PhysicalActivityandStrokeOutcome/EudraCT reporting/eudract.Rmd

---
title: "TALOS eudract AE reporting - example"
author: "AGDamsbo"
date: "Knitted: `r format(Sys.time(), '%d %B %Y')`"
output:
  pdf_document: default
  html_document: default
toc: TRUE
---

```{r setup, include=FALSE}
knitr::opts_chunk$set(echo = TRUE, message = FALSE, warning = FALSE)
```

# Intro

This is the data management workflow for the reporting of AEs in the TALOS trial to the EudraCT database.
A dedicated package is used to format data after a longer process of editing data to conform.
The source data is not perfectly formatted, and during the process, a few manual steps are necessary.
- Advise number one: Make sure to format data according to the desired format for reporting.

This is "page" 2 of 2. All cleaning is performed in "TALOS AE cleaning.Rmd"


```{r}
setwd("/Volumes/Data/TALOS/")
```

```{r}
library(haven)
library(dplyr)
# https://www.rdocumentation.org/packages/eudract/versions/0.9.3
library(eudract)
```

# Data import

Data set
```{r}
d<-as_factor(read_dta("/Volumes/Data/TALOS/talos_ae_clean.dta"))
write.csv(head(d,100),"sample_ae.csv")
d<-read.csv("sample_ae.csv")
```

## Modified trial-specific adjudication list with added maddra codes from eudract
```{r}
library(readxl)
adj_tbl<-read_xlsx("adjudication_table.xlsx")
# write.csv(soc_code,"soc_code.csv") # Export af soc_code til manuel kodning af oprindelig Adjudication List
head(adj_tbl)
head(soc_code)
```

# Formatting to EUDRACT
```{r}
# Included data example in the eudract-package
head(safety)
```

## Adding columns according to safety-format

### Fatal outcome
The "d$status" contains final status of every event instance, with those marked Dødelig being used. 
The other option would be to code according to adj_code, but the "d$status" was GCP monitored and is used.
```{r}
# for (i in 1:nrow(d)){
#   d$fatal[i]<-ifelse(str_contains(d$description[i],adj_tbl$code[adj_tbl$soc_term=="Death"],logic = "or"),1,0)
#   }

d$fatal<-ifelse(d$status=="Dødelig",1,0)
table(factor(d$fatal),factor(d$rtreat))
```

### Related
All events coded with either of the three categories a considered related in this binary form.
```{r}
d$related_bin<-ifelse(d$related=="Mulig"|
                          d$related=="Sandsynlig"|
                          d$related=="Afgjort relateret",
                        1,0)
```

### Serious
Only SAEs are occuring, no SAR or SUSAR
```{r}
ser<-c("SAE","SAR","SUSAR")
d$serious<-ifelse(d$CLFint %in% ser,1,0)
```

### Randomisation
Group naming according to groups defined on the EudraCT page.
```{r}
d$group<-ifelse(d$rtreat=="Placebo","Placebo","Active")
```


### SOC kode og term/subcat
```{r}
ls<-list()
for (i in 1:nrow(d)){
  # Text string split at ":", " ", "+" or "(" and constrained to first three digits. 
  # The last step as a security against a missing " " following the adjudication code or similar.
  v<-substr(unlist(strsplit(d$description[i],"[: +(]")),1,3) 
  # vector elements contained in adj_tbl$code are subset and added to list
  ls[[paste0("index", i)]] <- grep(paste(adj_tbl$code,collapse="|"),v,value = TRUE)
}
```

Splitting each list element into different columns, length(ls) equals nrow(d)
```{r}
for (i in 1:length(ls)){
  # Subsets liste efter navngivning i forrige loop
  v<-ls[[paste0("index", i)]]            
  for (j in 1:length(v)){
    # Føjer til eksisterende, tilføjer ekstra kolonner ved behov
    d[i,paste0("adj_code_", j)]<-v[j]  
  }
}
```

