Disagreements in meta-analyses using outcomes measured on continuous or rating scales: observer agreement study