Performance metrics for the 2-stage classifier based on maximum VAF and binary mutation profiles
| Metric . | Median, lower CI, upper CI, and CI width based on 1000 bootstrap replicates . | |||
|---|---|---|---|---|
| Outer model (myeloid vs no myeloid malignancy) . | Inner model (MDS vs non-MDS malignancy) . | |||
| VAF (17 genes) . | BIN (17 genes) . | VAF (10 genes) . | BIN (7 genes) . | |
| Sensitivity | 0.66, 0.53, 0.76, and 0.23 | 0.65, 0.54, 0.76, and 0.22 | 0.70, 0.45, 0.87, and 0.42 | 0.73, 0.48, 0.9, and 0.42 |
| Specificity | 0.92, 0.88, 0.95, and 0.07 | 0.91, 0.87, 0.94, and 0.07 | 0.66, 0.45, 0.84, and 0.39 | 0.58, 0.33, 0.76, and 0.43 |
| Accuracy | 0.81, 0.77, 0.85, and 0.08 | 0.81, 0.77, 0.84, and 0.07 | 0.67, 0.58, 0.75, and 0.17 | 0.66, 0.57, 0.73, and 0.16 |
| PPV | 0.84, 0.78, 0.89, and 0.11 | 0.82, 0.77, 0.88, and 0.11 | 0.71, 0.61, 0.82, and 0.21 | 0.68, 0.59, 0.77, and 0.18 |
| NPV | 0.8, 0.75, 0.86, and 0.11 | 0.80, 0.74, 0.85, and 0.11 | 0.64, 0.51, 0.77, and 0.26 | 0.63, 0.48, 0.78, and 0.3 |
| F0.5 | 0.79, 0.75,0.83, and 0.08 | 0.78, 0.73, 0.82, and 0.09 | 0.70, 0.62, 0.78, and 0.16 | 0.69, 0.6, 0.76, and 0.16 |
| AUROC | 0.85, 0.82, 0.88, and 0.06 | 0.85, 0.81, 0.88, and 0.07 | 0.73, 0.66, 0.8, and 0.14 | 0.69, 0.61, 0.77, and 0.16 |
| Percent selection MDS∗ | 0.68, 0.57, 0.76, and 0.19 | 0.69, 0.6, 0.77, and 0.17 | n/a | n/a |
| Metric . | Median, lower CI, upper CI, and CI width based on 1000 bootstrap replicates . | |||
|---|---|---|---|---|
| Outer model (myeloid vs no myeloid malignancy) . | Inner model (MDS vs non-MDS malignancy) . | |||
| VAF (17 genes) . | BIN (17 genes) . | VAF (10 genes) . | BIN (7 genes) . | |
| Sensitivity | 0.66, 0.53, 0.76, and 0.23 | 0.65, 0.54, 0.76, and 0.22 | 0.70, 0.45, 0.87, and 0.42 | 0.73, 0.48, 0.9, and 0.42 |
| Specificity | 0.92, 0.88, 0.95, and 0.07 | 0.91, 0.87, 0.94, and 0.07 | 0.66, 0.45, 0.84, and 0.39 | 0.58, 0.33, 0.76, and 0.43 |
| Accuracy | 0.81, 0.77, 0.85, and 0.08 | 0.81, 0.77, 0.84, and 0.07 | 0.67, 0.58, 0.75, and 0.17 | 0.66, 0.57, 0.73, and 0.16 |
| PPV | 0.84, 0.78, 0.89, and 0.11 | 0.82, 0.77, 0.88, and 0.11 | 0.71, 0.61, 0.82, and 0.21 | 0.68, 0.59, 0.77, and 0.18 |
| NPV | 0.8, 0.75, 0.86, and 0.11 | 0.80, 0.74, 0.85, and 0.11 | 0.64, 0.51, 0.77, and 0.26 | 0.63, 0.48, 0.78, and 0.3 |
| F0.5 | 0.79, 0.75,0.83, and 0.08 | 0.78, 0.73, 0.82, and 0.09 | 0.70, 0.62, 0.78, and 0.16 | 0.69, 0.6, 0.76, and 0.16 |
| AUROC | 0.85, 0.82, 0.88, and 0.06 | 0.85, 0.81, 0.88, and 0.07 | 0.73, 0.66, 0.8, and 0.14 | 0.69, 0.61, 0.77, and 0.16 |
| Percent selection MDS∗ | 0.68, 0.57, 0.76, and 0.19 | 0.69, 0.6, 0.77, and 0.17 | n/a | n/a |
BIN, input matrix based on 0 and 1 encoding any variant presence /absence in a gene; VAF, input matrix based on maximum variant allele frequency.
Percent selection of subjects with MDS diagnosed by the outer model as “myeloid malignancy = yes,” which was then used as input for the inner model.