Figure 2 shows the survival curves for the unadjusted association between each of the molecular biomarkers and time to incident CDR > 0 for the 664 participants. Because non-missing data on all 13 candidate variables for stepwise selection were required, Cox proportional hazards analyses consisted of smaller subsamples (n = 286 for PIB and n = 302 for CSF). Forty-two participants in the PIB, and 39 in the CSF analyses developed CDR > 0. Molecular (PIB, CSF amyloid-β₄₂, tau, tau/amyloid-β₄₂, ptau/amyloid-β₄₂), psychometric (SRTFREE), and structural (normalized hippocampal volume) biomarkers combined to predict first time to cognitive impairment in the models (Table 2). Trail Making B also independently contributed to prediction of CDR > 0 in the model testing PIB (P = 0.024). Age did not independently predict CDR > 0 once normalized hippocampal volume was included in the models. Supplementary Table 1 shows the results when age, gender and education were first forced into the models, and then the stepwise selection procedure was used to choose among the remaining candidate variables for model entry.