Extended Data Fig. 8: Proteomic predictive performance on the dementia-free test set.
From: Dissecting the genetic and proteomic risk factors for delirium
Receiver Operating Characteristic (ROC) (a) and Precision–Recall (PR) curves (b) showing performance metrics of different models for predicting incident delirium on the held-out test set (n_cases = 105; n_controls = 6,378). The ‘proteomic’ logistic regression model is fit with the 14 stability-selected proteins, the ‘basic’ model is fit using only age at protein collection, sex and BMI and the ‘proteomic+basic’ includes both. The ‘APOE+proteomic+basic’ model additionally includes APOE-ε4 haplotype count as predictor. The dotted black lines indicate the performance of a randomly classifying model, that is the diagonal of the ROC curve (a) and the disease prevalence horizontal line in the PR curve (b). AUC: Area Under the Curve. ROC and PR plots and AUC calculations are made using the yardstick R package (v1.3.0).
