Table 1.

Comparison of Selected Extrapolation Methods

Method	Endpoints Modelled	Description and Clinical Assumptions	Strengths	Limitations
Parametric survival modelling	OS	No other clinical assumption in addition to the shape of the hazard of death function	Easy to communicate	Relies on clinical assumptions Limited by parametric form Heterogeneity of the population not considered
Spline models	OS	Introduce flexibility to a distribution’s shape by using “knots” to indicate moments where the features of the distribution changes. Position of knots determines survival	Highly flexible	Timepoints and location of knots are not necessarily clinically meaningful
Cure models	OS	Assume that a fraction of the population never experiences the event of interest (eg, OS, PFS). A proportion of the population is assumed to not be associated with a disease-specific risk of event. Mixture cure models consider this fraction of the population (π), while non-mixture cure models do not differentiate these groups; non-mixture models define the point where cause-specific mortality is zero as time moves towards infinity (an asymptote) for the survival function	Both mixture cure and non-mixture cure models can be applied to different sub-populations within a trial with different survival profiles. Heterogeneity of the population considered Based on clinical assumption that can easily be validated/communicated	Medium-term health status determines long-term outcome
Landmark analysis	OS from landmark timepoint	Survival of patients is dependent on their response status at a landmark point and can also account for differences in patient survival by treatment	Heterogeneity of the population considered Based on clinical assumption that can easily be validated/communicated	Short-term health status determines long-term survival
Markov models	PFS, time to progression, post-progression survival	Progression and PPS, whenever it occurs, determines survival. OS is obtained indirectly based on probabilities calculated using the pre-progression survival, TTP, and PPS	Clinical events explicitly and structurally related	Competing risks or multi-state modelling models to estimate transition probabilities (more complex) Challenge to model post progression survival due to dependent censoring Challenging to achieve reasonable fit to OS

Abbreviations: OS, Overall survival; PFS, Progression-free survival; PPS, post-progression survival; TTP, Time to progression.