Table 4.
Prediction performance measures of the final selected models
| Model specification | LR, p-value | BCV AUROC (97.5% CI) | BCV Brier score (%) | R2 (%) |
|---|---|---|---|---|
| FMNS | 25.91, 0.0176 | 57.49(45.38–68.38) | 12.10 | 1.00 |
| SCOV | 31.08, 0.0175 | 58.38(45.38–68.72) | 12.00 | 2.00 |
| NBC | 22.41, 0.0077 | 61.51(48.87–70.71) | 11.72 | 4.00 |
FMNS = PMLE.model + logIgGAMA1 + logIgG1AMA1 + dominant c.233 with 5 knots restricted cubic spline, Model SCOV = Slope Corrected final model with van Houwelingen-Le Cessie heuristic estimate
NBC = No baseline variable included: only dominant c.233 + logIgGAMA1 + logIgG1AMA1
LR, BCV, RMSE, R2, AUROC, BS, represent the Likelihood Ratio test statistic, bootstrap cross-validation, Root Mean Squared Error, the proportion of explained variation in predicted risk and Area Under the Receiver Operating Characteristic curve, Brier score, respectively