Parametric proportional hazards fitting with left truncation and
right censoring for common families of distributions, piecewise constant
hazards, and discrete models. Parametric accelerated failure time models
for left truncated and right censored data. Proportional hazards
models for tabular and register data. Sampling of risk sets in Cox
regression, selections in the Lexis diagram, bootstrapping.
Broström (2012)
New function: mpch gives the mean of a piecewise constant hazards distribution.
The function plot.Surv is removed, since it now is a plot method from the survival package, which eha depends on. Note though that this implies a new set of arguments. See the help page of survival:::plot.survfit.
New versioning system (Hadley Wickham's suggestion).
Documentation 'roxygenized'.
Development moved to gitHub.
The package glmmML was merged into eha in version 2.0. This has turned out to be a very bad idea ("double entry"), and all relevant functions are now Defunct and removed. Use glmmML instead.
glmmML()
, glmmboot()
, ghq()
, glmmbootFit()
, glmmML.fit()
are all Defunct and removed. No reverse dependency used any of them. Those functions are all found in the package glmmML, on CRAN.
phreg()
: Start values by 'coxreg' (bad idea) removed.
aftreg()
: in 'addMeans' and 'aftreg1', an error
(spotted by Jingchunzi Shi) with fixed scale corrected.
init.c: Added after R-3.4.0.
coxreg()
: "Geometric bootstrap" removed (it never worked and was
never used).
cal.window()
, `age.window(: Works now with 'tibbles'.
rpch: Bug fixed (spotted by Brigitte Mueller).
plot.phreg: parameters 'changeLog' and '...' enhanced. Documentation updated.
plot.coxreg: Bug concerning 'printLegend' fixed.
weibreg: Note added in documentation: "Use phreg instead". weibreg will soon be deprecated.
coxreg: Changed the default for 'hazards' to TRUE.
getHaz: Modified to handle missing strats and score.
print.risksets: Added '...' to argument list.
coxreg: Fitting a NULL model didn't produce enough information like log likelihood ('loglik'), for comparison of models. Fixed.
glmmboot: 'frail' set to Inf or -Inf for appropriate clusters.
plot.hazdata: Updated to acknowledge the argument 'hazards' that was included in 'coxreg' in version 2.4-2.
plot.coxreg: Updated to acknowledge the argument 'hazards' that was included in 'coxreg' in version 2.4-2. Errors in the manual page corrected.
risksets: Risk set sampling did not always work (memory leak in C code); fixed.
risksets: New value in the Value list: sample_fraction.
plot.aftreg: Now registered as an S3 method.
print.aftreg: Scale and shape parameters are only printed on the log scale.
src: 'abs' corrected to 'fabs' in two places in C code.
coxreg: Argument 'hazards' added, with default = FALSE.
plot.Surv: Argument 'printLegend' added.
coxreg: Changed labels in case of ordered factors.
Foreign calls: 'DUP = FALSE' removed everywhere.
print.pch: Bug preventing printing(!) fixed.
plot.pch: Bug preventing plotting(!) fixed.
coxreg: Call to survival::[agreg,coxph].fit in "standard" cases reintroduced; Thanks to Terry Therneau, who exports these functions from survival_2.37-6. So this version of 'eha' requires survival (>= 2.37-6). Argument 'center' reintroduced, with slightly different meaning.
plot.coxreg: argument 'newdata' not used.
phreg: Stratified analysis in the piecewise constant hazards model introduced. Argument 'center' reintroduced, with slightly different meaning.
plot.phreg: Argument 'newdata' not used.
plot.coxreg: added the argument col, for color line(s).
plot.hazdata: Ditto.
coxreg: Restoring equity.
glmmML: Removed spurious printing in 'glmmml.c'.
coxreg: 'Downgraded' due to limitations in using non-exported function (agreg) from 'survival'. This will be taken care of ASAP. Meanwhile, live with slightly slower 'coxreg' and choking on huge data sets.
