Table 3 Model choices in INES and possible alternative configurations of a PSM
From: INES: Interactive tool for construction and extrapolation of partitioned survival models
Configuration in INES | Possible alternative modelling choices | Reason for not employing the alternative configuration in INES |
|---|---|---|
The hazard function in the intervention is the same as the control | Allow independent hazard functions in each treatment group | Complexity: Would require the user to choose 4 parametric functions rather than 2 Validation: Predictions made by the model cannot be explained with reference to published HR |
Proportional hazards are assumed | Allow the HR to change over time | Complexity: Would require the user to specify a function for how the HR changes over time (continuously or piecewise) during and after the clinical study. These parameters are rarely estimated in clinical studies |
INES provides the option of estimating the hazard ratios from the observed data. These HR are estimated using the Cox model | Estimate the hazard ratios using the corresponding parametric model (flexsurvreg) rather than the Cox model | Validation: The HR would be different for each parametric model, and might be substantially different to the value published in a clinical study, which is usually also estimated by a Cox model |
The utility weights are constant over time within treatment groups | Allow utility weights to vary over time, for example, to take account of adverse treatment-related events during chemotherapy | Work in progress: This may be considered for a future release of the model |
No probabilistic sensitivity analysis | Allow probabilistic sensitivity analysis, including model averaging of different options for the survival function | Work in progress: This may be considered for a future release of the model |