1_read_data_redcap.R

# 1_read_data_redcap.R
# read the LOCATE data from REDCap using the API; calculate additional variables; add information from study scans
# May 2023
library(dplyr)
library(tidyr)
library(janitor)
library(redcapAPI) # to directly get data from REDCap
library(RCurl) # for making the API request
library(jsonlite) # for converting JSON data
library(stringr)
library(eq5d)
library(readxl)
library(openxlsx) # for exporting data for checking
source('99_functions.R')
source('API/config.R') # for key

## Section 1: REDCap data dictionary ##
dictionary = read.csv('data/LOCATENAFLD_DataDictionary_2023-05-04.csv', stringsAsFactors = FALSE) %>%
  clean_names() %>%
  filter(!str_detect(variable_field_name, pattern='_instr_'), # remove instructions that are not variables
         !variable_field_name == 'audit_score_result') # remove this derived variable
# get variables of particular types:
dates = filter(dictionary, text_validation_type_or_show_slider_number=='date_dmy') %>% pull(variable_field_name)
numbers1 = filter(dictionary, text_validation_type_or_show_slider_number=='number') %>% pull(variable_field_name)
numbers2 = filter(dictionary, str_detect(string=variable_field_name, pattern='eq5d_')) %>% pull(variable_field_name)
numbers = union(numbers1, numbers2)
yes_no1 = filter(dictionary, choices_calculations_or_slider_labels=='1, Yes | 0, No') %>% pull(variable_field_name)
yes_no2 = filter(dictionary, field_type=='yesno') %>% pull(variable_field_name)
yes_no = c(yes_no1, yes_no2)
yes_no_unclear = filter(dictionary, choices_calculations_or_slider_labels=='1, Yes | 0, No | 2, Unclear') %>% pull(variable_field_name)
drinks = filter(dictionary, choices_calculations_or_slider_labels=='0, Never | 1, Less than monthly | 2, Monthly | 3, Weekly | 4, Daily or almost daily') %>% pull(variable_field_name)
audit = c('how_often','standard_drinks_day','six_or_more','not_stop_once_started','failed_normal_expectations','morning_drinking','guilt_remorse','unable_to_remember','you_someone_else_injured','relative_hp_concerned')

## Section 2: study scans data
load('data/0_scans.RData')
scans = select(scans, record_id, exam_date, e_median_k_pa) # just keep the exam date and KPA

##
## Section 3: get data from REDCap via API ##
rcon = redcapConnection(url = api_url, token = api_token_locate)
result = exportRecords(rcon, forms = NULL) 

