Apples, Arsenic, and Risk - Part 11: Type 2 diabetes - 80th vs the 20th percentiles

How to make a weak association sound strong:

1. Present the data in a format to disguise the weakness (2% instead of 1.02)
2. Compare percentiles to increase the ORs reported.

Now please don't get me wrong here, there is a perfectly good reason to compare percentiles.  Was this one of them?  I don't think it was done for any other reason than to bolster the paper's claim that there is a positive association between arsenic and type 2 diabetes prevalence.

But let's assume that there is no sneaky reason for presenting this data in her report.

If that's the case, then let me ask this question: Does the OR for the 20%/80% percentile support Consumer Report's claim that there is a link between "low-level arsenic exposure with the prevalence of type 2 diabetes in the United States?" And, if so, will the concentration of arsenic they found in the apple juice they tested be high enough to increase the risk for developing type 2 diabetes?  That is, after all, what is meant by "there is a positive association between arsenic and type 2 diabetes prevalence."

What do we know so far from Dr. Ana Navas-Acien's study on arsenic and type 2 diabetes?  Well we know that for the 93 participants in the study who had type 2 diabetes, their average total arsenic was 6.2 ug/L as you can see in Table 2:

Source

We also know that the third model "was further adjusted for urine arsenobetaine and blood mercury levels" to provide "estimates for the association of inorganic arsenic not derived from seafood and for arsenobetaine."

And we know that the ratio of arsenic concentrations comparing participants with type 2 diabetes for Model 3 was 1.26 (1.02-1.56) as can be seen in Table 2 above.

So...an average of 6.2 ug/L total arsenic generated an OR of 1.26 (1.02-1.56) when it was further adjusted for urine arsenobetaine and blood mercury levels.

What's the average urinary total arsenic they found in the 2003-2004 NHAMES study?

Source

We're going to use this Table again in an upcoming post, it's important here because we need to understand what is meant by the term "low-level arsenic exposure."

There is a weak association when you look at Model 3 because the range is just barely above "1" - (1.02-1.56).  Because that range does not include the number "1", Dr. Ana Navas-Acien can report a "positive" association:

This finding supports the hypothesis that low levels of exposure to inorganic arsenic in drinking water, a widespread exposure worldwide, may play a role in diabetes prevalence.

and...

These results support our hypothesis that exposure to inorganic arsenic, which in this population was most likely derived from drinking water, is associated with an increased risk of diabetes.

and again...

We found a positive association between total urine arsenic, likely reflecting inorganic arsenic exposure from drinking water and food, with the prevalence of type 2 diabetes in a population with low to moderate arsenic exposure.

Which brings me once again to my question: how much confidence in that number do you have?

Had the paper stopped at:

After adjustment for diabetes risk factors and markers of seafood intake, participants with type 2 diabetes had a 26% higher level of total arsenic...

...we would only have an argument on including the word "weak" to describe the association.  Attach a verb, don't attach a verb, toe-mate-o, toe-maut-o.

Now look at the rest of the "results" paragraph in the abstract at the beginning onf the paper:

After similar adjustment, the odds ratios for type 2 diabetes comparing participants at the 80th vs the 20th percentiles were 3.58 for the level of total arsenic (95% CI, 1.18-10.83).

1.26 now jumps to 3.58.  And the 95% CI range is now (1.18-10.83).

Stout! ...or is it robust?

There's one slight problem here.  Are we still talking about "low-level arsenic exposure?"

Look at Table 3....

Source

Notice what the urinary total arsenic concentration is for the 80th percentile?  16.5 ug/L compared to the 20th percentile of 3.0 ug/L.  What's the geometric mean - average - they report?  6.2 ug/L compared to 7.3 ug/L.  Hmmm...interesting.

So let's reword that results statement to reflect what is really going on here.

When Dr. Ana Navas-Acien writes:

After similar adjustment, the odds ratios for type 2 diabetes comparing participants at the 80th vs the 20th percentiles were 3.58 for the level of total arsenic (95% CI, 1.18-10.83

What she really is stating is this:

After similar adjustment, the odds ratios for type 2 diabetes comparing participants with a  urinary total arsenic of 16.5 ug/L (almost three times the average) with those without type 2 diabetes and a urinary total arsenic of 3.0 ug/L (less than half the average), were 3.58 (1.18-10.83)

That statement above is factual and accurately describes the results from Table 3.

To get that high of an odds ratio the arsenic has to be almost three times the average level normally found in the population.  And, to top it off, it will only be that high if it is compared to a population that has 2 time less arsenic than the population as a whole.

16.5 ug/L is not a low level of arsenic.  It is above the MCL of 10 ug/L and way above the 4 ug/L Consumer Reports found in the samples of apple juice they tested.

Well Mr. Smarty-Pants, doesn't it also show that increasing the level of arsenic increases the risk of type 2 diabetes!  Explain that you wet blanket!

Yeah, it does show that, but it is based on a sample size which, surprise! surprise! is not referenced (at least I can't find it).

What is the "n" making up the 80th percentile of those with type 2 diabetes and what is the "n" for the 20th percentile for those without?  Remember, we started with 93/695 and then adjusted it once - Model 1, then again, Model 2, then even more, Model 3.  What's the "n" of two populations used in the OR for Model 3?  And then from that population they separated it further into percentiles, what's the "n" of the 80th and the "n" of the 20th?

In statistics, the error - bias - is reduced with more "n"s.

There is, however, another interesting bit of information you can get from Table 3.  Look at the next three columns called "Tertile"

What's a "tertile?":

Any of the two points that divide an ordered distribution into three parts, each containing a third of the population.

You know something you can get from those three tertile columns?  The total "n" of each population that made up that tertile.

So if the 80th percentile is 16.5 ug/L and the highest third of all the data is >10.8 ug/L, how many of the 34 "n" to in the third tertile of Model 3 are in the 80th percentile?

What's also interesting is that when  Dr. Ana Navas-Acien compared the lowest population, the middle, and the highest, there was no statistical significance in any of the pairings.  None.  They all contain the number "1" in the ranges.

I think I'm done looking at this paper.  It has too many issues to accept its claim that:

This finding supports the hypothesis that low levels of exposure to inorganic arsenic in drinking water, a widespread exposure worldwide, may play a role in diabetes prevalence.

So, low-levels of arsenic - below the MCL and at the levels found in apple juice - do not support a link to cancer, hyperkeratosis, and type 2 diabetes.  Is that it?  Can I move on to something else, you know, like hazardous waste being wacky?

What about Consumer Reports telling its readers about a 2011study in the International Journal of Environmental Research and Public Health that examined the long-term effects of low-level exposure on more than 300 rural Texans whose groundwater?

Consumer Reports claims that these Texans were estimated to have exposure to arsenic at median levels below the MCL and when tested showed poor scores in language, memory, and other brain functions.  What about low levels of arsenic and diminished intelligence?  Isn't that a problem?

Rule number one: Always read the report the findings/recommendations were based on.

Onward...to infinity and..../sigh

Next Post: Apples, Arsenic, and Risk - Part 12: Correlation does not imply causation.


.