Study design and assumptions
In this study, we are not interested in a potential causal effect of the predictors on timing of antenatal care initiation via a mechanism of timing of pregnancy recognition, which is what would be obtained when using mediation methods. From a public health point of view, it would be too complicated or even impossible to intervene on age, relationship status, migration background or socioeconomic position for people to initiate antenatal care earlier. Early pregnancy recognition might be a factor that may have a causal relationship with timing of antenatal care initiation, is amenable to intervention, and may reduce inequalities in antenatal care initiation. To answer this research question, we assessed the effect of early pregnancy recognition on timing of antenatal care initiation, and we used this information to estimate the reduction in inequalities in antenatal care initiation if everyone in our study population had recognized the pregnancy early. This design is also called a hypothetical intervention,32 because we used data from an observational study to estimate the counterfactuals, i.e. what would have happened if we had intervened in such a way that all participants would recognize the pregnancy within 6 weeks of gestation.
Three causal assumptions apply in respect to the intervention and the outcome.33 First, exchangeability refers to no residual confounding or bias in the association between pregnancy recognition and timing of antenatal care initiation. We therefore adjusted for all predictors as potential confounders. Second, positivity means that the probability of receiving the intervention (early pregnancy recognition) conditional on the covariates must be greater than 0. This was tested by estimating propensity scores in a model of the non-imputed dataset with early pregnancy recognition as dichotomous outcome and all the covariates as predictors. As the participant with the smallest propensity score had a score of 0.217 to receive the intervention, the positivity assumption was satisfied. Third, consistency requires the intervention to be clearly defined. In practice, our intervention would require sexually active women of reproductive age to take a pregnancy test in case of pregnancy suspicion (e.g. their menstruation did not start in the week that they expected). This would be complicated for those with an irregular cycle, thereby potentially violating the consistency assumption. Therefore, in order to relax this assumption, we repeated the analyses in only those with a regular cycle (28 days plus minus 4 days) in sensitivity analyses.34