Assessing reproducibility: agreement vs correlation.

I have read with interest the paper by SENÉTERRE et al. [1]. However, perhaps the method used to assess the reproducibility of the measurements is not the best one. The authors use Spearman's rank test, and they find an "excellent and highly significant" reproducibility, since r-value is 0.99 in all cases. Considering that correlation is different from agreement, BLAND and ALTMAN [2] have proposed a method for assessing observer agreement and variability in continuous measurements. It would be interesting to know if such high reproducibility would have been obtained had the proper statistical method been used, most of all because from their figure 4 [1] it seems that in several cases the difference between first and second slice, or between first and second observer, may be as high as about 500 pixels. Should a high reproducibility be obtained, this test must be correlated with physiological tests, but if reproducibility is low this technique cannot be recommended for clinical use in bronchoprovocation tests.