Two sequential stages. The first derives a baseline probability from age, sex, hemoglobin, hematocrit, white blood cell count and platelet count, with white blood cells, platelets and age modelled as restricted cubic splines. The second updates that probability with the erythropoietin category — low (<4.3 mIU/mL), normal (reference) or high (>29 mIU/mL); a normal value leaves the baseline unchanged. Scoring runs on the server and the model coefficients are not present in this page.
Leave-one-centre-out internal–external validation: each centre was held out in turn, with the model, penalisation and threshold derived only from the remaining centres. Development cohort n=1,258, 94 JAK2-positive, four centres.
| Measure | Value |
|---|---|
| AUROC (95% CI) | 0.971 (0.953–0.989) |
| Calibration slope / intercept | 0.958 / +0.039 |
| Brier score | 0.024 |
| Sensitivity | 98.9% — 1 missed |
| Specificity | 67.7% |
| JAK2 testing rate | 37.3% |
| Tests avoided per 1,000 patients | 627 |
| PPV / NPV | 19.8% / 99.87% |