Fig 1.
This flowchart shows an overview of the methodology used in developing the PRS and models.
Table 1.
Baseline characteristics of our study cohort.
Fig 2.
ROC curve for PAD detection model.
ROC curve comparing the clinical-only and combined-PRS model showing a slightly larger AUC with the PRS model.
Table 2.
NRI Result for the clinical (initial) and combined PAD detection models.
Fig 3.
Decision curve analysis plot for the two models.
The vertical axis shows the net benefit for using the models and the horizontal axis displays the threshold probability values for detecting PAD. Screen all is shown by the blue line which indicates a case in which all patients that present will be screened, while the screen none line shows the alternate scenario. We can see that the purple line illustrating the combined model with PRS dominates the clinical model across all the thresholds. Therefore, the preferred model would be the model with PRS.
Table 3.
Net reclassification improvement results for the MACCE prediction model.
Fig 4.
ROC plots for the RF models for MACCE (left) and AE (right), with and without PRS.
Fig 5.
Feature importance plots for the RF models (using SNP data) to predict MACCE (left) and AE (right).
We see that both models prioritize very similar genetic information, alongside the well-known clinical variables.