Summarise and analyse clinical trial information
Ralf Herold
2024-12-11
Source:vignettes/ctrdata_summarise.Rmd
ctrdata_summarise.RmdGeneral information on the ctrdata package is available
here: https://github.com/rfhb/ctrdata.
Remember to respect the registers’ terms and conditions (see
ctrOpenSearchPagesInBrowser(copyright = TRUE)). Please cite
this package in any publication as follows: Ralf Herold (2024). ctrdata:
Retrieve and Analyze Clinical Trials in Public Registers. R package
version 1.18.0. https://cran.r-project.org/package=ctrdata
Preparations
Here using MongoDB, which is faster than SQLite, can handle credentials, provides access to remote servers and can directly retrieve nested elements from paths. See README.md and Retrieve clinical trial information for examples using SQLite. Also PostgreSQL can be used as database, see Install R package ctrdata.
db <- nodbi::src_mongo(
url = "mongodb://localhost",
db = "my_database_name",
collection = "my_collection_name"
)
db
# MongoDB 7.0.8 (uptime: 265661s)
# URL: mongodb://localhost
# Database: my_database_name
# Collection: my_collection_name
# empty collection if exists
nodbi::docdb_delete(db, db$collection)See Retrieve clinical trial information for more details.
#
library(ctrdata)
# These two queries are similar, for completed interventional (drug)
# trials with children with a neuroblastoma from either register
ctrLoadQueryIntoDb(
# using queryterm and register...
queryterm = "query=neuroblastoma&age=under-18&status=completed",
register = "EUCTR",
euctrresults = TRUE,
con = db
)
# [...]
# $n
# [1] 226
ctrLoadQueryIntoDb(
# ...or using full URL of search results
queryterm = paste0(
"https://classic.clinicaltrials.gov/ct2/results?",
"cond=neuroblastoma&recrs=e&age=0&intr=Drug"),
con = db
)
# * Appears specific for CTGOV Classic website
# Since 2024-06-25, the classic CTGOV servers are no longer available. Package ctrdata has translated
# the classic CTGOV query URL from this call of function ctrLoadQueryIntoDb(queryterm = ...) into a
# query URL that works with the current CTGOV2. This is printed below and is also part of the return
# value of this function, ctrLoadQueryIntoDb(...)$url. This URL can be used with ctrdata functions.
# Note that the fields and data schema of trials differ between CTGOV and CTGOV2.
#
# Replace this URL:
#
# https://classic.clinicaltrials.gov/ct2/results?cond=neuroblastoma&recrs=e&age=0&intr=Drug
#
# with this URL:
#
# https://clinicaltrials.gov/search?cond=neuroblastoma&intr=Drug&aggFilters=ages:child,status:com
#
# * Found search query from CTGOV2: cond=neuroblastoma&intr=Drug&aggFilters=ages:child,status:com
# * Checking trials using CTGOV REST API 2.0, found 222 trials
# (1/3) Downloading in 1 batch(es) (max. 1000 trials each; estimate: 22 Mb total)
# Download status: 1 done; 0 in progress. Total size: 9.79 Mb (615%)... done!
# (2/3) Converting to NDJSON...
# (3/3) Importing records into database...
# JSON file #: 1 / 1
# = Imported or updated 222 trial(s)
# Updated history ("meta-info" in "my_collection_name")
# $n
# [1] 222
dbQueryHistory(con = db)
# query-timestamp query-register query-records query-term
# 1 2024-06-29 09:01:16 EUCTR 226 query=neuroblastoma&age=under-18&status=completed
# 2 2024-06-29 09:02:12 CTGOV2 222 cond=neuroblastoma&intr=Drug&aggFilters=ages:child,status:comFind fields / variables of interest
Specify a part of the name of a variable of interest; all variables
including deeply nested variable names are searched. Set
sample = TRUE (default) to rapidly execute the function in
large databases.
#
dbFindFields(namepart = "date", sample = FALSE, con = db)