## process and tidy-up the data
data = clean_names(result) %>%
  select(-name, -address, -phone_number, -participant_contacted, -interest_to_participate, -patient_gp, -ends_with('complete'), -pregnant, -leave_area) %>% # variable not needed; no patients were pregnant; none leave area
  separate(record_id, into=c('centre','num'), sep = '-', remove=FALSE) %>% # extract centre number; keep original record id variable
  mutate(centrename = centre_name(centre),
         year_referral = as.numeric(format(date_of_referral, '%Y')),
         year_today = as.numeric(format(date_today, '%Y'))) %>%
  mutate(referral_time = as.numeric(date_returned - date_of_referral), # time between randomised and GP referral
         bmi_calc = weight / ((height/100)^2), # calculated BMI
         bmi_calc = as.numeric(bmi_calc), # remove label
         body_mass_index = as.numeric(body_mass_index), # remove label
         bmi = coalesce(body_mass_index, bmi_calc), # is sometimes missing for both, so combine two sources
         year_now = coalesce(year_referral, year_today),
         year_born = ifelse(is.na(year_born_baseline) == TRUE, year_born_referral, year_born_baseline), # get year born from two source
         age = year_now - year_born,
         age = as.numeric(age)) %>% # remove label
  select(-starts_with('year_'), -body_mass_index, -bmi_calc, -audit_score_result) # drop some variables not needed
## combine two tick box results into one variable from referral letter
data = mutate(data,
              bmi_6m = case_when(
                bmi_6m_0 == 'Checked' ~ 1,
                bmi_6m_1 == 'Checked' ~ 0,
                bmi_6m_0 == 'Unchecked' & bmi_6m_1 == 'Unchecked' ~ NA_real_
              ),
              fbc_6m = case_when(
                fbc_6m_0 == 'Checked' ~ 1,
                fbc_6m_1 == 'Checked' ~ 0,
                fbc_6m_0 == 'Unchecked' & fbc_6m_1 == 'Unchecked' ~ NA_real_
              ),
              liver_6m = case_when(
                liver_6m_0 == 'Checked' ~ 1,
                liver_6m_1 == 'Checked' ~ 0,
                liver_6m_0 == 'Unchecked' & liver_6m_1 == 'Unchecked' ~ NA_real_
              ),
              hep_6m = case_when(
                hep_6m_0 == 'Checked' ~ 1,
                hep_6m_1 == 'Checked' ~ 0,
                hep_6m_0 == 'Unchecked' & hep_6m_1 == 'Unchecked' ~ NA_real_
              ),
              iron_6m = case_when(
                iron_6m_0 == 'Checked' ~ 1,
                iron_6m_1 == 'Checked' ~ 0,
                iron_6m_0 == 'Unchecked' & iron_6m_1 == 'Unchecked' ~ NA_real_
              ),
              us_6m = case_when(
                us_6m_0 == 'Checked' ~ 1,
                us_6m_1 == 'Checked' ~ 0,
                us_6m_0 == 'Unchecked' & us_6m_1 == 'Unchecked' ~ NA_real_
              )) %>%
  select(-ends_with('_0'), -'bmi_6m_1', -'fbc_6m_1', -'liver_6m_1', -'hep_6m_1', -'iron_6m_1', -'us_6m_1') # can't use general '_1' because of data in notes form
# convert variables
data = mutate(data,
         across(any_of(dates), as.Date), # convert dates
         across(any_of(numbers), as.numeric), # convert to numbers
         across(any_of(yes_no), yesno), # convert yes/no
         how_often = audit_score(how_often), # audit not working in `across`
         standard_drinks_day = audit_score(standard_drinks_day),
         six_or_more = audit_score(six_or_more),
         not_stop_once_started = audit_score(not_stop_once_started),
         failed_normal_expectations = audit_score(failed_normal_expectations),
         morning_drinking = audit_score(morning_drinking),
         guilt_remorse = audit_score(guilt_remorse),
         unable_to_remember = audit_score(unable_to_remember),
         you_someone_else_injured = audit_score(you_someone_else_injured),
         relative_hp_concerned = audit_score(relative_hp_concerned)
)

## export liver size for hand cleaning, just do once ...
export_liver = function(){
  export = select(data, record_id, liver_size) %>%
    filter(!is.na(liver_size),
           liver_size != 'not stated') %>%
    mutate(size = NA)
  wb <- createWorkbook()
  addWorksheet(wb, 1)
  writeData(wb, sheet = 1, export)
  saveWorkbook(wb, "data/liver_size.xlsx", overwrite = TRUE)
}
# ... now read in cleaned liver data and merge with original data
data = select(data, -liver_size) # remove liver size from original data
cleaned = read_excel('data/liver_size_24042023.xlsx', sheet='for_r') %>%
  clean_names() %>%
  select(-liver_size) %>% # remove old version
  rename('liver_size' = 'size_cm') # rename new version
data = mutate(data, record_id = as.character(record_id)) # for merge
data = left_join(data, cleaned, by='record_id')

## AUDIT alcohol score
audit_score = select(data, record_id, all_of(audit)) %>%
  mutate_if(is.factor, as.numeric) %>%
  pivot_longer(cols = all_of(audit)) %>%
  filter(!is.na(value)) %>% # ignore missing, create score from what is available
  group_by(record_id) %>%
  summarise(n = n(), audit = sum(value)) %>%
  ungroup() %>%
  filter(n >= 4) %>% # must have answered 4 or more questions (most were fully complete)
  select(-n)
data = left_join(data, audit_score, by='record_id')

