White matter analysis of the extremely preterm born adult brain

Highlights • We investigated the long-term effect of extreme prematurity on structural and microstructural characteristics of the brain network.• The hierarchical organisation of the extremely preterm adult brain is intact; however, the extremely preterm brain has significantly reduced structural connectivity and neurite density compared to the term brain.• The most significant alterations are observed in the somatosensory and motor cortical areas, and in the deep nuclei.

: Results of the statistical and effect size analysis of the weighted betweenness centrality B i of the structural brain networks of EP and FT.
2 Distribution of brain volume 8 Figure 1: The distribution of the Total Brain Volume (TBV) for Extremely preterm (EP) and Full-Term (FT) born subjects. The samples are displayed with different colours to differentiate by sex and characters to separate between gestational age in completed Weeks (W). The FT group has higher TBV compared to EP group.
3 Linear regression: impact of adding the volume of ROIs as 9 a regressor 10 In the main manuscript, we examine the effect of TBV and prematurity on each graph measure by 11 using a linear model in which, for each brain region, the graph theory measure is the dependent with and support these findings, as expected.

29
Figure 2: Significance map of the effect of extreme preterm birth (prematurity), volume of individual brain regions (ROIV), and Total Brain Volume (TBV) on each graph metric. The colour scale for statistical significance ranges from grey when the node is not statistically significance, light red when the node is significant at p−value lower than 0.05, to dark red when the node is significant after Bonferroni correction. Figure 3: Maps of variance explained by extreme preterm birth (prematurity), volume of individual brain regions (ROIV), and Total Brain Volume (TBV) on each graph metric. The colour bar shows that the brighter the colour, the higher the variance explained.
The linear model 2, containing the volume of ROI as a regressor, might suffer from collinearity issues.

30
This problem occurs when two or more regressors are correlated as the TBV and the ROI volume 31 are. Although multicollinearity might not be a concern for the model's power of prediction, it inflates 32 the variance of the affected variables, namely, TBV and ROIs volume. As bigger TBV directs to a 33 bigger volume of ROI, these two variables move in the same direction. This pattern makes it hard 34 for the model to tease apart the single effect, which drives the standard error high. To quantitatively 35 analyse the potential multicollinearity, we investigate the correlation between TBV and the volume 36 of ROIs, and estimate how the TBV regressor is affected by adding the ROI volume as a regressor. 37 Figure 4.A below shows that the correlation between TBV and ROIs' volume is undoubtedly high 38 (above 0.5 for 90% of the regions). Although there is no clear rule when a correlation is too high to 39 lead to multicollinearity, these values indicate that there is risk of multicollinearity. To investigate 40 how adding ROI's volume as a regressor affects TBV regressor, we compare the ratio between the coefficient and standard error of TBV in model 1 and model 2, as the standard error would be inflated 42 in the case of multicollinearity. We compare the ratio of the coefficient of TBV and its standard 43 error in model 1 and model 2. The results are shown in Figure 4.B below. The figure suggests that 44 the standard error of TBV in model 2 increases with respect to model 1. This result might hint 45 that, in model 2, it becomes harder to disentangle the effect of ROI's volume from TBV. For the 46 reasons listed above, we think that model 2 does not constitute an improvement for describing the relationship between prematurity, TBV, and graph measures.   Table 6: Results of covariate analysis for evaluating the effect of being born extremely preterm (Prematurity) and Total Brain Volume (TBV) on the values of Fractional Anisotropy (FA), Mean Diffusivity (MD), Neurite Density Index (NDI), and Orientation Dispersion Index (ODI) over the brain and along hub and peripheral sub-networks.