Type: Package
Title: Analysis of Kin-Cohort Studies
Version: 0.7
Date: 2015-08-15
Author: Victor Moreno, Nilanjan Chatterjee, Bhramar Mukherjee
Maintainer: Victor Moreno <v.moreno@iconcologia.net>
Depends: survival
Description: Analysis of kin-cohort studies. kin.cohort provides estimates of age-specific cumulative risk of a disease for carriers and noncarriers of a mutation. The cohorts are retrospectively built from relatives of probands for whom the genotype is known. Currently the method of moments and marginal maximum likelihood are implemented. Confidence intervals are calculated from bootstrap samples. Most of the code is a translation from previous 'MATLAB' code by N. Chatterjee.
License: GPL-2 | GPL-3 [expanded from: GPL (≥ 2)]
Packaged: 2015-08-28 10:33:32 UTC; h501uvma
NeedsCompilation: no
Repository: CRAN
Date/Publication: 2015-08-28 16:36:59

Internal functions for marginal method

Description

Internal functions for marginal method

Value

pyear

calculates number of events and person years

pwexp

estimates survival and hazard for piece-wise exponential model

mendelian

calculates the mendelian probabilities of carrying the mutation conditional on the proband genotype for 1 gene.

mendelian.combine

combines mendelian probabilities of carrying the mutation conditional on the proband genotype for 2 genes.

See Also

kc.marginal

Marginal Maximum Likelihood estimation of kin-cohort data

Description

This function estimates cumulative risk and hazard at given ages for carriers and noncarriers of a mutation based on the probands genotypes. It uses the Marginal Maximum Likelihood estimation method (Chatterjee and Wacholder, 2001). Piece-wise exponential distribution is assumed for the survival function.

Usage

kc.marginal(t, delta, genes, r, knots, f, pw = rep(1,length(t)), 
            set = NULL, B = 1, maxit = 1000, tol = 1e-5, subset,
            logrank=TRUE, trace=FALSE)

Arguments

t

time variable. Usually age at diagnosis or at last follow-up

delta

disease status (1: event, 0: no event

genes

factor or numeric vector (1 gene), matrix or dataframe (2 genes) with genotypes of proband numeric. factors and data.frame with factors are prefered in order to use user-defined labels. Otherwise use codes (1:noncarrier, 2: carrier, 3: homozygous carrier)

r

relationship with proband 1:parent, 2:sibling 3:offspring 0:proband. Probands will be excluded from analysis and offspring will be recoded 1 internally.

knots

time points (ages) for cumulative risk and hazard estimates

f

vector of mutation allele frequencies in the population

pw

prior weights, if needed

set

family id (only needed for bootstrap)

B

number of boostrap samples (only needed for bootstrap)

maxit

max number of iterations for the EM algorithm

tol

convergence tolerance

subset

logical condition to subset data

logrank

Perform a logrank test

trace

Show iterations for bootstrap

Value

object of classes "kin.cohort" and "chatterjee".

cumrisk

matrix with cumulative risk estimates for noncarriers, carriers and the cumulative risk ratio. Estimates are given for the times indicated in the knot vector

hazard

matrix with hazard estimates for noncarriers, carriers and the hazard ratio. Estimates are given for the times indicated in the knot vector

knots

vector of knots

conv

if the EM algorithm converged

niter

number of iterations needed for convergence

ngeno.rel

number of combinations of genotypes in the relatives

events

matrix with number of events and person years per each knot

logHR

mean log hazard ratio estimate (unweighted)

logrank

logrank test. If 2 genes, for the main effects, the cross-classification and the stratified tests

call

copy of call

if bootstrap confidence intervals are requested (B>1) then the returned object is of classes "kin.cohort.boot" and "chatterjee" with previous items packed in value estimate and each bootstrap sample packed in matrices.

Note

This function is best called by kin.cohort than directly

References

Chatterjee N and Wacholder S. A Marginal Likelihood Approach for Estimating Penetrance from Kin-Cohort Designs. Biometrics. 2001; 57: 245-52.

