This was partially controlled for in their analysis, but it's a cohort study meaning that people weren't randomized into groups. So certainly there could be some of that happening, but the analysis minimizes (but doesn't totally eliminate) how much that influences the outcome.

Behavioral science is hard, and messy.