10 Create biomarker trajectory data from UK biobank assessment center data

In this chapter, we create biomarker trajectory data using the UKB assessment center data. The biomarkers we phenotype are

platelet
urine albumin
urine creatinine
blood albumin
blood creatinine
cholesterol
cystatin C
blood glucose (random)
hba1c
HDL
LDL
triglycerides
systolic blood pressure (SBP)
diastolic blood pressure (DBP)
urine albumin to urine creatinine ratio (UACR)

In addition, we phenotype patient’s macroalbumuminuria or microalbuminuria status which will be used in phenotyping time-to-event data for diabetic kidney disease.

As we did in chapters 3 and 5, we exclude any subjects who have mismatch between self-reported sex and genetically determined sex. If biomarker measurements or their dates are missing, we exclude those measurements from the trajectory table as well.

Load packages and functions.

library(tidyverse)
library(lubridate)
source("functions.R")

Load reformatted UKB demographics and labs tables.

demog_ukb <- readRDS("generated_data/demog_UKB.RDS")
labs_ukb <- readRDS("generated_data/labs_UKB.RDS")
demog_sel <- readRDS("generated_data/demog_selected.RDS")

Define dictionary that maps field ID’s to biomarkers.

fieldID_key <- c(
  "30080" = "platelets",
  "30500" = "albumin_urine",
  "30510" = "creat_urine",
  "30600" = "albumin_blood",
  "30690" = "chol",
  "30700" = "creat_blood",
  "30720" = "cystatin_c",
  "30740" = "glucose_rand", #"Blood glucose"
  "30750" = "hba1c",
  "30760" = "hdl",
  "30780" = "ldl",
  "30870" = "trig",
  "4080" =  "sbp",
  "93" = "sbp",
  "4079" = "dbp",
  "94" = "dbp"
)

Create biomarker trajectory data for all biomarkers defined above.

biomarker_traj_tab_long <- 
  full_join(labs_ukb,demog_ukb) %>% 
  select(f.eid,matches(paste0(paste(paste0('f.',names(fieldID_key)),collapse = '|')))) %>%
  pivot_longer(cols = -f.eid, names_to = "fields", values_to = "measurement") %>%
  filter(!is.na(measurement)) %>%
  separate(fields, c('f',"biomarker_field_id", "visit", "array"), sep = "\\.") %>%
  mutate(biomarker = fieldID_key[biomarker_field_id]) %>% 
  left_join(demog_sel %>% select(f.eid,date_init,date_repeat)) %>%
  mutate(event_dt = as.Date(ifelse(visit == '0', date_init, 
                                   ifelse(visit=='1', date_repeat, NA)), 
                            origin=origin)) %>%
  select(f.eid,measurement,event_dt,visit,biomarker)

Compute urine albumin to urine creatinine ratio (UACR).

uacr_traj <- biomarker_traj_tab_long %>% filter(biomarker %in% c("albumin_urine","creat_urine")) %>%
  pivot_wider(id_cols = c(f.eid,event_dt), names_from = c(biomarker,visit), values_from = measurement) %>%
  mutate(uacr0 = albumin_urine_0/(creat_urine_0*10^-3), uacr1 = albumin_urine_1/(creat_urine_1*10^-3)) %>%
  select(f.eid,event_dt,uacr0,uacr1) %>%
  pivot_longer(cols = c(uacr0,uacr1), names_to = "visit", values_to = "measurement") %>%
  mutate(visit = ifelse(visit == "uacr0",'0','1')) %>%
  select(f.eid,measurement,event_dt,visit) %>%
  filter(!is.na(measurement)) %>%
  mutate(biomarker = "uacr")

Add UACR trajectory data to the biomarker_traj_tab_long.

biomarker_traj_tab_long <- bind_rows(biomarker_traj_tab_long,uacr_traj) %>% 
  filter(!is.na(event_dt),!is.na(measurement))

Create macro/microalbuminuria event table.

albuminuria_event_tab <- 
  biomarker_traj_tab_long %>% 
  filter(biomarker == "uacr") %>%
    mutate(macroalbuminuria = ifelse(measurement < 33.9,F,T),
           microalbuminuria = ifelse(measurement < 3.4,F,T)) %>%
    select(f.eid,event_dt,macroalbuminuria,microalbuminuria) %>%
    pivot_longer(cols=c(macroalbuminuria,microalbuminuria),values_to = "event", names_to = "type") %>%
  select(f.eid,event,event_dt,type)

If a subject was missing urine albumin or urine creatinine levels, we cannot compute urine albumin to creatinine ratio (UACR). Urine albumin levels were assumed to be undetectable (NAs) if urine albumin levels were below \(6.7\). Subjects with known detection issue had missing urine albumin measurements. However, we can still rule out micro/macroalbuminuria using the urine creatinine levels by computing the upper bound of UACR.

