This is the example code for Dunkler D, Sauerbrei W, Heinze G (2016). Global, Parameterwise and Joint Shrinkage Factor Estimation. Journal of Statistical Software. 69(8), 1-19. http://dx.doi.org/10.18637/jss.v069.i08.

Example R Code

################################################################################
### R code for
### "Global, Parameterwise and Joint Shrinkage Factor Estimation"
### written by Daniela Dunkler, March 2016
################################################################################

## works with R 3.2.3 & shrink-package 1.2.1

## load the packages
library("shrink")
library("survival")
library("mfp")
library("aod")                                                  # for wald-test
## 
## Attaching package: 'aod'
## The following object is masked from 'package:survival':
## 
##     rats
sessionInfo()
## R version 3.2.3 (2015-12-10)
## Platform: x86_64-w64-mingw32/x64 (64-bit)
## Running under: Windows 7 x64 (build 7601) Service Pack 1
## 
## locale:
## [1] LC_COLLATE=English_Canada.1252  LC_CTYPE=English_Canada.1252   
## [3] LC_MONETARY=English_Canada.1252 LC_NUMERIC=C                   
## [5] LC_TIME=English_Canada.1252    
## 
## attached base packages:
## [1] stats     graphics  grDevices utils     datasets  methods   base     
## 
## other attached packages:
## [1] aod_1.3         mfp_1.5.2       survival_2.38-3 knitr_1.12.3   
## [5] shrink_1.2.1   
## 
## loaded via a namespace (and not attached):
##  [1] Rcpp_0.12.2         formatR_1.3         RColorBrewer_1.1-2 
##  [4] plyr_1.8.3          tools_3.2.3         rpart_4.1-10       
##  [7] digest_0.6.9        evaluate_0.8.3      memoise_0.2.1      
## [10] polspline_1.1.12    nlme_3.1-125        gtable_0.2.0       
## [13] lattice_0.20-33     Matrix_1.2-3        yaml_2.1.13        
## [16] SparseM_1.7         mvtnorm_1.0-5       gridExtra_2.2.1    
## [19] stringr_1.0.0       cluster_2.0.3       MatrixModels_0.4-1 
## [22] devtools_1.10.0     grid_3.2.3          nnet_7.3-11        
## [25] foreign_0.8-66      rmarkdown_0.9       multcomp_1.4-4     
## [28] latticeExtra_0.6-28 TH.data_1.0-7       Formula_1.2-1      
## [31] magrittr_1.5        ggplot2_2.1.0       Hmisc_3.17-2       
## [34] scales_0.4.0        codetools_0.2-14    htmltools_0.2.6    
## [37] rms_4.4-2           MASS_7.3-45         splines_3.2.3      
## [40] colorspace_1.2-6    quantreg_5.21       sandwich_2.3-4     
## [43] stringi_1.0-1       acepack_1.3-3.3     munsell_0.4.3      
## [46] zoo_1.7-12

Section 2.1: Deep Vein Thrombosis Study

## Section 2.1: Deep Vein Thrombosis Study
## load data
data("deepvein")

# number of observations, events, median observation time, and names of
# variables
nrow(deepvein)
## [1] 929
deepvein$status2 <- abs(deepvein$status-1)
survfit(Surv(time, status2) ~ 1, data = deepvein)
## Call: survfit(formula = Surv(time, status2) ~ 1, data = deepvein)
## 
##       n  events  median 0.95LCL 0.95UCL 
##   929.0   782.0    37.8    35.6    41.7
table(deepvein$status)
## 
##   0   1 
## 782 147
round(100*table(deepvein$status)/nrow(deepvein), 2)
## 
##     0     1 
## 84.18 15.82
sort(names(deepvein))
##  [1] "age"      "bmi"      "durther"  "fiimut"   "fvleid"   "loc"     
##  [7] "log2ddim" "pnr"      "sex"      "status"   "status2"  "time"

Section 2.2: Breast Cancer Study

## Section 2.2: Breast Cancer Study
## load data
data("GBSG")


# number of observations, events, median observation time, and names of
# variables
nrow(GBSG)
## [1] 686
table(GBSG$cens)
## 
##   0   1 
## 387 299
round(100*table(GBSG$cens)/nrow(GBSG), 2)
## 
##     0     1 
## 56.41 43.59
GBSG$cens2 <- abs(GBSG$cens-1)
# median observation time is given in days here
survfit(Surv(rfst, cens2) ~ 1, data = GBSG)
## Call: survfit(formula = Surv(rfst, cens2) ~ 1, data = GBSG)
## 
##       n  events  median 0.95LCL 0.95UCL 
##     686     387    1645    1570    1717
# median observation time in months
1645 / 30.5
## [1] 53.93443
sort(names(GBSG))
##  [1] "age"      "cens"     "cens2"    "esm"      "htreat"   "id"      
##  [7] "menostat" "posnodal" "prm"      "rfst"     "tumgrad"  "tumsize"