Coxreg: Removed for the same reason as the downgrade of coxreg.
phreg: The lognormal and loglogistic distributions have an intercept added to the linear predictor; in effect extending these distributions to three-parameter families. A bug (found by Claire Williams) affecting the covariance matrix of the parameter estimates has been eliminated.
aftreg: The same bug as for phreg eliminated. The available parametrizations are changed to 'lifeAcc' and 'lifeExp'. For the Gompertz distribution, the 'canonical' representation is used regardless.
hlnorm, Hlnorm, hllogis, Hllogis: New parameter, 'prop', added. It simply multiplies the previous definitions by its value.
age.window, cal.window: speed up and handling of empty intervals
perstat: speed up.
plot.phreg, plot.coxreg: legend added for stratified fits.
Coxreg: New function designed to handle huge data sets. coxreg chokes on 'integer overflow', and memory overflow, when the collection of risk sets is created and saved. Coxreg is essentially a wrapper for 'coxph' in the 'survival' package, but with methods for nice(r) printing.
coxreg: Error in the calculation of 'total time at risk' fixed (gave negative values when no left truncation!).
phreg: The gompertz 'scale' parameter is now a real scale parameter through a reparametrization of the baseline gomperz distribution. It follows now the 'canonical' parametrization. There is also a new argument, 'param', which can take the values "canonical" (default) and "rate". Better start values for the Gompertz regressions.
check.dist : Has failed due to different scaling in coxreg and phreg since 2.2-2. Now fixed.
risksets: Added an argument, 'members', a logical with default value 'TRUE'. If TRUE, the members of all risk sets are returned. With huge data sets, this may be too much, and not wanted. In such cases, set 'members = FALSE'.
coxreg: Memory leaks fixed in getHaz.f, sizes.c and risk.c.
Gompertz distribution: new parameter 'param'; values 'default' (default, nothing changes), and 'canonical', which gives meaning 'PH' and 'AFT' to the shape and scale parameters, respectively.
aftreg: New input argument: 'param', with three values, 'default' (which is the default;), 'survreg', and 'canonical'. The value 'default' uses the standard parametrization, while 'survreg' uses the same parametrization as the survreg function in the survival package. 'canonical' is especially useful in the Gompertz distribution for separating PH and AFT parameters.
C code cleaned (avoiding some Warnings at CRAN).
check.dist: New argument, col, added. Default is now black-and-white plotting.
ltx: New function for printing coxreg fits in latex format.
plot.Surv: New arguments, col and lty, added. The default is now black-and-white plotting.
data: Two new data sets, "logrye" and "scania".
oldmort: Data set updated.
aftreg.fit: Errors in documentation corrected.
SurvSplit: Bug fixed. It works now.
glmmboot: Bug in calculation of 'frail' fixed.
plot.phreg: Bug in argument 'ylim' fixed. Bug with stratified fits fixed.
aftreg, phreg: summary method added (just printing).
glmmML: Merged into eha. This adds binary and count data regression for clustered data to eha. glmmML is still a standalone package on CRAN, for how long remains to be seen.
Pch: New class of distributions; Piecewise constant hazards (Pch).
phreg: piecewise constant baseline hazard function introduced.
phreg: 'center' reinstalled as a logical, default = TRUE, so nothing changes if 'center' is left untouched.
age.window: If the selected window is empty, an error was previously thrown. Now NULL is returned, with a warning.
cal.window: Now returns NULL, with a warning, if the selected window is empty.
aftreg: Gompertz distribution added (back again). Better start values.
phreg: Better start values for Gompertz distribution.
eha: Depends (R >= 2.13.0)
phreg, aftreg, weibreg: Errors in documentation corrected.
ChangeLog: Errors in 1.3-6 corrected.
coxph, phreg, aftreg: Bug fix: Logical covariates now OK. (converted to factor).
phreg: Removed "coxph" from "class".
aftreg: Ditto.
weibreg: Ditto.
oldmort: New data set.
coxreg: added return values 'df' and 'n'.
coxreg, phreg, aftreg: Added extractAIC and nobs methods.
toBinary: Now includes also risksets without survivors (if any).
fert: Row names changed to row numbers.
swe07: New data set.
