PharmaSUG 2020 - Paper EP-337

Generating ADaM Compliant ADSL Dataset Using R

Vipin Kumpawat, Eliassen Group Life Science, Somerset NJ, USA

Abstract
In this paper we show how to generate an ADaM compliant ADSL dataset using R. R packages such as Sas7bdat, Dplyr, Tidyr, and
Hmisc are used to generate the ADSL dataset. A procedure to set-up the R-environment for the process of generating the ADSL
dataset is shown. The typical steps used to create the ADSL dataset along with the derivation of numeric variables, flags, treatment
variables and trial dates are outlined. R procedures to attach labels to the variables are discussed. A side by side comparison of R
and SAS code is presented. A known weakness in R such as attaching labels to the variables [1] has been resolved in this work. The
challenges encountered in generating the ADSL dataset using R are discussed.

Introduction
SAS is widely used in clinical trials. Like SAS, R is a language and environment for statistical computing and graphics. R can be
considered as a viable alternative to SAS for generating specialized clinical trials datasets, tables, listings and figures. R is freely
available and an open source environment which is supported by R foundation of statistical computing. R has many specific
packages available for design, monitoring and analysis of clinical trials datasets, these include Sas7bdat, Dplyr, Tidyr, and Hmisc
that enable reading, merging, transposing, attaching label's to variables respectively [2].

ADaM and ADSL Background

ADaM datasets are classified in three structures: ADSL (Subject-Level Analysis Dataset), BDS (Basic Data Structure), and OCCDS
(Occurrence Data Structure), as shown in Figure 1 [3]. In this work we focus on the ADSL dataset, which contains key data on
demographics, exposure and disposition of a clinical trial. Regardless of the type of the clinical trial design, ADSL dataset contains
one record per subject. ADSL is a source for subject -level variables used in other ADaM datasets. ADSL dataset contains variables
that include information on demographics, randomization factors, planned and actual treatment, sub grouping, subject -level
population flags and important trial dates. The structure of ADSL dataset allows merging with other ADaM and SDTM dataset. The
ADSL dataset is used to provide the key facts about the subject, facts that facilitate analysis and interpretation of analysis.

Figure1. Classifications of Analysis Datasets

1
Steps To Generate ADSL Dataset
The process steps below outline the general steps needed to generate an ADSL dataset, beginning with the reading of the SDTM
datasets and ending with the generation of the ADSL dataset.

1: Reading Datasets

SDTM datasets which include DM, SUPPDM, EX, SUPPEX, DS and SUPPDS datasets are imported into the statistical
programming environment.

2: Sorting, Transposing and Merging DM and SUPPDM dataset

DM and SUPPDM datasets are sorted, and then SUPPDM dataset is transposed and finally merged with DM dataset.

3: Sorting, Transposing and Merging the DS and SUPPDS dataset.

DS and SUPPDS datasets are sorted, and then SUPPDS dataset is transposed and finally merged with DS dataset.

4: Sorting, Transposing and Merging the EX and SUPPEX dataset.

EX and SUPPEX datasets are sorted, and then SUPPEX dataset is transposed and finally merged with EX dataset.

5: Exposure Variables

Extracting Exposure related variables for period 1 and 2.

6: Combining all the datasets

Exposure and Disposition related variables are merged with DM dataset.

7: Generating Numeric Variables

Numeric variables are generated for RACE and SEX variables.

8: Generating Flags

Subject level population flags such as safety population flag (SAFFL), intent-to treat population flag (ITT) and enrollment
population flag (ENRFL) are generated.

9: Generating Treatment Variables

Treatment variables are generated for planned and actual treatment.

10: Generating Trial Dates

Trial dates such as treatment start date and treatment end date are generated.

11: Assigning Labels to the Variables

Labels are assigned to the variables.

