Figure 5l.
Of note, the mechanistic model is still more reliable when the
semi-mechanistic model is calibrated on more cycles. For example, the
mechanistic model fitted on the first cycle results in better prediction
of toxicities in cycles 4-6 than the SMM calibrated on all cycles prior
to the predicted one (see Figure 5). The superiority of the mechanistic
model is considerably larger when considering LDD statistics rather than
DD statistics (see Figure 6). Only the parameter values derived from the
regression of clinical factors resulted in comparable prediction
performance of the two models. All scenarios using cycle data resulted
in superior performance of the mechanistic model except for the highly
unbalanced situation where the SMM was calibrated on five cycles while
the mechanistic model was calibrated on only one cycle to predict the
toxicity of the last cycle.