The following several classes of frailty models using a penalized likelihood estimation on the hazard function but also a parametric estimation can be fit using this R package: 1) A shared frailty model (with gamma or log-normal frailty distribution) and Cox proportional hazard model. Clustered and recurrent survival times can be studied. 2) Additive frailty models for proportional hazard models with two correlated random effects (intercept random effect with random slope). 3) Nested frailty models for hierarchically clustered data (with 2 levels of clustering) by including two iid gamma random effects. 4) Joint frailty models in the context of the joint modelling for recurrent events with terminal event for clustered data or not. A joint frailty model for two semi-competing risks and clustered data is also proposed. 5) Joint general frailty models in the context of the joint modelling for recurrent events with terminal event data with two independent frailty terms. 6) Joint Nested frailty models in the context of the joint modelling for recurrent events with terminal event, for hierarchically clustered data (with two levels of clustering) by including two iid gamma random effects. 7) Multivariate joint frailty models for two types of recurrent events and a terminal event. 8) Joint models for longitudinal data and a terminal event. 9) Trivariate joint models for longitudinal data, recurrent events and a terminal event. Prediction values are available (for a terminal event or for a new recurrent event). Left-truncated (not for Joint model), right-censored data, interval-censored data (only for Cox proportional hazard and shared frailty model) and strata are allowed. In each model, the random effects have the gamma or normal distribution. Now, you can also consider time-varying covariates effects in Cox, shared and joint frailty models (1-5). The package includes concordance measures for Cox proportional hazards models and for shared frailty models.
Changes in Version 2.12.6 October 2017
o For joint nested and nested models martingale residuals are calculated at the lower level of clustering (individual level)
Changes in Version 2.12.3 July 2017
o Bug fixed in predictions for trivariate joint models.
Changes in Version 2.12.2 June 2017
o Bug fixed in predictions for joint nested frailty models.
Changes in Version 2.12.0 June 2017
o New model added: trivariate joint model for a longitudinal biomarker, recurrent events and a terminal event using a mechanistic approach for the biomarker (with analytical solution)
o Extension for functions longiPenal and trivPenal: up to three random effects for the biomarker can be applied
o Bug fixed : Estimation error in joint, joint general, joint nested, trivPenal and longiPenal models when data are unordered.
o Bug fixed for predictions using joint nested models - application of Gamma function
Changes in Version 2.11.1 March 2017
o Warnings about native subroutine registry fixed
Changes in Version 2.11.0 March 2017
o NEW : Marginal prediction method in the Joint Nested Frailty model, only for terminal event.
Changes in Version 2.10.6 March 2017
o Bug fixed in the conditional prediction for shared frailty model.
o New warning for the prediction in joint general model.
Changes in Version 2.10.5 February 2017
o Bug fixed : use of a subcluster/cluster covariate named with upper case in joint and joint nested models
o Bug fixed : Joint nested model estimation
o Bug fixed : Conditional prediction for recurrent events from a shared model.
o Bug fixed : examples of the documentation (plot of epoce, additive model, trivariate and joint nested model)
Changes in Version 2.10.4 January 2017
o Changes in the joint nested frailty model : add calculation of the bayesian frailties estimates (for families and for individuals)
o Problem fixed : survival() function in frailtypack can now be computed with a gamma shared frailty model with a piecewise baseline hazard function
o Changes in the prediction() function : 'group' argument removed and 'conditional' boolean argument added
o Changes in the conditional prediction method for shared modelling : possibility to compute prediction for more than one group
o Display of prediction's results : for each indivual you can see the true ID
o Problem fixed : prediction() function is now able to compute with disorderly individuals
o Changes in the prediction method : you can now use the prediction methods with time-dependant covariates
o NEW: Marginal prediction method in the shared modelling, for a recurrent event.