Figure2. ADSL dataset composition

2
Setting-up the R Environment
The first typical step of using any programming environment is to install the required packages and activate the libraries needed for
a program. R packages can be installed in R-Studio through the lower right pane of the R-Studio IDE (4th Quadrant). Figure 3 below
shows how to install the Dplyr package; the same procedure can be used to install other R packages namely Sas7bdat, Tidyr and
Hmisc shown in Table 1. R Libraries can be installed in R-Studio through console commands as shown in figure 4.

Figure3. R-Studio Snapshot to install R package

Figure4. R console command's to install R libraries

Table 1: R Packages and Libraries needed for generating ADSL dataset.

Procedure R Package R Library

Read SAS Dataset Sas7bdat Foreign
Merge Datasets Dplyr Dplyr
Transpose Dataset Tidyr Tidyr
Attach Label Hmisc Hmsic
Check attributes Hmisc Hmsic

R Code for generating the ADSL Dataset

After the R packages and libraries are installed, the R-environment is ready to read and process the input dataset's used to generate
the ADSL dataset. Step by step R code to generate the ADSL dataset is shown below. Table 2 summarizes the R functions used in
the processes of generating the ADSL dataset.

1. Reading Dm, Suppdm, Ds, Suppds, Ex and Suppex Datasets

dm<- read.sas7bdat("c:/sdtm/dm.sas7bdat")
suppdm<- read.sas7bdat("c:/sdtm/suppdm.sas7bdat")
ds<- read.sas7bdat("c:/sdtm/ds.sas7bdat")
suppds<- read.sas7bdat("c:/sdtm/suppds.sas7bdat")
ex<- read.sas7bdat("c:/sdtm/ex.sas7bdat")
suppex<- read.sas7bdat("c:/sdtm/suppex.sas7bdat")

3
2. Combining Dm and Suppdm Dataset by Sorting, Transposing and Merging Steps.

1. Sorting Dm and Suppdm Dataset

dm<- dm[order(usubjid), ]
suppdm<- suppdm[order(usubjid), ]
2. Transposing Suppdm Dataset
suppdm_ <- suppdm %>%
select(usubjid,qnam,qval) %>%
spread(.,qnam,qval)
3. Merging Dm and Suppdm Dataset
dm<- dm[order(usubjid), ]
suppdm_ <- suppdm_[order(usubjid), ]
dmall<- left_join(dm, suppdm_, by = "usubjid")

3. Combining Ds and Suppds Dataset by Sorting, Transposing and Merging Steps.

1. Sorting Ds and Suppds Dataset

ds<- ds[order(usubjid), ]
suppds<- suppds[order(usubjid), ]
2. Transposing Suppds Dataset
suppds_ <- suppds %>%
mutate(dsseq = as.numeric(idvarval)) %>%
select(usubjid,qnam,qval) %>%
spread(.,qnam,qval)
3. Merging Ds and Suppds Dataset
ds<- ds[order(usubjid, dsseq), ]
suppds_ <- suppds_[order(usubjid, dsseq), ]
dsall<- left_join(ds, suppds_, by = c("usubjid", “dsseq”)

4. Combining Ex and Suppex Dataset by Sorting, Transposing and Merging Steps.

1. Sorting Ex and Suppex Dataset

ex<- ex[order(usubjid), ]
suppex<- suppex[order(usubjid), ]
2. Transposing Suppex Dataset
suppex_ <- suppex %>%
mutate(exseq = as.numeric(idvarval)) %>%
select(usubjid,qnam,qval) %>%
spread(.,qnam,qval)
3. Merging Ex and Suppex Dataset
ex<- ex[order(usubjid, exseq), ]
suppex_ <- suppex_[order(usubjid, exseq), ]
exall<- left_join(ex, suppex_, by = c("usubjid", “exseq”)

5. Generating Exposure Variables For Period 1 And Period 2

ex1 <- subset(exall, exseq==1)

ex1 <- select(ex1, usubjid, extrt, exdose, exstdtc, exendtc)
ex2 <- subset(exall, exseq==2)
ex2 <- select(ex2, usubjid, extrt, exdose, exstdtc, exendtc)

