A professor of statistics says ...
Bravo, maria! Maria asked Henci Goer the question that sparked the debate on her site. She has not settled for the non-answer that Goer supplies or for the fact that my posts were deleted. She went to someone whom she believes is an independent expert.Ok, I posted this to a prof. in statistics and here is his response:
The study I am looking at is this study:
http://www.bmj.com/cgi/content/full/330/7505/1416
The following explanation was given by Johnson and Daviss about their study:
http://understandingbirthbetter.com/section.php?ID=31&Lang=En&Nav=Section
Some people say they have used the wrong comparison groups and that the correct comparison would prove that homebirth has triple the neonatality rate of hospital birth.
***
OK, here’s my take on it…
When I read the executive summary of the BMJ, I was struck by it’s modest claims in the results. By modest, I mean that it essentially reported the percentages of differing outcomes within it’s own data set. It was the conclusion, however, that struck me: it claims that their study group was similar to a group not in the study, namely, low risk hospital births in the US.
That seems to be the basis of the criticism. The comparison group has one obvious difference that masks for lots of other potential discrepancies: it was retrospective data. The authors of the study actually point this out in the study, however, so, to me, it isn’t fair to fault them for making the comparison. Perhaps they could have added a footnote to the conclusion in the exec summary, but that’s a bit picky. The disclaimer is clear in the discussion section:
"Regardless of methodology, residual confounding of comparisons between home and hospital births will always be a possibility. Women choosing home birth (or who would be willing to be randomised to birth site in a randomised trial) may differ for unmeasured variables from women choosing hospital birth…."
Consistent with this disclaimer, the biggest factor (in my opinion) is the demographics of their study group. This is visible in Table 1, which shows the characteristics of the mothers in the two groups:
- More women above the age of 25
- Likelihood of having already given birth is much higher
- Typical education levels are higher
- 95% had partners—which I would wager is significantly larger than the comparison group, whose rate is reported as N/A
Their study group is a self-selecting subpopulation of women—they are different from other women in ways that move them to choose a birth method that is out of the "main stream." This fact alone (supported by the items I just listed) suggests to me that they were better prepared for birth, and more aware of risks and of ways to handle them.
They did attempt to sort the data from the Nat. Center for Health Stats into "low risk" mothers, in order to make a better comparison. Assuming that sorting method valid, they arrive at the result that their group is, essentially, equivalent to the in hospital "low risk" group. Not shocking, given the kind of mom in their population.
I hope this is helpful.
***
Now my question is, what numbers did Amy Tuteur use to come to her comparison of homebirth being triple the risk of hospital birth. Where can I find these numbers and how are they a better comparison?
I think in the end, on one hand, eventhough this study has lots of merrit, the homebirth advocates should maybe not take it as a decisive study about the safety of homebirth, as they tend to do now, saying 'see!'
However, I do not think Amy's claims are grounded either so I would like to present to this prof. the numbers Amy is talking about and see what he comes up with.
Henci, would you please refer me to where I can find the numbers Amy is talking about? My apologies if they are posted here before!
Thanks!
maria.
ps: I asked two other people knowledgeable in statistics to look at this and I am waiting for their responses as well
****
Maria, I will refer you to the data and explain what I am talking about. Don't hesitate to ask additional questions or request additional data if you think it will be helpful.
The original problem:
According to Johnson and Davis, when analyzing the different intervention rates of home and hospital:
We compared medical intervention rates for the planned home births with data from birth certificates for all 3 360 868 singleton, vertex births at 37 weeks or more gestation in the United States in 2000, as reported by the National Center for Health Statistics [Births: final data for 2000. National vital statistics reports. Martin JA, Hamilton BE, Ventura SJ, Mencaker F, Park MM. Hyattsville, MD: National Center for Health Statistics, 2002;50(5)]
When analyzing the different mortality rate of home and hospital, Johnson and Davis used a group derived from out of date homebirth studies. I have always thought that was strange. Why not use the neonatal mortality data of the group that served as the comparison for interventions?
I went back and looked at the neonatal mortality data for this group, the EXACT group that Johnson and Daviss felt was the perfect comparison for intervention rates. I did this by reviewing the exact same paper that Johnson and Daviss used... Looking at the raw data we find a death rate of 0.9/1000 (white women, age 20-44, 37+ weeks, 2500+gm).
The hospital neonatal death rate for white babies at term of 0.9/1000 is not corrected for congenital anomalies, pre-existing medical conditions, pregnancy complications or multiple births. The true rate is substantially lower. Nonetheless, we can make an important comparison. Johnson and Daviss reported a neonatal death rate at homebirth of 2.7/1000 (uncorrected for congenital anomalies, breech or twins). The neonatal death rate in the comparison group THAT THEY USED was less than 0.9/1000.
So now we have an explanation for why Johnson and Daviss used two different comparison groups. They used one group (births in the year 2000) for comparing medical interventions. The neonatal death rate in that exact group was 0.9/1000, half the rate of neonatal deaths at homebirth. They supressed that information by using an entirely different group (drawn primarily from the 1970's and 1980's) instead of using the death rate from the year 2000.
Here's where you can find more about the new explanation: Johnson and Daviss acknowledge the validity of my criticism.
Johnson and Daviss have recently "re-analyzed" their own data and lowered the homebirth neonatal mortality rate: Johnson and Daviss: If at first you can't trick them, try, try again.
Labels: Henci Goer, Johnson and Daviss
<< Home