Panel A shows a receiver operating characteristic (ROC) curve with sensitivity on the y-axis and 1-specificity on the x-axis. It includes lines representing different groups: Combined, Eur, Amr, and Seropositive, each with their respective area under the curve (AUC) values. Panel B features violin plots comparing polygenic risk scores (PRS) between CeD patients and matched controls across different ancestry groups: Seropositive, Eur, Amr, and Afr. The y-axis represents the PRS score, and the x-axis lists the groups. Panel C displays calibration curves with observed CeD fraction on the y-axis and mean predicted risk on the x-axis. Different lines represent various groups, and Brier scores are provided for each. Panel D is a table summarizing the performance metrics of the PRS for predicting CeD across different subgroups, including total number, AUC, best threshold, sensitivity, specificity, positive predictive value (PPV), and negative predictive value (NPV).
Evaluation of a reported PRS prediction model in participants from the All of Us project with CeD participants. (A) Receiver operating characteristic (ROC) curves showing the performance of the PRS for distinguishing CeD cases from controls. The blue line represents the ROC curve obtained with seropositive CeD cases considered positive and their matched non-CeD controls considered negative. PRS performance is evaluated across different ancestry subgroups: European, admixed American, and the combined cohort (regardless of ancestry). (B) Violin plots comparing PRS distributions between CeD patients and matched controls. PRS scores from seropositive CeD patients or those with European (Eur), admixed American (Amr), or African (Afr) ancestry were compared to their respective matched non-CeD controls. Mean PRS values are indicated by white dots. The dashed line represents the optimal PRS threshold with the highest accuracy for distinguishing seropositive cases from non-CeD controls. Percentages reflect the proportion of individuals in each group with PRS values above this threshold. The black boxes represent the interquartile range (25th to 75th percentile) of PRS values. (C) Platt post hoc probability calibration was applied by fitting a logistic regression that maps the model’s raw score (logit) to calibrated event probabilities. Calibration performance is summarized by the Brier score (mean squared error between predicted probabilities and observed outcomes), with lower values indicating better probabilistic accuracy. Bins centered at higher mean predicted risk indicate that the model assigns higher average risk in that subgroup (e.g., mean predicted risk >0.8 in the seropositive group). (D) Performance of the PRS for predicting CeD across subgroups defined by genetic ancestry and serologic status. PPV, positive predictive value; NPV, negative predictive value.