Study Finds Higher Rates of Premature Birth Near Fracking Sites (jhsph.edu)

An anonymous reader writes: Researchers from the Johns Hopkins Bloomberg School of Public Health have published a study (abstract) noting that pregnant women are more likely to give birth prematurely if they live close to fracking sites. The researchers used data from 40 counties in Pennsylvania, in which 10,946 babies were born between January 2009 and January 2013. They compared the data with the fast spread of fracking sites across the state — thousands have been built since 2006.

"The researchers found that living in the most active quartile of drilling and production activity was associated with a 40 percent increase in the likelihood of a woman giving birth before 37 weeks of gestation (considered pre-term) and a 30 percent increase in the chance that an obstetrician had labeled their pregnancy "high-risk," a designation that can include factors such as elevated blood pressure or excessive weight gain during pregnancy. When looking at all of the pregnancies in the study, 11 percent of babies were born preterm, with the majority (79 percent) born between 32 and 36 weeks."

Comments

Re:Correlation vs cause

[belthize]

Sigh, I suspect they have as much or better clue than you do. It's entirely possible that there is no causal effect. Their study doesn't say there's a causal effect, it's says there's a correlated effect. Even the referenced press release states: "The researchers found that living in the most active quartile of drilling and production activity was associated with a 40 percent increase in the likelihood of a woman giving birth before 37 weeks of gestation."

Stop viewing science press releases through the filter of whether it conforms to your world view or your superficial understanding of correlation and causation. That whole correlation isn't causation crap is becoming a mantra around here. People parroting it without really understanding what it means or doesn't mean or whether it even f'ing applies to the article in question.

Re:Correlation is not causation

[AthanasiusKircher]

Science doesn't start with explaining the mechanisms behind unexplained phenomena - it starts with confimring that there *are* unexplained phenomena and then searching for the mechanisms.

This may be so. So let's look at the study's "confirmation" of "unexplained phenomena," shall we? Oh wait -- the study is behind a paywall, so I guess we'll just use what they tell us in the abstract:

In adjusted models, there was an association between unconventional natural gas development activity and preterm birth that increased across quartiles, with a fourth quartile odds ratio of 1.4 (95% confidence interval = 1.0, 1.9). There were no associations of activity with Apgar score, small for gestational age birth, or term birth weight (after adjustment for year).

Translation: We looked for 4 things, and only 1 thing was statistically significant. Even for the worst quartile (i.e., that with the most drilling), the effect was only an odds ratio of 1.4, though we have 95% confidence that it was between 1.0 and 1.9.

Let's note a few things here:

(1) Odds ratios are not the same as relative risk, which is the more intuitive way of understanding stats. If a study finds a relative risk of 2 for factor X, that means your chances of getting a condition with factor X are twice as much as if you didn't have X. A relative risk of 1.4 means a 40% increase in risk. Odds ratios are more complex and are used for various statistical reasons, but they often tend to exaggerate an effect -- and it's unclear from this abstract what the actual increased risk is. But it's likely less than the 40% listed in TFS.

(2) Statistically, they have a 95% confidence interval of 1.0 to 1.9. An odds ratio of 1.0 means there is no effect at all. Which means that there's probably a 5% chance the actual effect is outside of this range, possibly down to 1.0 (where there is no effect). The "no effect" line is drawn here where it's barely statistically significant (according to the typical 95% standard) for preterm birth.

(3) The study was a "retrospective cohort" study, which means that they looked at pre-existing data (rather than a "prospective cohort" which would look at a control group and a study group going forward in future). There are always dangers here in selecting a sample group that happens to line up with your analysis, since you get to pick the group you want. (Since I can't read the rest of the study, I don't know how "selective" they were in choosing which areas to study, for example.)

(4) The phrase "adjusted models" refers to earlier in the abstract where they talk about the various adjustments made for possible confounding variables and such. They also had a complex model for determining potential exposure based on "an inverse-distance squared model that incorporated distance to the mother's home; dates and durations of well pad development, drilling, and hydraulic fracturing; and production volume during the pregnancy." If that model is tweaked in various ways, it could probably completely change the study results. Anyhow, while such adjustments are important for modelling and confounding factors, they can be manipulated (often unintentionally) by researchers in all sorts of ways.

(5) They looked for FOUR things, but they only found a statistically significant effect for ONE of them. The chances of finding at least 1 out of 4 things to satisfy a 95% threshold is about 18.5%. So if they threw in random numbers here, at least one of these things would "flag positive" in nearly 1 out of 5 times.

(6) The abstract only reports the "worst quartile" as having this (already barely) statistically significant effect. Apparently other times of the year these effects were reduced (and possibly didn't even hit the barely statistically significant effect for the worst quartile)... which then leads to the question about how a 4-month window in a study may