Landscape Phase 1 Screening R1.Rmd

---
title: "Pilot Round 1"
author: "Meng Liu"
date: "2022-09-07"
output: html_document
editor_options: 
  chunk_output_type: console
---

Overview:
We finished Screening R1 in July 2023, the following steps were taken to prepare for the next stage (i.e., resolving inconsistencies):

1. downloaded the two google sheets (sheet1 and sheet2) and joined the sheets back to the full dataset

2. analysed the screening results and identified which records require round 2 screening (inconsistent records between the two coders)

3. marked records that require Screening R2 to set up a g-doc for crowdsourcing

The next step is to resolve all inconsitencies. 

# Load libraries 
```{r}
#devtools::install_github('Mikata-Project/ggthemr')
library(pacman)
p_load(tidyverse,rio)

theme_set(theme_minimal())

```

# Load data 
load the full record from the previous stage: 
```{r}
data <- read.csv("full_record_20220907.csv", encoding = "UTF-8")
```

# Screening 
## Load Sheets
```{r}
sheet1 <- read_csv("Screening R1/Round 1 Screening - Sheet1 (screener_id ODD).csv") %>% 
    rename(screening = `screening (include, exclude, uncertain)`,
           require_full_text = `require_full_text(1,0)`) %>% 
    filter(unique_id != "Example")

sheet2 <- read_csv("Screening R1/Round 1 Screening - Sheet2 (screener_id EVEN).csv") %>% 
    rename(screening = `screening (include, exclude, uncertain)`,
           require_full_text = `require_full_text(1,0)`) %>% 
    filter(unique_id != "Example")

```

## Corret typos
```{r}
# Identify typos
sheet1 %>% 
  count(screening) %>% pull(screening)

# Corret typos
sheet1 <- sheet1 %>% 
  mutate(screening = case_when(
    screening == "exlcude" ~ "exclude",
    screening == "incldue" ~ "include",
    screening == "unsure" ~ "uncertain",
    TRUE ~ screening
  ))

# Repeat the same for sheet2 
sheet2 %>% 
  count(screening) %>% pull(screening)

sheet2 <- sheet2 %>% 
  mutate(screening = case_when(
    screening == "exlcude" ~ "exclude",
    #screening == "incldue" ~ "include",
    screening == "unsure" ~ "uncertain",
    TRUE ~ screening
  ))

```


## Match Screening Results
```{r}
# Join the two sheets
sheets <- sheet1 %>% 
    left_join(sheet2 %>% select(unique_id,screening,screener_id,note,require_full_text), by = "unique_id")


# Match the screening results
sheets <- sheets %>% 
    mutate(decision = ifelse(
        screening.x == screening.y,screening.x, "inconsistent"
    ))
```

### Visualise the Results
```{r}
sheets %>% 
   count(decision) %>% 
   mutate(pct = n/sum(n)) %>%
   ggplot(aes(x="", y=pct, fill=decision)) +
   geom_col(width = 1) +
   coord_polar(theta = "y") +
   ggrepel::geom_label_repel(aes(label=paste(str_to_title(decision), scales::percent(pct))), position = position_stack(vjust = 0.5)) + 
   theme_void() + 
   theme(legend.position="none") +
   labs(fill = "Decision", title="Round 1 Screening", x="", y="",
        subtitle = paste0("n = ",nrow(sheets)))

ggsave("r1_screening_results.jpeg",width = 8,height =6 )
```

## Integrate to the Full Dataset
```{r}
# Extract the pilot screening results 

data <- read.csv("full_record_20220907.csv", encoding = "UTF-8")

pilot <-  data %>% filter(!r1_screening1=="")%>% 
          select(unique_id,starts_with("r1_")) %>% 
          # drop cols of pilot tagging 
          select(-starts_with("r1_tag")) %>% 
          select(-starts_with("r1_coder")) %>% 
          select(-starts_with("r1_require_more")) %>%
          select(-starts_with("r1_key")) %>%
          mutate(unique_id = as.integer(unique_id)) %>% 
          mutate(r1_screening1 = case_when(r1_screening1 == "unsure" ~ "uncertain",
                                           TRUE ~ r1_screening1),
               r1_screening2 = case_when(r1_screening2 == "unsure" ~ "uncertain",
                                         r1_screening2 == "exlude" ~ "exclude",
                                           TRUE ~ r1_screening2)  )


# Rename columns with r1 prefix
results <- sheets %>% 
  rename(r1_screening1 = screening.x,
         r1_screening2 = screening.y,
         r1_screener1 = screener_id.x,
         r1_screener2 = screener_id.y,
         r1_screen_note1 = note.x,
         r1_screen_note2 = note.y,
         r1_require_full_text1 = require_full_text.x,
         r1_require_full_text2 = require_full_text.y,
         r1_screening_decision = decision) %>% 
  select(unique_id,starts_with("r1_")) %>% 
  mutate(unique_id = as.integer(unique_id))


# Combine the results of piloting and screening R1
combined = rbind(results,pilot)
# Found a duplicate record in combined and remove it
combined[duplicated(combined$unique_id), ] %>% pull(unique_id)
combined <- distinct(combined, unique_id, .keep_all = TRUE)
# Integrate back to the full record

wos = data %>% 
          # remove all the manual coding cols to prepare for mergging
          select(-starts_with("r1_")) %>% 
          select(-starts_with("r1_tag")) %>% 
          select(-starts_with("r1_coder")) %>% 
          select(-starts_with("r1_require_more")) %>%
          select(-starts_with("r1_key")) %>%
          mutate(unique_id = as.integer(unique_id))

full_record =
  left_join(wos,combined,by="unique_id")


full_record = full_record %>% 
  mutate(action_required = case_when(
    r1_screening_decision%in%c("inconsistent","uncertain") ~ "Y",
    is.na(r1_screening_decision) ~ "Y",
    TRUE ~ "N")
         )

#saveRDS(full_record,"full_record_20230714.rds")
# the full record was saved as RDS to prevent data loss as some metadata from WoS were too long to be preseved as is in excel

r2_screening = full_record %>% 
select(unique_id, action_required, title, author_keywords, abstract, keywords_plus, doi,starts_with("r1_")) 

r2_screening %>% count(action_required)

#export(r2_screening,"r2_screening_sheet.csv")

```

# Calculate IRR for pilot + R1 
```{r}
# use the meta_rate programme by Reza Norouzian
source("meta-rate.r")

screener1 = full_record %>% 
            select(unique_id,r1_screening1) %>% 
            rename(study.name = unique_id,
                   code = r1_screening1) 
          
screener2 = full_record %>% 
            select(unique_id,r1_screening2) %>% 
            rename(study.name = unique_id,
                   code = r1_screening2)

meta_rate(screener1,screener2)

```



# Tabulation of screener contribution
```{r}
tb1 = full_record %>% 
  count(r1_screener1) %>% 
  rename(screener = r1_screener1)

tb2 = full_record %>% 
  count(r1_screener2) %>% 
  rename(screener = r1_screener2)

contribution = rbind(tb1,tb2) %>% 
  filter(!is.na(screener)) %>% 
  group_by(screener) %>% 
  summarise(n=sum(n),.groups = "drop") 

tibble(screener = 1:57) %>% 
  left_join(contribution,by="screener") %>% 

  export("screener_contribution.csv")

```