# Finding fields in database collection (may take some time) . . . . .
# Field names cached for this session.
# EUCTR
# "e231_full_title_date_and_version_of_each_substudy_and_their_related_objectives"
# EUCTR
# "e231_full_title_date_and_version_of_each_substudy_and_their_related_objectives_es"
# EUCTR
# "e231_full_title_date_and_version_of_each_substudy_and_their_related_objectives_it"
# EUCTR
# "n_date_of_competent_authority_decision"
# EUCTR
# "n_date_of_ethics_committee_opinion"
# EUCTR
# "p_date_of_the_global_end_of_the_trial"
# EUCTR
# "trialChanges.globalAmendments.globalAmendment.date"
# EUCTR
# "trialChanges.globalInterruptions.globalInterruption.date"
# EUCTR
# "trialChanges.globalInterruptions.globalInterruption.restartDate"
# EUCTR
# "trialInformation.analysisStageDate"
# EUCTR
# "trialInformation.globalEndOfTrialDate"
# EUCTR
# "trialInformation.primaryCompletionDate"
# EUCTR
# "trialInformation.recruitmentStartDate"
# EUCTR
# "x6_date_on_which_this_record_was_first_entered_in_the_eudract_database"
# CTGOV2
# "annotationSection.annotationModule.unpostedAnnotation.unpostedEvents.date"
# CTGOV2
# "derivedSection.miscInfoModule.submissionTracking.estimatedResultsFirstSubmitDate"
# CTGOV2
# "derivedSection.miscInfoModule.submissionTracking.firstMcpInfo.postDateStruct"
# CTGOV2
# "derivedSection.miscInfoModule.submissionTracking.firstMcpInfo.postDateStruct.date"
# CTGOV2
# "derivedSection.miscInfoModule.submissionTracking.firstMcpInfo.postDateStruct.type"
# CTGOV2
# "derivedSection.miscInfoModule.submissionTracking.submissionInfos.releaseDate"
# CTGOV2
# "derivedSection.miscInfoModule.submissionTracking.submissionInfos.resetDate"
# CTGOV2
# "documentSection.largeDocumentModule.largeDocs.date"
# CTGOV2
# "documentSection.largeDocumentModule.largeDocs.uploadDate"
# CTGOV2
# "protocolSection.statusModule.completionDateStruct"
# CTGOV2
# "protocolSection.statusModule.completionDateStruct.date"
# CTGOV2
# "protocolSection.statusModule.completionDateStruct.type"
# CTGOV2
# "protocolSection.statusModule.dispFirstPostDateStruct"
# CTGOV2
# "protocolSection.statusModule.dispFirstPostDateStruct.date"
# CTGOV2
# "protocolSection.statusModule.dispFirstPostDateStruct.type"
# CTGOV2
# "protocolSection.statusModule.dispFirstSubmitDate"
# CTGOV2
# "protocolSection.statusModule.dispFirstSubmitQcDate"
# CTGOV2
# "protocolSection.statusModule.lastUpdatePostDateStruct"
# CTGOV2
# "protocolSection.statusModule.lastUpdatePostDateStruct.date"
# CTGOV2
# "protocolSection.statusModule.lastUpdatePostDateStruct.type"
# CTGOV2
# "protocolSection.statusModule.lastUpdateSubmitDate"
# CTGOV2
# "protocolSection.statusModule.primaryCompletionDateStruct"
# CTGOV2
# "protocolSection.statusModule.primaryCompletionDateStruct.date"
# CTGOV2
# "protocolSection.statusModule.primaryCompletionDateStruct.type"
# CTGOV2
# "protocolSection.statusModule.resultsFirstPostDateStruct"
# CTGOV2
# "protocolSection.statusModule.resultsFirstPostDateStruct.date"
# CTGOV2
# "protocolSection.statusModule.resultsFirstPostDateStruct.type"
# CTGOV2
# "protocolSection.statusModule.resultsFirstSubmitDate"
# CTGOV2
# "protocolSection.statusModule.resultsFirstSubmitQcDate"
# CTGOV2
# "protocolSection.statusModule.startDateStruct"
# CTGOV2
# "protocolSection.statusModule.startDateStruct.date"
# CTGOV2
# "protocolSection.statusModule.startDateStruct.type"
# CTGOV2
# "protocolSection.statusModule.statusVerifiedDate"
# CTGOV2
# "protocolSection.statusModule.studyFirstPostDateStruct"
# CTGOV2
# "protocolSection.statusModule.studyFirstPostDateStruct.date"
# CTGOV2
# "protocolSection.statusModule.studyFirstPostDateStruct.type"
# CTGOV2
# "protocolSection.statusModule.studyFirstSubmitDate"
# CTGOV2
# "protocolSection.statusModule.studyFirstSubmitQcDate" The search for fields is cached and thus accelerated during the R
session; calling ctrLoadQueryIntoDb() or changing
sample = ... invalidates the cache.
Data frame from database
The fields of interest can be obtained from the database and are represented in an R data.frame, for example:
#
result <- dbGetFieldsIntoDf(
c(
# EUCTR protocol-related information
"f41_in_the_member_state",
"f422_in_the_whole_clinical_trial",
"a1_member_state_concerned",
"p_end_of_trial_status",
"n_date_of_competent_authority_decision",
"a2_eudract_number",
#
# EUCTR results-related information
"trialInformation.recruitmentStartDate",
"trialInformation.globalEndOfTrialDate",
#
# CTGOV2
"protocolSection.statusModule.overallStatus",
"protocolSection.statusModule.startDateStruct.date",
"trialInformation.recruitmentStartDate",
"protocolSection.statusModule.primaryCompletionDateStruct.date"
),
con = db
)Metadata from data frame
The objects returned by functions of this package include attributes with metadata to indicate from which database, table / collection and query details. Metadata can be reused in R.
attributes(result)
# [...]
#
# $class
# [1] "data.frame"
#
# $`ctrdata-dbname`
# [1] "my_database_name"
#
# $`ctrdata-table` <-- this attribute will be retired by end 2024
# [1] "my_collection_name"
#
# $`ctrdata-table-note`
# [1] "^^^ attr ctrdata-table will be removed by end 2024"
#
# $`ctrdata-collection`
# [1] "my_collection_name"
#
# $`ctrdata-dbqueryhistory`
# query-timestamp query-register query-records query-term
# 1 2024-06-29 09:01:16 EUCTR 226 query=neuroblastoma&age=under-18&status=completed
# 2 2024-06-29 09:02:12 CTGOV2 222 cond=neuroblastoma&intr=Drug&aggFilters=ages:child,status:comDe-duplicate records
In the database, the variable “_id” is the unique index for a record. This “_id” is the NCT number for CTGOV records (e.g., “NCT00002560”), and it is the EudraCT number for EUCTR records including the postfix identifying the EU Member State (e.g., “2008-001436-12-NL”).
