Longitudinal computed tomography-based delta-radiomics of visceral adipose tissue predicts infliximab secondary loss of response in Crohn’s disease patients

Table 3 Performance of models for predicting secondary loss of response in the training and validation cohorts

	AUC (95%CI)	Sensitivity	Specificity	PPV	NPV	P alue1	Accuracy
Training cohort (n = 112)
Clinical model	0.767 (0.678-0.857)	85.1% (63/74)	36.8% (14/38)	72.4% (63/87)	56.0% (14/25)	-	68.8% (77/112)
Radiomics model	0.816 (0.737-0.896)	86.5% (64/74)	47.4% (18/38)	76.2% (64/84)	64.3% (18/28)	0.133	73.2% (82/112)
Combined model	0.853 (0.786-0.921)	85.1% (63/74)	52.6% (20/38)	77.8% (63/81)	64.5% (20/31)	0.023	74.1% (83/112)
Validation cohort (n = 49)
Clinical model	0.707 (0.539-0.876)	96.97% (32/33)	31.3% (5/16)	74.4% (32/43)	83.3% (5/6)	-	79.9% (37/49)
Radiomics model	0.750 (0.605-0.895)	84.8% (28/33)	50.0% (8/16)	77.8% (28/36)	61.5% (8/13)	0.204	73.5% (36/49)
Combined model	0.812 (0.677-0.948)	93.9% (31/33)	50.0% (8/16)	79.5% (31/39)	80.0% (8/10)	0.048	79.6% (39/49)

¹P value indicates the significance level of the comparison of areas under the curves with the clinical model as the reference using DeLong test in the training and validation cohort, respectively.

SLR: Secondary loss of response AUC: Area under the curve NPV: Negative predictive value PPV: Positive predictive value.