Table 4 Performance of both models in the validation data set
Royal Marsden Model | Model derived from the decision tree analysis | |
|---|---|---|
Definition of high-risk patients | Patients with at least two of the following prognostic factors (albumin <35 g l−1, LDH > ULN and more than two metastatic sites) | Patients with albumin <33 g l−1 or patients with albumin ⩾33 g l−1 and with platelet counts ⩾400.000 mm−3 |
Evaluable patients | 172 (56.2%) | 172 (56.2%) |
Rate of early death in low-risk group | 1/103 (1.0% (0.1–6.0)) | 6/133 (4.5% (1.8–10.0)) |
Rate of early death in high-risk group | 14/69 (20.3% (12.0–32.0)) | 9/39 (23.1% (11.7–39.7)) |
Sensitivity | 0.93 (0.66–1.00) | 0.60 (0.33–0.82) |
Specificity | 0.65 (0.57–0.72) | 0.81 (0.73–0.86) |
Positive predictive value | 0.20 (0.12–0.32) | 0.23 (0.12–0.40) |
Negative predictive value | 0.99 (0.94–1.00) | 0.95 (0.90–0.98) |
Rate of well-classified patients | 0.67 (0.60–0.74) | 0.79 (0.73–0.85) |
Discriminative slope | 19.3% (9.2–29.4) | 18.0% (4.0–32.0) |
Brier score | 0.098 | 0.010 |