It is relevant to de-duplicate records because a trial can be registered in both CTGOV and EUCTR, and can have records by involved country in EUCTR.
De-duplication is done at the analysis stage because this enables to select if a trial record should be taken from one or the other register, and from one or the other EU Member State.
The basis of de-duplication is the recording of additional trial
identifiers in supplementary fields (variables), which are checked and
reported when using function dbFindIdsUniqueTrials():
# Obtain de-duplicated trial record ids
ids <- dbFindIdsUniqueTrials(
preferregister = "EUCTR",
con = db
)
# Searching for duplicate trials...
# - Getting all trial identifiers (may take some time), 479 found in collection
# - Finding duplicates among registers' and sponsor ids...
# - 192 EUCTR _id were not preferred EU Member State record for 59 trials
# - Keeping 57 / 0 / 0 / 0 / 211 records from EUCTR / CTGOV / ISRCTN / CTIS / CTGOV2
# = Returning keys (_id) of 268 records in collection "my_collection_nameThe unique ids can be used like this to de-duplicate the data.frame created above:
Reviewing a specific trial
It will often be useful to inspect all data for a single, for example to understand the meaning and relation of fields, and to see neighbouring elements to fields of interest.
#
# Adding some trials from CTIS for a search similar to above
ctrLoadQueryIntoDb(
queryterm = 'searchCriteria={"containAny":"neuroblastoma"}',
register = "CTIS",
con = db
)
# $n
# [1] 25
#
# $success
# [1] "2024-514917-36-00" "2024-518931-12-00" "2023-504246-64-02" "2024-517295-37-00" "2024-513470-22-00"
# [6] "2024-513843-10-00" "2022-501694-39-01" "2024-511404-17-00" "2024-511975-14-00" "2024-512102-25-00"
# [11] "2024-515552-21-00" "2024-511259-16-00" "2023-507418-28-00" "2023-509673-22-00" "2024-513141-37-00"
# [16] "2024-511071-16-00" "2024-512095-35-00" "2023-507178-41-00" "2023-506778-11-00" "2022-501725-21-00"
# [21] "2023-504880-18-00" "2023-508587-29-00" "2022-502668-20-00" "2023-504246-64-00" "2023-503684-42-00"
#
# $failed
# NULLIdentify a trial of interest by its _id, and use this in
any standard database function. Retrieve the trial’s full
JSON representation that ctrdata had loaded
into the database and visualise its nested structure of field names and
values.
#
# This example requires to install this package for visualisation:
# remotes::install_github("hrbrmstr/jsonview")
# Get a particular trial from any database supported
# by ctrdata (duckdb, SQLite, PostgreSQL, MongoDB):
oneTrial <- nodbi::docdb_query(
src = db,
key = db$collection,
query = '{"_id":"2022-501725-21-00"}',
limit = 1L
)
# Create an interactive widget where nodes can be expanded:
jsonview::json_tree_view(oneTrial)
CtisNested
Simple analysis of dates
In a data.frame generated with dbGetFieldsIntoDf(),
fields are typed as dates, logical, character or numbers.
#
str(result)
# 'data.frame': 268 obs. of 12 variables:
# $ _id : chr "2004-004386-15-DE" "2005-000915-80-IT" "2005-001267-63-IT" "2005-002089-13-GB" ...
# $ f41_in_the_member_state : int NA NA 5 37 70 24 100 35 10 24 ...
# $ f422_in_the_whole_clinical_trial : int 230 NA 12 67 70 NA 100 156 2230 NA ...
# $ a1_member_state_concerned : chr "Germany - BfArM" "...
# $ p_end_of_trial_status : chr "Completed" "Completed" "Completed" ...
# $ n_date_of_competent_authority_decision : Date, format: "2005-07-08" "2005-04-21" ...
# $ a2_eudract_number : chr "2004-004386-15" "2005-000915-80" ...
# $ trialInformation.recruitmentStartDate : Date, format: "2005-07-26" NA NA ...
# $ trialInformation.globalEndOfTrialDate : Date, format: "2006-10-27" NA NA ...
# $ protocolSection.statusModule.overallStatus : chr NA NA NA NA ...
# $ protocolSection.statusModule.startDateStruct.date : Date, format: NA NA NA ...
# $ protocolSection.statusModule.primaryCompletionDateStruct.date: Date, format: NA NA NA ...
# - attr(*, "ctrdata-dbname")= chr "my_database_name"
# - attr(*, "ctrdata-table")= chr "my_collection_name"
# - attr(*, "ctrdata-table-note")= chr "^^^ attr ctrdata-table will be removed by end 2024"
# - attr(*, "ctrdata-collection")= chr "my_collection_name"
# - attr(*, "ctrdata-dbqueryhistory")='data.frame': 2 obs. of 4 variables:
# ..$ query-timestamp: chr [1:2] "2024-06-29 09:01:16" "2024-06-29 09:02:12"
# ..$ query-register : chr [1:2] "EUCTR" "CTGOV2"
# ..$ query-records : int [1:2] 226 222
# ..$ query-term : chr [1:2] "query=neuroblastoma&age=under-18&status=completed"
# "cond=neuroblastoma&intr=Drug&aggFilters=ages:child,status:com"This typing facilitates using the respective type of data for analysis, for example of dates with base R graphics:
# Open file for saving
png("vignettes/nb1.png")
# Visualise trial start date
hist(
result[["n_date_of_competent_authority_decision"]],
breaks = "years"
)
box()
dev.off()
Histogram1
Merge corresponding fields from registers
The field “n_date_of_competent_authority_decision” used above exists
only in EUCTR, yet it corresponds to the field “start_date” in CTGOV.
