ToDo

Proper variance for yates survival curve

Robust variance for multi-state Cox survival

Reliability models vignette
Pseudo-values vignette

Update iteration for penalized models wrt the changes in version 2.44.
actually - rewrite the penalized models

add design of the survival library document to vignettes

include.lowest for tcut (first find out what the issue is)

add "fun" to summary.survfit (really want this?)
 
confint method for coxph objects that uses profile likelihood.  (But will it
actually matter?)

deal with variables that have spaces, e.g.,  Surv(time, status) ~ `bad name`

---
Cook and Lawless, Multistate models for life history data, page 104, just above 
section 3.5.2.  State that  
  $\hat \theta - \hat\theta_{(i)} \approx V U_i$
where $V$ is the Cox model variance.
This is my formula for the dfbeta residuals.  Thay also state that
 $LD_i = 2(loglik(\theta) - loglik(\theta_{(i)}) = U'_i V U_i $

This appears to be Han Van H's likelihood displacement residual and would lead
to the cross-validated log-likelhood?   They are talking about parametric
models though.

----------------

Depricate survConcordance.  Issues that may arise:

Functions that use survConcordance:
CatPredi: survConcordance.fit
messina:  survConcordance.fit
distcomp:  ""

These use survConcordance()$concordance
palasso
ePCR
ncvreg
MXM
compound.Cox
survminer

These include my survproto.h, but don't seem to use concordance1/2
KMgene, distcomp, seqMeta

survBootOutliers uses cox_object$concordance with C/(C+D), and does not notice
the issue of strata.  

---------
coxph still returns the Splus style assign.  It's time to give it up.

outside: a covariate with variance=0 in all the risk sets leads to a
	 "contact the author" message