o Correction applied on the mathematical expression of prediction method from shared model
Changes in Version 2.10.3 October 2016
o "na.pass" global function defined in the NAMESPACE file
o Update of the vignettes 'Package_summary.Rmd' in the 'inst/doc' directory
Changes in Version 2.10.2 October 2016
o Vignettes modified (legend in the title)
o 'event' legend deleted in the plot of a shared model
o Compiling warnings fixed
o Plot bug fixed in the plot of a shared model
Changes in Version 2.10.1 July 2016
o New prediction option for a new recurrent event.
o Bug fixed for the gfortran compilation
Changes in Version 2.9.4 July 2016
o Bug fixed for the vignettes builder
Changes in Version 2.9.3 July 2016
o New model added : Joint Nested frailty model for recurrent (with two clustering levels) and terminal events, accounts for two frailty terms.
o For all the plot methods of frailtypack : addition of 'Xlab' and 'Ylab' (labels for the X-axis and Y-axis)
o Warning added if left truncation with joint frailty model
o Warning added for the use of interval-censored data in joint frailty model, the option is not available for the model
o New option "initialize = TRUE" for fitting a joint frailty model to provide new initial values, before fitting the joint nested model.
o Bug fixed in models prediction with formulas defined separately
o Bug fixed for trivPenal and longiPenal for definition of individuals identificators
o Bug fixed for global Wald test for qualitative covariates in the nested and the joint frailty model
Changes in version 2.8.3 January 2016
o Bug fixed for predictions for frailty models
o Bug fixed for calculation of residuals for longitudinal biomarker in bivariate and trivariate models
Changes in version 2.8.2 December 2015
o Description of the different models and options in Frailtypack using a vignette ("Package_summary")
Changes in version 2.8 November 2015
o New models added: joint model for longitudinal data and a terminal event (longiPenal function) and trivariate joint model for longitudinal data, recurrent events and a terminal event (trivPenal function)
o For these models summary, print and plot methods are available as well as functions epoce, Diffepoce and predictions were adapted
o Functions form altered: all the character options start with a capital letter, eg. was: plot(x, type.plot = "hazard") is now plot(x, type.plot = "Hazard")
o Joint frailty models for clustered data now are modelled in a framework of semi-competing risks (the parameter alpha is not recommended in these semi-competing models)
o Interactions are now available for all the models (using "*" or ":")
Changes in version 2.7.6 August 2015
o New model added: Joint General frailty model for recurrent and terminal events with 2 covariates
Changes in Version 2.7.5 March 2015
o Bug fixed for Martingale residuals (in shared and joint models with log normal frailties)
Changes in Version 2.7.3 February 2015
o Prediction and Monte Carlo confidence bands added for shared and joint gaussian frailty models.
o Bug fixed for the prediction function with shared or Cox models (reading of survival times)
o Bug fixed for plotting the baseline hazard and survival functions in Weibull shared and joint models
o New functions to compute estimators of Expected Prognostic Observed Cross-Entropy (EPOCE) evaluating prediction accuracy in joint gaussian frailty models.
Changes in Version 2.7.1 October 2014
o Bug fixed for the multivariate Wald test for covariates with more than 3 categories.
o Bug fixed for EPOCE, definition of kappa.
Changes in Version 2.7 August 2014
o In 'frailPenal' and 'additivePenal' functions, no more 'kappa1', 'kappa2', 'nb.int1' and 'nb.int2'. Replaced by two vectors 'kappa' and 'nb.int'.
o More levels of stratification (up to 6) for shared frailty model.
o Now possible stratification in a joint frailty model for the recurrent event part (up to 6 levels).
o New construction of the dataframe when using 'prediction' function on a joint frailty model. Need now the event indicator variable.
Changes in Version 2.6.1 July 2014
o Different way to do Monte-Carlo method to compute confidence intervals in 'prediction' function giving less variability.
o Back to knots placed using equidistant by default for estimating baseline hazard function with splines. You can now use the option 'hazard="Splines-per"' in frailtyPenal in order to have knots placed using percentiles.
o Back to value 10-3 by default for the three convergence criteria.
o No longer need to use as.factor() in command to print Wald tests on covariates.
o Print p-value of one-sided Wald test for frailty parameter and two-sided Wald test for alpha parameter in joint model.
o New functions to compute estimators of Expected Prognostic Observed Cross-Entropy (EPOCE) evaluating prediction accuracy in joint model.