Section 5.1: Deep Vein Thrombosis Study

## Section 5.1: Deep Vein Thrombosis Study
# set the reference level for all categorical variables
deepvein$sex <- relevel(deepvein$sex, ref = "female")
deepvein$fiimut <- relevel(deepvein$fiimut, ref = "absent")
deepvein$fvleid <- relevel(deepvein$fvleid, ref = "absent")
deepvein$loc <- relevel(deepvein$loc, ref = "PE")
# contrasts(deepvein$loc)


## fit full model, and compute shrinkage factors (jackknife estimates and dfbeta
## approximations)
fitfull <- coxph(Surv(time, status) ~ log2ddim + bmi + durther + age +
                 sex + loc + fiimut + fvleid, data = deepvein, x = TRUE)
summary(fitfull)
## Call:
## coxph(formula = Surv(time, status) ~ log2ddim + bmi + durther + 
##     age + sex + loc + fiimut + fvleid, data = deepvein, x = TRUE)
## 
##   n= 929, number of events= 147 
## 
##                    coef exp(coef)  se(coef)      z Pr(>|z|)   
## log2ddim       0.219517  1.245475  0.085739  2.560  0.01046 * 
## bmi            0.005865  1.005883  0.019051  0.308  0.75817   
## durther        0.021881  1.022122  0.023681  0.924  0.35550   
## age           -0.003973  0.996035  0.006583 -0.603  0.54622   
## sexmale        0.495927  1.642019  0.189718  2.614  0.00895 **
## locdistal     -0.905095  0.404504  0.311078 -2.910  0.00362 **
## locproximal   -0.179351  0.835813  0.180336 -0.995  0.31996   
## fiimutpresent -0.162573  0.849954  0.390499 -0.416  0.67718   
## fvleidpresent -0.108886  0.896833  0.194228 -0.561  0.57506   
## ---
## Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
## 
##               exp(coef) exp(-coef) lower .95 upper .95
## log2ddim         1.2455     0.8029    1.0528    1.4734
## bmi              1.0059     0.9942    0.9690    1.0442
## durther          1.0221     0.9784    0.9758    1.0707
## age              0.9960     1.0040    0.9833    1.0090
## sexmale          1.6420     0.6090    1.1321    2.3816
## locdistal        0.4045     2.4722    0.2199    0.7442
## locproximal      0.8358     1.1964    0.5870    1.1902
## fiimutpresent    0.8500     1.1765    0.3954    1.8272
## fvleidpresent    0.8968     1.1150    0.6129    1.3123
## 
## Concordance= 0.61  (se = 0.026 )
## Rsquare= 0.028   (max possible= 0.855 )
## Likelihood ratio test= 26.2  on 9 df,   p=0.001894
## Wald test            = 23.07  on 9 df,   p=0.006032
## Score (logrank) test = 23.9  on 9 df,   p=0.004465
## wald-test for categorical predictor "loc"
wald.test(Sigma = vcov(fitfull)[c("locdistal", "locproximal"),
          c("locdistal", "locproximal")],
          b = fitfull$coeff[c("locdistal", "locproximal")], Terms = 1:2)
## Wald test:
## ----------
## 
## Chi-squared test:
## X2 = 8.6, df = 2, P(> X2) = 0.014
shrink(fitfull, type = "global", method = "jackknife")
## Shrinkage Factors (type=global, method=jackknife):
## [1] 0.6069174
## 
## Shrunken Regression Coefficients:
##      log2ddim           bmi       durther           age       sexmale     locdistal 
##   0.133228852   0.003559850   0.013279842  -0.002411042   0.300986429  -0.549317699 
##   locproximal fiimutpresent fvleidpresent 
##  -0.108851134  -0.098668180  -0.066084738
## apply backward elimination to full model*, and compute shrinkage factors to
## selected model
## * dummy variables of loc are jointly tested and selected
fitd <- step(fitfull, direction = "backward")
