survwrapper: Function for fitting and selecting hazard functions to...
In bokov/powertrip: Analysis of longevity data.

Description Usage Arguments Details Value Note Author(s) References See Also Examples

Takes one or two vectors of event times (numeric format) and optionally corresponding vectors of indicator variables to designate right-censored events. Fits several mortality models, selects the best fitting one/s, and if two vectors were given, tests hypotheses about the model parameters.

survwrapper(x, y = NULL, models = c("g", "gm", "l", "lm"), cx = rep(1, length(x)), cy = rep(1, length(y)), ext = F, n = length(c(x, y)), AIC = F, BIC = F, breakties = "AIC", compare.matrix = NULL, constraint.matrix = NULL, thresh = 0.05, smooth = 7)

`x`	A numeric vector of event times. For example, number of days an individual has survived.
`y`	An optional second numeric vector of event times, in the same units as `x`. Specifying this vector enables fitting joint models and hypothesis testing.
`models`	A character vector of model names: `g` = Gompertz, `gm` = Gompertz-Makeham, `l` = logistic, `lm` = logistic-Makeham, and `w` = Weibull. The default is to fit all the models except Weibull.
`cx`	A vector of 0 and 1 the same length as `x`. Ordinary events (e.g. death) are indicated by 1 and censored events (e.g. withdrawal from the study) are indicated by 0.
`cy`	A vector of 0 and 1 the same length as `y`. Ordinary events (e.g. death) are indicated by 1 and censored events (e.g. withdrawal from the study) are indicated by 0.
`ext`	Not implemented.
`n`	Total sample size. Should normally be left for the script to automatically calculate, but can be specified when survwrapper is called from another script repeatedly in order to speed up runtimes.
`AIC`	Whether to calculate the AIC (Akaike Information Criterion) for each candidate model.
`BIC`	Whether to calculate the BIC (Bayes Information Criterion) for each candidate model.
`breakties`	What criterion to use for choosing a model if more than one is justified by the comparisons.
`compare.matrix`	A matrix for specifying a customized comparison algorithm.
`constraint.matrix`	A matrix of 1's and 0's for specifying a customized set of parameter constraints to test.
`thresh`	Significance cutoff.
`smooth`	Not yet supported.

This function takes vectors (assumed to be times-to-event) and uses numerical methods to find maximum likelihood estimates of model parameters for a one or more models (by default, these are exponential, Gompertz, Gompertz-Makeham, logistic, and Logistic-Makeham). Censored events can be specified with the cx and cy arguments, which should be vectors of 1's and 0's the same length as x and y, respectively. If both x and y are specified, the best joint model/s are chosen (such that the same type of model is fitted to both populations, and the likelihood ratio between this joint model and the corresponding joint model but one fewer parameter is significantly different from 1 according to the chi-squared distribution). Then, for each parameter in the model/s so chosen, a test is performed on the null hypothesis that constraining the parameter such that it is identical between the two populations will result in a joint model that does not fit significantly worse than the full model. If this null hypothesis is rejected, the interpretation is that the corresponding parameter significantly differs between populations. The null hypothesis of all parameters not being significantly different is also tested by default. The user can also specify which hypotheses to test.

`x.m`	A data frame containing a column for the log-likelihood and each of the model parameter estimates. Each model fitted is represented by its own row.
`y.m`	If `y` was specified, a data frame like `x.m` but for the `y` population.
`xy.sm`	If `y` was specified, a data frame showing the results of a series of full-reduced tests where the reduced (`simple`) joint model is compared to the full (`complex`). `df` indicates the different in degrees between the complex and simple models. `LLc` and `LLs` are the log-likelihoods for the complex and simple models, respectively. `LR` is the likelihood ratio between the complex and simple models. `chi2` is the chi squared statistic calculated from this ratio. `npar` is the number of parameters in the complex model. `AIC` and `BIC` are respectively the Akaike and Bayes information criteria for the complex model. `p` is the P-value for the full-reduced test and `sig` is whether this was deemed significant, and `chosen` is whether, based on the overall pattern of significant and non-significant full-reduced tests, the complex joint model for that row was the one selected for testing hypotheses about parameter differences.
`par.differences`	IN PROGRESS
`x`	The original values in the `x` argument.
`y`	The original values in the `y` argument, if any.
`cx`	The original values in the `cx` argument.
`cy`	The original values in the `cy` argument, if any.
`x.d`	IN PROGRESS
`y.d`	IN PROGRESS
`suggested.models`	A string with the abbreviation of the joint model/s chosen for hypothesis testing.
`nx`	An integer representing the sample size of the `x` group.
`ny`	An integer representing the sample size of the `y` group, if any.

Uses Nelder-Mead algorithm to find maximum likelihood estimates of model parameters.

Alex F. Bokov (bokov@uthscsa.edu), Jon A. Gelfond

Pletcher,S.D., Khazaeli,A.A., and Curtsinger,J.W. (2000). Why do life spans differ? Partitioning mean longevity differences in terms of age-specific mortality parameters. Journals of Gerontology Series A-Biological Sciences and Medical Sciences 55, B381-B389

optim

## Generate two sets of survival times.
population1 <- simsurv(629,type='g',p=c(7.33e-4,0.1227,0,0));
population2 <-
  simsurv(574,type='lm',p=c(5.4818e-5,0.1543,0.0023,0.6018));

## Fit models to the populations and compare the parameters.
models1vs2 <- survwrapper(population1,population2);