Table 2. Admission Day Model Performance for patients with Acetaminophen-induced Acute Liver Failure.
Admission Model | Variables in Model | Dataset Used (Sample size) | Accuracy (95% CI) | Specificity (95% CI) | Sensitivity (95% CI) | AUROC (95% CI) |
---|---|---|---|---|---|---|
KCC | pH, INR, creatinine, coma grade (low/high) | All Day 1 with non-missing values for all variables (N = 679) | 0.692 (0.656–0.727) | 0.895 (0.864–0.922) | 0.272 (0.214–0.335) | 0.582 (0.551–0.613) |
KCC | pH, INR, creatinine, coma grade (low/high) | Day 1 Training Set (N = 249) | 0.578 (0.514–0.640) | 0.759 (0.624–0.865) | 0.528 (0.456–0.600) | 0.559 (0.528–0.590) |
KCC | pH, INR, creatinine, coma grade (low/high) | Day 1 Test Set (N = 424) | 0.804 (0.763–0.841) | 0.870 (0.831–0.902) | 0.292 (0.170–0.441) | 0.585 (0.554–0.616) |
KCC-CART | INR, creatinine, coma grade (low/high), pH | Day 1 Training Set (N = 288) | 0.722 (0.667–0.773) | 0.715 (0.634–0.787) | 0.729 (0.649–0.800) | 0.740 (0.712–0.767) |
KCC-CART | INR, creatinine, coma grade (low/high), pH | Day 1 Test Set (N = 515) | 0.658 (0.615–0.699) | 0.652 (0.605–0.696) | 0.670 (0.580–0.801) | 0.704 (0.675–0.732) |
NEW-CART | MELD, MV, lactate | Day 1 Training Set (N = 288) | 0.750 (0.696–0.800) | 0.771 (0.693–0.837) | 0.729 (0.649–0.800) | 0.791 (0.764–0.816) |
NEW-CART | MELD, MV, lactate | Day 1 Test Set (N = 515) | 0.718 (0.677–0.757) | 0.710 (0.666–0.752) | 0.767 (0.654–0.858) | 0.755 (0.727–0.781) |
Abbreviations
AUROC: Area under the receiver operator curve
CART: Classification and Regression Tree Analysis
CI: Confidence interval
Coma grade as defined by West Haven Criteria[21]: Low grade ~ Grade I or II, High grade ~ Grade III or IV
KCC: King’s College Criteria
KCC-CART: Classification and Regression Tree analysis using traditional King’s College Criteria Variables
INR: Internationalized Ratio; MELD: Model for End-Stage Liver Disease, MV: mechanical ventilation.
N: Number of patients in sample dataset with outcomes; 95% CI: 95% confidence intervals
NEW-CART: Classification and Regression Tree analysis using new Variables