90 Chapter 5 5.3.5 Statistical analysis T-tests, Mann-Whitney U tests, or chi-squared tests were performed for baseline characteristics in order to compare the group of participants who developed dementia at follow-up and the group of participants who did not develop dementia at follow-up. Cox proportional hazard models were used to estimate hazard ratios (HRs) for dementia risk associated with WMH shape markers and WMH volumes. Values of solidity, convexity, and eccentricity were inverted to aid comparability of HRs. Solidity was multiplied by 100 and natural log-transformed due to non-normal distribution. Z-scores with total mean (whole group) as reference were calculated for WMH shape markers to enable comparability. Cox regression was performed on Z-scores of WMH shape markers, controlled for age, sex, and cognitive status at baseline (mild cognitive impairment or normal cognition). WMH volumes were multiplied by 100 and natural log-transformed due to non-normal distribution before a Z-score was calculated. Cox proportional hazard models for WMH volumes were controlled for age, sex, cognitive status at baseline, and intracranial volume. Additionally, the area under the ROC curve was estimated per WMH marker. Moreover, a second Cox regression model was run for both WMH shape markers and WMH volumes, where the regression analyses were additionally adjusted for cardiovascular risk factors (BMI, hypertension, diabetes mellitus, cholesterol level, and smoking status). Additionally, a third Cox regression model controlled for cardiovascular risk factors, the occurrence of coronary artery disease, and brain infarcts was run. In secondary analyses, we examined whether there were regional differences in the brain in relation to the discovered associations with dementia outcome by repeating Model 1 per brain lobe for WMH volumes. In additional secondary analyses to test for WMH volume dependency of the discovered associations with WMH shape markers, Model 1 was repeated for the WMH shape markers additionally corrected for total WMH volume (as a percentage of intracranial volume). 5.3.6 Sensitivity analyses To confirm the robustness of our shape analysis method, all models were repeated with WMH shape markers recalculated by applying a volume-weighted approach, where each lesion shape marker was scaled according to lesion size before calculating an average per participant. To test whether the Cox regression outcome was affected by oversegmentation, 62 participants with partial oversegmentation in addition to multiple correct WMH segmentations were excluded and the analyses were rerun. To test whether the Cox regression outcome was affected by subcortical infarcts, 266 participants with subcortical infarcts were excluded and the analyses were rerun. To test whether non-linear age-related effects in dementia occurrence influenced our results, we repeated the Cox regression models in an age- and sex-matched subset of
RkJQdWJsaXNoZXIy MTk4NDMw