Studies have been conducted about this, and many times have shown at best very mild correlations with confounding not controlled for (one French study is an example of this).