Empirical methods for the validation of time-to-event mathematical models taking into account uncertainty and variability: application to EGFR + lung adenocarcinoma

Table 2 Results of the various validation methods applied to the full dataset. (PH = proportional hazards)

Method	Metric value (%)	Ratio of of samples where PH assumption is not met (%)
Raw coverage	57.96	NA
Juncture	87.71	NA
Bootstrapped LR (no OTU)	87.68	4.68
Bootstrapped LR (with OTU)	98.14	12.92
Bootstrapped weighted log-rank combo (no OTU)	64.9	4.62
Bootstrapped weighted log-rank combo (with OTU)	87.24	13.28

The acceptability threshold was set at 80%. Given the way all four metrics are defined, the higher the value, the closer the model predictions are to the observed values according to the validation assumptions

ISSN: 1471-2105