## Start:  AIC=1784.06
## Surv(time, status) ~ log2ddim + bmi + durther + age + sex + loc + 
##     fiimut + fvleid
## 
##            Df    AIC
## - bmi       1 1782.2
## - fiimut    1 1782.2
## - fvleid    1 1782.4
## - age       1 1782.4
## - durther   1 1782.9
## <none>        1784.1
## - log2ddim  1 1788.6
## - sex       1 1789.2
## - loc       2 1790.5
## 
## Step:  AIC=1782.15
## Surv(time, status) ~ log2ddim + durther + age + sex + loc + fiimut + 
##     fvleid
## 
##            Df    AIC
## - fiimut    1 1780.3
## - fvleid    1 1780.5
## - age       1 1780.5
## - durther   1 1781.0
## <none>        1782.2
## - log2ddim  1 1786.6
## - sex       1 1787.2
## - loc       2 1788.6
## 
## Step:  AIC=1780.34
## Surv(time, status) ~ log2ddim + durther + age + sex + loc + fvleid
## 
##            Df    AIC
## - fvleid    1 1778.7
## - age       1 1778.7
## - durther   1 1779.1
## <none>        1780.3
## - log2ddim  1 1784.9
## - sex       1 1785.3
## - loc       2 1786.7
## 
## Step:  AIC=1778.66
## Surv(time, status) ~ log2ddim + durther + age + sex + loc
## 
##            Df    AIC
## - age       1 1777.0
## - durther   1 1777.4
## <none>        1778.7
## - log2ddim  1 1783.2
## - sex       1 1783.4
## - loc       2 1785.0
## 
## Step:  AIC=1777.01
## Surv(time, status) ~ log2ddim + durther + sex + loc
## 
##            Df    AIC
## - durther   1 1775.8
## <none>        1777.0
## - log2ddim  1 1781.5
## - sex       1 1782.6
## - loc       2 1783.3
## 
## Step:  AIC=1775.77
## Surv(time, status) ~ log2ddim + sex + loc
## 
##            Df    AIC
## <none>        1775.8
## - log2ddim  1 1780.3
## - sex       1 1781.2
## - loc       2 1782.8
print(fitd)
## Call:
## coxph(formula = Surv(time, status) ~ log2ddim + sex + loc, data = deepvein, 
##     x = TRUE)
## 
## 
##                coef exp(coef) se(coef)     z      p
## log2ddim     0.2188    1.2446   0.0854  2.56 0.0104
## sexmale      0.4909    1.6338   0.1847  2.66 0.0079
## locdistal   -0.9224    0.3976   0.3101 -2.97 0.0029
## locproximal -0.2051    0.8146   0.1787 -1.15 0.2511
## 
## Likelihood ratio test=24.5  on 4 df, p=6.37e-05
## n= 929, number of events= 147
## wald-test for categorical predictor "loc"
wald.test(Sigma = vcov(fitd)[c("locdistal", "locproximal"),
          c("locdistal", "locproximal")],
          b = fitd$coeff[c("locdistal", "locproximal")], Terms = 1:2)
## Wald test:
## ----------
## 
## Chi-squared test:
## X2 = 9.1, df = 2, P(> X2) = 0.01
shrink(fitd, type = "global", method = "jackknife")
## Shrinkage Factors (type=global, method=jackknife):
## [1] 0.8076362
## 
## Shrunken Regression Coefficients:
##    log2ddim     sexmale   locdistal locproximal 
##   0.1767045   0.3964779  -0.7449390  -0.1656066
pwsf <- shrink(fitd, type = "parameterwise", method = "jackknife")
pwsf
## Shrinkage Factors (type=parameterwise, method=jackknife):
##    log2ddim     sexmale   locdistal locproximal 
##   0.7321036   0.8351074   0.8393993   0.1321006 
## 
## Shrunken Regression Coefficients:
##    log2ddim     sexmale   locdistal locproximal 
##  0.16017851  0.40996379 -0.77423621 -0.02708736
sqrt(diag(pwsf$ShrinkageFactorsVCOV))
##    log2ddim     sexmale   locdistal locproximal 
##   0.3703055   0.3699730   0.3268146   0.8696058
shrink(fitd, type = "parameterwise", method = "jackknife",
       join = list(c("locdistal", "locproximal")))
## Shrinkage Factors (type=parameterwise with join option, method=jackknife):
##    log2ddim     sexmale   locdistal locproximal 
##   0.7806284   0.8363976   0.8055175   0.8055175 
## 
## Shrunken Regression Coefficients:
##    log2ddim     sexmale   locdistal locproximal 
##   0.1707954   0.4105972  -0.7429847  -0.1651721
## dummy variables of loc are separately tested and selected
## generate dummy variables, fit full model and then apply backward elimination
deepvein <- cbind(deepvein, model.matrix( ~ loc, data = deepvein)[, -1])
fitfull2 <- coxph(Surv(time, status) ~ log2ddim + bmi + durther + age +
                  sex + locdistal + locproximal + fiimut + fvleid,
                  data = deepvein, x = TRUE)  # fitfull2 is identical to fitfull