#### Death only event subset and recoding - manual work
Originally a "continuation" variable was also included in the export for more information on the event, however, this variable has been excluded from the data set.
```{r}
## If only 1 code, it is in adj_code_1, test if this code is any categorised as "Death" 
# subset_death<-d[d$adj_code_1 %in% adj_tbl$code[adj_tbl$soc_term=="Death"] &  
#                     lengths(ls)==1,  ## Redundant test, that only one code was used
#                   c("description","continuation","event_id")]
## Adds an extra column for adding alternative code manually
# subset_death$add_code<-c("")
# write.csv(subset_death,"subset_death.csv")
```

Hand coded data set imported again
```{r}
head(subset_death_coded<-read_xlsx("subset_death_coded.xlsx")) ## Eight (8) cases, 5 had a new code added
## event_id 335 were not recoded, as two events (also event_id 333) are already created for this same death...
for (i in 1:nrow(d)){
  for (j in 1:nrow(subset_death_coded)){
    d$adj_code_2[i]<-ifelse(d$event_id[i]==subset_death_coded$event_id[j],
                              subset_death_coded$add_code[j],d$adj_code_2[i])
  }
}
```


#### Subset events coded with "801"
```{r}
# subset_801<-d[grepl("801",d$description),c("description","continuation","event_id")]
# write.csv(subset_801,"subset_801.csv")
```

Every event has been coded with soc_code alternative to 801 or NONE if deemed irrelevant based on other codes at same event.
```{r}
head(alt_801<-read_excel("subset_801_alt.xlsx") %>% na.omit)
```

### Converting to new, long data.frame
All events with bleeding (severity) or death are excluded. Death counts will be added later.
```{r}
library(tidyr)
# dput(names(d))
dta<-pivot_longer(select(d,!matches(c("description", "expected", "status","CLFint"))),
                  starts_with("adj_code_"),
                  names_to="adj_index",
                  values_to = "adj_code") %>%   # Pivotting to long format
  na.omit %>%  # Omitting NAs, result is complete case only
  filter(adj_code %in% adj_tbl$code[adj_tbl$soc_term!="Death"&adj_tbl$soc_term!="Bleeding"]) 
# Excluding events marked with Death or Bleeding, as these are additional codes not coresponding to soc_code terms
head(dta)
```

### Matching adj_code to soc_code
Adding soc_codes and manually coded alternative categories for 801 codes. 
```{r warning=FALSE}
for (i in 1:nrow(dta)){
  dta$soc[i]<-adj_tbl$meddra[adj_tbl$code==dta$adj_code[i]]
  for (j in 1:nrow(alt_801)){
    dta$soc[i]<-ifelse(dta$event_id[i]==alt_801$event_id[j]&dta$adj_code[i]=="801",
                       alt_801$alt_801[j],dta$soc[i])
  }
}
```

Few were not coded, omitting NAs.
```{r}
dta <- dta %>% na.omit  # Only keeping correctly coded cases
```

Adding term and subcat
```{r warning=FALSE}
# Subcategory name is the combined soc_term and adj_subcat for additional details in the final report.
for (i in 1:nrow(dta)){
  dta$term[i]<-soc_code$soc_term[soc_code$meddra==dta$soc[i]]
  dta$subcat[i]<-paste0(dta$term[i],": ",adj_tbl$subcat[adj_tbl$code==dta$adj_code[i]])
}
```

# Configuring XML

## Creating specified data frame
```{r}
df<-data.frame(subjid=dta$rnumb,
               term=dta$subcat,
               soc=as.integer(dta$soc),
               serious=dta$serious,
               related=dta$related_bin,
               fatal=dta$fatal,
               group=dta$group)
head(df)
```

## Handling Deaths
```{r}
# Deaths in named integer vector
# ae_deaths<-table(df$fatal,df$group)[2,] # No deaths included in the sample data set
# These are all the deaths observed within 6 months after randomisation, eg after intention-to-treat
all_deaths<-c("Active"=16,"Placebo"=12) 
excess_death<-all_deaths  #-ae_deaths
```

## Creating safety summary
```{r}
safe_sum<-safety_summary(data=df,
               exposed=c("Active"=319,"Placebo"=323),
               excess_deaths=excess_death)
```

```{r}
simple <- tempfile(fileext = ".xml")
eudract <- tempfile(fileext = ".xml")
simple_safety_xml(safe_sum, simple)
```


```{r}
eudract_convert(input=simple,
                output=eudract)
```