Changes in Version 2.6 March 2014
o NEW: Fit now a multivariate gaussian frailty model (two types of recurrent events and a terminal event).
o Major evolution of frailtyPenal function. 'Frailty' and 'joint' arguments removed.
o Now estimation of baseline hazard functions with splines, knots are placed using percentile (previously using equidistant intervals).
o Significant change of prediction function. You can compute predictions in two different ways: with a variable prediction time or a variable window of prediction.
o 'type' argument of prediction function removed. As long as there is a 'group' argument, for a shared model, computation of conditional predictions will be done.
o 'B' argument added in 2.4.1 to initialize regression coefficients was renamed 'init.B'
o Possibility to initialize the variance of the frailties with argument 'init.Theta' in shared and joint frailty models.
o Possibility to initialize the coefficient with argument 'init.Alpha' in joint frailty model.
o Moreover, with 'Alpha="none"', frailtyPenal can fit a joint model with a fixed alpha (=1).
o New argument: 'print.times', added in every model to print iteration process.
Changes in Version 2.5.1 February 2014
o Bug fixed about joint frailty model without any covariate.
Changes in Version 2.5 November 2013
o New dynamic tool of prediction added for Cox proportionnal hazard, shared and joint frailty model.
o Add IPCW estimation of concordance measures as Uno (Stat Med 2011). Significant changes in the printing of 'Cmeasures' function.
o Bug fixed about parametrical survival functions plotting with left truncated data.
o Bug fixed which allowed cross validation with interval-censored data.
o Possibility to print and change the three convergence criterions in frailtyPenal and additivePenal.
Changes in Version 2.4.1 April 2013
o Bug fixed about estimation of frailties in shared models using recurrentAG=TRUE.
o Printing bug about standard deviation of the random effet variance in a model without covariate.
o Possibility to initialize regression coefficients in shared and joint frailty models.
Changes in Version 2.4 April 2013
o Fit now a model with time-varying effects for covariates (only for Cox, shared gamma and a joint gamma frailty model).
Changes in Version 2.3 February 2013
o Fit now a Shared and a Joint Frailty model with a log-normal distribution for the random effects.
o "Breast cosmesis" dataset added for interval-censoring illustration ("Diabetes" dataset removed).
o Weibull hazard parameters bug fixed : shape and scale were reversed.
o Linear predictors : output reorganized.
o Plot options improved (now color is allowed).
o Use of 'SurvIC' function modified. Now for the left-truncated and interval-censored data we use : SurvIC(left-trunc-time,lower-time,upper-time,event).
o No need of the intcens argument to fit a model for interval-censored data anymore, 'SurvIC' function is enough.
Changes in Version 2.2-27 November 2012
o Fit now a Joint Frailty model for clustered data.
Changes in Version 2.2-26 October 2012
o Minor bug fixed about loglikelihood in Nested Frailty model.
o The package accepts samples unsorted on clusters.
Changes in Version 2.2-25 September 2012
o "Diabetes" dataset added for interval-censoring illustration.
Changes in Version 2.2-24 July 2012
o Fit a Shared Gamma Frailty or a Cox proportional hazard model for interval-censored data.
o No longer need to use cluster function for fitting a Cox proportional hazard model.
o Minor bug fixed in Nested Frailty model.
o Printing bug fixed in multivariate Wald test.
Changes in Version 2.2-22 March 2012
o Fit a Shared Gamma Frailty model using a parametric estimation.
o Fit Joint Frailty model for recurrent and terminal events using a parametric estimation.
o Fit a Nested Frailty model using a parametric estimation.
o Fit an Additive Frailty model using a parametric estimation.
o Concordance measures in shared frailty and Cox models (Cmeasures).
Changes in Version 2.2-10
o NEW VERSION OF FRAILTYPACK including Additive, Nested and Joint Frailty models
o Paper submitted to Journal of Stat Software