############################################################################################################
# External validation and recalibration of the ERASL risk scores                                           #
#                                                                                                          #
# BR Beumer; K Takagi ; B Vervoort; S Buettner; Y Umeda; T Yagi; T Fujiwara;EW Steyerberg; JNM IJzermans*  #                                                                            #
#                                                                                                          #
#   Affiliations:                                                                                          # 
#   * 	Department of Surgery, Division of HPB & Transplant Surgery, Erasmus MC, University Medical Centre # 
# Rotterdam, Rotterdam, The Netherlands Erasmus MC Rotterdam                                               # 
#                                                                                                          # 
# Corresponding author: Prof. Dr. J.N.M. IJzermans,                                                        # 
# Email: j.ijzermans@erasmusmc.nl                                                                          # 
############################################################################################################ 

library(readxl)
library(tableone)
library(dplyr)
library(tidyr)
library(mice)
library(VIM)
library(survival) # KM plots, baseline haz
library(rms)      # cph, validate, rcorr.cens
library(boot)     # generic nonparametric bootstrap
library(survAUC)  # GHCI,tdAUC

# survcomp install
#if (!requireNamespace("BiocManager", quietly = TRUE))
#  install.packages("BiocManager")
#BiocManager::install("survcomp")
library(survcomp) # Rd
library(SurvRegCensCov) # 
R.Version()$version.string

setwd("C:\\Users\\beren\\Dropbox\\Current projects\\9. External Validation johnson\\src")


#path = "E:\\7. ERASL validation\\data\\external\\HCC_EMC_JP(17-02-2020).xlsx"
path = "C:\\Users\\062010\\Desktop\\ERASL_val\\HCC_EMC_JP(18-02-2021).xlsx"
df <- read_excel(path, sheet="Source data")

#---------- Renaming and date format ----------------------------------------
# renaming columns
old = c("Patient ID", "Birth date", "Date of surgery","Child-pugh score", "Child-pugh class",  "Total bilirubin", 
        "Tumor size (cm)", "Tumor number", "Microvascular invasion")

new = c("Patient_ID", "Birth_date", "Date_of_surgery","Child_pugh_score", "Child_pug_class",  
        "Total_bilirubin", "Tumor_size", "Tumor_number", "Microvascular_invasion")

for(i in 1:length(old)){
  names(df)[names(df) == old[i]] = new[i]
} 

#df <- rename(df, D_r = T_r)
#df <- rename(df, D_d = T_d)

# date transformation
df$Birth_date <- as.Date(df$Birth_date)
df$Date_of_surgery <- as.Date(df$Date_of_surgery)
df$D_d <- as.Date(df$D_d)
df$D_r <- as.Date(df$D_r)

# mutations
df$AFP <- recode(df$AFP, '0' = 0.0001)
df$Age <- as.numeric((df$Date_of_surgery - df$Birth_date)/365.25)
df$T_r <- df$D_r - df$Date_of_surgery 
df$T_d <- df$D_d - df$Date_of_surgery

df <- mutate(df, ALBI_score = -0.085 *Albumin + 0.66*log10(Total_bilirubin))
df$ALBI_grade <-cut(df$ALBI_score, 
                        breaks= c(-500, -2.60,-1.39,500),
                        labels= c(1,2,3))

# solitary tumor
df <- mutate(df, Solitary = as.numeric(Tumor_number==1))

# recurrence <2yr
df <- mutate(df, Early = as.numeric((T_r<365.25*2)) & (E_r==1))

#---------- Inspection and selection -----------------------------------------
#head(df)
#names(df)
#summary(df)
#str(df)

# Nr patients per center
table(df$Center)

# Nr patients total and unique
length(df$Patient_ID)
length(unique(df$Patient_ID))


# Select patients before 05/02/2018 to assure FU >= 2yr
df_sel <- filter(df, df$Date_of_surgery < as.Date("2018-02-05"))
table(df_sel$Center) # nr patients per center


# Select relevant columns
col_sel <-  c("Patient_ID", "Sex", "Age","Birth_date", "Date_of_surgery", "HBV", "HCV", 
              "Child_pugh_score", "Child_pug_class",  "Albumin", "Total_bilirubin", "ALBI_score", "ALBI_grade",
              "AFP", "Tumor_size", "Tumor_number", "Solitary", "Microvascular_invasion", "Differentiation", 
              "D_r", "T_r", "E_r", "Early", "D_d", "T_d", "E_d", "Center", 
              "Extended hemihepatectomie rechts", "Extended hemihepatectomie links", "Hemihepatectomie rechts" , "Hemihepatectomie links", 
              "Segment resectie", "Wigresectie", "Open_RFA", "Resection_margin", "Adjuvant_chemotherapy", 
              "Intrahepatic_recurrence", "Extrahepatic_recurrence", 
              "Reoperation", "Salvage_LT", "Ablation", "TACE", "RTx", "Systemic_chemotherapy", "Best_supportive_care" 
              )


df_sel <- df_sel[, col_sel]

#---------- Baseline Table ------------------------------------------------

#### complete case #############################################
col_pre <- c('Sex', 'Albumin', 'Total_bilirubin', 'AFP', 'Tumor_size', 'Tumor_number', 'D_r', 'E_r')
df_sel <- df_sel[complete.cases(df_sel[,col_pre]), ]
table(df_sel$Center)
#
# in R niet gechecked voor tumor number 0,99
# in R niet gechecked voor tumor size > 0
#### #############################################

df_emc_baseline = filter(df_sel, df_sel$Center=='ErasmusMC')
df_jp_baseline = filter(df_sel, df_sel$Center=='Okayama')

# Number cases
print(paste('Totaal aantal cases :', dim(df_sel)[1]))
table(df_sel$Center)

# Gender
print('Gender - ErasmusMC')
t <- table(recode(df_emc_baseline$Sex, '0'='Male', '1'='Female'))
t
print(round(t[2]/(t[1] + t[2]),2))

print('Gender - Okayama')
t <- table(recode(df_jp_baseline$Sex, '0'='Male', '1'='Female'))
t
print(round(t[2]/(t[1] + t[2]),2))

# Age
print('Age - ErasmusMC')
round(as.numeric(mean(df_emc_baseline$Age, na.rm=TRUE)))
round(sd(df_emc_baseline$Age, na.rm=TRUE))

print('Age - Okayama')
round(mean(df_jp_baseline$Age, na.rm=TRUE))
round(sd(df_jp_baseline$Age, na.rm=TRUE))

# Hep B
print('Hep B - ErasmusMC')
t <- table(df_emc_baseline$HBV)
t
round(as.numeric(t[2]/(t[1] + t[2])),2)

print('Hep B - Okayama')
t <- table(df_jp_baseline$HBV)
t
round(as.numeric(t[2]/(t[1] + t[2])),2)

# Hep C
print('Hep C - ErasmusMC')
t <- table(df_emc_baseline$HCV)
t
round(as.numeric(t[2]/(t[1] + t[2])),2)


print('Hep C - Okayama')
t <- table(df_jp_baseline$HCV)
t
round(as.numeric(t[2]/(t[1] + t[2])),2)