4
 ex2<- ex2 %>%
rename(extrt2 = extrt, exdose2 = exdose, exstdtc2 =
exstdtc, exendtc2 = exendtc)

6. Combine Exposure and Disposition Datasets With Dm Dataset

dmexall<- left_join(dmall, ex1, by = "usubjid")

dmexall<- left_join(dmexall, ex2, by = "usubjid")
adsl<- left_join(dmexall, dsall, by = "usubjid")

7. Generate Numeric Variables for Sex and Race Variable

adsl$sexn[adsl$sex=="m"]<- 1
adsl$sexn[adsl$sex=="f"]<- 2
adsl$racen[adsl$race=="asian"]<- 1
adsl$racen[adsl$race=="other"]<- 2

8. Generate Flags (Saffl, Ittfl, Enrfl)

adsl$saffl <- ifelse(!is.na(adsl$exdose) & !is.na(adsl$exstdtc),

"Y", "N")
adsl$ittfl <- ifelse(!is.na(adsl$armcd), "Y", "N")
adsl$enrfl <- ifelse(!is.na(adsl$rfstdtc) & !is.na(adsl$rficdtc),
"Y", "N")

9. Generate Treatment Variables

1. Treatment Variables for Planned Treatment

adsl$trt01p[adsl$armcd=="xx"] <- "refe"

adsl$trt02p[adsl$armcd=="xx"] <- "test"
adsl$trt01pn[adsl$armcd=="xx"] <- 2
adsl$trt02pn[adsl$armcd=="xx"] <- 1
adsl$trt01p[adsl$armcd=="yy"] <- "test"
adsl$trt02p[adsl$armcd=="yy"] <- "refe"
adsl$trt01pn[adsl$armcd=="yy"] <- 1
adsl$trt02pn[adsl$armcd=="yy"] <- 2

2. Treatment Variables for Actual Treatment

adsl$trt01a[adsl$actarmcd=="xx"] <- "refe"

adsl$trt02a[adsl$actarmcd=="xx"] <- "test"
adsl$trt01an[adsl$actarmcd=="xx"] <- 2
adsl$trt02an[adsl$actarmcd=="xx"] <- 1
adsl$trt01a[adsl$actarmcd=="yy"] <- "test"
adsl$trt02a[adsl$actarmcd=="yy"] <- "refe"
adsl$trt01an[adsl$actarmcd=="yy"] <- 1
adsl$trt02an[adsl$actarmcd=="yy"] <- 2

5
 10. Generating Trial Dates

adsl$trt01sdt <- adsl$exstdtc

adsl$trt02sdt <- adsl$exstdtc2

11. Assigning Labels To The Variables

label(adsl$studyid) <- "Study Identifier"

label(adsl$usubjid) <- "Unique Subject Identifier"

Similarly attach all the labels to the remaining variables.

12. Checking the Labels

describe(adsl)

Table 2: R functions used to generate ADSL dataset.

Procedure R Function
Read SAS Dataset read.sas7bdat
Merge Datasets inner_join/left_join/right_ join/full_join
Transpose Dataset Spread
Attach Label Label
Variable Selection Select
Character to Numeric as.numeric
Numeric to Character as.character
Check attributes Content
Sort Dataset Order
Rename Variable Rename
Conditional operator ifelse()
Check Labels Describe
Check variable type Class

Comparison of SAS and R Code

There are some key differences between SAS and R programming language . The most basic difference is that SAS is case insensitive
while R is case sensitive. There are other key differences [4,5] that pertain to how each language sorts and rounds data and the way
each language handle’s missing values . SAS and R handle missing values differently in the sorting process. In SAS the missing
values will be sorted before the populated value for both numeric and character variable when using the Proc Sort procedure. In R,
the character column missing value will be sorted before the populated value while the numeric column missing value will be sorted
after populated value when using the order function, which is a Proc Sort equivalent of R. SAS and R handle rounding of numeric
values differently resulting in different values, this is due to the difference in the algorithm that each language uses for rounding
[4,5].

