Harmonizing multisite data with the ComBat method for enhanced Parkinson's disease diagnosis via DAT-SPECT

Area of Science:

• Neuroimaging
• Medical Statistics
• Neurology

Background:

• Dopamine transporter single-photon emission computed tomography (DAT-SPECT) is vital for Parkinson's disease (PD) evaluation.
• Inter-site variability in DAT-SPECT data limits its utility in multisite clinical trials.
• Prospective correction is effective but not always feasible; retrospective methods like ComBat are underexplored.

Purpose of the Study:

• To evaluate the effectiveness of retrospective ComBat correction for harmonizing DAT-SPECT data.
• To compare the diagnostic performance of original, prospectively corrected, and retrospectively corrected DAT-SPECT specific binding ratios (SBRs) in differentiating PD patients from healthy controls.

Main Methods:

• Analyzed DAT-SPECT SBRs from 72 healthy adults and 81 PD patients across four clinical sites.
• Applied prospective linear regression correction and retrospective ComBat correction to the SBRs.
• Assessed diagnostic accuracy using the area under the receiver operating characteristic curve (AUC-ROC).

Main Results:

• Both prospective and ComBat corrections improved the discrimination performance between PD patients and controls.
• ComBat-corrected SBRs showed comparable diagnostic performance (AUC-ROC = 0.987) to prospectively corrected SBRs (AUC-ROC = 0.992).
• ComBat correction resulted in substantial changes in mean SBR values with reduced variance.

Conclusions:

• Harmonization methods, particularly ComBat, are valuable for standardizing SBR-based PD diagnosis in multisite studies.
• Retrospective ComBat correction offers a viable alternative to prospective correction for DAT-SPECT data.
• Caution is advised when interpreting ComBat-corrected SBRs due to significant changes in mean values.