fitd2 <- step(fitfull2, direction = "backward")
## Start:  AIC=1784.06
## Surv(time, status) ~ log2ddim + bmi + durther + age + sex + locdistal + 
##     locproximal + fiimut + fvleid
## 
##               Df    AIC
## - bmi          1 1782.2
## - fiimut       1 1782.2
## - fvleid       1 1782.4
## - age          1 1782.4
## - durther      1 1782.9
## - locproximal  1 1783.1
## <none>           1784.1
## - log2ddim     1 1788.6
## - sex          1 1789.2
## - locdistal    1 1792.5
## 
## Step:  AIC=1782.15
## Surv(time, status) ~ log2ddim + durther + age + sex + locdistal + 
##     locproximal + fiimut + fvleid
## 
##               Df    AIC
## - fiimut       1 1780.3
## - fvleid       1 1780.5
## - age          1 1780.5
## - durther      1 1781.0
## - locproximal  1 1781.2
## <none>           1782.2
## - log2ddim     1 1786.6
## - sex          1 1787.2
## - locdistal    1 1790.6
## 
## Step:  AIC=1780.34
## Surv(time, status) ~ log2ddim + durther + age + sex + locdistal + 
##     locproximal + fvleid
## 
##               Df    AIC
## - fvleid       1 1778.7
## - age          1 1778.7
## - durther      1 1779.1
## - locproximal  1 1779.3
## <none>           1780.3
## - log2ddim     1 1784.9
## - sex          1 1785.3
## - locdistal    1 1788.7
## 
## Step:  AIC=1778.66
## Surv(time, status) ~ log2ddim + durther + age + sex + locdistal + 
##     locproximal
## 
##               Df    AIC
## - age          1 1777.0
## - durther      1 1777.4
## - locproximal  1 1777.8
## <none>           1778.7
## - log2ddim     1 1783.2
## - sex          1 1783.4
## - locdistal    1 1787.0
## 
## Step:  AIC=1777.01
## Surv(time, status) ~ log2ddim + durther + sex + locdistal + locproximal
## 
##               Df    AIC
## - durther      1 1775.8
## - locproximal  1 1776.2
## <none>           1777.0
## - log2ddim     1 1781.5
## - sex          1 1782.6
## - locdistal    1 1785.3
## 
## Step:  AIC=1775.77
## Surv(time, status) ~ log2ddim + sex + locdistal + locproximal
## 
##               Df    AIC
## - locproximal  1 1775.1
## <none>           1775.8
## - log2ddim     1 1780.3
## - sex          1 1781.2
## - locdistal    1 1784.8
## 
## Step:  AIC=1775.1
## Surv(time, status) ~ log2ddim + sex + locdistal
## 
##             Df    AIC
## <none>         1775.1
## - log2ddim   1 1778.9
## - sex        1 1780.7
## - locdistal  1 1782.8
summary(fitd2)
## Call:
## coxph(formula = Surv(time, status) ~ log2ddim + sex + locdistal, 
##     data = deepvein, x = TRUE)
## 
##   n= 929, number of events= 147 
## 
##               coef exp(coef) se(coef)      z Pr(>|z|)   
## log2ddim   0.20392   1.22621  0.08483  2.404   0.0162 * 
## sexmale    0.49535   1.64107  0.18496  2.678   0.0074 **
## locdistal -0.84053   0.43148  0.30277 -2.776   0.0055 **
## ---
## Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
## 
##           exp(coef) exp(-coef) lower .95 upper .95
## log2ddim     1.2262     0.8155    1.0384    1.4480
## sexmale      1.6411     0.6094    1.1421    2.3581
## locdistal    0.4315     2.3176    0.2384    0.7811
## 
## Concordance= 0.602  (se = 0.026 )
## Rsquare= 0.025   (max possible= 0.855 )
## Likelihood ratio test= 23.16  on 3 df,   p=3.744e-05
## Wald test            = 20.03  on 3 df,   p=0.0001673
## Score (logrank) test = 20.85  on 3 df,   p=0.0001132
shrink(fitd2, type = "global", method = "jackknife")
## Shrinkage Factors (type=global, method=jackknife):
## [1] 0.8439991
## 
## Shrunken Regression Coefficients:
##   log2ddim    sexmale  locdistal 
##  0.1721124  0.4180761 -0.7094099
shrink(fitd2, type = "parameterwise", method = "jackknife")
## Shrinkage Factors (type=parameterwise, method=jackknife):
##  log2ddim   sexmale locdistal 
## 0.7700839 0.8317218 0.8975722 
## 
## Shrunken Regression Coefficients:
##   log2ddim    sexmale  locdistal 
##  0.1570393  0.4119946 -0.7544400
## Table 2
t2_jack <- c(shrink(fitd, type = "global", method = "jackknife")$ShrinkageFactors,
             pwsf$ShrinkageFactors,
             shrink(fitd, type = "parameterwise", method = "jackknife",
                    join = list(c("locdistal", "locproximal")))$ShrinkageFactors,
             system.time(shrink(fitd, type = "parameterwise", method = "jackknife"))[1])

t2_dfbeta <- c(shrink(fitd, type = "global", method = "dfbeta")$ShrinkageFactors,
               shrink(fitd, type = "parameterwise", method = "dfbeta")$ShrinkageFactors,
               shrink(fitd, type = "parameterwise", method = "dfbeta",
                      join = list(c("locdistal", "locproximal")))$ShrinkageFactors,
               system.time(shrink(fitd, type = "parameterwise", method = "dfbeta"))[1])

t2 <- cbind(Jackknife = t2_jack, DFBETA = t2_dfbeta)

