This is an interesting finding. But, a reminder that this is epidemiological data. It is nearly impossible to "adjust" for all confounding variables in epidemiological studies. For example, you may expect a health user-bias among people who use solariums, sunlamps, or have high vitamin-D levels, and health at baseline might be effected by random fluctuations not accounting for all variance. I would look for evidence of randomized trials before taking this as the truth.