCDE	95% CI	RRCDE	95% CI
Stillbirth	29.31	27.98, 30.70	3.81	3.51, 4.14	29.31	27.98, 30.70	4.54	4.18, 4.92	29.31	27.98, 30.70	9.23	8.51, 10.02
Early neonatal	10.94	9.93, 12.06	2.65	2.40, 2.92	10.94	9.93, 12.06	3.15	2.86, 3.48	10.94	9.93, 12.06	6.41	5.81, 7.08
Late neonatal	3.95	3.29, 4.75	2.56	2.13, 3.08	3.95	3.29, 4.75	3.05	2.53, 3.67	3.95	3.29, 4.75	6.20	5.15, 7.45
Perinatal mortality	22.31	21.41, 23.26	3.32	3.18, 3.46	22.31	21.41, 23.26	3.95	3.18, 3.46	22.31	21.41, 23.26	8.04	7.70, 8.39

Abbreviations: CI, confidence interval; RR_CDE, the adjusted risk ratio for controlled direct effects; R_U, the effect if unmeasured confounding increased the likelihood of the mortality outcome by a factor of 1.5 or 6.

^aRisk ratios were adjusted for maternal age, liveborn parity, marital status, maternal race, smoking during pregnancy, and chronic hypertension through log-binomial regression models.

^bUnder the simplifying assumptions that the prevalence of U among term deliveries was 5% both with and without abruption, the controlled direct effect at term (i.e., mediator absent) is unaffected. Note, however, that this would change if the prevalence of U among term pregnancies were assumed to be different with and without abruption.