SAUG ADaM Standards

ADaM Standards
Wouter van Wyk
Why ADaM
– SDTM purpose is to provide collected data
• Not designed for ease of analysis
– ADaM purpose is to provide data that is ready for analysis
• Is designed for ease of analysis
• “1 proc away”
– ADaM helps the reviewer trace:
• What you said you would do
• What you did
– TRACEABILITY is of utmost importance
What is Traceability?
– Traceability permits the understanding of the relationship between the
analysis results, the analysis datasets and the SDTM datasets.
– Traceability is built by clearly establishing the path between an element and
its predecessor(s)
– Traceability establishes across-dataset relationships, as well as withindataset relationships.
– Traceability = Transparency
What is Traceability?
Analysis Result
Annotated shells
It should be easy to trace
where a value came from
ADaM Dataset
ADaM Specs
SDTM Dataset
(in dataset)
2 Kinds of Traceability
– Metadata traceability
• How was a variable or observation derived?
• ADaM Specifications, Define.xml
– Data point traceability
• Which observations were used to derive a value of observation?
• Traceability variables in the ADaM dataset
ADaM Rules for Analysis Datasets
– At least ADSL should be created
– Not all data from SDTM needs to go into ADaM
– There is not a 1:1 relationship between ADaM and SDTM
• 1 SDTM dataset could map to multiple ADaM Datasets
• Many SDTM datasets could map to the same ADaM Dataset
• Some of the SDTM data will be present in more than 1 ADaM Dataset
– Create the optimum number of ADaM datasets as required for analysis
• E.g. If Medical History is not summarized, you don’t need to create an ADaM
dataset for it.
ADaM Rules for Analysis Datasets
– If the variable is straight from SDTM it
• must have the same attributes
• must have the same contents
• may not be modified in any way
– See the ADaM Implementation Guide on naming conventions and controlled
Classes of ADaM structures
– Subject Level Structure (ADSL)
• Reserved dataset name ‘ADSL’
• One record per subject, regardless of study design
– Basic Data Structure (BDS)
• Designed with the majority of analysis files in mind
• One or more records per subject per analysis parameter, per analysis time point (if
• Most if not all the by Visit domains (Findings) from SDTM would fit in this structure
• Adverse Events does not fit the BDS Structure
– You can create your own class if these 2 are not appropriate
SDTM and ADaM “Core” Definitions
SDTM “Core” Definitions
ADaM “Core” Definitions
• Must be present and must be
• Must be present
• Any variable necessary to make a
record meaningful in the context of a
specific domain
• Must be present (but may be blank)
Conditionally Required if applicable
• Must be present if applicable (e.g.
“Actual Treatment” would be required
if patients did not receive the “Planned
• Any variable that is appropriate
when collected or derived (remove if
• May be present
Analysis Data Subject Level - ADSL
– Contains important information about a subject
– Structured as one record per subject
• Regardless of clinical trial design
• Allows for easy merging with any other ADaM or SDTM dataset
– Used as a source for
• Variables required in other analysis datasets
• Denominator values for populations of interest
– Contains some Required variables
Analysis Data Subject Level - ADSL
– Primary source of subject-level variables included in other analysis datasets
– Variables describe attributes of a subject as well as the subject’s experience
in trial (e.g. date of death)
– Not intended to be the only file that supports subject level data
• E.g. if efficacy analysis is time to event, create ADTTE file instead of putting
everything in ADSL
– Standard Disposition and Demographics tables should be able to be created
from ADSL
What goes into ADSL?
– Study and subject identifiers
– Subject demographics
– Population indicators
– Treatment variables
– Trial dates
– Stratification variables
– Treatment duration and possibly compliance variables
– See the ADaM IG for possible ADSL variables
Basic Data Structure - BDS
– Flexible enough for the majority of analyses
– Vertical structure: can be thought of as one or more records
per subject (USUBJID)
per analysis parameter (PARAM, PARAMCD)
per analysis time point (AVISIT, ADY)
– Even though similar in structure to the SDTM Findings domains (think of VS,
LB and EG) if is not a replica and could contain any type of data
An Event
A Finding
An Intervention
Completely derived observation that is created from multiple types of SDTM data
Basic Data Structure - BDS
– The BDS domains have an “Analysis-focused” design
• One proc away
– Supportive columns may be added as appropriate
– Named ADxxxxxx, where then ‘x’ denotes a name of your choosing
What goes into BDS?
– ADSL Variables
• Keep only what is needed
• STUDYID, USUBJID at minimum
BDS Analysis Parameter Variables
BDS Analysis Value Variables
BDS Analysis Timing Variables
BDS Analysis Descriptor Variables
BDS Indicator Variables
BDS Analysis Population Indicators
BDS Analysis Enabling Variables
BDS Data Point Traceability Variables
BDS Analysis Parameter Variables
– Variables to describe what is being analyzed
– Required
• PARAM - Parameter Description
– Value of PARAM should be what appears on statistical tables
– Uniquely describes what is contained in the Analysis variable
– Includes: test name, unit, position, location etc
• PARAMCD – 8 character version of PARAM
– Other
BDS Analysis Value Variables
– Variables that contain numeric of character values that are analyzed
– At least one of the following is required
• AVAL: numeric result as described by PARAM
• AVALC: character result as described by PARAM
– Other
BASE: numeric baseline result
BASEC: character baseline result
CHG: change from baseline
PCHG: Percent change from baseline
SHIFTy: function of defined pairs,
– Such as shift from baseline - ‘Normal to Abnormal’
Some BDS Analysis Descriptor Variables
– Use when there are derived observations within a PARAM
– Use where there is windowing
AWRANGE, AWLO, AWHI: Analysis window valid range
AWTARGET: Analysis window target day
AWDIFF: Difference from target day
– Use in time-to event analysis
STARTDT: Original date of risk for the TTE analysis
CNSR: Defines whether the event is censored
EVENTDESC: Description of the event of interest or censoring reason
The actual time to event result will be in AVAL
BDS Indicator Variables
– Conventions:
• Character variable will end in *FL
– Y/N/Null
• Numeric variable will end in *FN
– 1/0/Null
– Examples:
• ABLFL: Baseline flag
• ANLzzFL: Analyzable flag
• ONTRTFL: On-treatment flag
BDS Data Point Traceability Variables
– A variable is supportive if it is not required in order to perform an analysis but
is included in order to facilitate traceability
– Retaining certain SDTM variables
– Additional variables
• SRCDOM: Domain of origin for value contained in AVAL (e.g. ”LB”)
• SRCVAR: Variable name used to create AVAL (e.g. “LBSTRESN”)
• SRCSEQ: Sequence number to identify record used (e.g. = LBSEQ)
What to do with Adverse Events?
– Some kinds of data do not fit in the BDS Structure
• E.g. Adverse Events, Concomitant Medications
– Create your own structure based on ADaM principles
• Dataset name should start with “AD”
• SDTM variables should not be modified, rather create new variable and keep the
SDTM variable for traceability
• Dataset should be analysis ready, one proc away
• Follow naming conventions
• Ensure traceability back to the source data
– ADAE could mean you only need to merge ADSL with SDTM.AE and create
a few numeric dates as needed, but will be driven by what analyses need to
be done.
Some useful websites
Any Questions?

similar documents