Apples, Arsenic, and Risk - Part 9: Type 2 diabetes - A very weak link

Ana Navas-Acien, M.D., Ph.D., a physician & epidemiologist at Johns Hopkins University’s Bloomberg School of Public Health, was used by Consumer Reports in their January 2012 article called "Arsenic in your Juice".  According to Consumer Reports:

[Dr. Ana Navas-Acien] was the lead author of a 2008 study in the Journal of the American Medical Association that first linked low-level arsenic exposure with the prevalence of type 2 diabetes in the United States.

If you can remember the "ground rules" I discussed in my last post, we can look at her paper, her data, and her conclusion to see if there is, indeed, a "link" between low-level arsenic exposure and type 2 diabetes.

Let's first look at the paper's claim:

The positive association between total urine arsenic and diabetes after adjustment for markers of seafood intake was consistent for most subgroups examined, with somewhat greater associations in participants who were younger, overweight, and never smokers (FIGURE 2).

Source

We are all taught in our EPI classes, at least I was at Texas A&M's School of Rural Public Health, that if the range of your odds ratio (OR) includes the number one - "1" - there is no statistically significant association.  Period.

Now there is nothing wrong with Dr. Ana Navas-Acien including all her data in her report.  I would have left all the non-associations out, but that's just me.  What's troubling with her work is the graphic she includes in Figure 2.

Source

What can one reasonably conclude from looking at that graphic?  Look how all those black boxes fall on the "favors association" side of the graph.  What is Dr. Ana Navas-Acien trying to convey to her audience?

This is where the "whole truth" comes in.  Her graph makes it look like there is an association between total urine arsenic and diabetes every where she looked.  That's not true.  Look at the ranges.  All but four of them include the number "1."  Look at the P-values reported, all of them are above 0.05 for the interaction.

Remember, we want to reject the null hypothesis that there is ‘no effect of the intervention’ or ‘no differences in the effect of intervention between studies’ (no heterogeneity).  We can only reject the null if the P-Value is less than 0.05 (95%).  The null says there is no association, to reject that notion we need a P-value that is less than 0.05.

Now look at the four Adjusted Ratios in Table 2 that do not include the number "1" in their range:

• 1.71
• 1.35
• 1.55
• 1.54

How large of a difference is that?  They are all less than 2 to 1, so you can't say the association is twice as high.  But there's another factor to consider when looking at those four values.  How many participants - "n" - were in the populations compared?  I'll address that in an upcoming post, because its important.  But let's forget about those four for now and look at the overall "Adjusted Ratio" she reports:

1.26 (1.02-1.56)

How confident should we be in that number?  The range does not include a "1", so we can truthfully state that we see an association here:

After adjustment for biomarkers of seafood intake, total urine arsenic was associated with increased prevalence of type 2 diabetes.

The number spat out, "1.26," does not include the number "1" in its range therefore there is a positive association even though the lower range is only 2/100 higher than "1".

And the whole truth?  Here is what she should have concluded:

Our research shows a weak association between urinary arsenic that has been adjusted for biomarkers of seafood intake and the prevalence of type 2 diabetes.

We have two things working against us here when we calculate ORs and P-values.

1. Statistics looks at the data point as absolute
2. Sample and analytical data points are not absolute numbers - they are a statistical range that we report as a single number.

How much confidence should we have in this Odd Ratio: 1.26 (1.02-1.56)?

Consumer Reports believes there is enough confidence in that number to make the claim that Dr. Ana Navas-Acien "first linked low-level arsenic exposure with the prevalence of type 2 diabetes in the United States."  

That claim was backed up by this journal paper, and in that paper the OR of 1.26 (1.02-1.56) is reported and the conclusion 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." is made.

It all rests on this one bit of data: 1.26 (1.02-1.56).

How much confidence should we have in this Odd Ratio: 1.26 (1.02-1.56)?  It is that number and it's range that allows her to report 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.

Next Post: Apples, Arsenic, and Risk - Part 10: Type 2 diabetes - 1.26 based on what?

.