Bias in tensor based morphometry Stat-ROI measures may result in unrealistic power estimates

Abstract

A series of reports have recently appeared using tensor based morphometry statistically-defined regions of interest, Stat-ROIs, to quantify longitudinal atrophy in structural MRIs from the Alzheimer's Disease Neuroimaging Initiative (ADNI). This commentary focuses on one of these reports, Hua et al. (2010), but the issues raised here are relevant to the others as well. Specifically, we point out a temporal pattern of atrophy in subjects with Alzheimer's disease and mild cognitive impairment whereby the majority of atrophy in two years occurs within the first 6 months, resulting in overall elevated estimated rates of change. Using publicly-available ADNI data, this temporal pattern is also found in a group of identically-processed healthy controls, strongly suggesting that methodological bias is corrupting the measures. The resulting bias seriously impacts the validity of conclusions reached using these measures; for example, sample size estimates reported by Hua et al. (2010) may be underestimated by a factor of five to sixteen.

Fig. 1

Cumulative atrophy for TBM-derived Stat-ROI. Left panel: Mean percent change in TBM-derived Stat-ROI measures for 91 AD and 189 MCI subjects in ADNI (data taken from Hua et al. (2010) Fig. 3 p. 68). Right panel: Mean percent change in TBM-derived Stat-ROI measures for 81 AD, 198 MCI, and 160 HC subjects in ADNI (data publicly-available at http://www.loni.ucla.edu, current as of May 5, 2010).