Version 3.x of the survival package is now available from CRAN. This is a
major update whose primary goal is to make the analysis of multi-state
survival models as easy to do as simple Kaplan-Meier and Cox proportional
hazard fits. The primary changes are
-- in Surv(time1, time2, status) the status variable can now be a
factor that specifies which state a subject has transitioned into. From a
user's point of view this is the single key change.
-- survfit() will then create multi-state Pr(state | time) curves,
i.e., the Aalen-Johansen estimate. (Kaplan-Meier and cumulative incidence
are special cases of the AJ). Printout and plotting are unchanged.
-- coxph() will then create multi-state fits. There is an easy
formula-like syntax for specifying shared coefficients or baseline hazards.
-- all use standard data, i.e. the (time1, time2, status) format that
is common for time-dependent covariates; there is no need to create special
intermediate data sets. (An id variable is necessary to specify which rows
go with whom.)
-- robust variance estimates available throughout
Many other CRAN packages depend on survival, which drove a second goal of
"don't break it". Part of the testing involved running the test
suits of
all 679 reverse dependencies under the new version.
Terry Therneau
therneau at mayo.edu
[[alternative HTML version deleted]]