Table 2 Prediction accuracy of various models for glaucoma classification using logistic regression, lasso regression and XGBoost, with the upper row presenting AUC values for PRS, IDS, IDPs, IGSs, and clinical covariates in individuals with OCT images (N=55,469) and the lower row summarizing AUC values for PRS, IGSs, and clinical covariates in individuals without OCT images (N=402,847).
From: Multimodal deep learning approaches for improving polygenic risk scores with imaging data
Samples | Models | Logistic | Lasso | XGBoost |
|---|---|---|---|---|
Individuals with OCT images (N=55,469) | PRS | 0.6196 (0.0070) | 0.6196 (0.0070) | 0.6150 (0.0070) |
IDS | 0.7766 (0.0062) | 0.7766 (0.0062) | 0.7742 (0.0063) | |
IDPs | 0.6890 (0.0072) | 0.6896 (0.0072) | 0.7106 (0.0070) | |
IGSs | 0.6245 (0.0070) | 0.6155 (0.0070) | 0.6387 (0.0070) | |
Clinical | 0.6870 (0.0060) | 0.6822 (0.0060) | 0.6778 (0.0060) | |
PRS + IDS | 0.7531 (0.0068) | 0.7545 (0.0068) | 0.7881 (0.0061) | |
PRS + IDPs | 0.7115 (0.0069) | 0.7105 (0.0069) | 0.7314 (0.0068) | |
PRS + IGSs | 0.6620 (0.0069) | 0.6560 (0.0069) | 0.6733 (0.0067) | |
PRS + IGSs + Clinical | 0.7403 (0.0060) | 0.7343 (0.0060) | 0.7397 (0.0060) | |
PRS + IDS + IDPs + Clinical | 0.7867 (0.0060) | 0.7857 (0.0060) | 0.7921 (0.0060) | |
Individuals without OCT images (N=402,847) | PRS | 0.6322 (0.0026) | 0.6322 (0.0026) | 0.6301 (0.0026) |
IGSs | 0.5547 (0.0027) | 0.5531 (0.0027) | 0.5467 (0.0027) | |
Clinical | 0.6995 (0.0022) | 0.6903 (0.0022) | 0.6979 (0.0022) | |
PRS + IGSs | 0.6368 (0.0025) | 0.6362 (0.0025) | 0.6340 (0.0025) | |
PRS + IGSs + Clinical | 0.7382 (0.0022) | 0.7311 (0.0022) | 0.7397 (0.0022) |