Table2 <- cbind(t2, round((t2[, 2] - t2[, 1]) / t2[, 1] * 100, 1))
Table2[, 1:2] <- round(Table2[, 1:2], 4)
dimnames(Table2)[[2]][3] <- "Relative difference"
Table2 <- rbind(Table2[1,], rep(NA, 3), Table2[2:5,], rep(NA, 3), Table2[6:8,],
                rep(NA, 3), Table2[10,])
dimnames(Table2)[[1]][c(1:2, 7, 10, 12)] <- c("Global shrinkage",
          "Parameterwise shrinkage", "Joint shrinkage", "loc", "Computing time")
Table2
##                         Jackknife DFBETA Relative difference
## Global shrinkage           0.8076 0.8123                 0.6
## Parameterwise shrinkage        NA     NA                  NA
## log2ddim                   0.7321 0.7385                 0.9
## sexmale                    0.8351 0.8373                 0.3
## locdistal                  0.8394 0.8449                 0.7
## locproximal                0.1321 0.1470                11.2
## Joint shrinkage                NA     NA                  NA
## log2ddim                   0.7806 0.7864                 0.7
## sexmale                    0.8364 0.8386                 0.3
## loc                        0.8055 0.8111                 0.7
##                                NA     NA                  NA
## Computing time             1.4800 0.0100               -99.3
# max. difference
max(abs(Table2[1:10, 1] - Table2[1:10, 2]), na.rm=TRUE)
## [1] 0.0149
## Figure 1: Data simulated from deep vein thrombosis study: Comparison of Jackknife and
##           DFBETA-approximated global shrinkage factors for selected and unselected
##           models.

deepvein0 <- subset(deepvein, status == 0)
deepvein1 <- subset(deepvein, status == 1)

n0 <- nrow(deepvein0)
n1 <- nrow(deepvein1)

ratio <- 0.2                                     # based on n1 / n0
## Note: Running this code is time-consuming, since 2 * B * S data sets are analyzed.
## size <- seq(from = 200, to = 2000, length.out = 19)      # based on nrow(deepvein)
# B <- 100
## You may want to reduce B and length.out of size: e.g.
## size <- seq(from = 200, to = 2000, length.out = 4)
## B <- 3
#
# S <- length(size)
# 
# shrinkGJ <- shrinkGD <- shrinkGJsel <- shrinkGDsel <- matrix(NA, nrow = B, ncol = S)
# set.seed(954681)
# 
# for (s in 1:S) {
#   cat("\n", s)
#   for (b in 1:B) {
#     cat(".")
#     data <- rbind(deepvein0[sample(x = 1:n0, size = size[s]*(1-ratio), replace = TRUE),],
#                   deepvein1[sample(x = 1:n1, size = size[s]*ratio, replace = TRUE),])
#     
#     fitfull <- coxph(Surv(time, status) ~ log2ddim + bmi + durther + age + sex + loc +
#                        fiimut + fvleid, data = data, x = TRUE)
#     fitsel <- step(fitfull, direction = "backward", trace = 0)
#     
#     if (!is.null(fitsel$coefficients)) {                     # no null model selected
#       shrinkGJsel[b, s] <- shrink(fitsel, type = "global", method = "jackknife")$ShrinkageFactors
#       shrinkGDsel[b, s] <- shrink(fitsel, type = "global", method = "dfbeta")$ShrinkageFactors
#     }
#     
#     
#     fit <- coxph(Surv(time, status) ~ log2ddim + sex + loc, data = data, x = TRUE)
#     
#     shrinkGJ[b, s] <- shrink(fit, type = "global", method = "jackknife")$ShrinkageFactors
#     shrinkGD[b, s] <- shrink(fit, type = "global", method = "dfbeta")$ShrinkageFactors
#   }
# }
## In some smaller data sets (and especially when shrinkage factors are estimated with the
## Jackknife method) there might be issues with convergence in individual data sets.
## However, coxph will issue a warning.
# 
# 
# shrinkGDselm <- apply(shrinkGDsel, 2, median)
# shrinkGJselm <- apply(shrinkGJsel, 2, median)
# 
# cbind(n = size, Diff_J_D_sel = shrinkGJselm - shrinkGDselm)
# 
# shrinkGDm <- apply(shrinkGD, 2, median)
# shrinkGJm <- apply(shrinkGJ, 2, median)
# 
# cbind(n = size, Diff_J_D = shrinkGJm - shrinkGDm)
# 
# xrange <- c(0.5, 1)
## xrange <- range(shrinkGDselm, shrinkGJselm, shrinkGDm, shrinkGJm)

