Figure 6l.
In order to check, whether the selection of patients used for the primary data set impose a bias on the resulting parameter distribution, we fitted both models to the sparser data of additional 146 NHL patients fulfilling weaker inclusion criteria. Table S.1.2 in Supporting Information S.1 provides resulting individual parameters distributions. Results are compared with the corresponding distributions of the primary data set using Welch’s t-test (unequal sample size, unequal variances). No significant differences were detected for the parameters of both models indicating that bias caused by patient selection is small.
We analysed whether clinical risk factors are related to our individual parameter estimates derived from fitting all cycles (Supporting information S.5). Several significant associations were found. The pharmacodynamics effect of cyclophosphamide showed the strongest dependence on clinical factors. Most of these dependences are consistently observed for both data sets.
We finally determined sensitivity as well as the Youden index of both models to predict severe grade 3-4 thrombopoenia at the next cycle. The mechanistic model showed superior sensitivity (Figure 7) prediction performance throughout (Figure 8).