The Table 3 below compares the code in SAS and R to generate ADSL dataset. Table 3 is a comprehensive procedure step and code
reference for generating the ADSL dataset in SAS and R.

6
 Table 3: SAS and R code comparison.

No Procedure SAS code R code

1 Importing libname sdtm "c:\sdtm"; library(foreign)
and Reading data dm1; dm <- read.sas7bdat("c:/sdtm/
Dataset set sdtm.dm; dm.sas7bdat")
run;
 similarly read suppdm, ds, suppds,  similarly read suppdm, ds, suppds, ex, suppex
ex, suppex

2 Checking proc contents data = dm; library(hmisc)

Dataset run; content(dm)

3 Sorting proc sort data = suppdm; suppdm <- suppdm [order(usubjid), ]

Dataset by usubjid;
run;

4 Transposing proc transpose data = suppdm out library(tidyr)

Data = suppdmt(drop=_name_ _label_);
by usubjid; suppdm_ <- suppdm %>%
id qnam; var qval; select(usubjid, qnam, qval) %>%
idlabel qlabel; spread(., qnam, qval)
run;

5 Merging Dm, data dmall; library(dplyr)

Suppdm merge dm1(in=a) suppdmt (in=b); dm <- dm[order(usubjid), ]
by usubjid; suppdm_ <- suppdm_[order(usubjid), ]
if a; dmall <- left_join(dm, suppdm_ , by =
run; "usubjid")
 similarly dsall, exall datasets  similarly dsall, exall datasets generated
generated

6 Ex dataset data ex1; set exall;

for Period 1 if exseq = 1; ex1 <- subset(exall, exseq==1)
and 2 keep usubjid extrt exdose ex1 <- select(ex1, usubjid, extrt,
exstdtc exendtc; exdose, exstdtc, exendtc)
run;
ex2 <- subset(exall, exseq==2)
data ex2 ; set exall; ex2 <- select(ex2, usubjid, extrt,
if exseq = 2; exdose, exstdtc, exendtc)
extrt2 = extrt;
exdose2 = exdose; ex2 <- ex2 %>% rename(extrt2 = extrt,
exstdtc2 = exstdtc; exdose2 = exdose, exstdtc2 = exstdtc,
exendtc2 = exendtc; exendtc2 = exendtc)
keep usubjid extrt2 exdose2
exstdtc2 exendtc2; run;

7 Combining data adsl1; dmexall <- left_join(dmall, ex1, by =

Ex, Ds merge dmall(in=a) ex1 ex2 dsall; "usubjid")
datasets with by usubjid; dmexall <- left_join(dmexall, ex2, by =
Dm dataset if a; "usubjid")
run; adsl<- left_join(dmexall, dsall, by =
"usubjid")

8 Generating data adsl2; adsl$sexn[adsl$sex=="m"]<- 1

Numeric set adsl1; adsl$sexn[adsl$sex=="f"]<- 2
variable if sex = "m" then sexn = 1;
else if sex = "f" then sexn = 2; adsl$racen[adsl$race=="asian"]<- 1
if race = "asian" then racen =1; adsl$racen[adsl$race=="other"]<- 2
else if race = "other" then
racen = 2;run;