## pdf("compJvsD.pdf", width = 6, height = 4)
# par(oma = c(0, 0, 0.5, 0.5), mar = c(3.5, 4, 0, 0))
# plot(range(size), xrange, type = "n", las = 1, xlab = "", ylab = "", xaxt = "n")
# axis(1, at = size)
# mtext(side = 1, text = "Sample size", line = 2.3)
# mtext(side = 2, text = "Global shrinkage factor", line = 2.8)
# 
# points(size, shrinkGDselm, pch = 3, col = "darkolivegreen4", cex = 1.5)
# points(size, shrinkGJselm, pch = 1, col = "darkgoldenrod3", cex = 1.3)
# lines(size, shrinkGDselm, lty = 2, col = "darkolivegreen4", lwd = 2)
# lines(size, shrinkGJselm, lty = 3, col = "darkgoldenrod3", lwd = 2)
# 
# points(size, shrinkGDm, pch = 3, col = "firebrick3", cex = 1.5)
# points(size, shrinkGJm, pch = 1, col = "dodgerblue3", cex = 1.3)
# lines(size, shrinkGDm, lty = 2, col = "firebrick3", lwd = 2)
# lines(size, shrinkGJm, lty = 3, col = "dodgerblue3", lwd = 2)
# 
# legend(x = 1600, y = 0.62, legend = c("Jackknife", "DFBETA"), lwd = 2, title = "unselected",
#        col = c("dodgerblue3", "firebrick3"), inset = 0.05, bty = "n", pch = c(1, 3),
#        text.col = c("dodgerblue3", "firebrick3"), title.col = "black")
# 
# legend(x = 1000, y = 0.62, legend = c("Jackknife", "DFBETA"), lwd = 2, title = "selected",
#        col = c("darkgoldenrod3", "chartreuse4"), inset = 0.05, bty = "n", pch = c(1, 3),
#        text.col = c("darkgoldenrod3", "chartreuse4"), title.col = "black")
## dev.off()

Section 5.2: Breast Cancer Study

## Section 5.2: Breast Cancer Study
## define predictors as suggested by Sauerbrei (1999, Applied Statistics)
GBSG$enodes <- exp(-0.12*GBSG$posnodal)


# create ordinal dummy-coded variable tumgrad for grades 1 to 3
contrasts(GBSG$tumgrad) <- matrix(c(0, 1, 1, 0, 0, 1), ncol = 2, byrow = FALSE,
                                  dimnames = list(1:3, c("tumgrad1", "tumgrad2")))
contrasts(GBSG$tumgrad)
##   tumgrad1 tumgrad2
## 1        0        0
## 2        1        0
## 3        1        1
# for nicer variable names use default dimnames
contrasts(GBSG$tumgrad) <- matrix(c(0, 1, 1, 0, 0, 1), ncol = 2, byrow = FALSE)


## # model selection (backward elimination) and estimation
fitg <- mfp(Surv(rfst, cens) ~ fp(age) + fp(prm) + fp(esm) + fp(tumsize) +
              fp(enodes) + tumgrad + menostat + strata(htreat),
            family = cox, data = GBSG, alpha = 0.05, select = 0.05)
print(fitg)
## Call:
## mfp(formula = Surv(rfst, cens) ~ fp(age) + fp(prm) + fp(esm) + 
##     fp(tumsize) + fp(enodes) + tumgrad + menostat + strata(htreat), 
##     data = GBSG, family = cox, alpha = 0.05, select = 0.05)
## 
## 
## Deviance table:
##           Resid. Dev
## Null model    3198.026
## Linear model  3077.005
## Final model   3055.989
## 
## Fractional polynomials:
##           df.initial select alpha df.final power1 power2
## enodes             4   0.05  0.05        1      1      .
## prm                4   0.05  0.05        2    0.5      .
## tumgrad1           1   0.05  0.05        1      1      .
## tumgrad2           1   0.05  0.05        0      .      .
## tumsize            4   0.05  0.05        0      .      .
## menostat2          1   0.05  0.05        0      .      .
## age                4   0.05  0.05        4     -2     -1
## esm                4   0.05  0.05        0      .      .
## 
## 
## Transformations of covariates:
##                                  formula
## age      I((age/100)^-2)+I((age/100)^-1)
## prm                 I(((prm+1)/100)^0.5)
## esm                                 <NA>
## tumsize                             <NA>
## enodes                       I(enodes^1)
## tumgrad                          tumgrad
## menostat                            <NA>
## 
## Re-Scaling:
## Non-positive values in some of the covariates. No re-scaling was performed.
## 
##               coef exp(coef) se(coef)      z        p
## enodes.1   -1.9782   0.13832   0.2272 -8.706 0.00e+00
## prm.1      -0.5717   0.56456   0.1109 -5.154 2.55e-07
## tumgrad1.1  0.5137   1.67141   0.2495  2.059 3.95e-02
## age.1       0.6060   1.83303   0.1188  5.101 3.37e-07
## age.2      -2.6539   0.07037   0.5902 -4.496 6.91e-06
## 
## Likelihood ratio test=142  on 5 df, p=0 n= 686
## Table 3
## (p-values unconditional on the selected degree and power for prm, age, and
## enodes = fitg$pvalues)
varorder <- c("age.1", "age.2", "prm.1", "enodes.1", "tumgrad1.1")

