In CC-HIC/ccdata: The critical care clinical data toolbox

Load data

Usually a RData file which stores all the dataset will be given. A sample RData can be found in /doc/sample_ccd.RData.

library(cleanEHR)
data.path <- paste0(find.package("cleanEHR"), "/doc/sample_ccd.RData")
load(data.path)

Data overview

You can have a quick overview of the data by checking infotb. In the sample dataset, sensitive variables such as NHS number and admission time have been removed or twisted.

print(head(ccd@infotb))

The basic entry of the data is episode which indicates an admission of a site. Using episode_id and site_id can locate a unique admission entry. pid is a unique patient identifier.

# quickly check how many episodes are there in the dataset.
ccd@nepisodes

There are 263 fields which covers patient demographics, physiology, laboratory, and medication information. Each field has 2 labels, NHIC code and short name. There is a function lookup.items() to look up the fields you need. lookup.items() function is case insensitive and allows fuzzy search.

# searching for heart rate
lookup.items('heart') # fuzzy search

+-------------------+--------------+--------------+--------+-------------+
|     NHIC.Code     |  Short.Name  |  Long.Name   |  Unit  |  Data.type  |
+===================+==============+==============+========+=============+
| NIHR_HIC_ICU_0108 |    h_rate    |  Heart rate  |  bpm   |   numeric   |
+-------------------+--------------+--------------+--------+-------------+
| NIHR_HIC_ICU_0109 |   h_rhythm   | Heart rhythm |  N/A   |    list     |
+-------------------+--------------+--------------+--------+-------------+

Inspect individual episode

# check the heart rate, bilirubin, fluid balance, and drugs of episode_id = 7. 
# NOTE: due to anonymisation reason, some episodes data cannot be displayed
# properly. 
episode.graph(ccd, 7, c("h_rate",  "bilirubin", "fluid_balance_d"))

Non-longitudinal Data

sql.demographic.table() can generate a data.table that contains all the non-longitudinal variables. A demonstration of how to do some work on a subset of data.

# contains all the 1D fields i.e. non-longitudinal
tb1 <- sql.demographic.table(ccd)

# filter out all dead patient. (All patients are dead in the dataset.)
tb1 <- tb1[DIS=="D"]

# subset variables we want (ARSD = Advanced respiratory support days,
# apache_prob = APACHE II probability)
tb <- tb1[, c("SEX", "ARSD", "apache_prob"), with=F]
tb <- tb[!is.na(apache_prob)]

# plot
library(ggplot2)
ggplot(tb, aes(x=apache_prob, y=ARSD, color=SEX)) + geom_point()

Longitudinal data

To deal with longitudinal data, we need to first to transform it into a long table format.

Create a `cctable`

# To prepare a YAML configuration file like this. You write the following text
# in a YAML file. 
conf <- "
NIHR_HIC_ICU_0108:
  shortName: hrate
NIHR_HIC_ICU_0112:
  shortName: bp_sys_a
  dataItem: Systolic Arterial blood pressure - Art BPSystolic Arterial blood pressure
NIHR_HIC_ICU_0093:
   shortName: sex
"
library(yaml)
tb <- create.cctable(ccd, yaml.load(conf), freq=1)

# a lazy way to do that. 
tb <- create.cctable(ccd, list(NIHR_HIC_ICU_0108=list(), 
                         NIHR_HIC_ICU_0112=list(), 
                         NIHR_HIC_ICU_0093=list()), 
                     freq=1)
print(tb$tclean)

Manipulate on `cctable`

Get the mean heart rate of each patient.

tb$tclean[, mean(NIHR_HIC_ICU_0108, na.rm=T), by=c("site", "episode_id")]

Data cleaning

To clean the data, one needs to write the specification in the YAML configuration file.

conf <-"
NIHR_HIC_ICU_0108:
  shortName: hrate
  dataItem: Heart rate
  distribution: normal
  decimal_places: 0
  range:
    labels:
      red: (0, 300)
      amber: (11, 150)
    apply: drop_entry
  missingness: # remove episode if missingness is higher than 70% in any 24 hours interval 
    labels:
      yellow: 24
    accept_2d:
      yellow: 70 
    apply: drop_episode
"

ctb <- create.cctable(ccd, yaml.load(conf), freq=1)
ctb$filter.ranges("amber") # apply range filters
ctb$filter.missingness()
ctb$apply.filters()

cptb <- rbind(cbind(ctb$torigin, data="origin"), 
              cbind(ctb$tclean, data="clean"))


ggplot(cptb, aes(x=time, y=NIHR_HIC_ICU_0108, color=data)) + 
  geom_point(size=1.5) + facet_wrap(~episode_id, scales="free_x")