See Also

kin.cohort, print.kin.cohort, plot.kin.cohort

Examples

## Not run: 
data(kin.data)
attach(kin.data)
res.mml<- kc.marginal(age, cancer, gen1, rel, knots=c(30,40,50,60,70,80), f=0.02)
res.mml

## End(Not run)

Kin-cohort estimation of penetrance by the method of moments

Description

This function estimates cumulative risk and hazard at given ages for carriers and noncarriers of a mutation based on the probands genotypes. It uses the method of moments described by Wacholder et al (1998)

Usage

kc.moments(t, delta, genes, r, knots, f, pw = rep(1,length(t)), 
           set = NULL, B = 1, logrank = TRUE, subset, trace=FALSE)

Arguments

t

time variable. Usually age at diagnosis or at last follow-up

delta

disease status (1: event, 0: no event

genes

genotype of proband numeric. A factor is preferred, otherwise numeric code of genotypes (1: noncarrier, 2:carrier, [3: homozygous carrier])

r

relationship with proband 1:parent, 2:sibling 3:offspring 0:proband. Probands will be excluded from analysis and offspring will be recoded 1 internally.

knots

time points (ages) for cumulative risk and hazard estimates

f

mutation allele frequency in the population

pw

prior weights, if needed

set

family id (only needed for bootstrap)

B

number of boostrap samples (only needed for bootstrap)

logrank

if logrank test is desired

subset

logical condition to subset data

trace

Show iterations for bootstrap

Value

object of classes "kin.cohort" and "wacholder".

cumrisk

matrix of dimension (number of knots x 3) with cumulative risk festimates or noncarriers, carriers and the cumulative risk ratio

knots

vector of knots

km

object class survfit (package survival)

logrank

p-value of the logrank test

events

matrix with number of events and person years per each knot

call

copy of call

if bootstrap confidence intervals are requested (B>1) then the returned object is of classes "kin.cohort.boot" and "wacholder" with previous items packed in value estimate and each bootstrap sample packed in matrices.

Note

This function is best called by kin.cohort than directly

References

Wacholder S, Hartge P, Struewing JP, Pee D, McAdams M, Lawrence B, Tucker MA. The kin-cohort study for estimating penetrance. American Journal of Epidemiology. 1998; 148: 623-9.

See Also

kin.cohort, print.kin.cohort, plot.kin.cohort

Examples

## Not run: 
data(kin.data)
attach(kin.data)
res.km<- kc.moments(age, cancer, gen1, rel, knots=c(30,40,50,60,70,80), f=0.02)
res.km

## End(Not run)

Analysis of kin-cohort data

Description

This function estimates cumulative risk at given ages for carriers and noncarriers of a mutation based on the probands genotypes. It can use the Marginal Maximum Likelihood estimation method (Chatterjee and Wacholder, 2001) or the method of moments (Wacholder et al, 2001). Bootstrap confidence intervals can be requested.

Usage

kin.cohort(..., method = c("marginal", "mml", "chatterjee", 
                             "moments",  "km",  "watcholder"))

Arguments

...

see kc.marginal and kc.moments for details

method

choose estimation method: Marginal Maximum Likelihood (selected by "marginal", "mml", "chatterjee") or method of moments (selected by "moments", "km", "watcholder")

Details

This function is a wrapper that will call kc.marginal or kc.moments depending on the argument method.

Author(s)

Victor Moreno, Nilanjan Chatterjee, Bhramar Mukherjee.

Maintainer: Victor Moreno <v.moreno@iconcologia.net>

References

Wacholder S, Hartge P, Struewing JP, Pee D, McAdams M, Lawrence B, Tucker MA. The kin-cohort study for estimating penetrance. American Journal of Epidemiology. 1998; 148: 623-9.

Chatterjee N and Wacholder S. A Marginal Likelihood Approach for Estimating Penetrance from Kin-Cohort Designs. Biometrics. 2001; 57: 245-52.