## calculate EQ-5D-3L 
# a) at baseline
scores.df = select(data, record_id, contains('eq5d')) %>%
  rename('MO' = 'eq5d_mb_b', # mobility
         'SC' = 'eq5d_sc_b', # self-care
         'UA' = 'eq5d_ua_b', # usual activities
         'PD' = 'eq5d_pd_b', # pain/discomfort
         'AD' = 'eq5d_ad_b') %>% # anxiety/depression
  select(record_id, MO, SC, UA, PD, AD) %>%
  drop_na() # remove all missing
scores.df.baseline = mutate(scores.df,
  eq5d_b = eq5d(scores.df, country="Australia", version="3L", type="TTO"),
  eq5d_b = eq5d_b*100 #  make eq-5d on 0 to 100 scale
  ) %>%
  select(record_id, eq5d_b)
# b) at follow-up
scores.df = select(data, record_id, contains('eq5d')) %>%
  rename('MO' = 'eq5d_mb_12m', # mobility
         'SC' = 'eq5d_sc_12m', # self-care
         'UA' = 'eq5d_ua_12m', # usual activities
         'PD' = 'eq5d_pd_12m', # pain/discomfort
         'AD' = 'eq5d_ad_12m') %>% # anxiety/depression
  select(record_id, MO, SC, UA, PD, AD) %>%
  drop_na() # remove all missing
# 
scores.df.fu = mutate(scores.df,
      eq5d_12m = eq5d(scores.df, country="Australia", version="3L", type="TTO"),
      eq5d_12m = eq5d_12m*100 #  make eq-5d on 0 to 100 scale
    ) %>%
  select(record_id, eq5d_12m)
# add scores back to data
data = left_join(data, scores.df.baseline, by='record_id')
data = left_join(data, scores.df.fu, by='record_id') # 

## secondary outcome, date of successful fibroscan, 
# take earliest date from patient notes or study data; keep kpa data associated with scan
notes = select(data, record_id, fibroscan_date_2, kpa) %>%
  mutate(kpa = as.numeric(kpa)) %>%
  rename('exam_date' = 'fibroscan_date_2')
both_scans = rename(scans, 'kpa' = 'e_median_k_pa') %>%
  bind_rows(notes, .id = 'source') %>%
  filter(!is.na(exam_date), !is.na(kpa)) %>% # must have date and KPA result
  arrange(record_id, exam_date) %>%
  group_by(record_id) %>%
  slice(1) %>% # take earliest reading
  ungroup() %>%
  select(-source) %>% # no longer needed
  rename('KPA' = 'kpa',
         'scan_date' = 'exam_date')
# now add this information to the main data
data = left_join(data, both_scans, by = "record_id") %>% 
  mutate(kpa = as.numeric(kpa),
         kpa2 = coalesce(kpa, KPA)) %>% # take whichever KPA reading is available
  select(-kpa, -KPA) %>%
  rename('kpa' = 'kpa2')

## primary outcome using fibroscan date and KPA data
# first flag missing scan dates
miss = filter(data, 
              fibroscan == 'Yes (successful)' & 
              !is.na(randomised), # must have been randomised
              is.na(fibroscan_date_2)) %>%
  select(record_id)
miss # should be empty
# clinically significant fibrosis is over 8
data = mutate(data,
              primary_date = case_when(
                is.na(scan_date) ~ NA_real_,
                is.na(kpa) ~ NA_real_,
                !is.na(scan_date) & !is.na(kpa) & kpa < 8 ~ NA_real_,
                !is.na(scan_date) & !is.na(kpa) & kpa >= 8 ~ as.numeric(scan_date) # 8 and above is high risk
              ),
              primary_date = as.Date(primary_date, origin='1970-01-01'))
#

## fixes to exclusions and a few other small edits
data = mutate(data,
  record_id = as.character(record_id), # remove label for merging
  # randomised but excluded - set randomised to blank
  randomised = as.character(randomised),
  randomised = ifelse(!is.na(randomised) & !is.na(exclude_reason), NA, randomised),
  # a few more who never returned questionnaire but not marked in REDCap
  exclude_reason = as.character(exclude_reason),
  exclude_reason = ifelse(is.na(randomised) == TRUE & withdrew_or_excluded ==TRUE, 'Could not be contacted' , exclude_reason),
  withdrew_reason = as.character(withdrew_reason),
  reason = coalesce(exclude_reason, withdrew_reason),
  reason = ifelse(withdrew_or_excluded == 'Neither', 'Could not be contacted', reason), # add people who could not be contacted
  phone_time = as.numeric(phone_date - date_returned)/(365.25/12) ) # time between randomisation and telephone follow-up (months)

## fixes to form (patient notes) data
# ED presentation
index = data$ed_pres_1 < data$date_returned # presentations before randomised
index[is.na(index)] = FALSE
data$ed_pres_1[index] = data$ed_pres_2[index] # shift date from next ED presentation ...
data$ed_pres_2[index] = NA # ... and blank second
data$ed_reason_1[index] = data$ed_reason_2[index] # shift reason from next ED presentation ...
data$ed_reason_2[index] = NA #... and blank second
if(data$ed_presentations[index] == 'Yes - once'){
  data$ed_presentations[index] = 'No'
}

