Skip to main content
. Author manuscript; available in PMC: 2020 Nov 2.
Published in final edited form as: Neuroimage. 2020 Jul 5;220:117129. doi: 10.1016/j.neuroimage.2020.117129

Table 1.

Comparison of harmonization methods in ADNI data.

 AD × time
 AD × time
 AD × time
 LMCI × time
 LMCI × time
 LMCI × time
# significant
 # coef/p
 # coef /p
 # significant
 # coef/p
 # coef/p
Method features < Cross < Unharm features < Cross < Unharm
LongComBatREML, no scanner 30 17/16 13/2 10 2/3 3/1
LongComBatREML, with scanner 29 16/13 12/0 10 2/1 3/0
LongComBatMSR, no scanner 33 15/25 13/21 15 2/8 3/8
LongComBatMSR, with scanner 31 14/24 12/16 14 2/7 3/4
CrossComBat, with scanner 27 11
Unharmonized, with scanner 31 11
Shared significant features 25 9

Notes: Table shows number of Bonferroni-corrected significant features for each method. Coefficient estimate and p-value comparisons between methods only include shared significant features. LongComBatREML: longitudinal ComBat, restricted maximum likelihood method; LongComBatMSR: longitudinal ComBat, mean squared residual method; CrossComBat, Cross: cross-sectional ComBat; Unharm: unharmonized; coef: estimated coefficient; AD: Alzheimer’s disease; LMCI: late mild cognitive impairment.