SDTM vs ADaM vs Define-XML: How Clinical Trial Data Moves From Collection to Analysis
Understand how clinical trial data move from collection through SDTM tabulation datasets into ADaM analysis datasets, and how Define-XML documents those structures. The Resource also explains ADSL, traceability, and why standardized submission data are not automatically analysis-ready.
Clinical trial data do not usually move directly from a case report form into a statistical model. Between collection and analysis, the data are reorganized into standardized structures, transformed into analysis-oriented datasets, and accompanied by metadata that explains what the datasets and variables mean and how they were created. In the workflow described by Case and Tian (2022), three concepts are central: SDTM for standardized tabulation data, ADaM for analysis data, and Define-XML for metadata describing those data structures.
The conceptual sequence is:
case report form / collected data → SDTM tabulation data → ADaM analysis data → statistical analyses and outputs
Define-XML describes the SDTM and ADaM datasets and variables rather than functioning as another analysis dataset (Case & Tian, 2022).
Understanding this separation answers an important practical question: why are standardized submission/tabulation datasets not automatically the same datasets that statisticians need for analysis?
The workflow starts with what the trial needs to collect
A clinical trial protocol establishes the study plan, while the case report form (CRF) is designed to record information required according to that protocol. The International Encyclopedia of Statistical Science similarly describes the CRF as the document designed to record the required information to be reported according to the protocol (Lovric, 2011).
Case and Tian (2022) place the CRF explicitly at the beginning of the standardized-data programming workflow. The blank CRF is used to collect information for individual patients, and its development should ensure that information needed for subsequent analysis is captured. The annotated CRF then helps connect collected CRF information to SDTM domains.
Important distinction: The CRF is fundamentally organized around data collection. SDTM, by contrast, reorganizes those collected observations into standardized tabulation structures.
What is SDTM?
The Study Data Tabulation Model (SDTM) provides a standard structure for human clinical trial data. Case and Tian (2022) describe SDTM datasets as tabulation data built around observations collected about subjects participating in a clinical study. SDTM domains are organized around topics and commonly use two-letter domain identifiers—for example, DM for Demographics and AE for Adverse Events (Case & Tian, 2022).
The practical purpose of SDTM is therefore standardization of the collected clinical data into defined tabular structures. It provides an organized representation of observations rather than being designed primarily around the calculations required for a particular statistical analysis (Case & Tian, 2022).
SDTM domains
Case and Tian (2022) distinguish special-purpose domains from general-observation domains. Examples they discuss include:
DM
Demographics
AE
Adverse Events
DS
Disposition
CM
Concomitant Medications
EX
Exposure
EG
ECG Test Results
LB
Laboratory Test Results
They also describe trial-design domains including TA, TE, TI, TS, and TV (Case & Tian, 2022).
General-observation domains are grouped into Interventions, Events, and Findings, while special-purpose domains have distinct structures (Case & Tian, 2022).
Variables have defined roles within SDTM
SDTM is not simply a convention for dataset names. Variables within observations have defined roles. Case and Tian (2022) describe five major roles:
- Identifier variables, which identify the study, subject, domain, or record.
- Topic variables, which identify the focus of an observation.
- Timing variables, which describe when an observation or event occurred.
- Qualifier variables, which provide results or other characteristics of the observation.
- Rule variables, used in trial-design domains to describe aspects of the study and its conduct.
SDTM variables are also characterized as required, expected, or permissible, with those categories affecting whether variables must appear and whether null values are allowed (Case & Tian, 2022).
The result is a standardized representation of collected clinical observations. But standardized representation is not the same objective as statistical analysis.
What is ADaM?
The Analysis Data Model (ADaM) is a standard analysis-data structure built from SDTM. Its purpose is to create datasets that support the summarization and analysis of clinical trial data (Case & Tian, 2022).
The key distinction in SDTM vs ADaM:
SDTM organizes clinical observations into standardized tabulations; ADaM reorganizes and derives information so that it can support the planned analyses.
Clinical-trial analyses can involve continuous, categorical, time-to-event, and other endpoint structures, and the analytical datasets therefore need to provide the variables and organization required by the corresponding analysis (Dmitrienko & Koch, 2017).
ADaM analysis datasets
Case and Tian (2022) describe three standard classes of ADaM datasets:
ADSL
Subject-Level Analysis Dataset
BDS
Basic Data Structure
Examples discussed include laboratory, ECG, vital-sign, and drug-accountability analysis datasets.