Thus, to give the start of the trial, the two fields can be merged for
analysis, using the convenience function
dfMergeVariablesRelevel() in ctrdata
package:
# Merge two variables into a new variable:
result$trialstart <- dfMergeVariablesRelevel(
result,
colnames = c(
# EUCTR
"n_date_of_competent_authority_decision",
# CTGOV2
"protocolSection.statusModule.startDateStruct.date"
)
)
# Plot from both registers
png("vignettes/nb2.png")
hist(
result[["trialstart"]],
breaks = "years"
)
box()
dev.off()
Histogram2
In a more sophisticated use of
dfMergeVariablesRelevel(), values of the original variables
can be mapped into new values of the merged variable, as follows:
# First, define how values of the new, merged variable
# (e.g., "ongoing") will result from values of the
# original variable (e.g, "Recruiting):
mapped_values <- list(
"ongoing" = c(
# EUCTR
"Recruiting", "Active", "Ongoing",
"Temporarily Halted", "Restarted",
# CTGOV
"Active, not recruiting", "Enrolling by invitation",
"Not yet recruiting",
# CTGOV2
"ACTIVE_NOT_RECRUITING", "ENROLLING_BY_INVITATION",
"RECRUITING", "TEMPORARILY_NOT_AVAILABLE",
# CTIS
"Ongoing, recruiting", "Ongoing, recruitment ended",
"Ongoing, not yet recruiting", "Authorised, not started"
),
#
"completed" = c(
"Completed", "COMPLETED", "Ended"),
#
"other" = c(
"GB - no longer in EU/EEA", "Trial now transitioned",
"Withdrawn", "Suspended", "No longer available",
"SUSPENDED", "NO_LONGER_AVAILABLE", "WITHDRAWN",
"WITHHELD", "UNKNOWN",
"Terminated", "TERMINATED", "Prematurely Ended",
"Under evaluation")
)
# Secondly, use the list of mapped
# values when merging two variable:
tmp <- dfMergeVariablesRelevel(
result,
colnames = c(
# EUCTR
"p_end_of_trial_status",
# CTGOV2
"protocolSection.statusModule.overallStatus"
),
levelslist = mapped_values
)
table(tmp)
# ongoing completed other
# 5 256 6 User annotations
When using ctrLoadQueryIntoDb(), ctrdata
adds to each record the fields annotation and
record_last_import. The annotation field is a single string
that the user specifies when retrieving trials (Retrieve clinical trial information).
The user can specify to append, prefix or replace any existing
annotations when a trial record is loaded again, see example below. The
last date and time when the trial record was imported is updated
automatically when using ctrLoadQueryIntoDb(). These fields
can also be used for analysis. For example, string functions can be used
for annotations, e.g. to split it into components. Since no annotations
were specified when retrieving the trials in the steps above, there are
so far no annotation fields and stopifnodata is set to
FALSE to avoid the function raises an error to alert
users:
#
ctrLoadQueryIntoDb(
queryterm = "query=neuroblastoma&resultsstatus=trials-with-results",
register = "EUCTR",
euctrresults = TRUE,
annotation.text = "test annotation",
annotation.mode = "append",
con = db
)
result <- dbGetFieldsIntoDf(
fields = c(
"annotation",
"record_last_import"
),
con = db
)
str(result)
# 'data.frame': 552 obs. of 3 variables:
# $ _id : chr "2004-004386-15-DE" "2004-004386-15-ES" ...
# $ record_last_import: Date, format: "2024-06-29" "2024-06-29" ...
# $ annotation : chr "test annotation" "test annotation" ...Analysing sample sizes in historic versions of trial records
Historic versions can set to be retrieved for CTGOV2 by specifying
ctgov2history = <...> when using
ctrLoadQueryIntoDb(); this functionality was added in
ctrdata version 1.18.0. The versions include all trial data
available at the date of the respective version.
For CTGOV2 records, the historic versions are added as follows into
the ctrdata data model of a trial record, where the
ellipsis ... represents all trial data fields:
{"_id":"NCT01234567", "title": "Current title", ..., "history": [{"history_version": {"version_number": 1, "version_date": "2020-21-22 10:11:12"}, "title": "Original title", ...}, {"history_version": {"number": 2, "date": "2021-22-23 11:13:13"}, "title": "Later title", ...}]}
The example shows how planned or realised number of participants
(sample size) changed over time for individual trials, using available
data (that is, only from CTGOV2; historic versions were
available for CTIS only until its relaunch on
2024-06-17).
