96 Chapter 4 all N biomarkers we used, GFAP was associated most strongly with Aβ, which could explain the attenuated estimates when Aβ was added as covariate. MTA was not associated with clinical progression after correcting for covariates, nor with MMSE decline. Although we previously showed a dose response pattern with MTA as N (20), the small variability in MTA within cognitively normal individuals makes it too crude a measure to accurately predict decline. Overall, we show there is room for improved prediction beyond Aβ and p-tau, using HV, NfL and GFAP as N biomarkers. Limitations of the present study include that the list of N biomarkers examined is not exhaustive. For example, FDG-PET or other MRI atrophy measures have also been suggested as suitable N markers. Although the list of putative N biomarkers is long, we chose to use a variety of N biomarkers obtained by three different modalities that are widely used in literature, which makes our study relevant to the field. Another limitation is that the sample sizes somewhat differed for each N biomarker. This might have led to differences in outcome. However, when we repeated the analyses in the sample with complete data, results were similar, indicating their robustness (eTables 1 and 2). Furthermore, our sample consisted of individuals with SCD presenting at a memory clinic, and the results might not be directly translatable to a community based setting or to other disease stages. Nonetheless, individuals with SCD can be considered an especially clinically relevant group, that might particularly benefit from the AT(N) classification system to grade their degree of underlying pathology. These are the individuals who present to a memory clinic because of worries about their cognition, and for this group AT(N) prediction modelling can make a relevant contribution. Another limitation is the lack of optimal cut-off values for HV, NfL and GFAP. Instead, we pragmatically used cut-off values obtaining a 10% and 25% N positivity rate, to provide a range of the true effect sizes. Additionally, we used continuous N biomarkers in all models. However, different cut-off values would probably have resulted in slightly different results. Last, we had a mean follow-up duration of 3.8 years and our sample had a relatively young age. Together, this could explain the low percentage of individuals with clinical progression to MCI or dementia, which limits the power to detect associations with N biomarkers. Furthermore, MMSE has a ceiling effect in cognitively normal individuals and perhaps our relatively short follow-up time hampered the finding of associations. Since all N biomarkers reflect different aspects of neurodegeneration, they could also have different associations with cognitive tests measuring specific cognitive domains. It would be interesting to investigate associations with other neuropsychological tests, but that is beyond the scope of this study since our aim was to assess the association between N biomarkers and disease progression in general. Strengths include the relatively large sample size of this well-defined cohort.
RkJQdWJsaXNoZXIy MjY0ODMw