clippda (version 1.22.0)

fisherInformation: A generic function to compute the heterogeneity correction factor in sample size calculations

Description

This generic function computes the inverse of the expected 'Fisher' information matrix, I^-1/theta (see the definition of theta in section 2.3 of Nyangoma et al., 2009). The second diagonal element of this matrix is the variance of the mean difference between cases and controls, having adjusted for the effect of confounders. The elements of I are sums of the proportions of samples having given attributes, or sums of proportions of class memberships of given conditional contingency tables obtained from the cross tabulated the attributes of the samples under study. Thus it contains information on the heterogeneity in the data due to imbalances in the proportions of samples having given attributes.

Usage

fisherInformation(Data, ...)

Arguments

Data

An object of aclinicalProteomicsData class. It extracts and uses a dataframe of the clinical information about the samples from a slot in the Data.

...

Some methods for this generic function may take additional, optional arguments. At present none do.

Value

This function returns a matrix: and its second diagonal element (divided by 2) is the quantity called Z (or the heterogeneity-correction factor) in the sample size calculation function, sampleSize.

Details

Note that continuous variables must first be discretized, and the variable names must coincide with the column names of PhenoInfo extracted from the object. Currently this function only accepts a maximum of three binary variables. The existing methods (e.g. Diggle et al. 1997, page 31) for continuous repeated data consider only a single exposure variable. We recommend that some form of variable selection be used to determine which covariates to include in the analysis.

References

1. Nyangoma SO, Ferreira JA, Collins SI, Altman DG, Johnson PJ, and Billingham LJ (2009): Sample size calculations for planning clinical proteomic profiling studies using mass spectrometry. Bioinformatics (Submitted)

2. Diggle PJ, Heagerty P, Liang K.-Y and Zeger SL. (2002). Analysis of Longitudinal Data (second edition). Oxford: Oxford University Press

Examples

#########################################################################################
#The matrices of interest are of the form (see eq. 15, 18 and 22 Nyangoma et al. (2009))
#########################################################################################

#Examples are:

###################
# 1 binary variable
###################

data.frame(x1=c(1,'b'),x2=c('b','b'))

#####################
# 2 binary variables
#####################
data.frame(x1=c(1,'b','c'),x2=c('b','b','d'),x3=c('c','d','c'))

##########################################
# 3 binary variables

data.frame(x1=c(1,'b','c','d'),x2=c('b','b','e','f'),x3=c('c','e','c','g'),x4=c('d','f','g','d'))

##############################################################################
##############################################################################
# Data # pheno_urine
# the phenotypic information of the urine cancer patients and normal controls.
#####
# I have discretized protein concentration
# concentration<=70 and concentration>70
##########################################
##########################################

#data(pheno_urine)
#PhenoInfo <- pheno_urine

#variables <- c('Tumor','Sex','Protein_concIndex')

#variables=c('Tumor','Sex')
#variables=c('Tumor')


# Tumor must contain characters "c" and "n"

#Protein_concIndex <- pheno_urine[!(pheno_urine$stage == 'late'),]$Protein_conc

#Protein_concIndex[Protein_concIndex<=70] <- 0
#Protein_concIndex[Protein_concIndex>70] <- 1
#Protein_concIndex=as.factor(Protein_concIndex)

#PhenoInfo <- data.frame(pheno_urine[!(pheno_urine$stage == 'late'),],Protein_concIndex)

#FisherInformation(PhenoInfo,variables)

data(liverdata)

data(liver_pheno)

OBJECT=new("aclinicalProteomicsData")

OBJECT@rawSELDIdata=as.matrix(liverdata)
OBJECT@covariates=c("tumor" ,    "sex")
OBJECT@phenotypicData=as.matrix(liver_pheno)
OBJECT@variableClass=c('numeric','factor','factor')
[e-mail]=53

inversefisherinformation <- fisherInformation(OBJECT)

inversefisherinformation