This is pretty worthless study. They are doing multiple comparisons without doing Bonferroni correction. Once you do that, there's no significant finding. On top of that, they have stupidly large 95% confidence intervals, their test metrics are pretty weak, and they only have blood levels of Hg of only a fraction of study participants. They seem to attempt to control for confounds, which is commendable, but with their sample being rather small, there's little hope of obtaining significant finding after controlling and correcting for multiple comparisons.
Controling for confounds here is the most important part really. It's like all the studies finding that people drinking wine have better health outcomes -- it turns out that wealthy people both are more likely to drink more wine, and have better health outcomes for reasons other than drinking wine. If you don't control for that, you'll find spurious associations.