Table 3.
Diagnostic statistics for clinical predication models
| Model fixed at 90% specificity | Model fixed at maximum Youden’s Index* | |||||||||
|---|---|---|---|---|---|---|---|---|---|---|
| AUROC (95% CI) | Sens | PPV† | NPV‡ | Sens | Spec | PPV† | NPV‡ | |||
| Outcome: Histologic Improvement | ||||||||||
| Clinical Model | 0.89 (0.82 – 0.95) | 0.67 | 0.74 | 0.87 | 0.84 | 0.84 | 0.69 | 0.92 | ||
| Cross-validated Model | 0.85 (0.77 – 0.92) | 0.65 | 0.74 | 0.86 | 0.81 | 0.80 | 0.64 | 0.91 | ||
| Model validated using TONIC dataset | 0.84 (0.77 – 0.91) | 0.56 | 0.79 | 0.77 | 0.85 | 0.72 | 0.65 | 0.89 | ||
| Clinical Model: log(P/1-P)= −8.64 – 0.06 × baseline GGT (U/L) −0.08 × change in GGT (U/L) + 1.84 × lobular inflammation grade + 0.02 × baseline LDL (mg/dL) + 0.008 × baseline alkaline phosphatase (U/L) + 1.48 if hypertensive + 0.20 × age (years) | ||||||||||
| Outcome: Resolution of Zone 1, periportal pattern | ||||||||||
| Clinical Model | 0.91 (0.83 – 0.99) | 0.42 | 0.80 | 0.69 | 0.95 | 0.82 | 0.78 | 0.96 | ||
| Cross-validated Model | 0.83 (0.71 – 0.95) | 0.32 | 0.75 | 0.66 | 0.79 | 0.82 | 0.75 | 0.85 | ||
| Model validated using TONIC dataset | 0.92 (0.81 – 1.00) | 0.10 | 0.67 | 0.46 | 1.00 | 0.82 | 0.88 | 1.00 | ||
| Clinical model: log(P/1-P)= 6.23 – 0.05 × baseline ALT (U/L) – 0.07 × change in ALT (U/L) − 0.06 × baseline LDL (mg/dL) + 3.10 if white | ||||||||||
| Outcome: Fibrosis Improvement | ||||||||||
| Clinical Model | 0.89 (0.83 – 0.94) | 0.63 | 0.77 | 0.83 | 0.90 | 0.77 | 0.65 | 0.94 | ||
| Cross-validated Model | 0.85 (0.78 – 0.91) | 0.56 | 0.75 | 0.81 | 0.94 | 0.64 | 0.56 | 0.96 | ||
| Model validated using TONIC dataset | 0.82 (0.75 – 0.89) | 0.43 | 0.74 | 0.72 | 0.87 | 0.69 | 0.64 | 0.90 | ||
| Clinical model: log(P/1-P)= 0.09 – 0.03 × baseline ALT (U/L) – 0.02 × change in ALT (U/L) + 1.71 × baseline fibrosis stage – 1.16 × baseline BMI z-score – 0.68 × baseline ballooning score + 0.92 × baseline lobular inflammation score + 0.95 if white | ||||||||||
Youden’s Index=sensitivity+specificity-1
PPV = Positive Predictive Value: the probability that the disease is present when the test is positive
NPV = Negative Predictive Value: the probability that the disease is not present when the test is negative.
In CyNCh, 43/146 (29%) had histologic improvement; in TONIC 56/147 (38%) had histologic improvement.
In CyNCh, 19/46 (41%) had resolution of borderline Zone 1 steatohepatitis; in TONIC 22/40 (55%) had resolution of borderline, zone 1 steatohepatitis.
In CyNCh, 48/146 (33%) had fibrosis improvement; in TONIC 58/146 (40%) had fibrosis improvement.
TONIC clinical model for P, probability of histologic improvement: log(P/1-P)= −1.22 – 0.04 × baseline GGT (U/L) −0.06 × change in GGT (U/L) + 0.54 × lobular inflammation grade + 0.008 × baseline LDL (mg/dL) + 0.007 × baseline alkaline phosphatase (U/L) – 0.46 if hypertensive − 0.13 × age (years)
TONIC clinical model for P, probability of resolution of zone 1, periportal pattern: log(P/1-P)=5.88 – 0.09 × baseline ALT (U/L) – 0.08 × change in ALT (U/L) + 0.005 × baseline LDL (mg/dL) – 1.56 if white
TONIC clinical model for P, probability of fibrosis improvement: log(P/1-P)= +0.79 – 0.02 × baseline ALT (U/L) – 0.01 × change in ALT (U/L) + 1.13 × baseline fibrosis stage – 0.37 × baseline BMI z-score – 0.35 × baseline ballooning score + 0.47 × baseline lobular inflammation score – 1.30 if whit