### fixes to employment as we did not have retired as a category
## baseline employment
data = mutate(data, employment = as.character(employment)) # change to character
index = str_detect(tolower(data$employment_other), 'retired|pension|super')
index[is.na(index)] = FALSE
data$employment_other[index] = ''
data$employment[index] = 'Retired'
# more consistent and shift to category
index = str_detect(tolower(data$employment_other), '\\bcarer\\b')
data$employment[index] = 'Carer'
data$employment_other[index] = ''
# more consistent and shift to category
index = str_detect(tolower(data$employment_other), '\\bfull.time\\b')
data$employment[index] = 'Full-time'
data$employment_other[index] = ''
# more consistent and shift to category
index = str_detect(tolower(data$employment_other), 'casual|supply')
data$employment[index] = 'Part-time'
data$employment_other[index] = ''
# more consistent and shift to category
index = str_detect(data$employment_other, 'Stay at home mother')
data$employment[index] = 'Home duties'
data$employment_other[index] = ''
## 12 months employment
data = mutate(data, employment_12 = as.character(employment_12)) # change to character
index = str_detect(tolower(data$employment_other_12), 'retired|pension|super')
index[is.na(index)] = FALSE
data$employment_other_12[index] = ''
data$employment_12[index] = 'Retired'
# fix typo
index = str_detect(data$employment_other_12, 'Volunteer worjk')
data$employment_other_12[index] = 'Volunteer work'
# more consistent
index = str_detect(data$employment_other_12, 'Contract work')
data$employment_other_12[index] = 'Contract work'
# more consistent and shift to category
index = str_detect(data$employment_other_12, 'Carer')
data$employment_12[index] = 'Carer'
data$employment_other_12[index] = ''
# more consistent and shift to category
index = str_detect(data$employment_other_12, 'Stay at home mother')
data$employment_12[index] = 'Home duties'
data$employment_other_12[index] = ''
# more consistent and shift to category
index = str_detect(tolower(data$employment_other_12), 'casual|supply')
data$employment_12[index] = 'Part-time'
data$employment_other_12[index] = ''

## add per protocol for new model of care; they must have had a scan
data = left_join(data, scans, by='record_id') %>% # scans data loaded above
  mutate(pp = case_when(
                randomised == 'Usual care' ~ TRUE, # include all as PP does not apply because "treatment" is usual care, so it is whatever they did
                randomised == 'New model of care' & !is.na(exam_date) ~ TRUE, #  completed the treatment protocol originally allocated
                randomised == 'New model of care' & is.na(exam_date) ~ FALSE
         ))
#with(data, table(randomised, is.na(exam_date))) # check

## if age/sex is missing, use referral data
data = mutate(data,
              sex = as.character(sex),
              sex_referral = as.character(sex_referral),
              sex = ifelse(is.na(sex)==TRUE, sex_referral, sex)) %>%
  select(-sex_referral)

## Edit from Ingrid: combine 'Yes, dietician' and 'Yes, both', 'Yes, nutritionist'
data = mutate(data,
              dietician_baseline = case_when(
                dietician_baseline == 'No' ~ 'No',
                dietician_baseline == 'Yes, both' ~ 'Yes',
                dietician_baseline == 'Yes, dietician' ~ 'Yes',
                dietician_baseline == 'Yes, nutritionist' ~ 'Yes'
              ),
              dietician_12m = case_when(
                dietician_12m == 'No' ~ 'No',
                dietician_12m == 'Yes, both' ~ 'Yes',
                dietician_12m == 'Yes, dietician' ~ 'Yes',
                dietician_12m == 'Yes, nutritionist' ~ 'Yes'
              ))

## there were no admissions to hospital, so can delete date variables
data = select(data,
              -'admission_date_1',
              -'discharge_1',
              -'hospitalisation_usage',
              -'admission_date_2',
              -'discharge_2',
              -'hospitalisation_usage_2',
              -'admission_date_3',
              -'discharge_3',
              -'hospitalisation_usage_3')
dictionary = filter(dictionary,
                    !variable_field_name %in% c('admission_date_1', 'discharge_1',
              'hospitalisation_usage', 'admission_date_2', 'discharge_2',
              'hospitalisation_usage_2', 'admission_date_3', 'discharge_3',
              'hospitalisation_usage_3','sex_referral'))


## save ##
save(data, dictionary, file = 'data/1_redcap_data.RData')

#
# filter(data, centre=='122', randomised==1) %>% pull(record_id)