Modeling the relationship between women’s perceptions and future intention to use institutional maternity care in the Western Highlands of Guatemala

A study conducted in Bangladesh found the odds of giving birth in a health facility among women who intended to give birth in a health facility were 14.4 times higher compared to women who did not intend to give birth in a health facility [15]. Further, a study from Ethiopia asked women where they intended to give birth and found at follow-up post-delivery that 80% who intended to have an institutional birth actually did [16].

Guatemala is divided into 22 departments (similar to provinces) and 334 municipalities (similar to counties).

Proximity of villages (e.g. proximate, intermediate, and remote) were determined by the local study team based on their experience traveling to villages within the three municipalities. In very rough terms, the proximate villages were about 10-15 min vehicle ride from the municipal centers, the intermediate villages about 30 min and the remote villages were up to an hour plus ride by car with some villages lacking vehicle access (and in one case electricity).

Estimates of marginal (in the case of continuous covariates) and discrete (in the case of categorical covariates) change in probability of an outcome provide a more precise and intuitive way to interpret marginal effects in non-linear models such as those presented here than either model coefficients or odds-ratios. Parameter estimates from non-linear models are not in the metric of the original dependent variable and interpretation is difficult. Transformation of estimates to odds-ratios, though a conventional strategy to facilitate interpretation, is little better. Odds-ratios are not measures of differences in relative likelihood (or relative risk), but of relative odds, which are themselves non-linear transformations of differences in likelihood, dependent on the (unobserved) probability of the outcome in the reference category and equally problematic to interpret. A simple solution becoming common in the social and health sciences is to transform coefficient estimates directly to marginal or discrete change in probability of the outcome. Simply put, a one unit difference in independent variable x can be interpreted as associated with a difference of b in the probability of observing the outcome y, in the same way model estimates are interpreted in the linear regression, conditional on stated levels of other covariates in the model.

The “don’t know” responses were predominantly among the home group (42) compared to the facility group (3).

There was concern that linguistically and culturally there is little difference between “very good” and “good,” which is why they are coded together and not separated.

Only 16 women (from both groups) intended to give birth in a health facility, but would not recommend it to others. Explanations for their unwillingness to recommend centered on feeling uncomfortable telling others what to do.

Women have confidence with their decision to give birth at home because their comadrona (traditional midwife) was seen as capable and treated her well/with kindness, was with her at every moment and never left. Some cited the benefit of the comadrona being able to speak her language. Many women mentioned having the freedom to ask for food, drink or to use the temazcal (traditional sauna) whenever they desired. The minority of the facility birth group who would not recommend a facility birth to someone else cited specific reasons for not liking the care they received, such as being cold, no temazcal, no food, not allowing her family in, a lack of privacy, and getting an episiotomy without asking her permission first.