# Child pugh
print('Child pugh - ErasmusMC')
table(df_emc_baseline$Child_pug_class)

print('Child pugh - Okayama')
table(df_jp_baseline$Child_pug_class)

# Albi grade
print('ALBI grade - ErasmusMC')
table(df_emc_baseline$ALBI_grade)

print('ALBI grade - Okayama')
table(df_jp_baseline$ALBI_grade)

#Albumin
print('Albumin - Erasmus MC')
round(mean(df_emc_baseline$Albumin, na.rm=TRUE))
round(sd(df_emc_baseline$Albumin, na.rm=TRUE), 1)

print('Albumin - Okayama')
round(mean(df_jp_baseline$Albumin, na.rm=TRUE))
round(sd(df_jp_baseline$Albumin, na.rm=TRUE), 1)

#Bilirubin
print('Bilirubin - Erasmus MC')
round(median(df_emc_baseline$Total_bilirubin, na.rm=TRUE))
round(quantile(df_emc_baseline$Total_bilirubin, na.rm=TRUE))[c(2,4)]

print('Bilirubin - Okayama')
round(median(df_jp_baseline$Total_bilirubin, na.rm=TRUE))
round(quantile(df_jp_baseline$Total_bilirubin, na.rm=TRUE))[c(2,4)]

#AFP
print('AFP - Erasmus MC')
round(median(df_emc_baseline$AFP, na.rm=TRUE))
round(quantile(df_emc_baseline$AFP, na.rm=TRUE))[c(2,4)]

print('AFP - Okayama')
round(median(df_jp_baseline$AFP, na.rm=TRUE))
round(quantile(df_jp_baseline$AFP, na.rm=TRUE))[c(2,4)]

#Tumor size
print('Tumor size - Erasmus MC')
round(median(df_emc_baseline$Tumor_size, na.rm=TRUE) * 10)
round(quantile(df_emc_baseline$Tumor_size, na.rm=TRUE) * 10)[c(2,4)]

print('Tumor size - Okayama')
round(median(df_jp_baseline$Tumor_size, na.rm=TRUE) * 10)
round(quantile(df_jp_baseline$Tumor_size, na.rm=TRUE) * 10)[c(2,4)]

#Solitary tumor
print('Solitary tumor - Erasmus MC')
t <- table(df_emc_baseline$Solitary)
t
round(as.numeric(t[2]/(t[1] + t[2])),2)

print('Solitary tumor - Okayama')
t <- table(df_jp_baseline$Solitary)
t
round(as.numeric(t[2]/(t[1] + t[2])),2)

#Microvascular invasion
print('MVI - Erasmus MC')
t <- table(df_emc_baseline$Microvascular_invasion)
t
round(as.numeric(t[2]/(t[1] + t[2])),2)

print('MVI - Okayama')
t <- table(df_jp_baseline$Microvascular_invasion)
t
round(as.numeric(t[2]/(t[1] + t[2])),2)

#Recurrence <2 yr
print('rec <2yr - Erasmus MC')  ### afwijkend
t <- table(df_emc_baseline$Early)
t
round(as.numeric(t[2]/(t[1] + t[2])),2)

print('rec <2yr - Okayama')    
t <- table(df_jp_baseline$Early)
t
round(as.numeric(t[2]/(t[1] + t[2])),2)


#Median recurrence free survival
print('Median RFS - Erasmus MC')
surv.obj.emc  <- Surv(time = (df_emc_baseline$T_r/365.25*12), event = df_emc_baseline$E_r)
fit <- survfit(surv.obj.emc ~ 1, data=df_emc_baseline)
fit

print('Median RFS - Japan')
surv.obj.jp  <- Surv(time = (df_jp_baseline$T_r/365.25*12), event = df_jp_baseline$E_r)
fit <- survfit(surv.obj.jp ~ 1, data=df_jp_baseline)
fit

# Median followup
print('Median recurrence free survival in months - Erasmus MC (pre)')
surv.obj.emc  <- Surv(time = (df_emc_baseline$T_r/365*12), event = !df_emc_baseline$E_r)
fit <- survfit(surv.obj.emc ~ 1, data=df_emc_baseline)
summary(fit)

print('Median recurrence free survival in months - Erasmus MC (post)')
surv.obj.emc.post  <- Surv(time = (df_emc_baseline$T_r/365*12), event = !df_emc_baseline$E_r)
fit <- survfit(surv.obj.emc ~ 1, data=df_emc_baseline)
summary(fit)

print('Median recurrence free survival in months - Okayama')
surv.obj.jp  <- Surv(time = (df_jp_baseline$T_r/365*12), event = !df_jp_baseline$E_r)
fit <- survfit(surv.obj.jp ~ 1, data=df_jp_baseline)
summary(fit)

#"Extended hemihepatectomie rechts", "Extended hemihepatectomie links", "Hemihepatectomie rechts" , "Hemihepatectomie links", 
#"Segment resectie", "Wigresectie", "Open_RFA", "Resection_margin", "Adjuvant_chemotherapy", 
#"Intrahepatic_recurrence", "Extrahepatic_recurrence", 
#"Reoperation", "Salvage_LT", "Ablation", "TACE", "RTx", "Systemic_chemotherapy", "Best_supportive_care", 


## Type of resection
print('Extended hemi right (EMC-JP)')
t <- table(df_emc_baseline["Extended hemihepatectomie rechts"])
t
#round(as.numeric(t[2]/(t[1] + t[2])),2)

t <- table(df_jp_baseline["Extended hemihepatectomie rechts"])
t
#round(as.numeric(t[2]/(t[1] + t[2])),2)

print('Extended hemi left (EMC-JP)')
t <- table(df_emc_baseline["Extended hemihepatectomie links"])
t
#round(as.numeric(t[2]/(t[1] + t[2])),2)

t <- table(df_jp_baseline["Extended hemihepatectomie links"])
t
#round(as.numeric(t[2]/(t[1] + t[2])),2)

print('Hemi right (EMC-JP)')
t <- table(df_emc_baseline["Hemihepatectomie rechts"])
t
#round(as.numeric(t[2]/(t[1] + t[2])),2)

t <- table(df_jp_baseline["Hemihepatectomie rechts"])
t
#round(as.numeric(t[2]/(t[1] + t[2])),2)

print('Hemi left (EMC-JP)')
t <- table(df_emc_baseline["Hemihepatectomie links"])
t
#round(as.numeric(t[2]/(t[1] + t[2])),2)

t <- table(df_jp_baseline["Hemihepatectomie links"])
t
#round(as.numeric(t[2]/(t[1] + t[2])),2)

print('segment resectie (EMC-JP)')
t <- table(df_emc_baseline["Segment resectie"])
t
#round(as.numeric(t[2]/(t[1] + t[2])),2)

t <- table(df_jp_baseline["Segment resectie"])
t
#round(as.numeric(t[2]/(t[1] + t[2])),2)

print('Wigresectie(EMC-JP)')
t <- table(df_emc_baseline["Wigresectie"])
t
#round(as.numeric(t[2]/(t[1] + t[2])),2)

t <- table(df_jp_baseline["Wigresectie"])
t
#round(as.numeric(t[2]/(t[1] + t[2])),2)

print('Open_RFA (EMC-JP)')
t <- table(df_emc_baseline["Open_RFA"])
t 
#round(as.numeric(t[2]/(t[1] + t[2])),2)

t <- table(df_jp_baseline["Open_RFA"])
t
#round(as.numeric(t[2]/(t[1] + t[2])),2)