#
# load some trials from CTGOV2 specifying that
# for each trial, 5 versions should be retrieved
ctrLoadQueryIntoDb(
queryterm = "cond=neuroblastoma&aggFilters=phase:3,status:com",
register = "CTGOV2",
con = db,
ctgov2history = 5L
)
# * Appears specific for CTGOV REST API 2.0
# * Found search query from CTGOV2: cond=neuroblastoma&aggFilters=phase:3,status:com
# * Checking trials using CTGOV REST API 2.0, found 24 trials
# (1/3) Downloading in 1 batch(es) (max. 1000 trials each; estimate: 2.4 Mb total)
# Download status: 1 done; 0 in progress. Total size: 1.11 Mb (482%)... done!
# (2/3) Converting to NDJSON...
# (3/3) Importing records into database...
# JSON file #: 1 / 1
# * Checking and processing historic versions...
# Download status: 24 done; 0 in progress. Total size: 1.35 Mb (515%)... done!
# - Downloading 117 historic versions (estimate: 4.4 MB total)...
# Download status: 117 done; 0 in progress. Total size: 4.14 Mb (532%)... done!
# - Merging trial versions . . . . . . . . . . . . . . . . . . . . . . . .
# - Updating trial records . . . . . . . . . . . . . . . . . . . . . . . .
# Updated 24 trial(s) with historic versions
# = Imported or updated 24 trial(s)
# Updated history ("meta-info" in "my_collection_name")
# get relevant fields
result <- dbGetFieldsIntoDf(
fields = c(
# CTGOV2
"history.protocolSection.designModule.enrollmentInfo.count",
"history.history_version.version_date"
),
con = db
)
# helpers
library(dplyr)
library(tidyr)
library(ggplot2)
# mangle and plot
result %>%
unnest(cols = starts_with("history.")) %>%
group_by(`_id`) %>%
ggplot(
mapping = aes(
x = history.history_version.version_date,
y = history.protocolSection.designModule.enrollmentInfo.count,
colour = `_id`)
) +
geom_step() +
geom_point() +
theme_light() +
guides(colour = "none") +
labs(
title = "Sample sizes in trials including patients with a neuroblastoma",
subtitle = "Source: CTGOV2 records labelled as phase 3 and completed",
caption = Sys.Date()
)
ggsave("vignettes/samplesizechanges.png", width = 6, height = 4)
Sample size changes
Analysing nested fields such as trial results
The registers represent clinical trial information by nesting fields (e.g., several reporting groups within several measures within one of several endpoints). A visualisation of this hierarchical representation for CTGOV2 follows. Compare this with the outcome measures presented here: https://clinicaltrials.gov/study/NCT02139397?tab=results#outcome-measures, specifically “3. Determine the Progression Free Survival (PFS)…”
# Prepare
# remotes::install_github("https://github.com/hrbrmstr/jsonview")
# Find top level nodes within resultsSection
dbFindFields("^resultsSection[.][^.]+$", db, sample = FALSE)
# Using cache of fields.
# CTGOV2
# "resultsSection.adverseEventsModule"
# CTGOV2
# "resultsSection.baselineCharacteristicsModule"
# CTGOV2
# "resultsSection.moreInfoModule"
# CTGOV2
# "resultsSection.outcomeMeasuresModule"
# CTGOV2
# "resultsSection.participantFlowModule"
# Get relevant data
result <- dbGetFieldsIntoDf("resultsSection.outcomeMeasuresModule", db)
# Create interactive widget
jsonview::json_tree_view(result[result[["_id"]] == "NCT02139397", -1])
Ctgov2NestedResults
The analysis of nested information such as the highlighted duration
of response is facilitated with ctrdata as follows. The
main steps are:
Create a data from fields identified as shown in previous sections (using
dbGetFieldsIntoDf())Transform nested information to a long, name-value data frame (using
dfTrials2Long())Identify the measures of interest (e.g. PFS, blue circle above) by specifying the name and value of these fields (
wherename,wherevaluein functiondfName2Value()) andObtain values by specifying the name(s) of its value field(s) (red and green circles in figure above;
valuenamein functiondfName2Value()).
This is put together in the following example. Note that
CTGOV fields are no longer downloadable (see NEWS.md) but
may exist in previously created databases.
#
#### 1. Create data frame from results fields
#
# These are key results fields from
# CTGOV2, CTGOV and from EUCTR:
result <- dbGetFieldsIntoDf(
fields = c(
# EUCTR - note this requires to set parameter
# euctrresults = TRUE in ctrLoadQueryIntoDb()
# as shown above in section "User annotations"
"trialInformation.populationAgeGroup",
"subjectDisposition.recruitmentDetails",
"baselineCharacteristics.baselineReportingGroups.baselineReportingGroup",
"endPoints.endPoint",
"subjectAnalysisSets",
"adverseEvents.seriousAdverseEvents.seriousAdverseEvent",
# CTGOV2
"resultsSection.outcomeMeasuresModule",
"protocolSection.designModule.designInfo.allocation",
"resultsSection.participantFlowModule",
# CTGOV
"clinical_results.baseline.analyzed_list.analyzed.count_list.count",
"clinical_results.baseline.group_list.group",
"clinical_results.baseline.analyzed_list.analyzed.units",
"clinical_results.outcome_list.outcome",
"study_design_info.allocation"
),
con = db
)
# Keep only unique trial records
result <- result[result[["_id"]] %in% dbFindIdsUniqueTrials(con = db), ]
#### 2. All nested data are transformed to a long,
# name-value data.frame (resulting in several
# hundred rows per trial record):
#
long_result <- dfTrials2Long(df = result)
# Total 174667 rows, 157 unique names of variables
# explore
result <- dbGetFieldsIntoDf("endPoints", db)
jsonview::json_tree_view(result[result[["_id"]] == "2010-019348-37-IT", -1])