t3_beta_j <- c(NA, round(fitg$coefficients, 3)[varorder])
t3_df <- c(fitg$df.initial[7, ], NA, NA, fitg$df.initial[c(2, 1, 3), ])
t3_p <- c(round(fitg$pvalues[7, "p.null"], 3), NA, NA,
          round(fitg$pvalues[c(2, 1, 3), "p.null"], 3))
t3_pwsf0 <- shrink(fitg, type = "parameterwise", method = "jackknife")
t3_pwsf <- round(c(NA, t3_pwsf0$ShrinkageFactors[varorder]), 3)
t3_pwsfse <- round(c(NA, sqrt(diag(t3_pwsf0$ShrinkageFactorsVCOV))[varorder]), 3)
t3_jointsf0 <- shrink(fitg, type = "parameterwise", method = "jackknife",
                      join=list(c("age.1", "age.2")))
t3_jointsf <- round(c(t3_jointsf0$ShrinkageFactors[varorder[1]], NA, NA,
                      t3_jointsf0$ShrinkageFactors[varorder[-c(1:2)]]), 3)
t3_jointsfse <- round(c(sqrt(diag(t3_jointsf0$ShrinkageFactorsVCOV))["join.age.1"],
                        NA, NA,
                        sqrt(diag(t3_jointsf0$ShrinkageFactorsVCOV))[varorder[-c(1:2)]]), 3)

Table3 <- cbind("beta_j" = t3_beta_j, "df" = t3_df, "p value" = t3_p,
                "PWSF" = t3_pwsf, "PWSF SE" = t3_pwsfse,
                "Joint shrinkage" = t3_jointsf,
                "Joint shrinkage SE" = t3_jointsfse)
dimnames(Table3)[[1]][1] <- "age"
Table3
##            beta_j df p value  PWSF PWSF SE Joint shrinkage Joint shrinkage SE
## age            NA  4   0.001    NA      NA           0.876              0.188
## age.1       0.606 NA      NA 0.811   0.236              NA                 NA
## age.2      -2.654 NA      NA 0.782   0.277              NA                 NA
## prm.1      -0.572  4   0.000 0.978   0.189           0.982              0.189
## enodes.1   -1.978  4   0.000 0.987   0.116           0.986              0.116
## tumgrad1.1  0.514  1   0.028 0.811   0.453           0.809              0.452
## compute shrinkage factors
globalsf <- shrink(fitg, type = "global", method = "jackknife")
globalsf
## Shrinkage Factors (type=global, method=jackknife):
## [1] 0.9526639
## 
## Shrunken Regression Coefficients:
##   enodes.1      prm.1 tumgrad1.1      age.1      age.2 
## -1.8845360 -0.5446421  0.4893540  0.5772861 -2.5282947
sqrt(globalsf$ShrinkageFactorsVCOV)
##            global
## global 0.08062613
pwsf <- shrink(fitg, type = "parameterwise", method = "jackknife")
pwsf
## Shrinkage Factors (type=parameterwise, method=jackknife):
##   enodes.1      prm.1 tumgrad1.1      age.1      age.2 
##  0.9874769  0.9777288  0.8108045  0.8108243  0.7822610 
## 
## Shrunken Regression Coefficients:
##   enodes.1      prm.1 tumgrad1.1      age.1      age.2 
## -1.9534020 -0.5589719  0.4164852  0.4913355 -2.0760587
round(sqrt(diag(pwsf$ShrinkageFactorsVCOV)), 3)
##   enodes.1      prm.1 tumgrad1.1      age.1      age.2 
##      0.116      0.189      0.453      0.236      0.277
round(cov2cor(pwsf$ShrinkageFactorsVCOV), 3)
##            enodes.1  prm.1 tumgrad1.1  age.1  age.2
## enodes.1      1.000 -0.055     -0.078  0.030  0.021
## prm.1        -0.055  1.000     -0.200  0.026  0.032
## tumgrad1.1   -0.078 -0.200      1.000 -0.040 -0.035
## age.1         0.030  0.026     -0.040  1.000  0.984
## age.2         0.021  0.032     -0.035  0.984  1.000
jointsf <- shrink(fitg, type = "parameterwise", method = "jackknife",
                  join=list(c("age.1", "age.2")))
jointsf
## Shrinkage Factors (type=parameterwise with join option, method=jackknife):
##   enodes.1      prm.1 tumgrad1.1      age.1      age.2 
##  0.9862779  0.9816686  0.8094979  0.8763299  0.8763299 
## 
## Shrunken Regression Coefficients:
##   enodes.1      prm.1 tumgrad1.1      age.1      age.2 
## -1.9510302 -0.5612242  0.4158141  0.5310300 -2.3257103
round(sqrt(diag(jointsf$ShrinkageFactorsVCOV)), 3)
## join.age.1   enodes.1      prm.1 tumgrad1.1 
##      0.188      0.116      0.189      0.452
## Figure 2: selected model: Log hazard relative to 50 years
# refit model fitg explicitly including the two dummy variables of
# tumgrad
GBSG <- cbind(GBSG, model.matrix(~tumgrad, data = GBSG)[, -1])

