To test the association between each factor and PPH
maternal mortality, we used a series of random-intercept
hierarchical logistic regression models to take into account
the dependence of observations within hospitals.
A three-stage statistical procedure was used. First, a
series of univariate hierarchical logistic regressions were
performed to identify which factor was associated with
PPH maternal mortality. The association was quantified
with unadjusted odds ratios and their 95 % confidence
intervals. A critical p value inferior to 20 % was chosen
for variable selection in further analyses.