## Margin
print('Margin status (EMC-JP')
t <- table(df_emc_baseline["Resection_margin"])
t
round(as.numeric(t[2]/(t[1] + t[2])),2)

t <- table(df_jp_baseline["Resection_margin"])
t
round(as.numeric(t[2]/(t[1] + t[2])),2)


## Type of recurrence 
print('Intrahepatic recurrence (EMC-JP')
t <- table(df_emc_baseline["Intrahepatic_recurrence"])
t
round(as.numeric(t[2]/(t[1] + t[2])),2)

t <- table(df_jp_baseline["Intrahepatic_recurrence"])
t
round(as.numeric(t[2]/(t[1] + t[2])),2)


print('Extrahepatic recurrence (EMC-JP')
t <- table(df_emc_baseline["Extrahepatic_recurrence"])
t
round(as.numeric(t[2]/(t[1] + t[2])),2)

t <- table(df_jp_baseline["Extrahepatic_recurrence"])
t
round(as.numeric(t[2]/(t[1] + t[2])),2)

sum(df_emc_baseline$E_r)

View(df_emc_baseline %>% filter(E_r == 1) %>% select(c(Patient_ID, Extrahepatic_recurrence, Intrahepatic_recurrence)))

## treatment of recurrence
print('reoperation (EMC-JP')
t <- table(df_emc_baseline["Reoperation"])
t
round(as.numeric(t[2]/(t[1] + t[2])),2)

t <- table(df_jp_baseline["Reoperation"])
t
round(as.numeric(t[2]/(t[1] + t[2])),2)


print('Salvage_LT  (EMC-JP')
t <- table(df_emc_baseline["Salvage_LT"])
t
round(as.numeric(t[2]/(t[1] + t[2])),2)

t <- table(df_jp_baseline["Salvage_LT"])
t
round(as.numeric(t[2]/(t[1] + t[2])),2)

print('Ablation  (EMC-JP')
t <- table(df_emc_baseline["Ablation"])
t
round(as.numeric(t[2]/(t[1] + t[2])),2)

t <- table(df_jp_baseline["Ablation"])
t
round(as.numeric(t[2]/(t[1] + t[2])),2)



print('TACE  (EMC-JP')
t <- table(df_emc_baseline["TACE"])
t
round(as.numeric(t[2]/(t[1] + t[2])),2)

t <- table(df_jp_baseline["TACE"])
t
round(as.numeric(t[2]/(t[1] + t[2])),2)



print('RTx  (EMC-JP')
t <- table(df_emc_baseline["RTx"])
t
round(as.numeric(t[2]/(t[1] + t[2])),2)

t <- table(df_jp_baseline["RTx"])
t
round(as.numeric(t[2]/(t[1] + t[2])),2)



print('Chemo  (EMC-JP')
t <- table(df_emc_baseline["Systemic_chemotherapy"])
t
round(as.numeric(t[2]/(t[1] + t[2])),2)

t <- table(df_jp_baseline["Systemic_chemotherapy"])
t
round(as.numeric(t[2]/(t[1] + t[2])),2)



print('Bests supportive care  (EMC-JP')
t <- table(df_emc_baseline["Best_supportive_care"])
t
round(as.numeric(t[2]/(t[1] + t[2])),2)

t <- table(df_jp_baseline["Best_supportive_care"])
t
round(as.numeric(t[2]/(t[1] + t[2])),2)




## complete case analysis ###
col_pre <- c('Sex', 'Albumin', 'Total_bilirubin', 'AFP', 'Tumor_size', 'Tumor_number', 'D_r', 'E_r')
df_complete <- df_sel[complete.cases(df_sel[,col_pre]), ]

# nr per center
table(df_complete$Center)



#---------- Calculate ERASL scores ---------------------------------------------

# reformat data
df_complete <- mutate(df_complete, 
                        #gender = as.numeric(Sex == 0),
                        gender = as.numeric(Sex == 'M'),
                        albigrade_1 = as.numeric(as.numeric(ALBI_grade) >1),
                        log_afp = log(AFP),
                        log_diamet = log(Tumor_size))

df_complete$aantaltum_23[df_complete$Tumor_number >= 4] <- 2
df_complete$aantaltum_23[(df_complete$Tumor_number == 2) |(df_complete$Tumor_number == 3)] <- 1
df_complete$aantaltum_23[df_complete$Tumor_number <2] <- 0  

df_complete <- mutate(df_complete, pre_score = 0.818*gender +
                                               0.447*albigrade_1 +
                                               0.100*log_afp +
                                               0.580*log_diamet +
                                               0.492*aantaltum_23)

# ERASL pre
df_complete$pre_risk <-cut(df_complete$pre_score, 
                    breaks= c(-500, 2.558, 3.521,500),
                    labels= c('low', 'intermediate', 'high'))

df_complete$pre_risk_n <-cut(df_complete$pre_score, 
                              breaks= c(-500, 2.558, 3.521,500),
                              labels= c(1, 2, 3))

df_complete <- mutate(df_complete, 
                      pre_low = as.numeric(pre_risk_n==1), 
                      pre_intermediate = as.numeric(pre_risk_n==2),
                      pre_high = as.numeric(pre_risk_n==3)
)

# ERASL post
df_complete <- mutate(df_complete, post_score = 0.677*gender +
                                                0.458*albigrade_1 +
                                                0.082*log_afp +
                                                0.451*log_diamet +
                                                0.379*aantaltum_23 +
                                                0.661*Microvascular_invasion)

df_complete$post_risk <-cut(df_complete$post_score, 
                           breaks= c(-500, 2.332, 3.445,500),
                           labels= c('low', 'intermediate', 'high'))

df_complete$post_risk_n <-cut(df_complete$post_score, 
                            breaks= c(-500, 2.332, 3.445,500),
                            labels= c(1, 2, 3))

df_complete <- mutate(df_complete, 
                      post_low = as.numeric(post_risk_n==1), 
                      post_intermediate = as.numeric(post_risk_n==2),
                      post_high = as.numeric(post_risk_n==3)
                      )



# split emc and jp describe erasl pre_risk and pre_scores (pooling?)
df_emc_erasl <- filter(df_complete, Center == 'ErasmusMC')                        
print('ERASL pre - Erasmus MC')

mean(df_emc_erasl$pre_score, na.rm=TRUE)                
sd(df_emc_erasl$pre_score, na.rm=TRUE)
table(df_emc_erasl$pre_risk)

print('ERASL post - Erasmus MC')
mean(df_emc_erasl$post_score, na.rm=TRUE)
sd(df_emc_erasl$post_score, na.rm=TRUE)
table(df_emc_erasl$post_risk)

df_jp_erasl  <- filter(df_complete, Center == 'Okayama')
print('ERASL pre - Okayama')
mean(df_jp_erasl$pre_score, na.rm=TRUE)
sd(df_jp_erasl$pre_score, na.rm=TRUE)
table(df_jp_erasl$pre_risk)

print('ERASL post - Okayama')
mean(df_jp_erasl$post_score, na.rm=TRUE)
sd(df_jp_erasl$post_score, na.rm=TRUE)
table(df_jp_erasl$post_risk)            