fitg <- mfp(Surv(rfst, cens) ~ fp(age) + fp(prm) + fp(esm) + fp(tumsize) +
              fp(enodes) + tumgrad1 + tumgrad2 + menostat + strata(htreat),
            family = cox, data = GBSG, alpha = 0.05, select = 0.05)

globalsf <- shrink(fitg, type = "global", method = "jackknife")
pwsf <- shrink(fitg, type = "parameterwise", method = "jackknife")
jointsf <- shrink(fitg, type = "parameterwise", method = "jackknife",
                  join=list(c("age.1", "age.2")))

# define new data for which prediction is requested
# newdatref is the new data set for the reference observation
age <- 30:75
newdat <- data.frame(age = age, enodes = 1, prm = 1, tumgrad1 = 0,
                     htreat = factor(1))
newdatref <- data.frame(age = 50, enodes = 1, prm = 1, tumgrad1 = 0,
                        htreat = factor(1))


xaxis <- seq(from = min(age), to = max(age), by = 5)


# pdf("plotgbsg.pdf", width = 6, height = 6)
par(fig = c(0, 1, 0, 0.3), new = FALSE, oma = c(0, 0, 0.5, 0.5),
    mar = c(3.5, 4, 0, 0))
hist(GBSG$age[(GBSG$age >= min(age)) & (GBSG$age <= max(age))],
     breaks = seq(from = min(age), to = max(age), by = 2.5), main = "",
     xlim = range(age), xaxs = "r", yaxs = "r", xlab = "", ylab = "",
     axes = FALSE, col = "gray88")
axis(side = 1, at = xaxis)
axis(side = 2, at = seq(from = 0, to = 60, by = 20), las = 2)
mtext(side = 1, text = "Age", line = 2.3)
mtext(side = 2, text = "Frequency", line = 2.8)
box()

par(fig = c(0,1,0.3,1), new = TRUE, oma = c(0, 0, 0.5, 0.5),
    mar = c(1, 4, 0, 0))
plot(range(age), c(-0.05, 0.8), xlab = "", ylab = "", type = "n", yaxs = "i",
     xaxt = "n", yaxt = "n")
axis(side = 1, at = xaxis, labels = rep("", times = length(xaxis)))
axis(side = 2, at = seq(from = 0, to = 0.7, by = 0.1), las = 2,
     labels = seq(from = 0, to = 0.7, by = 0.1))
mtext(side = 2, text = "Log hazard relative to 50 years", line = 2.8)
lines(x = age, y = predict(fitg, newdata = newdat, type = "lp") -
        predict(fitg, newdata = newdatref, type = "lp"), lty = 1, col = "gray25",
      lwd = 2)
lines(x = age, y = predict(globalsf, newdata = newdat, type = "lp") -
        predict(globalsf, newdata = newdatref, type = "lp"), lty = 5,
      col = "forestgreen", lwd = 2)
lines(x = age, y = predict(pwsf, newdata = newdat, type="lp") -
        predict(pwsf, newdata = newdatref, type = "lp"), lty = 2,
      col = "dodgerblue3", lwd = 2)
lines(x = age, y = predict(jointsf, newdata = newdat, type = "lp") -
        predict(jointsf, newdata = newdatref, type = "lp"), lty = 4,
      col = "firebrick3", lwd = 2)

legend("topright", inset = 0.02, bty = "n", lty = c(1, 5, 4, 2),
       title = "SHRINKAGE",
       legend = c("No", "Global", "Joint", "Parameterwise"),
       col = c("gray25", "forestgreen", "firebrick3", "dodgerblue3"), lwd = 2,
       text.col = c("gray25", "forestgreen", "firebrick3", "dodgerblue3"))

# dev.off()