SurvivalBase
The SurvivalBase.jl package is intended to hold internals and shared infrastructure for several other front-facing packages in the JuliaSurv organization, while having the minimal set of dependencies. Please do not use it directly and instead used other front-facing packages and interfaces from the organization.
Right now, the main infrastructure consists of special functions Surv and Strata used in StatsModels.jl's @formulas to be able to express as formulas strandard survival models.
The current implementation only provides bindings for right censored survivals times (with Surv(T,Δ), left-hand-side of formulas) and stratified estimators (with Strata(x), right-hand-side of formulas), but extensions to more complex survival times (e.g. truncation and/or censoring from right and/or left) is planned (note: it would be straightforward and non-breaking).
SurvivalBase.Strata — MethodStrata(x::Symbol)Create a StrataTerm object that holds a variable in the construction of a formula to stratify an estimator.
The behavior of the stratification depends on the fitted estimator: is is usally used on the right hand side of formulas from StatsModels.jl as
@formula(Surv(time,status) ~ Strata(covariate1) + covariate 2)It then has a behavior that unufortunately heavily depends on the estimator, e.g. stratification of a log-rank-type test.
SurvivalBase.Surv — MethodSurv(T::Symbol, Δ::Symbol)Create a SurvTerm object that holds a tuple of variables (T,Δ) in the construction of a formula, usually on the left-hand-side, representing a right-censored output time T with its status indicatrix Δ: Δ being true means that the time was indeed observed, while false means censoring.
This information, usually on the left hand side a formula, has then a behavior that may depend on the estimator. It allows to use formulas from StatsModels.jl as follow:
@formula(Surv(time,status) ~ covariate1 + covariate 2)This usage is common in, e.g., cox models. However, note that the meaning of the formula heavily depends on the model : for hazard regession and maximum likelyhood estimation, this does not mean the same thing.