#### 3. Obtain values for measures of interest
#
# The parameters can be regular expressions.
clinicalDuration <- dfName2Value(
df = long_result,
wherename = paste0(
"endPoints.endPoint.title|",
"resultsSection.outcomeMeasuresModule.outcomeMeasures.title"
),
wherevalue = paste0(
"duration of response|DOR|",
"free survival|DFS|PFS|EFS"
),
valuename = paste0(
"resultsSection.*outcomeMeasures.classes.categories.measurements.value|",
"endPoints.*armReportingGroup.tendencyValues.tendencyValue.value|",
"resultsSection.outcomeMeasuresModule.outcomeMeasures.unitOfMeasure|",
"endPoints.endPoint.unit|",
"resultsSection.outcomeMeasuresModule.outcomeMeasures.groups.title|",
"endPoints.*armReportingGroup.armId"
)
)
# Returning values for 45 out of 131 trials
# Duration has been reported with various units:
sort(unique(clinicalDuration[grepl("unit", clinicalDuration$name), "value"]))
# For convenience, `dplyr` and related functions can be used, as follows:
library(dplyr)
library(tidyr)
clinicalDuration %>%
as_tibble() %>%
mutate(
group_id = paste0(`_id`, "_", sub("([0-9]+)[.]?.*", "\\1", identifier)),
name_short = sub(".*[.](.+)", "\\1", name),
name_short = if_else(name_short == "unitOfMeasure", "unit", name_short)
) %>%
group_by(group_id) %>%
mutate(
is_duration = any(grepl("day|month|week|year", value, ignore.case = TRUE))) %>%
ungroup() %>%
filter(is_duration) %>%
select(name_short, value, where, group_id) %>%
pivot_wider(id_cols = c(group_id, where), names_from = name_short, values_fn = list) %>%
unnest(c(value, unit)) %>%
filter(!grepl("999[9]*", value)) %>%
rowwise() %>%
mutate(
value = as.numeric(value),
arm_names = paste(armId, title, collapse = " / "),
) %>%
ungroup() %>%
mutate(
days = case_when(
grepl("[wW]eek", unit) ~ value * 7,
grepl("[mM]onth", unit) ~ value * 30,
grepl("[yY]ear", unit) ~ value * 30,
.default = value
)) %>%
select(!c(value, unit, armId, title)) -> clinicalDuration
clinicalDuration
# # A tibble: 163 x 4
# group_id where arm_names days
# <chr> <chr> <chr> <dbl>
# 1 2010-019348-37-IT_8 Duration of response by treatment "Arm-235 211.
# 2 2010-019348-37-IT_8 Duration of response by treatment "Arm-235 242.
# 3 2010-019348-37-IT_8 Duration of response by treatment "Arm-235 146.
# 4 2010-022951-49-LV_2 Progression-free Survival "Arm-648 108
# 5 2010-022951-49-LV_2 Progression-free Survival "Arm-648 51
# 6 2010-022951-49-LV_4 Progression Free Survival (PFS) in Participants wit "Arm-648 156
# 7 2010-022951-49-LV_4 Progression Free Survival (PFS) in Participants wit "Arm-648 51
# 8 2011-004168-30-DE_2 Progression-free survival PP population "Arm-937 159
# 9 2011-004168-30-DE_2 Progression-free survival PP population "Arm-937 132
# 10 2012-000510-10-GB_2 Progression Free Survival "Arm-782 NA Analysing primary endpoints
Text analysis has to be used for many fields of trial information from the registers. Here is an example to simply categorise the type of primary endpoint. In addition, the number of subjects are extracted and compared by type of primary endpoint.
#
result <- dbGetFieldsIntoDf(
c(
# CTGOV
"resultsSection.outcomeMeasuresModule.outcomeMeasures.title",
"protocolSection.designModule.enrollmentInfo.count",
# EUCTR
"e51_primary_end_points",
"f11_number_of_subjects_for_this_age_range"
),
con = db
)
# De-duplicate
result <- result[
result[["_id"]] %in%
dbFindIdsUniqueTrials(con = db),
]
# Merge primary endpoint (pep)
result$pep <- dfMergeVariablesRelevel(
df = result,
colnames = c(
"resultsSection.outcomeMeasuresModule.outcomeMeasures.title",
"e51_primary_end_points"
)
)
# Merge number of subjects
result$nsubj <- dfMergeVariablesRelevel(
df = result,
colnames = c(
"protocolSection.designModule.enrollmentInfo.count",
"f11_number_of_subjects_for_this_age_range"
)
)
# For primary endpoint of interest,
# use regular expression on text:
result$pep_is_efs <- grepl(
pattern = "((progression|event|relapse|recurrence|disease)[- ]free)|pfs|dfs|efs)",
x = result$pep,
ignore.case = TRUE
)
# Tabulate
table(result$pep_is_efs)
# FALSE TRUE
# 221 35
# Plot
library(ggplot2)
ggplot(
data = result,
aes(
x = nsubj,
y = pep_is_efs
)
) +
geom_boxplot() +
scale_x_log10()
