For the perinatal epidemiologists.
In working with a large population-based perinatal registry dataset, I've assigned a missing value to gestational age and birth weight variables for observations that are categorized as stillbirth. This is so that they are not included when calculating counts and rates of preterm birth, low birth weight and SGA, as well as from other analyses concerning with them as dependent variables.
I'm not particularly interested in stillbirth itself except as a DV at times; however, have I made an error in removing birth weight and gestational age values for stillbirths?
Are there any instances where one would want to include stillbirths in an analysis using gestational age or birth weight as either the DV or IV?
In hindsight, I could have kept those values in and just remember to include "if stillbirth != 1"(not equal 1) in the code, but would be a hassle and a risk of forgetting at times. Thoughts?
As with many questions, the answer is, 'it depends'. Many reports from public health agencies focus only on live births, in which case these events would be excluded. But it makes more sense to provide the total number of outcomes, divided by the sum of live births and fetal deaths (or stillbirths if all of these events are enumerated).
Thanks Russell,
for the majority of the research I'm doing, I think I'm correct by having the stillbirths excluded from my analyses, but may eventually need to append that data back in.
I agree with the 'it depends' response. If the exposure causes severe fetal growth restriction (say due to placental problems) that in turn causes stillbirth then it might be best to include them. However, in my research (and from what I read your research too) I don't think it makes much of a difference because a larger proportion of stillbirth would be due to exposures other than those for which we are primarily interested. For preterm, it might be that excluding stillbirths would mean you would ignore differential censoring at early gestational ages.
As an additional note, I'm told that some software includes the missing value in the degrees of freedom. This won't make much of a difference because stillbirth numbers tend to be low. I have not encountered the problem as I suspect SAS/R would be okay following your approach. I find it simpler to define a study dataset and work off that one.
Good luck with your research.
thanks Gavin for your answers.
I've heard that as well re: missing values and DoF. I will have to remember to check if that's the case with Stata, I think it's fine though. When working with such large datasets (>200K), it's easy to overlook.
I agree for the 'it depends' response (Gavin) but you must remembre that many "risk factors" for preterm birth,LBW or fetal growth restrition are also RF for still birth. Indeed, an important information of quality care at population level, is the perinatal mortality ratio for those fetus with >= 3000 g and, specially, at labor.
Thanks Carlos,
that's a good point and an interesting statistic to calculate at the sub-regional level (Local Health Areas here in BC).
In the past S/B were excluded however to have accurate rates SB's should be included, as per Carlos' response above. Use a 'fetus at risk approach' - ie include gestation into the model.As Gavin said above SB is a very small proportion and so in pop health data in/or out the effect sizes will not change (much). Its also a bit more politically correct to include SB eg Lancet (Lawn).
One other issue that you may want to consider is that what is counted as a stillbirth varies to a surprising degree. See Gibson E, Culhane J, Saunders T, Webb D, Greenspan J: Effect of nonviable infants on the infant mortality rate. American Journal of Public Health 90(5), pp1-4, 2000.
- Badges
- Science topic