Wednesday, April 15, 2009

New study on home birth vs. hospital birth - deconstructed!

Jill at The Unnecesarean posted today about a new study out comparing home and hospital birth outcomes in the Netherlands. I was, to put it mildly, excited. I have read a lot of the critiques of previous home vs. hospital comparison studies. I wanted to see if this one was going to overcome some of those limitations. The BBC article she linked to doesn't exactly go in-depth, epidemiologically, so it was one of those moments when I was so glad for my full-text access! Being a student has its perks. (If you want the article, let me know.)

I knew I was going to take this study apart anyway and thought I might as well make a blog post out of it. I also thought I'd see if I can explain some of the epidemiological/statistical terms and principles that are important to studies like this. Let me know if I'm talking too far down or too far up.

First off, the study methods. This was a nationwide, retrospective cohort study. Let's break that epidemiological talk down. "Nationwide": the researchers used the records all the women who delivered in the Netherlands over a 6-year period (Jan 1 2000 to Dec 31 2006). That gives the advantage of a very large sample size and correspondingly high statistical power to detect differences between groups. The "retrospective" piece has the disadvantage that researchers had to use data collected in the past (no chance to go back and fix problems with data collection), and not for the exact purposes of this study.

The "cohort" piece means this was not a randomized controlled trial. The researchers took did not assign women randomly to place of birth; women made their own choices about place of birth. This is a major objection to almost any study of place of birth. Maybe women who choose a particular place of birth also tend to have other economic, social, and education indicators that are associated with better or worse birth outcomes. We can try to adjust for those indicators (as this study does) but we can't be sure we're catching all the confounding factors in the study. Maybe there are intangible factors we can't measure, like how well a woman takes care of herself, or how confident she is, that also influence both what birth place she chooses, and her birth outcomes. An RCT is by far the best way to assess risks.

Unfortunately, it's really hard to find participants for an RCT of place of birth. So we're mostly stuck with cohort studies. Back to the article, and how well they do in getting the most accurate results possible.

The researchers excluded births outside of low-risk definitions: multiples, non-cephalic (breech or transverse) presentations, VBACs (we could debate this, but), prolonged rupture of membranes (>24 hours) without contractions, congenital abnormalities, and intrauterine fetal demise before onset of labor. They also excluded women who were cared for by obstetricians, even if they were low-risk. So this is really, best said, a comparison of planned home vs. hospital births for low-risk women under midwifery-led care in the Netherlands.

To classify women into groups, the researchers used an intent-to-treat model. That is, women were assigned based on their intended place of birth. This is appropriate because the goal of the study is to determine whether it's equally safe to plan a home or hospital birth. If you judge it based on where people end up giving birth, you run the risk of skewing to high-risk cases ending up in the hospital, or unplanned home births going awry. If a person plans a hospital birth but ends up giving birth in her front hallway on the way out the door, or if someone plans a home birth but is transferred to the hospital for fetal distress, each should still be classified based on intent.

The researchers also collected information on maternal age, ethnicity, gestational age of the baby at birth, socio-economic status, and parity (primiparous or multiparous). All of these may be associated with both choosing a particular place of birth, and to perinatal outcomes. Again, because you can't randomize people, collecting this kind of data is important for statistical adjustment later on.

Because of the retrospective nature of the data collection, the researchers found that some records did not have the intended place of birth recorded. The midwives in those cases may not have recorded it, or the women may not have decided until after labor began. So there were 3 comparison groups, Home (60.7%), Hospital (30.8%), and Unknown (8.5%).

To compare the perinatal mortality outcomes of the two groups, the researchers calculated crude and adjusted relative risks of perinatal mortality and NICU admission. The "hospital" group was set as the norm (1.0). This is a good example of where statistical adjustment comes into play. The crude relative risk for home birth was .90, with a confidence interval of .69 to 1.17. That is to say, it's trending towards home birth being slightly less risky, but because the confidence interval crosses the mean (1.0), it's not statistically significant. The adjusted relative risk is 1.02 (confidence interval: .77-1.36). Now it trends slightly more risky - because they compensated for the fact that a statistically lower-risk group of women chose home birth - but again, the difference is not statistically significant.

In the end, the only statistically significant relative risk that researchers found was that babies in the "Unknown" group were more likely to be admitted to the NICU (RR 1.33, CI: 1.07-1.65). There was no difference between groups for perinatal mortality, and no difference in NICU admission between home and hospital groups. They also did not find any interaction between risk factors (like parity) and planned place of birth (in other words, there's a higher risk of perinatal mortality for babies of primiparous women, but that risk is the same regardless of where they give birth).

Overall, this struck me as a fairly well-done study. It has the advantages of a very large sample and a comparison of "apples to apples": low-risk women receiving midwifery care. My main question about this study would be whether there are other confounding factors that might be missing from the statistical adjustment. An important limitation of this study is its generalizability. The Netherlands has a well-integrated system of midwifery care and home birth, and can provide continuity of care from home to hospital. While this is an excellent argument that home birth can be as safe as hospital birth, it doesn't necessarily translate that home birth is as safe as hospital birth in every setting.

Still, it's exciting and probably the largest validation of home birth safety thus far. I hope it has some good effects around the world, as well as here in the U.S. (although Amy at Giving Birth With Confidence thinks not.)

Questions? Thoughts? This is super long, I know. At least it helped me think the study through!

2 comments:

Amy Romano said...

Great summary and I'm glad you broke it down - strengths and weaknesses. There will be pros and cons to any study design on home birth. As for the feasibility of an RCT, we'll never have one with the power to detect differences in perinatal mortality, even if women were keen to enroll in such a study, which they're not. (Another recent paper by some of the same authors of this Dutch study explore the reasons why. http://www.ncbi.nlm.nih.gov/pubmed/19250365) For these and other reasons, I don't think we should necessarily strive for an RCT, but should think about the best ways to get clinically meaningful data with other study designs, including population level research.

Glad to have found your blog! I'm guest blogging at Giving Birth with Confidence, where you found me. I'm launching a new blog on Lamaze.org called "Science & Sensibility" and will continue to critique new research over there. I'll add you to my blogroll and hope you'll do the same!

Rebecca said...

I'd love to! Looking forward to your new blog.

I agree that a large enough RCT is likely never going to happen (even if you found enough women willing to be randomized, I'd wonder if their willingness to be randomized could itself be connected to a confounding factor, given how few women are in the general population.) This seems like one of the best studies done to date.