#---------- Survival analsis ERASL risk groups --------------------
col = c('Patient_ID', 'T_r', 'E_r', 'pre_score', 'pre_risk', 'pre_risk_n', 'post_score', 'post_risk', 'post_risk_n', 'Center')
df_tmp <- df_complete[, col]

df_tmp_emc_pre = subset(df_tmp, df_tmp$Center == "ErasmusMC")
df_tmp_emc_post = subset(df_tmp_emc_pre, complete.cases(df_tmp_emc_pre))

df_tmp_jp_pre = subset(df_tmp, df_tmp$Center == "Okayama")
df_tmp_jp_post = subset(df_tmp_jp_pre, complete.cases(df_tmp_jp_pre))

print('emc pre')
fit <- survfit(Surv(time = df_tmp_emc_pre$T_r/365.25*12, event = df_tmp_emc_pre$E_r) ~ df_tmp_emc_pre$pre_risk)
fit

print('jp pre')
fit <- survfit(Surv(time = df_tmp_jp_pre$T_r/365.25*12, event = df_tmp_jp_pre$E_r) ~ df_tmp_jp_pre$pre_risk)
fit

print('emc post')
fit <- survfit(Surv(time = df_tmp_emc_post$T_r/365.25*12, event = df_tmp_emc_post$E_r) ~ df_tmp_emc_post$post_risk)
fit

print('jp post')
fit <- survfit(Surv(time = df_tmp_jp_post$T_r/365.25*12, event = df_tmp_jp_post$E_r) ~ df_tmp_jp_post$post_risk)
fit

#---------- FU truncation after 24 months --------------------------
# set E for ( E = 1 and T > 24) --> E = 0 
# set T for (T> 24)             --> T = 24

df_compl_trunc = df_complete

df_compl_trunc$E_r[(df_compl_trunc$E_r == TRUE) & (df_compl_trunc$T_r) > 2*365] <- 0
df_compl_trunc$T_r[df_compl_trunc$T_r > 365*2] <- 365*2
df_compl_trunc$T_r <- as.numeric(df_compl_trunc$T_r)


col_pre <- c('Patient_ID', 'pre_score', 'pre_risk', 'pre_risk_n', 'pre_low', 'pre_intermediate', 'pre_high',
             'T_r', 'E_r', 
             'gender', 'albigrade_1', 'log_afp', 'log_diamet', 'aantaltum_23', 'HCV', 'HBV')

col_post <- c('Patient_ID','post_score', 'post_risk', 'post_risk_n','post_low', 'post_intermediate', 'post_high',
              'T_r', 'E_r',
              'gender', 'albigrade_1', 'log_afp', 'log_diamet', 'aantaltum_23', 'Microvascular_invasion', 'HCV', 'HBV')

df_pre_emc <- filter(df_compl_trunc, df_compl_trunc$Center == 'ErasmusMC')[,col_pre]
df_post_emc <- filter(df_compl_trunc, df_compl_trunc$Center == 'ErasmusMC')[,col_post]
df_post_emc <- df_post_emc[complete.cases(df_post_emc[, col_post]), ]
dim(df_post_emc)

df_pre_jp <- filter(df_compl_trunc, df_compl_trunc$Center == 'Okayama')[,col_pre]
df_post_jp <- filter(df_compl_trunc, df_compl_trunc$Center == 'Okayama')[,col_post]

write.csv(df_pre_emc, "E:\\7. ERASL validation\\data\\intrim\\pre_ERASL_emc.csv")
write.csv(df_pre_jp, "E:\\7. ERASL validation\\data\\intrim\\pre_ERASL_jp.csv")
write.csv(df_post_emc, "E:\\7. ERASL validation\\data\\intrim\\post_ERASL_emc.csv")
write.csv(df_post_jp, "E:\\7. ERASL validation\\data\\intrim\\post_ERASL_jp.csv")

pre.surv.obj.emc  <- Surv(time = df_pre_emc$T_r/365.25*12, event = df_pre_emc$E_r)
post.surv.obj.emc <- Surv(time = df_post_emc$T_r/365.25*12, event = df_post_emc$E_r)

pre.surv.obj.jp  <- Surv(time = df_pre_jp$T_r/365.25*12, event = df_pre_jp$E_r)
post.surv.obj.jp <- Surv(time = df_post_jp$T_r/365.25*12, event = df_post_jp$E_r)


#---------- Misspecification ------------------------------------
# offset regressions
pre.refit = coxph(pre.surv.obj.emc~gender+albigrade_1+log_afp+log_diamet+aantaltum_23+offset(pre_score), data=df_pre_emc)
print('EMC pre')
pre.refit

pre.refit = coxph(pre.surv.obj.jp~gender+albigrade_1+log_afp+log_diamet+aantaltum_23+offset(pre_score), data=df_pre_jp)
print('Japan pre')
pre.refit

post.refit = coxph(post.surv.obj.emc~gender+albigrade_1+log_afp+log_diamet+aantaltum_23+Microvascular_invasion+offset(post_score), data=df_post_emc)
print('emc post')
post.refit

post.refit = coxph(post.surv.obj.jp~gender+albigrade_1+log_afp+log_diamet+aantaltum_23+Microvascular_invasion+offset(post_score), data=df_post_jp)
print('jp post')
post.refit

# Inspection of proportional hazards assumption
# emc pre
pre.refit = coxph(pre.surv.obj.emc~pre_score, data=df_pre_emc)
tmp <- cox.zph(pre.refit, transform="identity", global=T)           # transform='km'
print(tmp)
plot(tmp)
abline(h=0, lty=3)

pre.refit = coxph(pre.surv.obj.emc~gender+albigrade_1+log_afp+log_diamet+aantaltum_23, data=df_pre_emc)
tmp <- cox.zph(pre.refit, transform="identity", global=T)
print(tmp)
plot(tmp)

# jp pre
pre.refit = coxph(pre.surv.obj.jp~pre_score, data=df_pre_jp)
tmp <- cox.zph(pre.refit, transform="identity", global=T)          #transform='km'
print(tmp)
plot(tmp)
abline(h=0, lty=3)

pre.refit = coxph(pre.surv.obj.jp~gender+albigrade_1+log_afp+log_diamet+aantaltum_23, data=df_pre_jp)
tmp <- cox.zph(pre.refit, transform="identity", global=T)
print(tmp)
plot(tmp)


# emc post
tmp <- cox.zph(post.refit, transform="identity", global=T)
print(tmp)
plot(tmp)
abline(h=0, lty=3)

post.refit = coxph(post.surv.obj.emc~post_score + post_score*T_r, data=df_post_emc)
post.refit

post.refit = coxph(post.surv.obj.emc~gender+albigrade_1+log_afp+log_diamet+aantaltum_23+Microvascular_invasion, data=df_post_emc)
tmp <- cox.zph(post.refit, transform="identity", global=T)
print(tmp)
plot(tmp)