Spelling errors (some!) corrected in documentation.
male.mortality: Old name (male.mortality) of data frame mort reinstalled. Now known under two names. Overlapping spells no longer exist.
fert: Documentation corrected.
toDate: Now works as intended.
* dllogis: Bug fixed.
* hllogis: Bux fixed.
phreg: Fixed error and (some) numerical instability in the estimation of the Gompertz proportional hazards model (seems to be very dependent on good start values!).
plot.phreg: Fixed 'ylim' for cumulative hazards plot.
male.mortality: Data set renamed to
mort: The new name. (Just for convenience!)
coxreg: Major revision, calls 'coxph' in 'survival' (thanks to Terry Therneau!) for ordinary requests. Reason: speed. The argument 'center' is deprecated, centering of covariates is now routinely made. Affects the estimation of the baseline hazard, which is done at the means of the columns of the original design matrix.
phreg: Argument 'center' is deprecated. Centering is routinely performed, so the baseline hazard is estimated at the means of the columns of the original design matrix.
aftreg: Corrected error in documentation and made the argument 'center' deprecated.
aftreg: Estimation of intercept ("scale") is now comparable with corresponding estimate of Intercept in 'survreg'. 'shape' in aftreg is still '1 / scale' in survreg. Regression coefficients still have different signs but the same absolute values.
join.spells: a few bugs fixed, now allows for factor covariates.
male.mortality: data set "joined".
fert: is a new data set. Contains data on 19th century marital fertility from a parish in northern Sweden.
aftreg.fit: Check of data consistency in case of more than one record per individual added.
aftreg.fit: Fixed memory allocation bug in C code. This bug was not fatal (too much memory was allocated), but caused a severe performance problem, especially on Windows.
aftreg: 'dist = "gompertz"' should now work (realizing that it is a special case of 'dist = "EV").
phreg: Fixed bugs for the cases with fixed shape (includes 'dist = "gompertz"'.
aftreg: 'offset' works now.
phreg: Ditto.
check.dist: Bug fixed for 'dist = "gompertz"' (again!)
plot.phreg: Ditto.
aftreg: argument 'id' now works as intended, i.e., is looked for in 'data' (or 'parent.frame()')
print: functions return invisibly 'x'.
coxreg: argument 'weights' now works as intended, i.e., is looked for in 'data' (or 'parent.frame()')
check.dist: Bad 'ylim' fixed.
aftreg: Gompertz distribution had a bug; now fixed.
coxreg: Added (time-varying) weighted Cox regression. Returns error code 1 if failure, with a warning.
check.dist: Can now handle stratified models.
Bugs in qEV, rEV, all Gompertz functions, plus documentation of EV and Gompertz fixed. Note that these bugs did not affect aftreg or phreg! However,
check.dist had to be revised due to the above bugs.
Bug in coxreg, method = "ml" fixed.
Bug in coxreg, calculation of hazard, fixed.
Several errors in the documentation fixed.
* Added 'risktime' to toBinary output.
Added full profiling to the "ml" and "mppl" methods in coxreg.
mlreg is now deprecated. Use coxreg with method = ml or = mppl
table.events: New parameter: 'strict = TRUE'.
make.communal: In case of 'communal=FALSE', spells are no longer truncated. Instead a value of NA is given if birthdate is out of range. Previous behavior was unpredictable when birthdate was out of range.
mlreg: Geometric distribution (i.e., constant baseline discrete hazard) added. Not for frailty models, yet (on the TODO list).
mlreg: New argument, 'n.points', added to 'control'. Controls the number of points used in the Gauss-Hermite quadrature.
mlreg: Stricter control of numerical problems, especially in the frailty fit.
clean: Replaced by check.spells and join.spells.
Return values changed to conform with R-1.8.0 (and later).