To elaborate, suppose a subject’s urine albumin level \(a\) is below the detection level, i.e. \(a < 6.7\). Then, \(ACR = \frac{a}{c \times 10^{-3}} < \frac{6.7}{c\times 10^{-3}}\) where \(c:=\) urine creatinine levels. Thus, \(\frac{6.7}{c\times 10^{-3}}\) is an upper bound of ACR for the subject. Thus

-if the upper bound of \(ACR\) is less than 33.9, rule out macroalbuminuria

-if the upper bound of \(ACR\) is less than 3.4, rule out microalbuminuria.

subjects_with_urine_albumin_below_detection_limit <- 
  labs_ukb %>% 
  rename(below_detection_visit0 = f.30505.0.0, below_detection_visit1 = f.30505.1.0) %>%
  select(f.eid,contains("below")) %>%
  pivot_longer(contains("below"),names_to = "visit",values_to = "below_detection_limit") %>%
  mutate(visit = ifelse(visit == "below_detection_visit0",0,1),
         below_detection_limit = ifelse(is.na(below_detection_limit),NA,T)) %>%
  filter(below_detection_limit == T) %>% 
  select(f.eid,visit)

albuminuria_event_tab_inferred <- 
  biomarker_traj_tab_long %>% filter(biomarker == "creat_urine") %>%
  filter(f.eid %in% subjects_with_urine_albumin_below_detection_limit$f.eid) %>%
  mutate(ACR_upper_bound = 6.7/(measurement*10^-3)) %>%
  mutate(macroalbuminuria = ifelse(ACR_upper_bound < 33.9,F,NA),
         microalbuminuria = ifelse(ACR_upper_bound < 3.4,F,NA)) %>% 
  select(f.eid,event_dt,macroalbuminuria,microalbuminuria) %>%
  pivot_longer(cols=c(macroalbuminuria,microalbuminuria),values_to = "event", names_to = "type") %>%
  filter(!is.na(event)) %>%
  select(f.eid,event,event_dt,type)

Combine inferred albuminuria event table with albuminuria_event_tab

albuminuria_event_tab <- bind_rows(albuminuria_event_tab,albuminuria_event_tab_inferred) %>% distinct()

Load demographics table and ID’s of subjects with sex mismatch.

demog <- readRDS("generated_data/demog_selected.RDS")
sex_mismatch_subject_ids <- readRDS("generated_data/sex_mismatch_subject_ids.RDS")

Filter out the following from the biomarker trajectory table and albuminuria event table:

any subjects with sex mismatch
any unknown measurements
any measurements with unknown dates

biomarker_traj_tab_long <- biomarker_traj_tab_long[!(biomarker_traj_tab_long$f.eid %in% sex_mismatch_subject_ids),]
biomarker_traj_tab_long <- biomarker_traj_tab_long %>% left_join(demog %>% select(f.eid,DOB),by="f.eid")
biomarker_traj_tab_long$event_dt <- cleandates(biomarker_traj_tab_long$event_dt,biomarker_traj_tab_long$DOB,T) %>% as.Date()
biomarker_traj_tab_long <- biomarker_traj_tab_long %>% select(-DOB)
biomarker_traj_tab_long <- biomarker_traj_tab_long[!is.na(biomarker_traj_tab_long$event_dt),]
biomarker_traj_tab_long <- biomarker_traj_tab_long[!is.na(biomarker_traj_tab_long$measurement),]

albuminuria_event_tab <- albuminuria_event_tab[!(albuminuria_event_tab$f.eid %in% sex_mismatch_subject_ids),]
albuminuria_event_tab <- albuminuria_event_tab %>% left_join(demog %>% select(f.eid,DOB),by="f.eid")
albuminuria_event_tab$event_dt <- cleandates(albuminuria_event_tab$event_dt,albuminuria_event_tab$DOB,T) %>% as.Date()
albuminuria_event_tab <- albuminuria_event_tab %>% select(-DOB)
albuminuria_event_tab <- albuminuria_event_tab[!is.na(albuminuria_event_tab$event_dt),]

Save the biomarker trajectory table and albuminuria event table extracted from the UKB asssessment center data.

saveRDS(biomarker_traj_tab_long,"generated_data/biomarker_trajectory_ukb.RDS")
saveRDS(albuminuria_event_tab,"generated_data/albuminuria_event_tab_ukb.RDS")