# jp post
post.refit = coxph(post.surv.obj.jp~gender+albigrade_1+log_afp+log_diamet+aantaltum_23+Microvascular_invasion, data=df_post_jp)
tmp <- cox.zph(post.refit, transform="identity", global=T)
print(tmp)
plot(tmp)

#---------- Discrimination ---------------------------------

# Calibration slope - Harrel's C
###  documentation
#cph:       https://www.rdocumentation.org/packages/rms/versions/5.1-3.1/topics/cph
#validate:  https://www.rdocumentation.org/packages/rms/versions/5.1-3.1/topics/validate.cph

#### pre operative risk score ErasmusMC
print("ERASMUSMC")
print("calibration slope") # regression on LP

cox.mod.pre <- cph(pre.surv.obj.emc ~ pre_score, data = df_pre_emc, init = 1, x=T, y=T, surv = T)
cox.mod.pre

print("optimism corrected harrel's C")
set.seed(1)
Dxy = validate(cox.mod.pre, method='boot', B=200)[29]
Hc = Dxy/2 + 0.5;
print(Hc)

print("uncorrected harrel's C")
Dxy = validate(cox.mod.pre, method='boot', B=200)[1]
Hc = Dxy/2 + 0.5;
print(Hc)

harrel_c <- function(data, indices){
  d <- data[indices,]
  boot.surv <- Surv(time = d$T_r, event = d$E_r)
  
  Dxy = unname(rcorr.cens(x=d$pre_score, S=boot.surv)[2])
  harrel_c = -Dxy/2 + 0.5;
  return(harrel_c)
}

# ordinary non parametric bootstrap
# note no optimism correction
boot.obj <- boot(data=df_pre_emc, statistic=harrel_c, R=200)
print(boot.obj)


#### pre operative risk score Okayama
print("OKAYAMA")
print("calibration slope")

cox.mod.pre <- cph(pre.surv.obj.jp ~ pre_score, data = df_pre_jp, init = 1, x=T, y=T, surv = T)
cox.mod.pre

print("optimism corrected harrel's C")
set.seed(1)
Dxy = validate(cox.mod.pre, method='boot', B=200)[29]
Hc = Dxy/2 + 0.5;
print(Hc)

print("uncorrected harrel's C")
Dxy = validate(cox.mod.pre, method='boot', B=200)[1]
Hc = Dxy/2 + 0.5;
print(Hc)

harrel_c <- function(data, indices){
  d <- data[indices,]
  boot.surv <- Surv(time = d$T_r, event = d$E_r)
  
  Dxy = unname(rcorr.cens(x=d$pre_score, S=boot.surv)[2])
  harrel_c = -Dxy/2 + 0.5;
  return(harrel_c)
}

# ordinary non parametric bootstrap
# note no optimism correction
boot.obj <- boot(data=df_pre_jp, statistic=harrel_c, R=200)
print(boot.obj)

#### post operative risk score ErasmusMC
print("calibration slope ErasmusMC")
cox.mod.post <- cph(post.surv.obj.emc ~ post_score, data = df_post_emc, init = 1, x=T, y=T, surv = T)
cox.mod.post

print("optimism corrected harrel's C")
set.seed(1)
Dxy = validate(cox.mod.post, method='boot', B=200)[29]  #[29] optimism correction [1] origional 
harrel_c = Dxy/2 + 0.5;
print(harrel_c)

harrel_c <- function(data, indices){
  d <- data[indices,]
  boot.surv <- Surv(time = d$T_r, event = d$E_r)
  
  Dxy = unname(rcorr.cens(x=d$post_score, S=boot.surv)[2])
  harrel_c = -Dxy/2 + 0.5;
  return(harrel_c)
}

# ordinary non parametric bootstrap
# note no optimism correction
boot.obj <- boot(data=df_post_emc, statistic=harrel_c, R=200)
print(boot.obj)


#### post operative risk score Okayama
print("calibration slope Okayama")
cox.mod.post <- cph(post.surv.obj.jp ~ post_score, data = df_post_jp, init = 1, x=T, y=T, surv = T)
cox.mod.post

print("optimism corrected harrel's C")
set.seed(1)
Dxy = validate(cox.mod.post, method='boot', B=200)[29]  #[29] optimism correction [1] origional 
harrel_c = Dxy/2 + 0.5;
print(harrel_c)

harrel_c <- function(data, indices){
  d <- data[indices,]
  boot.surv <- Surv(time = d$T_r, event = d$E_r)
  
  Dxy = unname(rcorr.cens(x=d$post_score, S=boot.surv)[2])
  harrel_c = -Dxy/2 + 0.5;
  return(harrel_c)
}

# ordinary non parametric bootstrap
# note no optimism correction
boot.obj <- boot(data=df_post_jp, statistic=harrel_c, R=200)
print(boot.obj)


# Gonen and heller concordance index
gh <- function(data, indices){
  d <- data[indices]
  gh_stat <- GHCI(d)
  return(gh_stat)
}

print('ErasmusMC pre')
boot.gh.pre <- boot(data=df_pre_emc$pre_score, statistic=gh, R=200)
print(boot.gh.pre)

print('Okayama Pre')
boot.gh.pre <- boot(data=df_pre_jp$pre_score, statistic=gh, R=200)
print(boot.gh.pre)

print('ErasmusMC Post')
boot.gh.post <- boot(data=df_post_emc$post_score, statistic=gh, R=200)   ## NA/NaN/Inf in foreing function call (arg1)
print(boot.gh.post)

print('Okayama Post')
boot.gh.post <- boot(data=df_post_jp$post_score, statistic=gh, R=200)
print(boot.gh.post)

## Royston and Sauerbrei's D
kap = sqrt(8/pi)
sd = pi^2/6

Rd_pre <- function(data, indices){
  d <- data[indices,]
  D = D.index(d$pre_score, d$T_r, d$E_r)$coef
  Rd = (D^2)/(kap^2)/(sd + (D^2/(kap^2)))
  return(Rd)
}

# ErasmusMC pre
boot.Rd.pre <- boot(data=df_pre_emc, statistic=Rd_pre, R=200)
print('ErasmusMC pre')
print(boot.Rd.pre)

# Okayama pre
boot.Rd.pre <- boot(data=df_pre_jp, statistic=Rd_pre, R=200)
print('Okayama pre')
print(boot.Rd.pre)

Rd <- function(data, indices){
  d <- data[indices,]
  D = D.index(d$post_score, d$T_r, d$E_r)$coef
  Rd = (D^2)/(kap^2)/(sd + (D^2/(kap^2)))
  return(Rd)
}

