84 Chapter 4 (29), and left and right sides were averaged. All images were visually inspected for registration or segmentation errors. Non-fasted serum samples (n=296 (72%)) were obtained through venipuncture and centrifuged on average within 2 hours from collection, at 1800g, 10 minutes at room temperature, before immediate storage at -80 oC until analysis. Serum GFAP and NfL levels were measured using the commercially available SimoaTM GFAP Discovery Kit (Quanterix) and the SimoaTM NF-Light Advantage Kit (Quanterix) according to manufacturer’s instructions and with on-board automated sample dilution (4). All samples were measured in duplicates with good average intra-assay %CV. Standard protocol approvals, registrations, and patient consents The research is conducted in accordance with ethical consent by VU University and the Helsinki Declaration of 1975. For all individuals included in the study, written informed consent was available. Statistics All analyses were performed in R version 4.0.3. We first used all biomarkers as continuous measures (Aβ, p-tau, t-tau, MTA, HV, NfL and GFAP). Since the AT(N) classification is based on dichotomous variables, we repeated all analyses with dichotomized biomarkers (A, T, Nt-tau, NMTA, NHV25, NHV10, NNfL75, NNfL90, NGFAP75, NGFAP90). CSF p-tau, t-tau, serum NfL and GFAP were log transformed due to non-normality. For Cox proportional hazards models and linear mixed models, continuous predictors were transformed to z-scores for comparability of effect sizes, and HV was inverted, so that for all variables higher values indicates worse. We first compared demographic and clinical variables between individuals that remained stable, and those that progressed to MCI or dementia during follow-up, using t-test, Mann-Whitney U test and chi-square where appropriate. To assess correlations between biomarkers, we used Pearson correlation analysis (CSF Aβ, p-tau and t-tau, MTA score, HV, and serum NfL and GFAP). We additionally used partial correlation to adjust for age and sex. We then investigated the associations between biomarkers and clinical progression using Cox proportional hazards analyses, with progression to MCI or dementia as outcome. We ran four different models, with a cumulative number of predictors. We first ran analyses with continuous N biomarkers as single predictors (model 1). We then added age and sex as covariates (model 2). Then we added CSF Aβ as covariate (model 3), and finally, also CSF p-tau (model 4). In models with MTA and HV, scanner