See Also

kc.marginal, kc.moments

Examples

## Not run: 
data(kin.data)
attach(kin.data)

res.k<-   kin.cohort(age, cancer, gen1, rel, knots=c(30,40,50,60,70,80), f=0.02, 
                     method="km")
res.k          
plot(res.k)
plot(res.k,what="crr")

set.seed(1)
res.k.b<- kin.cohort(age, cancer, gen1, rel, knots=c(30,40,50,60,70,80), f=0.02, 
                     set=family, method="km", B=10)
res.k.b
plot(res.k.b)
plot(res.k.b,what="crr")

res.m<-   kin.cohort(age, cancer, gen1, rel, knots=c(30,40,50,60,70,80), f=0.02, 
                    method="mml")
res.m
plot(res.m)
plot(res.m, what="hr")

res.m2<-  kin.cohort(age, cancer, data.frame(gen1,gen2), rel, 
                     knots=c(30,40,50,60,70,80), f=c(0.02,0.01), method="mml")
res.m2
plot(res.m2)
plot(res.m2, what="hr")

set.seed(1)
res.m.b<- kin.cohort(age, cancer, gen1, rel, knots=c(30,40,50,60,70,80), f=0.02, 
                     set=family, method="mml", B=10)
res.m.b
plot(res.m.b)
plot(res.m.b, what="hr")

## End(Not run)

sample data for kin-cohort analysis

Description

Simulated data of a study on the penetrance of breast cancer for carriers 2 mutations.

Usage

data(kin.data)

Format

A data frame with 15341 observations on the following 5 variables.

age

age at diagnosis or at last follow-up

cancer

disease status (1: breast cancer, 0: no breast cancer

gen1

gen1 genotypes of proband

gen2

gen2 genotypes of proband

rel

relationship with proband 1:parent or offspring, 2:sibling

family

family id

Examples

data(kin.data)

methods for print and plot

Description

Functions to print a formatted output and produce plots

Usage

## S3 method for class 'kin.cohort'
print(x, descriptive = TRUE, cumrisk = TRUE, hazard = FALSE, survival = FALSE, 
        logrank = TRUE, HR = TRUE, digits = 5, ...)

## S3 method for class 'kin.cohort.boot'
print(x, cumrisk = TRUE, hazard = FALSE, HR = TRUE, conf = 0.95,
        digits = 5, show = TRUE, logrank = TRUE, ...)

## S3 method for class 'kin.cohort'
plot(x, what = c("cr", "hr", "crr"), min.age = min(x$knots), 
      max.age = max(x$knots), max.y, type, add=FALSE, color, line,  ...)

## S3 method for class 'kin.cohort.boot'
plot(x, conf = 0.95, what = c("cr", "hr", "crr"), min.age = min(x$knots), 
      max.age = max(x$knots), age.start = 0, start.ref, max.y, type,
      median = FALSE, add = FALSE, color, line, ...)

Arguments

x

object to be printed or plotted

descriptive

print table with number of events and person-years

cumrisk

print cumulative risk

hazard

print hazard

survival

print survival

HR

print harard ratios

logrank

print logrank p value

digits

digits for rounding

show

do not print

conf

coverage for confidence intervals

what

type of plot desired: cumulative risk ("cr"), hazard ratio ("hr", for marginal method only), cumulative risk ratio ("crr", for moments method only)

min.age

Minimal age for plots

max.age

Maximal age for plots

age.start

initial age value (x) for plots

start.ref

initial risk value (y) for plots

max.y

Max value for y axis

type

type of line in plots

add

If TRUE, then lines are added to current plot. Useful to compare analyses.

color

change line colors using a vector of values

line

change line width using a vector of values

median

plot median of bootstrap samples instead of point estimates

...

additional arguments for print or plot

Details

Specific output and plot types can be selected with arguments

simulation of kin cohort studies

Description

Functions to simulate data for kin-cohort analysis

Usage


kc.simul(nfam, f, hr, rand = 0, mean.sibs = 2, mean.desc = 1.5, 
         a.age = 8, b.age = 80, a.cancer = 3, b.cancer = 180 )

sample.caco(object, p.cases = 1, caco.ratio = 1, verbose = TRUE)

