Research Paper Volume 14, Issue 4 pp 1691—1712

Figure 2. ROC curves of GrimAA (A) SPARE-BAA (B) and their joint modeling (C) in predicting global cognitive status at Y30. The median of the first principal component of Stroop, RAVLT (long delay recall), and DSST test scores (i.e., global cognitive function) measured at Y30 was used to dichotomize the global cognitive status into low (denote by 1) vs. high (denote by 0). The ROC curves were generated using 80/20 training/testing sets with 5-fold cross-validation to avoid overfitting. Associations between two aging markers and Y30 cognitive status evaluated by logistic regression were presented as odds ratio (OR) with every one year greater in GrimAA/SPARE-BAA, adjusting for age, sex, race, study fields, and education. The p-values of AUC were calculated by comparing with the chronological age benchmark AUC curve. GrimAA: GrimAge acceleration; SPARE-BAA: SPARE-BA acceleration; OR: odds ratio; AUC: area under the ROC curve.