MMMS (version 0.1)

MMMS: Subgroup identification using a multi-marker molecular signature approach

Description

Identification of a treatment-specific subgroup for time-to-event outcomes using a multi-marker molecular signature approach

Usage

MMMS(time, event, treat, bio, covar = NULL, pct.lb = 20, pct.ub = 80, n.boot = 1000,  pos.direction = FALSE, nfolds = 5, alpha = 0.5, verbose = FALSE, seed = NULL)

Arguments

time

A numeric vector containing the follow up time for right censored data.

event

A numeric vector containing the status indicator, normally 0=alive, 1=dead.

treat

A numeric vector containing the treatment indicator: 1=treatment of interest, 0=alternative treatment (e.g. placebo or standard of care).

bio

A numeric data frame or matrix containing biomarker values.

covar

A numeric matrix containing clinical covariates. Default is NULL for not including any covariates.

pct.lb

A scalar indicating the lower bound of the search range for desired subgroup sizes in percentage (e.g. 20 means 20%).

pct.ub

A scalar indicating the upper bound of the search range for desired subgroup sizes in percentage (e.g. 80 means 80%).

n.boot

A scalar indicating the number of bootstraps for calculating the bootstrap p-value for the subgroup effect. Default is 1000, which is a recommended value.

pos.direction

A logical value indicating whether a subgroup with hazard ratio > 1 is desirable. Default is FALSE, i.e. a hazard ratio < 1 is desirable.

nfolds

The number of folds for cross validation in choosing tuning parameters. The function cv.glmnet() in the ``glmnet'' package is called, which requires cross validation to choose the tuning parameter ``lambda''. Default is 5.

alpha

A scalar for the elasticnet mixing parameter as in the ``glmnet'' package (0=ridige, 1=lasso). A fixed value is supposed to be used, without searching for the optimal alpha value. Default is 0.5.

verbose

A logical value indicating whether bootstrap progress should be printed. Default is FALSE.

seed

An integer for setting random seed, if provided. Default is NULL for not setting any seed.

Value

A list with the following elements:

score.obj

The object returned by get.score().

score

The composite scores based on interaction terms.

score.main

The composite scores based on main-effect terms.

coefs

Elnet coefficient estimates for interaction terms.

coefs.main

Elnet coefficient estimates for main-effect terms.

fit

The glmnet fitted object for obtaining the MMMS composite scores.

lambda

The optimal lambda value chosen for obtaining the MMMS composite scores.

alpha

The alpha value used for obtaining the MMMS composite scores.

subgrp.obj

The object returned by get.subgroup().

subgrp.size

The size (in percentage) of the optimal subgroup identified.

subgrp.fit

The fitted model object for the optimal subgroup identified.

subgrp.cut

The cutpoint of the composite score score for the optimal subgroup identified.

subgrp.pval

The p-value of the treatment-by-subgroup effect based on n.boot bootstraps. NA is returned if n.boot=0.

n.boot

The number of bootstraps considered for calculating subgrp.pval.

Details

MMMS() calls several functions that could also be used separately: get.score(), get.score.main(), get.subgroup(), etc.

As is described in Li et al. (2014), the bootstrap p-value is based on a statistically valid test whose type I error is approximately controlled at the nominal level. However, caution is needed for interpreting the estimates of subgroup size and treatment-by-subgroup interaction effect, as bias has been observed in these estimates. Approaches for correcting bias in the estimates may be implemented in future versions of the ``MMMS'' package.

References

Lin Li, Tobias Guennel, Scott Marshall, Leo Wang-Kit Cheung (2014) A multi-marker molecular signature approach for treatment-specific subgroup identification with survival outcomes. The Pharmacogenomics Journal. http://dx.doi.org/10.1038/tpj.2014.9

See Also

get.score, get.subgroup.

Examples

# load the dataset
data(simdat)
    
# estimate the MMMS (No bootstrap is considered for a quick illustration)
mmms = with(simdat,MMMS(time,event,treat,bio,covar,pct.lb=20,pct.ub=80,n.boot=0,
                        pos.direction=FALSE,nfolds=5,alpha=0.5,verbose=TRUE,seed=12345))