The December 31, 2015 issue of the New England Journal of Medicine published an article by Snowden et al. that compared outcomes for births planned at a hospital vs. at home or at a freestanding birth center. I’ll discuss the findings and identification in a little bit (you can see the NYT article by Pam Belluck here). But, I actually want to discuss the characteristics of women who plan their births at a hospital vs. elsewhere.
It’s easy to listen to the charged debates about home births and characterize the women who choose to plan out-of-hospital births as privileged: a look at the data (see Table 1 below) suggests that only two thirds of the women who plan to give birth at a hospital are white, compared with nearly 90% for those who plan to give birth at home. Similarly, those that plan home births are more likely to have some college education or higher.
However, a look at the insurance status of these two groups reveals something else: those who plan to give birth at home are also less likely to have insurance. Almost 100% of women who plan to give birth at a hospital have private or public insurance while more than a third of those who planned out-of-hospital births were categorized as “self-pay.” The article does not present joint distributions of these characteristics but it’s hard to imagine that the presumably richer and more educated white women who plan home births are also those who pay out of pocket rather than through insurance.
What is worse is that there are a small number of cases (about one sixth of planned out-of-hospital births, but less than 1% of all births) where the planned out-of hospital birth was transferred to a hospital, presumably due to complications. Women with low education and no insurance account for a very small percentage of such transfers. If the group giving birth at home consisted of only richer and educated white females who live closer and have easier access to a hospital, we should probably not see self-paying women accounting for 45% of planned home births but only 8% of intrapartum transfers.
The complications matter because as Tables 3 (below) and 4 make clear, the contribution of this paper is in its ability to disentangle hospital births that were planned at a hospital vs. planned at home and transferred. Without such reclassification, there is no difference in perinatal deaths between the two groups; with reclassification, about 2 out of 1,000 planned hospital births end up in a perinatal death vs. 4 out of 1,000 among those planned at home. The adverse outcomes account for such a disproportionately larger share of births that start outside a hospital but transferred to one that the share of perinatal deaths moves from 0.19 to 0.39 with the reclassification of transfers. Furthermore, even though these transfers are a tiny percentage of planned hospital births, they cause a small uptick in perinatal deaths at hospitals; meaning that there are a lot of instances, where even a transfer is unsuccessful in avoiding a perinatal death.
The study, as novel as it is descriptively, does not have an experimental or quasi-experimental causal identification strategy: it compares these two groups and controls for a rich set of variables to avoid confounding. The authors are very forthcoming about this limitation. However, these controls (Table 4) don’t seem to make any difference to the unadjusted comparisons in Table 3: the p-value for perinatal deaths is 0.003 both with and without controls. I find this discomforting – those characteristics should have a lot of explanatory power. Similarly, propensity score matching does not alter any of the findings (see here why Gary King thinks you should not use PSM). However, my worry is about the heterogeneity in the group who planned out-of-hospital births. As discussed above, they might be a combination of highly educated and richer white females with access to hospitals as a backup and poorer, uninsured women with less choice. If that is true, then the possibility of unobservable heterogeneity yielding these findings rather than causal effect of planned location of birth is higher: linear controls for observable characteristics and matching might not be enough – we don’t know (I thank Sarah Baird, who helped me see this point). However, post-hoc heterogeneity analysis (see Figure 1 in the paper) does not find any significant interactions with these baseline variables – other than mother’s age.
In the end, despite the fact that home births are not so easy to characterize, the descriptive evidence clearly shows that most fetal and neonatal outcomes are worse in planned out-of-hospital births while most maternal outcomes are better: induction and augmentation of labor, as well as cesarean deliveries are way more likely to be administered at hospitals than at home. So, regardless of home births being a luxury some can afford or otherwise, the end result is a tradeoff between perinatal and maternal outcomes.
P.S. Read Aaron Carroll (of the Incidental Economist) discuss how to make home births a safer option here.
P.P.S. See a new paper that just came out today about how physicians themselves are less likely to receive cesarean deliveries.