Skip to contents

Recoding CHMS medication variables

Source: vignettes/recoding_medications.qmd

1. Load packages 

# Install release version from CRAN
install.packages("chmsflow")

# Install the most recent version from GitHub
devtools::install_github("Big-Life-Lab/chmsflow")

2. Recode medication variables for individual cycles

The medication data object must always be named cyclex_meds for the recoding scripts to function correctly. All other data for the same cycle should be stored separately under the name cyclex, as specified in recoding-variables.qmd.

# Load recodeflow and dplyr
library(recodeflow)
library(dplyr)

# Recoding basic variables from cycles 1 and 3
cycle2_medication_data <- rec_with_table(cycle2_meds, c("clinicid", "atc_101a", "mhr_101b"), variable_details = variable_details)
head(cycle2_medication_data)

cycle3_medication_data <- rec_with_table(cycle3_meds, c("clinicid", "meucatc", "npi_25b"), variable_details = variable_details)
head(cycle3_medication_data)

# Recoding derived variable acemed for cycle2. Select vars by role used to avoid writing all med variables in one line
cycle2_ace_medication_data <- rec_with_table(cycle2_meds, c("clinicid", "acemed", recodeflow:::select_vars_by_role("Drugs", variables)), variable_details = variable_details)
head(select(cycle2_ace_medication_data, clinicid, acemed))

# Recoding derived variable acemed for cycle3. Much simpler function used here.
cycle3_ace_medication_data <- rec_with_table(cycle3_meds, c("clinicid", "meucatc", "npi_25b", "acemed"), variable_details = variable_details)
head(cycle3_ace_medication_data)

3. Merge recoded medication data from different cycles 

# Aggregating recoded cycle3 data by clinicid
cycle3_ace_medication_data <- cycle3_ace_medication_data %>%
  group_by(clinicid) %>%
  summarize(
    meucatc = paste(unique(meucatc), collapse = ", "), # Concatenate unique values of meucatc
    npi_25b = paste(unique(npi_25b), collapse = ", "), # Concatenate unique values of npi_25b
    acemed = max(as.numeric(as.character(acemed))) # Find maximum of acemed
  )
head(cycle3_ace_medication_data)

# De-factorizing acemed variable in recoded cycle2 data
cycle2_ace_medication_data$acemed <- as.numeric(as.character(cycle2_ace_medication_data$acemed))

# Merging recoded cycle2 and cycle3 data, and ensuring one acemed column in the end
cycles2and3_ace_medication_data <- merge(cycle2_ace_medication_data, cycle3_ace_medication_data, by = "clinicid")

cycles2and3_ace_medication_data <- cycles2and3_ace_medication_data %>%
  mutate(acemed = pmax(acemed.x, acemed.y)) %>%
  select(-c(acemed.x, acemed.y))
head(cycles2and3_ace_medication_data)

4. Example: Determine hypertension status by recoding medications

If your analysis requires medication variables (particularly when deriving other variables), always perform medication recoding first, before recoding any other variables.

# Cycles 1-2
# Recode medication variables first - anymed for hypertension and diab_drug for diabetes which is involved in determining hypertension status
cycle2_htn_medication_data <- rec_with_table(cycle2_meds, c("clinicid", recodeflow:::select_vars_by_role("Drugs", variables)), variable_details = variable_details)

# Create dummy medication variable by duplicating recoded medication variable
cycle2_htn_medication_data$anymed2 <- as.numeric(as.character(cycle2_htn_medication_data$anymed))
cycle2_htn_medication_data$diab_drug2 <- as.numeric(as.character(cycle2_htn_medication_data$diab_drug))
cycle2_htn_medication_data <- select(cycle2_htn_medication_data, clinicid, anymed2, diab_drug2)

# Merge medication data with rest of data
cycle2 <- merge(cycle2, cycle2_htn_medication_data, by = "clinicid")

# Recode hypertension (and diabetes) status using recoded, dummy medication variables
cycle2_htn_data <- rec_with_table(cycle2, c(recodeflow:::select_vars_by_role("Test", variables)), variable_details = variable_details)
head(cycle2_htn_data)

# Cycles 3-6
# Recode medication variables first - - anymed for hypertension and diab_drug for diabetes which is involved in determining hypertension status
cycle3_htn_medication_data <- rec_with_table(cycle3_meds, c("clinicid", "meucatc", "npi_25b", "anymed", "diab_drug"), variable_details = variable_details)

# Aggregating recoded cycle3 data by clinicid
cycle3_htn_medication_data <- cycle3_htn_medication_data %>%
  group_by(clinicid) %>%
  summarize(
    meucatc = paste(unique(meucatc), collapse = ", "), # Concatenate unique values of meucatc
    npi_25b = paste(unique(npi_25b), collapse = ", "), # Concatenate unique values of npi_25b
    anymed = max(as.numeric(as.character(anymed))), # Find maximum of anymed
    diab_drug = max(as.numeric(as.character(diab_drug))) # Find maximum of diab_drug
  )

# Create dummy medication variable by duplicating recoded medication variable
cycle3_htn_medication_data$anymed2 <- as.numeric(as.character(cycle3_htn_medication_data$anymed))
cycle3_htn_medication_data$diab_drug2 <- as.numeric(as.character(cycle3_htn_medication_data$diab_drug))
cycle3_htn_medication_data <- select(cycle3_htn_medication_data, clinicid, anymed2, diab_drug2)

# Merge medication data with original data
# Call overall data "cyclex" in order to recode other variables afterwards
cycle3 <- merge(cycle3, cycle3_htn_medication_data, by = "clinicid")

# Recode hypertension (and diabetes) status using recoded, dummy medication variables
cycle3_htn_data <- rec_with_table(cycle3, c(recodeflow:::select_vars_by_role("Test", variables)), variable_details = variable_details)
head(cycle3_htn_data)