I also wonder why they did simple case-control matching rather than propensity scores. The number of unobserved confounders you could imagine is high in an analysis like this.