## S3 method for class 'kin.cohort.sample'
summary(object,...)

Arguments

nfam

number of families to be generated

f

allele frequency

hr

hazard ratio for disease carriers relative noncarriers

rand

variance of random effect for cancer incidence (ratio of hr)

mean.sibs

mean number of sibllings and descendants (~Poisson)

mean.desc

mean number of sibllings and descendants (~Poisson)

a.age

shape parameter for age (~Weibull)

b.age

scale parameter for age (~Weibull)

a.cancer

shape parameter for cancer incidence (~Weibull)

b.cancer

scale parameter for cancer incidence (~Weibull)

object

object of class kin.cohort.sample and data.frame

p.cases

proportion of cases (affected) to include in sample. if more than 1, the exact number is assumed

caco.ratio

ratio of controls per case to include in sample

verbose

show the number of cases and controls sampled

...

additional arguments

Details

kc.simul will generate a cohort of probands of size nfam. Default parameters simulate a typical cancer study. Each of them will be assigned: a carrier status with probability f^2+2f(1-f); a current age drawn from a Weibull distribution with parameters a.age and b.age; an age at diagnosis (agecancer) drawn from a Weibull distribution with parameters a.cancer and b.cancer, if noncarrier. For carries, the scale (b.cancer) is shifted to get the desired hazard ratio (hr). If rand>0, then a family specific random effect is also added, drawn from a normal distribution with mean 0 and sd rand. If agecancer< age then the disease status (cancer) will be 1, 0 otherwise.

First degree relatives are generated for each proband: two parents, a random number of sibblings (drawn from a Poisson withe mean mean.sibs), and a random number of descendants (drawn from a Poisson with mean mean.desc). Each of them is assiggned a carrier status with probability according to mendelian transmission conditional of the proband carrier status. Current age for relatives are generated conditional on the proband's age, with random draws from normal distribution. Age at diagnosis (agecancer) is assumed independent, except for the optional family random effect. Gender is assigned at random with probability 0.5 for all individuals.

Note that the simulation of residual familial correlation with a random effect (rand$>0) does not mantain the desired hazard ratio (hr).

The generic function summary will show the number and proportion of carriers and affected subjects in the sample.

sample.caco will sample (from a simulation generated by kc.simul) a subset of cases (afected probands) and controls (unaffected probands) and their relatives. Currently only random sampling of controls is implemented (no matching). Sampling fraction is controled by caco.ratio.

Currently, only one gene and one disease are simulated.

Value

object of class kin.cohort.sample and data.frame with fields

famid

family id

rel

relative type (0=proband, 1=parents, 2=sibblings, 3=descendants)

age

current age of each subject

gender

gender (0=male, 1=female)

carrier

carrier status of proband (0=noncarrier, 1=carrier), common for all family members

cancer

affected (0=no, 1=yes)

agecancer

age at diagnosis or current age if not affected

real.carrier

carrier status or relatives (0=noncarrier, 1=carrier )

Examples

## Not run: 
set.seed(7)
## cohort 
s<-kc.simul(4000, f=0.03, hr=5)
summary(s)

## exclude probands
m.coh<- kc.marginal(s$agecancer, s$cancer, factor(s$carrier), s$rel,
                    knots=c(30,40,50,60,70,80,90), f=0.03)
m.coh

## relatives only
r.coh<- coxph(Surv(agecancer,cancer)~real.carrier, data=s)
print(exp(coef(r.coh)))

## probands only
p.coh<- coxph(Surv(agecancer,cancer)~carrier, data=s)
print(exp(coef(p.coh)))

## case-control
s.cc<- sample.caco(s)
summary(s.cc)

## exclude probands
m.caco<- kc.marginal(s.cc$agecancer, s.cc$cancer, factor(s.cc$carrier), 
                     s.cc$rel, knots=c(30,40,50,60,70,80,90), f=0.03)
m.caco

## relatives only
r.caco<- glm(cancer~real.carrier, family=binomial, data=s.cc, subset=(s.cc$rel!=0))
print(exp(coef(r.caco)[2]))

## probands only
p.caco<- glm(cancer~carrier, family=binomial, data=s.cc, subset=(s.cc$rel==0))
print(exp(coef(p.caco)[2]))

## End(Not run)