7
 9 Generating data adsl3; adsl$saffl <-
Safety Flag, set adsl2; ifelse( !is.na(adsl$exdose) & !
ITT Flag, if exdose ^= . and exstdtc ^= ‘’ is.na(adsl$exstdtc), "Y", "N")
Enrolment then saffl = "Y";
Flag else saffl = "N"; adsl$ittfl <-
if armcd ^= ‘’ then ittfl = ‘Y’; ifelse( !is.na(adsl$armcd), "Y", "N")
else ittfl = 'N';
if rfstdtc ^= " " and rficdtc ^= adsl$enrfl <-
" " then enrfl = ‘Y’; ifelse( !is.na(adsl$rfstdtc) & !
else enrfl = 'N'; is.na(adsl$rficdtc), "Y", "N")
run;
12 Treatment data adsl4; Treatment variables for planned treatment
Variable :- set adsl3;
if armcd = "xx" then do; adsl$trt01p[adsl$armcd=="xx"] <- "refe"
TRT01P trt01p = " refe"; adsl$trt02p[adsl$armcd=="xx"] <- "test"
TRT02P trt02p = " test ";
TRT01PN trt01pn = 2; adsl$trt01pn[adsl$armcd=="xx"] <- 2
adsl$trt02pn[adsl$armcd=="xx"] <- 1
TRT02PN trt02pn = 1;
end;
adsl$trt01p[adsl$armcd=="yy"] <- "test"
TRT01A if armcd = "yy" then do; adsl$trt02p[adsl$armcd=="yy"] <- "refe"
TRT02A trt01p = " test";
TRT01AN trt02p = " refe"; adsl$trt01pn[adsl$armcd=="yy"] <- 1
TRT02AN trt01pn = 1; adsl$trt02pn[adsl$armcd=="yy"] <- 2
trt02pn = 2;
end; run; Similarly for actual treatment
Similarly for actual treatment

13 Trial Dates data adsl5; adsl$trt01sdt <- adsl$exstdtc

set adsl4;
trt01sdt = exstdtc ; adsl$trt02sdt <- adsl$exstdtc2
trt02sdt = exstdtc2 ;
run;

14 Attaching proc sql ; library(hmisc)

Labels create table adsl as
select label(adsl$studyid) <- "Study Identifier"
studyid "study identifier", .
. .
. label(adsl$trt02sdt) <- "P-02 Start date"
trt02sdt "P-02 start date "
from adsl5; describe(adsl) #To check the labels
quit;

Results and Discussion

The key issue that this paper addresses is attaching labels to the variables. In R, labels can be attached to the variable by using
“Hmisc” package. After loading Hmisc package we start calling Hmisc library by following code.

library(hmisc)
Using label function in Hmisc, we add the label to each variable by following code.

label(adsl$studyid) <- "Study Identifier"

label(adsl$usubjid) <- "Unique Subject Identifier"

“Describe(adsl)” code can be used to check the labels attached to the variables. Below is the snapshot of the dataset which display
the labels with the variable name.

8
Two key challenges faced when coding in R were (1) R does not provide a log like SAS, so code debugging is difficult in R
compared to SAS. (2) We were able to attach label to the variables but were not able to attach label to the dataset.

Conclusion
In this paper we generated an ADaM compliant ADSL dataset using the R programming language. We demonstrate that R can be
used as an effective alternative to create the clinical trial dataset. We provided a step by step process to set-up the R environment
and the R code for reading the input dataset and processing the data to generate the ADSL dataset. We also compared the SAS and
R code for the process, and discussed challenges encountered and addressed issues like attaching labels to variables.

Reference
1. Prasanna Murugesan, 2018, “Clinical Trial Datasets (CDISC - SDTM/ADaM) Using R”, Phuse US Connect.
2. Monika M. Wahi, Peter Seebach, 2018, “Analyzing Health Data in R for SAS Users”, Boca Raton, Florida, Taylor &
Francis.
3. CDISC Analysis Data Model Team, 2016, “Analysis Data Model Implementation Guide Version 1.1”.
4. Ali Dootson, 2020,TFL Programming in R versus SAS, d-wise, https://www.d-wise.com/blog/tfl-programming-in-r-
versus-sas
5. Amol Waykar, Kevin Kramer, Kalyani Komarasetti, Andrew Miskell, 2020, Generating TFLs in R - Challenges and
Successes compared to SAS, Phuse US Connect.

Acknowledgement
I would like to thank Nagadip Rao, Associate Director of Eliassen Group Life Science for reviewing this paper and providing
valuable comments. I would also like to thank Lalitkumar Bansal of Statum Analytics for technical discussions and editorial inputs
to this paper.

Contact Information
Vipin Kumpawat
Eliassen Group Life Science
Somerset New Jersey USA
[email protected]
[email protected]