Automated and code-free development of a risk calculator using ChatGPT-4 for predicting diabetic retinopathy and macular edema without retinal imaging

Table 4 Prediction performance of developed algorithms and diabetes indicators for diabetic retinopathy

	ROC-AUC [95% CI]	Accuracy (%), [95% CI]	Sensitivity (%), [95% CI]	Specificity (%), [95% CI]	P-value*
Internal validation
LR (ChatGPT-4)	0.747 [0.703, 0.786]	72.0 [67.6, 76.1)	64.2 [53.7, 73.8)	74.1 [69.2, 78.7]	Reference
RF (R)	0.746 [0.703, 0.786]	73.3 [68.9, 77.4]	58.9 [48.3, 68.9]	77.3 [72.5, 81.6]	0.995
GBM (Orange)	0.737 [0.694, 0.778]	68.4 [63.8, 72.7]	71.6 [61.4, 80.3]	67.5 [62.3, 72.4]	0.169
SVM (MATLAB)	0.695 [0.650, 0.738]	64.1 [59.4, 68.5]	68.4 [58.1, 77.6]	62.9 [57.6, 68.0]	0.008
HbA1c	0.660 [0.616, 0.706]	69.5 [65.0, 73.8]	55.8 [45.2, 65.9]	73.2 [68.2, 77.8]	0.002
DM duration	0.682 [0.637, 0.725]	63.4 [58.7, 67.9]	65.2 [54.8, 74.7]	62.9 [57.6, 68.0]	0.004
External validation
LR (ChatGPT-4)	0.786 [0.726, 0.837]	74.4 [68.3, 79.9]	77.2 [62.1, 88.5]]	73.7 [66.7, 79.9]	Reference
RF (R)	0.800 [0.742, 0.851]	74.0 [67.8, 79.6]	81.8 [67.3, 91.8]	72.1 [65.0, 78.5]	0.194
GBM (Orange)	0.771 [0.710, 0.824]	70.9 [64.6, 76.7]	79.5 [64.7, 90.2]	68.8 [61.6, 75.5]	0.168
SVM (MATLAB)	0.714 [0.650, 0.772]	69.6 [63.1, 75.5]	70.4 [54.8, 83.2]	69.4 [62.1, 75.9]	0.051
HbA1c	0.699 [0.635, 0.758]	57.7 [51.0, 64.2]	79.5 [64.7, 90.2]	52.5 [44.9, 59.9]	0.034
DM duration	0.697 [0.632, 0.756]	62.9 [56.3, 69.3]	77.2 [62.1, 88.5]	59.5 [52.1, 66.7]	0.029

CI confidence interval, DM diabetes mellitus, GBM gradient boosting machine, LR logistic regression, ROC-AUC area under the receiver operating characteristic curve, SVM support vector machine
^*Differences in ROC-AUC values compared to the logistic regression performed using ChatGPT-4

ISSN: 2056-9920