Statistical non-independence of brain metabolite concentrations whether normalized to creatine or water.

Abstract

1H-MRS investigators studying brain metabolite concentrations often attribute biological significance to correlations between calculated metabolite values within the same voxel. A recent report in this journal provides a valuable perspective on how statistical non-independence of such values can undermine biological interpretations of their correlations. However, careful examination of this issue suggests their critical analysis does not go far enough. Hong et al. claim that appropriate water normalization, unlike creatine normalization, eliminates the problem of spurious correlation. Both logical and empirical considerations show this is not the case. Correlations between water-normalized metabolite values are also prone to substantial spurious correlations.