94 Chapter 4 DISCUSSION In a sample of cognitively normal individuals with SCD, we found modest to moderate correlations and low concordance between the N biomarkers t-tau, MTA, HV, NfL and GFAP. N biomarkers HV, NfL and GFAP each predicted clinical progression, and had predictive value in addition to Aβ and p-tau. Therefore, we recommend HV, NfL or GFAP as biomarkers for N. The tight correlation between t-tau and p-tau precludes the use of the former as a marker of a different biomarker category than the latter. We extend on former observations that different markers of N are not necessarily closely correlated. The low correlation between N biomarkers likely contributes to the often discordant biomarker results in the AT(N) classification (6, 9, 12, 15). We add blood-based biomarkers to the comparison, showing similarly modest associations with the N biomarkers in other modalities, and also similarly strong associations with clinically relevant outcomes. Although at a population level, the overall qualitative pattern of biomarker frequencies remains rather stable regardless of the type of biomarkers used (9), it becomes problematic when researchers and clinicians treat the different N biomarkers as if they were identical. For prediction modeling at the individual patient level, the prognosis for an individual will vary considerably depending on the choice of N biomarker. The choice of N biomarker will also have an effect on the design of therapeutic trials, as well as the potential implementation of the AT(N) classification in the clinic. Studies investigating the AT(N) classification that use different definitions of their biomarkers, cannot be directly compared. We found low to modest correlations and low concordance between different N biomarkers, which is largely in line with literature (6, 10, 12, 30, 31). One possible explanation for this is that although all N biomarkers capture a certain aspect of neurodegeneration, the underlying biological processes that lead to specific N biomarker abnormalities are far from identical. T-tau and NfL reflect the severity of neuroaxonal injury, atrophy on MRI reflects loss of the neuropil, and GFAP reflects astrocyte activity (2, 32-34). Literature suggests these processes all have a different longitudinal trajectory, for example, NfL and t-tau abnormality likely precede HV abnormality and t-tau eventually reaches a plateau (35-38). This means correlations between N biomarkers of different processes are probably dependent on disease stage. However, MTA and HV were also poorly correlated, which is remarkable considering both HV and MTA aim to measure a similar process. We found a correlation coefficient of -0.24, which is relatively low and slightly lower than coefficients found in literature (range r -0.27 to -0.54) (39-41). This low correlation could be due to the fact that the MTA score is partly influenced by the volume of