Testing the agreement of medical instruments: overestimation of bias in the Bland-Altman analysis.