BRCA - the Cox regression merging

Agnieszka Sitko

2019-06-27

library(factorMerger)
library(survival)

if (!require("forcats")) {
    install.packages("forcats")
}
library(forcats)

data("BRCA")

brcaSurv <- Surv(time = BRCA$time, event = BRCA$vitalStatus)

drugName <- BRCA$drugName # drug name
drugName <- fct_lump(as.factor(drugName), prop = 0.05) 

drugNameFM <- mergeFactors(response = brcaSurv[!is.na(drugName)], 
                           factor = drugName[!is.na(drugName)], 
                           family = "survival")

plot(drugNameFM, nodesSpacing = "effects", gicPanelColor = "grey2")

anova(coxph(brcaSurv[!is.na(drugName)] ~ drugName[!is.na(drugName)]))
#> Analysis of Deviance Table
#>  Cox model: response is brcaSurv[!is.na(drugName)]
#> Terms added sequentially (first to last)
#> 
#>                             loglik  Chisq Df Pr(>|Chi|)
#> NULL                       -131.16                     
#> drugName[!is.na(drugName)] -129.59 3.1525  4     0.5326
anova(coxph(brcaSurv[!is.na(drugName)] ~ cutTree(drugNameFM)))
#> Analysis of Deviance Table
#>  Cox model: response is brcaSurv[!is.na(drugName)]
#> Terms added sequentially (first to last)
#> 
#>                      loglik  Chisq Df Pr(>|Chi|)
#> NULL                -131.16                     
#> cutTree(drugNameFM) -129.93 2.4724  1     0.1159

subtype <- BRCA$histologicalType
subtype <- fct_lump(as.factor(subtype), prop = 0.05) 

subtypeFM <- mergeFactors(response = brcaSurv[!is.na(subtype)], 
                          factor = subtype[!is.na(subtype)],
                           family = "survival")

plot(subtypeFM) 

patCat <- BRCA$pathologicCategory %>% substr(1, 2)

patCatFM <- mergeFactors(response = brcaSurv[!is.na(patCat)],
                         factor = patCat[!is.na(patCat)],
                         family = "survival")

plot(patCatFM, responsePanel = "frequency", gicPanelColor = "red")

anova(coxph(brcaSurv[!is.na(patCat)] ~ patCat[!is.na(patCat)]))
#> Analysis of Deviance Table
#>  Cox model: response is brcaSurv[!is.na(patCat)]
#> Terms added sequentially (first to last)
#> 
#>                         loglik  Chisq Df Pr(>|Chi|)
#> NULL                   -539.39                     
#> patCat[!is.na(patCat)] -536.59 5.5993  4     0.2311
anova(coxph(brcaSurv[!is.na(subtype)] ~ cutTree(subtypeFM)))
#> Analysis of Deviance Table
#>  Cox model: response is brcaSurv[!is.na(subtype)]
#> Terms added sequentially (first to last)
#> 
#>                     loglik  Chisq Df Pr(>|Chi|)   
#> NULL               -539.39                        
#> cutTree(subtypeFM) -534.89 8.9946  1   0.002708 **
#> ---
#> Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1