Meta analyses look across published studies and essentially do statistics on these. The claim of the paper is that the studies that show adverse effects are explainable by fluke, and presumably are outnumbered by studies that don't show adverse effects.