highly significantP<10 −6
Box 1 Umbrella review evidence rating by strength of association Convincing Highly significant associations (P<10 −6 ) Cases of n >1000 I 2 <50% 95% prediction intervals excluding the null value Largest study nominally significant (P<0.05) No evidence of small study effects No evidence of excess significance bias Highly suggestive Highly significant associations (P<10 −6 ) Cases n >1000 Larg