# Warning message:
# Removed 25 rows containing non-finite values (`stat_boxplot()`).
ggsave("vignettes/boxpep.png", width = 6, height = 4)
BoxPEP
Analysis methods and p values
# helper
normalise_string <- function(x) {
x <- gsub(",", "", x)
x <- gsub("-", " ", x)
x <- tolower(x)
x <- tools::toTitleCase(x)
x <- gsub("[ ]+", " ", x)
x <- trimws(x)
x
}
# get trials into database
ctrLoadQueryIntoDb(
queryterm = paste0(
"https://clinicaltrials.gov/search?start=2010-01-01_2012-12-31", "
&intr=Drug&aggFilters=ages:child,phase:2 3,results:with"),
con = db)
# see definitions linked in help("ctrdate-registers") or directly go to
# https://clinicaltrials.gov/data-api/about-api/study-data-structure
# get result set
result <- dbGetFieldsIntoDf(c(
"protocolSection.armsInterventionsModule.armGroups.type",
"protocolSection.designModule.designInfo.allocation",
"protocolSection.designModule.designInfo.interventionModel",
"protocolSection.designModule.designInfo.maskingInfo.masking",
"protocolSection.designModule.enrollmentInfo.count",
"protocolSection.statusModule.startDateStruct.date",
"protocolSection.outcomesModule.primaryOutcomes.measure",
"resultsSection.outcomeMeasuresModule.outcomeMeasures.analyses.statisticalMethod",
"resultsSection.outcomeMeasuresModule.outcomeMeasures.analyses.pValue"
),
con = db)
# # number of participants (last number is typically all groups summed up)
# # also see README.Rmd for alternative way, summing up non-total groups
# # result$totalparticipants <- vapply(
# # result[["clinical_results.baseline.analyzed_list.analyzed.count_list.count.value"]],
# # FUN = function(x) rev(x)[1], numeric(1L))
#
# View(result)
#
# result$totalparticipants <- result$protocolSection.designModule.enrollmentInfo.count
# first reported p value for primary endpoint analysis
result$pvalueprimaryanalysis <- vapply(
result[["resultsSection.outcomeMeasuresModule.outcomeMeasures.analyses.pValue"]],
FUN = function(x) as.numeric(trimws(gsub("[<>=]", "", strsplit(x, " / ")[[1]][1]))), numeric(1))
# statistical method used for primary endpoint analysis
result$methodprimaryanalysis <- vapply(
result[["resultsSection.outcomeMeasuresModule.outcomeMeasures.analyses.statisticalMethod"]],
FUN = function(x) normalise_string(strsplit(x, " / ")[[1]][1]), character(1))
# keep randomised, parallel-group, placebo-controlled, blinded clinical trials
result$prct <-
grepl("PLACEBO|NO_INT", result$protocolSection.armsInterventionsModule.armGroups.type) &
grepl("^RANDOM", result$protocolSection.designModule.designInfo.allocation) &
grepl("^PARALLEL", result$protocolSection.designModule.designInfo.interventionModel) &
!grepl("NONE", result$protocolSection.designModule.designInfo.maskingInfo.masking)
#
table(result$prct)
# FALSE TRUE
# 923 281
#
result <- subset(result, prct == TRUE)
# helper
library(ggplot2)
# http://varianceexplained.org/statistics/interpreting-pvalue-histogram/
# http://www.pnas.org/content/100/16/9440.full
# plot p values
ggplot(
result,
aes(pvalueprimaryanalysis)) +
stat_ecdf(geom = "step") +
labs(
title = paste0(
"Paediatric phase 2 or 3 parallel-group interventional trials\n",
"with randomisation to placebo or to no intervention"),
x = "Range of p values",
y = "Empirical cumulative density of p values\nof primary endpoint results") +
geom_vline(
xintercept = 0.05,
linetype = 3)
ggsave("vignettes/phase23_paed_p_values.png", width = 6, height = 4)
# plot sample size v p value
ggplot(
result,
aes(
x = protocolSection.designModule.enrollmentInfo.count,
y = pvalueprimaryanalysis)) +
geom_point() +
ylim(0, 1) +
xlim(0, 1000) +
scale_x_log10() +
geom_hline(yintercept = 0.05, linetype = 3)
ggsave("vignettes/phase23_paed_p_values_participants.png", width = 6, height = 4)
# statistical method used for primary endpoint analysis
tmp <- table(result$methodprimaryanalysis)
tmp <- tmp[rev(order(tmp))]
tmp <- data.frame(tmp)
knitr::kable(tmp[1:10,])| Var1 | Freq |
|---|---|
| Ancova | 49 |
| Cochran Mantel Haenszel | 17 |
| Mixed Models Analysis | 14 |
| Anova | 11 |
| Chi Squared | 10 |
| Fisher Exact | 9 |
| t Test 2 Sided | 8 |
| Wilcoxon (Mann Whitney) | 5 |
| Mixed Model Repeated Measures (Mmrm) | 4 |
| Mixed Model for Repeated Measures (Mmrm) | 3 |
phase23_paed_p_values
phase23_paed_p_values
Investigational or authorised medicinal product?
The information about the status of authorisation (licensing) of a
medicine used in a trial is recorded in EUCTR in the field
dimp.d21_imp_to_be_used_in_the_trial_has_a_marketing_authorisation.
A corresponding field in CTGOV is not known. The status is in the tree
starting from the dimp element.
