Fig. 3: Smoothed calibration curve comparing the observed graft survival and the predicted graft survival of each physicians’ and the artificial intelligence-based prediction system. | Communications Medicine

Fig. 3: Smoothed calibration curve comparing the observed graft survival and the predicted graft survival of each physicians’ and the artificial intelligence-based prediction system.

From: Comparison of artificial intelligence and human-based prediction and stratification of the risk of long-term kidney allograft failure

Fig. 3

n = 400 patients, nine transplant physicians and the iBox system. Calibration plots at seven years post risk evaluation for the four hundred patients. Vertical axis is observed proportion of grafts surviving at seven years. Average predicted probability (predicted survival; x-axis) was plotted against Kaplan–Meier estimate (observed overall survival; y-axis). Gray line represents perfectly calibrated model. Except the dark blue line which represents the smoothed iBox predictions, each plot represents Physicians’ predictions. The intercept and slope were 0.135 and 0.87 for the iBox system, 0.48 and 0.46 for Resident #1, 0.00 and 1.14 for Resident #2, 0.31 and 0.78 for Resident #3, 0.47 and 0.51 for Fellow #1, 0.48 and 0.66 for Fellow #2, 0.32 and 0.65 for Fellow #3, 0.22 and 0.83 for Senior #1, 0.33 and 0.79 for Senior #2, 0.55 and 0.44 for Senior #3.

Back to article page