SPINS: FA site/scanner effect (pre-harmonization)

Description. Before harmonization of DWI data across sites, Slicer tractography shows a large effect of site and scanner in FA. However, we do not see a site effect within a subset of sites with PRISMA scanners (i.e., CMP, MRP, ZHP). These results preclude us from naively combining data across all sites – harmonization is required.

Note: The mean FA values are shown before the removal of problematic tracts and participants (below segmentation rate thresholds) and before multivariate imputation of missing FA values across all participants. However, it does not show include participants who failed QC. This analysis therefore shows data from n=73 tracts and 407 participants.

Visualizations. Circle geoms indicate non-PRISMA, triangles indicate PRISMA.

FA by group

FA by site

FA by group and site

Density plot

Non-linear effect

Statistics.

All scanners

#effect of diagnosis on FA
anova(lm(meanFA ~ group + age + sex + education, data = df))

#effect of site on FA
anova(lm(meanFA ~ site + age + sex + education, data = df))

#effect of scanner on FA (PRISMA vs. non-PRISMA)
anova(lm(meanFA ~ scanner + age + sex + education, data = df))

#effect of diagnosis on FA, when also controlling for site
anova(lm(meanFA ~ group + site + age + sex + education, data = df))

#effect of diagnosis on FA, when also controlling for scanner
anova(lm(meanFA ~ group + scanner + age + sex + education, data = df))

#effect of diagnosis on FA, when controlling for site and scanner
anova(lm(meanFA ~ group + site + scanner + age + sex + education, data = df))

PRISMAs only

#effect of diagnosis on FA
anova(lm(meanFA ~ group + age + sex + education, data = df_PRISMA))

#effect of site on FA
anova(lm(meanFA ~ site + age + sex + education, data = df_PRISMA))