survkit

A unified, extensible toolkit for time-to-event analysis: one interface over the established R survival stack, with cross-method triangulation as the consistency check on top.

survkit puts one clean interface over the established R survival stack — parametric accelerated-failure-time models, Cox proportional hazards, competing risks, recurrent events, frailty, flexible splines, and a non-parametric Kaplan-Meier baseline — driven by an extensible method registry so new estimators slot in without an API change. Every fit returns a single typed object with consistent prediction, summary and tidying methods.

On top of the fits sits the value layer that makes survkit a methodology rather than a wrapper:

Cross-method triangulation (survkit_triangulate()) that reads disagreement between methods as a model-misspecification signal.
Decision-relevant estimands — restricted mean survival time (survkit_rmst()) and cumulative incidence (survkit_cuminc()).
Predictive scoring (survkit_score()) — concordance and the censoring-weighted Brier score with its index of prediction accuracy.
An events-per-variable power gate that flags models too small to identify.
A registry conformance contract (survkit_validate_method()).
An optional ensemble-manifest emitter for composition with a wider analytics stack.

It stands alone — siblings are optional (Suggests, behind requireNamespace() guards) and R CMD check is clean with none installed.

Install

# from a local clone
devtools::install("survkit")

survival is the only modelling dependency; flexsurv, cmprsk, coxme, flexsurvcure and icenReg are optional backends, pulled only when a method needs them.

Quick start

library(survkit)

# what can it fit?
survkit_methods()

# a Weibull accelerated-failure-time model
fit <- survkit(survival::Surv(time, status) ~ age + sex, survival::lung,
               method = "weibull")
fit
survkit_tidy(fit)                      # estimates, ratios, CIs
survkit_curve(fit, times = c(100, 300, 500))

# Cox proportional hazards on the same data
cox <- survkit(survival::Surv(time, status) ~ age + sex, survival::lung,
               method = "cox")

# competing risks (Fine-Gray subdistribution hazard)
# event is a factor: first level = censoring
mg <- survival::mgus2
mg$etime <- with(mg, ifelse(pstat == 1, ptime, futime))
mg$event <- with(mg, factor(ifelse(pstat == 1, "pcm",
                            ifelse(death == 1, "death", "censor")),
                            levels = c("censor", "pcm", "death")))
survkit(survival::Surv(etime, event) ~ age + sex, mg, method = "fine_gray",
        cause = "pcm")

The small-sample gate

Survival models identify from events, not rows. survkit() runs an events-per-variable check on every fit and warns when a model is under-powered, so you can simplify, penalise, or abstain rather than trust over-fit estimates:

small_data <- survival::lung[1:25, ]
survkit_power(survival::Surv(time, status) ~ age + sex + ph.ecog, small_data)
#> $verdict        "underpowered"
#> $events_per_variable  7.3

Triangulate across methods

Fit the same question under several methods and let their disagreement flag a model worth a closer look:

survkit_triangulate(survival::Surv(time, status) ~ age + sex, survival::lung,
                    methods = c("weibull", "cox", "km"), tau = 500)
#> <survkit_triangulation> methods: weibull, cox, km
#>   restricted mean at tau = 500
#>    method tau     rmst
#>   weibull 500 311.6
#>       cox 500 312.7
#>        km 500 310.3
#>   verdict: agreement -- methods concur within threshold

Decision-relevant summaries

fit <- survkit(survival::Surv(time, status) ~ age, survival::lung, method = "km")
survkit_rmst(fit, tau = 365)                 # restricted mean survival time
survkit_score(cox, survival::lung, times = c(200, 400, 600))  # C-index, Brier, IPA

Extending it

A new estimator is one registry call — no change to survkit():

survkit_register("my_method", kind = "parametric", backend = "somepkg",
                 fit = function(formula, data, ...) {
                   raw <- somepkg::fit(formula, data, ...)
                   list(raw = raw, coef = coef(raw), vcov = vcov(raw),
                        loglik = as.numeric(logLik(raw)),
                        n = nrow(data), n_events = NA_integer_,
                        kind = "parametric", backend = "somepkg")
                 })

Contributing

Bug reports and suggestions are welcome through the issue tracker.

Citation

citation("survkit")