# ErasmusMC post
boot.Rd.post <- boot(data=df_post_emc, statistic=Rd, R=200)
print('ErasmusMC post')
print(boot.Rd.post)

# Okayama post
boot.Rd.post <- boot(data=df_post_jp, statistic=Rd, R=200)
print('Okayama post')
print(boot.Rd.post)

## Cumulative/dynamic tdAUC
tdauc <- function(data, indices){
  d <- data[indices,]
  
  d.surv.obj <- Surv(time = d$T_r, event = d$E_r)
  
  res.AUC.cd <- AUC.cd(Surv.rsp = d.surv.obj,
                       Surv.rsp.new = d.surv.obj,
                       lp           = d$pre_score,
                       lpnew        = d$pre_score,
                       times        = 0:2 * 365.25
  )
  return(res.AUC.cd$iauc)}

boot.tdauc.pre <- boot(data=df_pre_emc, statistic=tdauc, R=200)
print(boot.tdauc.pre)

boot.tdauc.pre <- boot(data=df_pre_jp, statistic=tdauc, R=200)
print(boot.tdauc.pre)

tdauc <- function(data, indices){
  d <- data[indices,]
  
  d.surv.obj <- Surv(time = d$T_r, event = d$E_r)
  
  res.AUC.cd <- AUC.cd(Surv.rsp = d.surv.obj,
                       Surv.rsp.new = d.surv.obj,
                       lp           = d$post_score,
                       lpnew        = d$post_score,
                       times        = 0:2 * 365.25
  )
  return(res.AUC.cd$iauc)}

boot.tdauc.post <- boot(data=df_post_emc, statistic=tdauc, R=200)
print(boot.tdauc.post)

boot.tdauc.post <- boot(data=df_post_jp, statistic=tdauc, R=200)
print(boot.tdauc.post)

## kaplan meier survival plots grouped by risk group
# ErasmusMC post
#tmp <- df_pre_emc[df_pre_emc$pre_risk_n %in% c(1,2),]
#tmp.surv.obj <- Surv(time = tmp$T, event = tmp$E)
fit <- survfit(pre.surv.obj.emc ~ pre_risk_n, data=df_pre_emc)
plot(fit, col=c("green", "orange", "red"), xlab = "time in days", ylab="Survival rate", main="Progression free survival ERASL-pre score (EMC)", las=1)
legend("topright", legend=c("low", "intermediate", "high"), col=c("green", "orange", "red"), lwd=2)

# Okayama pre
fit <- survfit(pre.surv.obj.jp ~ pre_risk_n, data=df_pre_jp)
plot(fit, col=c("green", "orange", "red"), xlab = "time in days", ylab="Survival rate", main="Progression free survival ERASL-pre score (JP)", las=1)
legend("topright", legend=c("low", "intermediate", "high"), col=c("green", "orange", "red"), lwd=2)

# ErasmusMC post
fit <- survfit(post.surv.obj.emc ~ post_risk_n, data=df_post_emc)
plot(fit, col=c("green", "orange", "red"), xlab = "time in days", ylab="Survival rate", main="Progression free survival ERASL-post score (EMC)", las=1)
legend("topright", legend=c("low", "intermediate", "high"), col=c("green", "orange", "red"), lwd=2)

# Okayama post
fit <- survfit(post.surv.obj.jp ~ post_risk_n, data=df_post_jp)
plot(fit, col=c("green", "orange", "red"), xlab = "time in days", ylab="Survival rate", main="Progression free survival ERASL-post score (JP)", las=1)
legend("topright", legend=c("low", "intermediate", "high"), col=c("green", "orange", "red"), lwd=2)


## Hazard Ratio between risk groups
# Erasmus MC pre
cox.mod <- coxph(pre.surv.obj.emc ~ pre_intermediate + pre_high, data=df_pre_emc)
print('Erasmus mc - pre')
cox.mod  ## alsnog transformeren se(coef)
round(exp(confint(cox.mod, level = 0.95)), 2)

# Okayama pre
cox.mod <- coxph(pre.surv.obj.jp ~ pre_intermediate + pre_high, data=df_pre_jp)
print('Okayama - pre')
cox.mod
round(exp(confint(cox.mod, level = 0.95)), 2)

# Erasmus MC post
sum(df_post_emc$post_high)
print(' ^ check sum post high')

cox.mod <- coxph(post.surv.obj.emc ~ post_intermediate +post_high, data=df_post_emc)
print('ErasmusMC - post')
cox.mod
round(exp(confint(cox.mod, level = 0.95)), 2)

# Okayama post
cox.mod <- coxph(post.surv.obj.jp ~ post_intermediate + post_high, data=df_post_jp)
print('Okayama - post')
cox.mod
round(exp(confint(cox.mod, level = 0.95)), 2)

## Baseline hazard function: 

df_pre_emc$T_r
# erasmus mc
cox.mod.pre <- cph(pre.surv.obj.emc ~ offset(pre_score), x=T, y=T, data=df_pre_emc)
cox.mod.post <- cph(post.surv.obj.emc ~ offset(post_score), x=T, y=T, data=df_post_emc)

# cumulative baseline hazard H0(t)
write.csv(basehaz(cox.mod.pre, centered=T), "E:\\7. ERASL validation\\data\\intrim\\H0_pre_emc.csv")
write.csv(basehaz(cox.mod.post, centered=T), "E:\\7. ERASL validation\\data\\intrim\\H0_post_emc.csv")

# japan
cox.mod.pre <- cph(pre.surv.obj.jp ~ offset(pre_score), x=T, y=T, data=df_pre_jp)
cox.mod.post <- cph(post.surv.obj.jp ~ offset(post_score), x=T, y=T, data=df_post_jp)

write.csv(basehaz(cox.mod.pre, centered=T), 'E:\\7. ERASL validation\\data\\intrim\\H0_pre_jp.csv')
write.csv(basehaz(cox.mod.post, centered=T), 'E:\\7. ERASL validation\\data\\intrim\\H0_post_jp.csv')

#---------- Calibration -----------------------------------------------------------

# Appologies for the jump to python. At the moment of writing this script I'm a beginner in R and after wrestling
# with the visualisations for a few hours it was more convienant to load the data into python, as I could make the graphs more clear. 
# The python code is included in the suplementary files aswell.

#--------------------------------------------------------------------------------------
# compute - interpolated baseline survival function to_s0
# transform - empirical baseline hazard function to S0 at median erasl

# for PRE and POST
# - plot aprroximated published baseline survival curve
# - plot empirical baseline curves of EMc/JP

# for PRE and POST
# compute ERASL prediction of RFS prob at time t for grid of n points between t=0, t=24
# Average the RFS probs at t per risk group
# Plot average risk over time for each risk group
# Super impose kaplan meier plot per risk group

# For JP also construct alternative calibration plot
# at t=12,24
# extract km survival probabilities for each risk group
# extract predicted survival probabilities for each risk group
# transform these to log(-log()) scale
# make the plot
#----------------------------------------------------------------------------------------

## Construct baseline hazard function from derivation set

t0 <-c(0,1,3,6,12,18,24)
s_pre  <- c(0.9999,0.998, 0.975, 0.915, 0.822, 0.753, 0.694)
s_post <- c(0.9999,0.998, 0.975, 0.917, 0.826, 0.757, 0.698)

to_H0 <- function(x, median)
  {
  s0 = exp((log(x))/(exp(median)))
  H0 = -log(s0)
  return(H0)
  }

