| | HaBCS | GeparSepto paclitaxel arm | GeparSepto nab-paclitaxel arm | GeparOcto ETC arm |
|---|
AUC (95% CI)* | 0.827 (0.779 to 0.871) | 0.766 (0.704 to 0.822) | 0.795 (0.746 to 0.840) | 0.754 (0.695 to 0.811) |
Calibration intercept (95% CI)a | 0.30 (− 0.06 to 0.65) | − 0.78 (− 1.10 to − 0.47) | 0.05 (− 0.25 to 0.35) | − 0.27 (− 0.56 to 0.02) |
Calibration slope (95% CI)a | 1.06 (0.79 to 1.33) | 0.68 (0.46 to 0.90) | 0.74 (0.55 to 0.92) | 1.08 (0.70 to 1.46) |
Model update necessaryb | No | Yes | Yes | No |
- HaBCS, hannover breast cancer study; pCR, pathological complete response; AUC, area under the receiver operating characteristic curve
- * The 95% CI was estimated using 10,000 bootstrap samples
- a95% CIs were calculated using regression coefficients and standard errors of the calibration model (logistic regression model)
- b"Yes" means that, according to the specified criteria, the final prediction model needs to be recalibrated before it is applied to treatment arm–like patients. To recalibrate a model, replace the linear term X in the formula for the predicted pCR (see the footnote in Table 3) with the calibration intercept plus X multiplied by the calibration slope. For instance, replace X with − 0.78 + 0.68X for patients treated similarly to those in the GeparSepto paclitaxel arm. The AUC is not affected by recalibration
