clippda (version 1.22.0)

sampleSizeContourPlots: A function to construct a grid with contours for calculating sample size in multi dimensional parameter space

Description

This function draws a grid for calculating the sample size based on the clinically important values of variances versus differences. Based on the analysis of data from past proteomic profiling studies of cancer, we define the clinically important parameters as the summary statistics of the intensities of the peaks with medium biological variation. On the grid, you may display the parameter values from a wide range of real-life data from past proteomic profiling studies, including: data from urine and serum samples of early- and late-stage colorectal cancer patients; serum samples of colorectal cancer patients assayed on four SELDI chip-types (IMAC, H50, Q10 and CM10); plasma samples from Limanda limanda fish; and urine samples of colorectal cancer patients analysed using both SELDI and MALDI sample processing protocols. These values may be used as guidelines for choosing the sample size calculation parameters. If your study involves profiling samples from late-stage disease or sera assayed on the IMAC chip, then the sample size is probably a value close to that of the outer left contour. The urine profiling studies require more samples to detect differences and the value of the contours to the right of grid may be used as bounds. You may also display parameters and sample size from your pilot study in this grid by inputting a vector (observedPara) of consensus values of the variance and the corresponding difference, or rbind several vectors of such parameters into a matrix/dataframe if you have multiple pilots.

Usage

sampleSizeContourPlots(Z,m,DIFF,VAR,beta,alpha,observedPara,Indicator)

Arguments

Z

the heterogeneity correction factor.

m

the number of replicates.

DIFF

the clinically important difference.

VAR

the protein variance.

beta

the power to estimate the clinically important difference.

alpha

the significance level.

observedPara

a vector or a matrix/dataframe (if there is more than one pilot study) containing the variance(s) and the clinically important difference(s) observed from your pilot data. The first element (column) of the vector (matrix) contains the observed variances, while the second contains the information on the clinically important difference(s).

Indicator

an indicator variable. If it is set to 1, then the results of previous proteomic profiling studies together with the results of your pilot study are included in the plot. If it is set to 0, it leads to a plot of only the latter.

Value

Plots of grids of variance versus the clinically important differences with sample size contours superimposed on it.

References

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

2. Nyangoma SO, Collins SI, Douglas GW, Altman DG, Johnson PJ, and Billingham LJ: Issues in sample size calculations for designing cancer proteomic profiling studies. BMC Bioinformatics, 2009, Submitted

Examples

# The plot will be saved in your working directory.
# On the grid, we have plotted a number of sample sizes we computed from real life data.
#From these values you can gauge how many samples you may need.
# Fewer samples than 50, will not result in any meaningful estimation of differences.
# For late-stage cancer you  need the fewest samples, even from a very variable sample such as urine.
# You need more samples, over 200, to estimate differences between early stage cancer and 
#noncancer controls.
#etc.
m <- 2

DIFF <- seq(0.1,0.50,0.01) # 0.01

VAR <- seq(0.2,4,0.1)

beta <- c(0.90,0.80,0.70)

alpha  <-  1 - c(0.001, 0.01,0.05)/2

Corr <- c(0.70,0.90) #intraclass correlation also fixed

Z <- 2.6   # fix at 2.6 or use FisherInformation(???)

# You may input parameters from your pilot study. Suppose they are: 

#observedPara=c(1,0.4) #the variance you computed from pilot data

observedPara <- data.frame(var=c(0.7,0.5,1.5),DIFF=c(0.37,0.33,0.43))

# you may set these values to 0, if you do not have pilot data
#observedVAR=0
#observedDIFF=0
# in this case the values computed from my pilot studies (dotted on the plot) 
# may be used as guidelines.

Indicator <- 0 #1

sampleSizeContourPlots(Z,m,DIFF,VAR,beta,alpha,observedPara,Indicator)