# convert
H0_pre <- sapply(s_pre, to_H0, median=2.558)
H0_post <- sapply(s_post, to_H0, median=2.332)

f_H0_pre <- smooth.spline(t0, H0_pre)
f_H0_post <- smooth.spline(t0, H0_post)

df_pre_emc$T_pre_star_lit  = predict(f_H0_pre, as.numeric(df_pre_emc$T_r)/365.25*12)$y
df_pre_jp$T_pre_star_lit  = predict(f_H0_pre, as.numeric(df_pre_jp$T_r)/365.25*12)$y
df_post_emc$T_post_star_lit  = predict(f_H0_post, as.numeric(df_post_emc$T_r)/365.25*12)$y
df_post_jp$T_post_star_lit  = predict(f_H0_post, as.numeric(df_post_jp$T_r)/365.25*12)$y

# reinforce lower bound for ln()
df_pre_emc$T_pre_star_lit[df_pre_emc$T_pre_star_lit <=0] <- 0.0000001
df_post_emc$T_post_star_lit[df_post_emc$T_post_star_lit <=0] <- 0.0000001
df_pre_jp$T_pre_star_lit[df_pre_jp$T_pre_star_lit <=0] <- 0.0000001
df_post_jp$T_post_star_lit[df_post_jp$T_post_star_lit <=0] <- 0.0000001


## Weibull calibration function
# EMC - pre
fit <- survreg(Surv(T_pre_star_lit, E_r) ~ pre_score, data=df_pre_emc, dist="weibull")
summary(fit)

# jp - pre
fit <- survreg(Surv(T_pre_star_lit, E_r) ~ pre_score, data=df_pre_jp, dist="weibull")
summary(fit)

# EMC - post
fit <- survreg(Surv(T_post_star_lit, E_r) ~ post_score, data=df_post_emc, dist="weibull")
summary(fit)

# jp - post
fit <- survreg(Surv(T_post_star_lit, E_r) ~ post_score, data=df_post_jp, dist="weibull")
summary(fit)

## Exponential calibration function
# EMC - pre
fit <- survreg(Surv(T_pre_star_lit, E_r) ~ pre_score, data=df_pre_emc, dist="exponential")
summary(fit)

# jp - pre
fit <- survreg(Surv(T_pre_star_lit, E_r) ~ pre_score, data=df_pre_jp, dist="exponential")
summary(fit)

# EMC - post
fit <- survreg(Surv(T_post_star_lit, E_r) ~ post_score, data=df_post_emc, dist="exponential")
summary(fit)

# jp - post
fit <- survreg(Surv(T_post_star_lit, E_r) ~ post_score, data=df_post_jp, dist="exponential")
summary(fit)



### Explain 10. checking calibration results
# Analaysis is a coppy of the one done before the weibull recalibration
# The calibration of the survival function is done by: exp(-exp(1/y*(log(-log(f_s0_pre(t)))-a-b*score)))
# With f_so_pre the baseline survival function for the pre erasl, and a, b, y the calibration parameters obtained from the weibull calibration
###


# model extension with HBV, HCV

# pre emc 
cox.mod.pre <- coxph(pre.surv.obj.emc ~ offset(pre_score) + HBV, data = df_pre_emc)
cox.mod.pre

cox.mod.pre <- coxph(pre.surv.obj.emc ~ offset(pre_score) + HCV, data = df_pre_emc)
cox.mod.pre

# pre jp 
cox.mod.pre <- coxph(pre.surv.obj.jp ~ offset(pre_score) + HBV, data = df_pre_jp)
cox.mod.pre

cox.mod.pre <- coxph(pre.surv.obj.jp ~ offset(pre_score) + HCV, data = df_pre_jp)
cox.mod.pre

# post emc
cox.mod.post <- coxph(post.surv.obj.emc ~ offset(post_score) + HBV, data = df_post_emc)
cox.mod.post

cox.mod.post <- coxph(post.surv.obj.emc ~ offset(post_score) + HCV, data = df_post_emc)
cox.mod.post

# post jp
cox.mod.post <- coxph(post.surv.obj.jp ~ offset(post_score) + HBV, data = df_post_jp)
cox.mod.post

cox.mod.post <- coxph(post.surv.obj.jp ~ offset(post_score) + HCV, data = df_post_jp)
cox.mod.post

#--------------- Forward selection - re-estimation_calibration ----------------------------

##forward selection Erasmus MC pre
#round1
cox.mod.pre <- coxph(pre.surv.obj.emc ~ pre_score, data = df_pre_emc)
cox.mod.pre
cox.mod.pre <- coxph(pre.surv.obj.emc ~ pre_score + gender , data = df_pre_emc)
cox.mod.pre
cox.mod.pre <- coxph(pre.surv.obj.emc ~ pre_score + albigrade_1 , data = df_pre_emc)
cox.mod.pre
cox.mod.pre <- coxph(pre.surv.obj.emc ~ pre_score + log_afp , data = df_pre_emc)
cox.mod.pre
cox.mod.pre <- coxph(pre.surv.obj.emc ~ pre_score + log_diamet , data = df_pre_emc)
cox.mod.pre
cox.mod.pre <- coxph(pre.surv.obj.emc ~ pre_score + aantaltum_23 , data = df_pre_emc)
cox.mod.pre
cox.mod.pre <- coxph(pre.surv.obj.emc ~ pre_score + HBV , data = df_pre_emc)
cox.mod.pre
cox.mod.pre <- coxph(pre.surv.obj.emc ~ pre_score + HCV , data = df_pre_emc)
cox.mod.pre

#round2
cox.mod.pre <- coxph(pre.surv.obj.emc ~ pre_score + log_afp + gender , data = df_pre_emc)
cox.mod.pre
cox.mod.pre <- coxph(pre.surv.obj.emc ~ pre_score + log_afp +albigrade_1 , data = df_pre_emc)
cox.mod.pre
cox.mod.pre <- coxph(pre.surv.obj.emc ~ pre_score + log_afp + log_diamet , data = df_pre_emc)
cox.mod.pre
cox.mod.pre <- coxph(pre.surv.obj.emc ~ pre_score + log_afp + aantaltum_23 , data = df_pre_emc)
cox.mod.pre
cox.mod.pre <- coxph(pre.surv.obj.emc ~ pre_score + log_afp + HBV , data = df_pre_emc)
cox.mod.pre
cox.mod.pre <- coxph(pre.surv.obj.emc ~ pre_score + log_afp + HCV , data = df_pre_emc)
cox.mod.pre