OCCDS
Occurrence Data Structure
Examples discussed include adverse events and concomitant medications.
This structure illustrates why “analysis-ready” does not simply mean “cleaned SDTM.” Analysis datasets can combine source information, retain variables required for interpretation, and add derived variables needed for the planned analysis.
What is ADSL?
ADSL is the subject-level analysis dataset. Case and Tian (2022) describe it as containing one record per unique subject and combining variables that describe the subject's experience in the clinical trial. These can include identifiers, demographic information, treatment variables, dose information, population flags, and other subject-level variables (Case & Tian, 2022).
Some ADSL variables come directly from SDTM—for example, subject-level information from the DM domain—while other variables can be derived using information from domains such as EX and DS (Case & Tian, 2022).
ADSL is therefore a useful example of the difference between tabulation and analysis structures. Subject information that is distributed across standardized source domains can be brought together into a subject-level dataset designed to support analysis.
Traceability is the bridge between SDTM and ADaM
Transforming data for analysis creates an obvious risk: once variables are derived, combined, or reorganized, can an analyst or reviewer determine where they came from?
For Case and Tian (2022), traceability is a fundamental principle of ADaM. ADaM variables should be traceable to corresponding SDTM domains and, through SDTM, back toward the collected eCRF information. Traceability establishes relationships both across datasets and within datasets and helps explain how analysis variables relate to their source datasets and variables (Case & Tian, 2022).
Important rule: Where an ADaM variable has the same name as an SDTM variable, Case and Tian state that it should be a direct copy whose label, meaning, and values are not modified (Case & Tian, 2022).
Conceptually, traceability preserves the chain:
collected observation → standardized SDTM representation → ADaM analysis variable → analysis
The purpose is not to prevent derivation. Derivation is often necessary for analysis. The purpose is to make the relationship between derived analytical information and its source understandable.
Why SDTM and ADaM are not the same thing
A common conceptual mistake in the SDTM ADaM workflow is to treat one as a newer or more processed version of the other without considering their different purposes.
They solve different problems.
SDTM asks
How should the collected clinical observations be represented in a standardized tabulation structure?
ADaM asks
How should information from those standardized data be organized or derived to support the planned analyses?
Case and Tian (2022) explicitly describe ADaM as being built from SDTM and used to develop datasets for summarizing and analyzing clinical data. ADaM specifications can therefore contain both information originating in SDTM domains and new variables or derivations required for analysis (Case & Tian, 2022).
This separation is methodologically useful because a structure optimized for representing collected observations is not necessarily a structure optimized for answering an analytical question. Statistical analysis may require subject-level population indicators, derived analysis values, treatment variables, baseline or visit-related organization, and other analysis-specific information that need not exist in the same form in the tabulation datasets.
The transformation should nevertheless remain traceable.
What is Define-XML in clinical trials?
Define-XML is metadata. It describes tabular dataset structures such as SDTM and ADaM rather than replacing those datasets (Case & Tian, 2022).
Case and Tian (2022) describe Define-XML as a machine-readable representation of specifications that provides information about how SDTM and ADaM datasets were created, including information needed to understand complex derivations.
At the dataset level, Define-XML can describe information such as the domain or dataset name, purpose, structure, location, and key variables. At the variable level, metadata can include the variable name, type, length, label, and comments. The book also describes metadata for controlled terminology and computation methods (Case & Tian, 2022).
Main metadata components
Case and Tian identify five main metadata components in their Define-XML creation example:
- Header metadata
- Table-of-contents or dataset-level metadata
- Variable-level metadata
- Controlled-terminology metadata
- Computation-method metadata
Within the XML structure they describe, dataset/domain-level metadata and variable-level metadata are represented separately, and external files or documents can also be linked (Case & Tian, 2022).
Define-XML answers a different question:
What exactly are these datasets and variables, and how should someone understand their structure and derivation?
