Figure 4l.
Comparison of prediction performance
We then fitted the model to a reduced number of cycles (1-5) and
predicted the respective next cycle toxicity. We also used the
individual parameters estimated by the regression models of clinical
risk factors to predict toxicity of cycles without utilizing any
platelet data under therapy. Results of our outcome
measures {MDDr,k}
and {SDDr,k} 1≤r<k≤6 for the mechanistic and
semi-mechanistic models are displayed in Figure 5. The mechanistic model
shows clearly better predictive power than the semi-mechanistic model
for predicting cycles 2-6. The predictive performance of the parameters
obtained from regression analysis of clinical factors are comparable for
the two models. Prediction performance of the semi-mechanistic model
improves only slightly when adding the cycle data. In contrast, the
prediction performance of the mechanistic model clearly improves when
adding cycle information, especially after adding at least one cycle.
The same applies when considering LDD rather than the DD as primary
measure of prediction performance (see Figure 6).