#
library(dplyr)
# Get results
result <- dbGetFieldsIntoDf(
fields = c(
"a1_member_state_concerned",
"n_date_of_competent_authority_decision",
"dimp.d21_imp_to_be_used_in_the_trial_has_a_marketing_authorisation",
"x6_date_on_which_this_record_was_first_entered_in_the_eudract_database",
"f422_in_the_whole_clinical_trial",
"a2_eudract_number"
),
con = db
)
# Find first date of authorisation in EU member state
tmp <- aggregate(
result[["n_date_of_competent_authority_decision"]],
by = list(result[["a2_eudract_number"]]),
FUN = function(x) min(x)
)
result <- merge(
x = result,
y = tmp,
by.x = "a2_eudract_number",
by.y = "Group.1",
all.x = TRUE
)
result %>%
rowwise() %>%
mutate(startdatefirst = min(
x, x6_date_on_which_this_record_was_first_entered_in_the_eudract_database, na.rm = TRUE)
) -> result
# Now de-duplicate
result <- result[
result[["_id"]] %in%
dbFindIdsUniqueTrials(
include3rdcountrytrials = FALSE,
con = db),
]
# How many of the investigational medicinal product(s)
# used in the trial are authorised?
number_authorised <- sapply(
result[["dimp.d21_imp_to_be_used_in_the_trial_has_a_marketing_authorisation"]],
function(i) length(i[i])
)
table(number_authorised, exclude = "")
# number_authorised
# 0 1 2 3 4 6 7 8 9 15
# 26 18 7 5 3 2 1 1 1 1
result[["any_authorised"]] <- number_authorised > 0L
# Helper
library(ggplot2)
library(scales)
# Plot
ggplot(
data = result,
aes(
x = startdatefirst,
fill = any_authorised
)
) +
scale_x_date(
breaks = breaks_width(width = "2 years"),
labels = date_format("%Y")
) +
geom_histogram(binwidth = 2 * 365.25) +
labs(
title = "Neuroblastoma trials in EU",
x = "Year of trial authorisation (or entered in EUCTR)",
y = "Number of trials",
fill = "Medicine\nauthorised?"
)
ggsave("vignettes/nbtrials.png", width = 6, height = 4)
HistogramNBtrials
Analyses using aggregation pipeline and mapreduce
Here is an example of analysis functions that can be run directly on a MongoDB server, which are fast and do not consume R resources.
# Load library for database access
library(mongolite)
# Create R object m to access the
# collection db created above:
m <- mongo(
url = paste0(db[["url"]], "/", db[["db"]]),
collection = db[["collection"]]
)
# Number of records in collection:
m$count()
# [1] 2627
# Number of EUCTR records, using JSON for query:
m$count(query = '{"_id": {"$regex": "[0-9]{4}-[0-9]{6}-[0-9]{2}", "$options": "i"}}')
# [1] 1422
# Alternative:
m$count(query = '{"ctrname": "EUCTR"}')
# [1] 272
# Number of CTGOV records:
m$count(query = '{"_id": {"$regex": "NCT[0-9]{8}", "$options": "i"}}')
# [1] 1204
# Alternative:
m$count(query = '{"ctrname": "CTGOV2"}')
# [1] 1204
# To best define regular expressions for analyses, inspect the field:
head(
m$distinct(
# key = "resultsSection.outcomeMeasuresModule.outcomeMeasures.title",
key = "protocolSection.outcomesModule.primaryOutcomes.measure",
query = '{"ctrname": "CTGOV2"}'
)
)
# [1] "\"Off\" Time"
# [2] "% Calories Taken Orally"
# [3] "% Change in Frequency of Handwaving Episodes"
# [4] "% of Patients in Whom a Sentinel Lymph Node is Identified"
# [5] "% of Subjects With Successful Preparation Rated by Colonoscopist on a 4 Point Scale (1=Poor to 4 = Excellent)"
# [6] "'Well-controlled Asthma Week' - a Derived Binary Variable (Yes/No)"Aggregation
The following example uses the aggregation pipeline in MongoDB. See here for details on mongo’s aggregation pipleline: https://docs.mongodb.org/manual/core/aggregation-pipeline/
#
# Total count of PFS, EFS, RFS or DFS
out <- m$aggregate(
# Count number of documents in collection that
# matches in primary_outcome.measure the
# regular expression,
pipeline =
'[{"$match": {"protocolSection.outcomesModule.primaryOutcomes.measure":
{"$regex": "(progression|event|relapse|recurrence|disease)[- ]free",
"$options": "i"}}},
{"$group": {"_id": "null", "count": {"$sum": 1}}}]'
)
out
# _id count
# 1 null 56
# List records of trials with overall survival
# as primary endpoint, and list start date
out <- m$aggregate(
pipeline =
'[{"$match": {"protocolSection.outcomesModule.primaryOutcomes.measure":
{"$regex": "overall survival", "$options": "i"}}},
{"$project": {"_id": 1, "protocolSection.statusModule.startDateStruct.date": 1}}]'
)
head(out)
# _id date
# 1 NCT00793845 2008-08
# 2 NCT02918292 2017-07-03
# 3 NCT01237327 2001-11
# 4 NCT00112827 2004-11
# 5 NCT00515073 2001-04
# 6 NCT00027560 2001-07Mapreduce
Since Mapreduce is deprecated starting in MongoDB 5 (https://docs.mongodb.com/manual/core/map-reduce/), use an aggregation pipeline:
# Count number of trials by number of study
# participants in bins of hundreds of participants:
m$aggregate(pipeline = '
[
{
"$project": {
"flooredNumber": {
"$multiply": [
{
"$floor": {
"$divide": [
{
"$toInt": "$protocolSection.designModule.enrollmentInfo.count"
},
100
]
}
},
100
]
}
}
},
{
"$group": {
"_id": "$flooredNumber",
"count": {
"$count": {}
}
}
},
{
"$sort": {
"_id": 1
}
}
]
')
# _id count
# 1 NA 350
# 2 0 3096
# 3 100 800
# 4 200 429
# 5 300 320
# 6 400 182
# 7 500 155
# 8 600 126
# 9 700 88
# 10 800 73
# 11 900 39
# 12 1000 32
# [...]