SDTM vs ADaM vs Define-XML
| Feature | SDTM | ADaM | Define-XML |
|---|---|---|---|
| Primary purpose | Standardize clinical trial observations into tabulation datasets | Organize and derive data to support analysis | Describe the datasets, variables, terminology, and derivations through metadata |
| Relationship to collected data | Represents observations originating from collected clinical data | Built from SDTM and may contain direct-source and derived analytical information | Describes SDTM and ADaM structures |
| Typical organization | Topic-oriented domains | Analysis-oriented datasets | Machine-readable metadata |
| Examples | DM, AE, DS, CM, EX, EG, LB | ADSL, ADLB, ADEG, ADAE, ADCM | Dataset-level, variable-level, controlled-terminology, and computation-method metadata |
| Main question answered | “How is this clinical observation represented consistently?” | “What data structure is needed for the planned analysis?” | “What do these datasets and variables mean, and how were they constructed?” |
| Role in analysis | Provides standardized source tabulations | Provides analysis datasets and derived variables | Documents and explains the data structures rather than serving as the analytical dataset |
| Traceability relationship | Connects standardized observations toward collected source information | Maintains relationships back to SDTM source variables and datasets | Helps document structures and derivations used in SDTM and ADaM |
(Case & Tian, 2022).
The SDTM–ADaM workflow as a chain of research decisions
A useful way to understand clinical trial data standards is not as three competing file formats but as a sequence of different data-management decisions.
1. Collection
What information must the study capture?
The protocol defines the study plan, and the CRF records the required participant information according to that plan (Case & Tian, 2022; Lovric, 2011).
2. Tabulation
How should those observations be represented consistently?
Collected information is mapped into SDTM domains and variables. The annotated CRF and SDTM specifications help document that mapping and the construction of SDTM variables (Case & Tian, 2022).
3. Analysis preparation
What does the planned analysis need?
ADaM datasets are built from SDTM and can contain both retained source variables and derived analytical variables. Their structure reflects the needs of analysis rather than simply reproducing the collection structure (Case & Tian, 2022).
4. Traceability
Can the analysis information be related back to its source?
ADaM preserves relationships to SDTM, helping users understand where analysis variables originated and how they relate to source variables and datasets (Case & Tian, 2022).
5. Metadata
Can another user understand the datasets without guessing?
Define-XML supplies machine-readable metadata about the dataset and variable structures, controlled terminology, and computation or derivation information (Case & Tian, 2022).
The complete conceptual workflow
Protocol and CRF
↓ define and collect study information
Collected clinical data
↓ standardize observations
SDTM domains
↓ select, combine, retain, and derive information for analysis
ADaM analysis datasets
↓
Statistical analyses, tables, figures, and listings
Alongside SDTM and ADaM:
Define-XML → metadata explaining the datasets, variables, terminology, and derivations
(Case & Tian, 2022).
Practical checks when reviewing an SDTM–ADaM workflow
When assessing whether a clinical-trial data workflow is conceptually coherent, check the chain rather than inspecting only the final analysis dataset:
Collection
Is the information needed for the intended analysis actually captured?
SDTM mapping
Can collected information be related to the appropriate standardized domains and variables?
Analysis purpose
Does the ADaM structure contain the information and derivations needed for the planned analysis?
ADSL
Are relevant subject-level characteristics, treatments, and population indicators represented at the subject level where appropriate?
Traceability
Can important analytical variables be related back to their SDTM predecessors?
Derivations
Are derived analytical variables distinguishable from variables copied directly from SDTM?
Metadata
Does the metadata explain dataset structures, variables, terminology, and relevant computation or derivation information?
Interpretability
Could another analyst understand how the analysis data relate to the standardized source data without reconstructing the workflow by guesswork?
These checks follow the separation of collection, standardization, analysis preparation, and metadata documentation described by Case and Tian (2022).
Key takeaway
The simplest way to understand SDTM vs ADaM is to ask what job each component performs.
SDTM standardizes what was observed and collected. ADaM structures and derives what is needed for analysis. Define-XML describes the resulting datasets and variables so their structure and derivations can be understood. The three therefore complement rather than replace one another (Case & Tian, 2022).
The critical connection is traceability. Clinical trial data may change structure as they move from collection to standardized tabulation and then to analysis, but the analytical information should remain meaningfully connected to its source. That is why submission/tabulation data and analysis data serve different purposes—and why metadata is needed to make the complete data pathway interpretable (Case & Tian, 2022).
References
Case, T., & Tian, Y. (2022). An introduction to creating standardized clinical trial data with SAS. SAS Institute.
Dmitrienko, A., & Koch, G. G. (Eds.). (2017). Analysis of clinical trials using SAS: A practical guide (2nd ed.). SAS Institute.
Lovric, M. (Ed.). (2011). International encyclopedia of statistical science. Springer. https://doi.org/10.1007/978-3-642-04898-2
Need help with a similar research question?
Share a short, non-confidential summary of your study and the decision you need to make.