##forward selection Erasmus MC post
#round 1
cox.mod.post <- coxph(post.surv.obj.emc ~ post_score, data = df_post_emc)
cox.mod.post
cox.mod.post <- coxph(post.surv.obj.emc ~ post_score + gender , data = df_post_emc)
cox.mod.post
cox.mod.post <- coxph(post.surv.obj.emc ~ post_score + albigrade_1 , data = df_post_emc)
cox.mod.post
cox.mod.post <- coxph(post.surv.obj.emc ~ post_score + log_afp , data = df_post_emc)
cox.mod.post
cox.mod.post <- coxph(post.surv.obj.emc ~ post_score + log_diamet , data = df_post_emc)
cox.mod.post
cox.mod.post <- coxph(post.surv.obj.emc ~ post_score + aantaltum_23 , data = df_post_emc)
cox.mod.post
cox.mod.post <- coxph(post.surv.obj.emc ~ post_score + Microvascular_invasion , data = df_post_emc)
cox.mod.post
cox.mod.post <- coxph(post.surv.obj.emc ~ post_score + HBV , data = df_post_emc)
cox.mod.post
cox.mod.post <- coxph(post.surv.obj.emc ~ post_score + HCV , data = df_post_emc)
cox.mod.post

#round 2
cox.mod.post <- coxph(post.surv.obj.emc ~ post_score + Microvascular_invasion + gender , data = df_post_emc)
cox.mod.post
cox.mod.post <- coxph(post.surv.obj.emc ~ post_score + Microvascular_invasion + albigrade_1 , data = df_post_emc)
cox.mod.post
cox.mod.post <- coxph(post.surv.obj.emc ~ post_score + Microvascular_invasion + log_afp , data = df_post_emc)
cox.mod.post
cox.mod.post <- coxph(post.surv.obj.emc ~ post_score + Microvascular_invasion + log_diamet , data = df_post_emc)
cox.mod.post
cox.mod.post <- coxph(post.surv.obj.emc ~ post_score + Microvascular_invasion + aantaltum_23 , data = df_post_emc)
cox.mod.post
cox.mod.post <- coxph(post.surv.obj.emc ~ post_score + Microvascular_invasion + HBV , data = df_post_emc)
cox.mod.post
cox.mod.post <- coxph(post.surv.obj.emc ~ post_score + Microvascular_invasion + HCV , data = df_post_emc)
cox.mod.post


# round 3
cox.mod.post <- coxph(post.surv.obj.emc ~ post_score + Microvascular_invasion + log_afp + gender , data = df_post_emc)
cox.mod.post
cox.mod.post <- coxph(post.surv.obj.emc ~ post_score + Microvascular_invasion + log_afp + albigrade_1 , data = df_post_emc)
cox.mod.post
cox.mod.post <- coxph(post.surv.obj.emc ~ post_score + Microvascular_invasion + log_afp + log_diamet , data = df_post_emc)
cox.mod.post
cox.mod.post <- coxph(post.surv.obj.emc ~ post_score + Microvascular_invasion + log_afp + aantaltum_23 , data = df_post_emc)
cox.mod.post
cox.mod.post <- coxph(post.surv.obj.emc ~ post_score + Microvascular_invasion + log_afp + HBV , data = df_post_emc)
cox.mod.post
cox.mod.post <- coxph(post.surv.obj.emc ~ post_score + Microvascular_invasion + log_afp + HCV , data = df_post_emc)
cox.mod.post


#forward selection Okayama pre
cox.mod.pre <- coxph(pre.surv.obj.jp ~ pre_score, data = df_pre_jp)
cox.mod.pre

#round 1
cox.mod.pre <- coxph(pre.surv.obj.jp ~ pre_score + gender , data = df_pre_jp)
cox.mod.pre
cox.mod.pre <- coxph(pre.surv.obj.jp ~ pre_score + albigrade_1 , data = df_pre_jp)
cox.mod.pre
cox.mod.pre <- coxph(pre.surv.obj.jp ~ pre_score + log_afp , data = df_pre_jp)
cox.mod.pre
cox.mod.pre <- coxph(pre.surv.obj.jp ~ pre_score + log_diamet , data = df_pre_jp)
cox.mod.pre
cox.mod.pre <- coxph(pre.surv.obj.jp ~ pre_score + aantaltum_23 , data = df_pre_jp)
cox.mod.pre
cox.mod.pre <- coxph(pre.surv.obj.jp ~ pre_score + HBV , data = df_pre_jp)
cox.mod.pre
cox.mod.pre <- coxph(pre.surv.obj.jp ~ pre_score + HCV , data = df_pre_jp)
cox.mod.pre

#round 2
cox.mod.pre <- coxph(pre.surv.obj.jp ~ pre_score + gender + albigrade_1 , data = df_pre_jp)
cox.mod.pre
cox.mod.pre <- coxph(pre.surv.obj.jp ~ pre_score + gender + log_afp , data = df_pre_jp)
cox.mod.pre
cox.mod.pre <- coxph(pre.surv.obj.jp ~ pre_score + gender + log_diamet , data = df_pre_jp)
cox.mod.pre
cox.mod.pre <- coxph(pre.surv.obj.jp ~ pre_score + gender + aantaltum_23 , data = df_pre_jp)
cox.mod.pre
cox.mod.pre <- coxph(pre.surv.obj.jp ~ pre_score + gender + HBV , data = df_pre_jp)
cox.mod.pre
cox.mod.pre <- coxph(pre.surv.obj.jp ~ pre_score + gender + HCV , data = df_pre_jp)
cox.mod.pre



#forward selection Okayama post
#round 1
cox.mod.post <- coxph(post.surv.obj.jp ~ post_score + gender , data = df_post_jp)
cox.mod.post
cox.mod.post <- coxph(post.surv.obj.jp ~ post_score + albigrade_1 , data = df_post_jp)
cox.mod.post
cox.mod.post <- coxph(post.surv.obj.jp ~ post_score + log_afp , data = df_post_jp)
cox.mod.post
cox.mod.post <- coxph(post.surv.obj.jp ~ post_score + log_diamet , data = df_post_jp)
cox.mod.post
cox.mod.post <- coxph(post.surv.obj.jp ~ post_score + aantaltum_23 , data = df_post_jp)
cox.mod.post
cox.mod.post <- coxph(post.surv.obj.jp ~ post_score + Microvascular_invasion , data = df_post_jp)
cox.mod.post
cox.mod.post <- coxph(post.surv.obj.jp ~ post_score + HBV , data = df_post_jp)
cox.mod.post
cox.mod.post <- coxph(post.surv.obj.jp ~ post_score + HCV , data = df_post_jp)
cox.mod.post


#round 2
cox.mod.post <- coxph(post.surv.obj.jp ~ post_score + gender + albigrade_1 , data = df_post_jp)
cox.mod.post
cox.mod.post <- coxph(post.surv.obj.jp ~ post_score + gender + log_afp , data = df_post_jp)
cox.mod.post
cox.mod.post <- coxph(post.surv.obj.jp ~ post_score + gender + log_diamet , data = df_post_jp)
cox.mod.post
cox.mod.post <- coxph(post.surv.obj.jp ~ post_score + gender + aantaltum_23 , data = df_post_jp)
cox.mod.post
cox.mod.post <- coxph(post.surv.obj.jp ~ post_score + gender + Microvascular_invasion , data = df_post_jp)
cox.mod.post
cox.mod.post <- coxph(post.surv.obj.jp ~ post_score + gender + HBV , data = df_post_jp)
cox.mod.post
cox.mod.post <- coxph(post.surv.obj.jp ~ post_score + gender + HCV , data = df_post_jp)
cox.mod.post