The Analysis Data Model, internationally known as the Analysis Data Model or ADaM, is the CDISC standard for datasets used to calculate and validate statistical analyses of clinical trials. It defines both data structures and metadata so that the content, origin, and analytical purpose of each dataset remain traceable. ADaM thus translates the tabular representation of a study into a form suitable for analysis.
This makes scientific analysis decisions visible as verifiable data attributes, not merely as program code.
From tabulation dataset to analysis
As a rule, ADaM follows the preparation of clinical data according to SDTM. While SDTM structures observations as documented, ADaM consolidates them for a specific analysis question, adds derived variables, and provides analysis populations. A subject-level dataset may, for example, include treatment assignment and population flags; other structures can provide measurements, changes from baseline, or time-to-event information.
Its construction is guided by the protocol and the Statistical Analysis Plan. Each derivation must be described in such a way that the analysis result can be traced from the ADaM dataset back to the SDTM variables used. This traceability enables not only rerunning a program, but also a scientific review of which inputs and rules underpin a reported metric.
Data structures and explanatory metadata
ADaM v2.1 describes core principles for all analysis datasets, including the Subject-Level Analysis Dataset and the Basic Data Structure. Such standard structures specify how information can be organised at the subject level or at the parameter-by-timepoint level. Additional implementation guides and addenda address specific analysis cases and provide conformance rules for technical checks.
The associated metadata must clearly indicate what a dataset contains, which input data were used, and what purpose it serves. Define-XML is used as the metadata format for submission; Analysis Results Metadata can additionally describe relationships to analyses. Technical rule checks are useful, but they do not replace a scientific review of whether the selected population, endpoint definition, and derivations align with the analysis plan.
Submitting analysis-ready data makes it possible to trace results from the clinical study report and key safety or efficacy analyses in a targeted manner. The FDA Study Data Technical Conformance Guide distinguishes clinical tabulations from analysis datasets and expects ADaM datasets to support key analyses. Programs should therefore not only output a value, but also clearly link their data inputs, derived variables, and analysis attributes.
A common source of error is discrepancies between a rule defined in the analysis plan and its data implementation, for example for reference dates, treatment periods, or censoring. ADaM metadata make such decisions visible, but cannot replace a scientific review. At the same time, they must be readable for statistical reviewers and sufficiently precise for technical reproduction of the results.
Distinction: data model rather than analysis method
ADaM is not a statistical method and does not determine whether, for example, a Cox model, an analysis of covariance, or a nonparametric analysis is appropriate. These decisions belong in the Statistical Analysis Plan. Rather, ADaM provides the standardised data basis from which the calculations described in the plan can be generated reproducibly.
An analysis population is also not synonymous with ADaM. It is a scientifically defined group of participants, such as the safety population or the randomised population; ADaM can represent the corresponding flags and rules. Compared with SDTM, ADaM is the downstream processing layer: derivations or imputations required for analyses may appear there, provided they are described transparently.
The naming of datasets and variables should make their analytical purpose clear and distinguish them from identically named SDTM objects. This makes it clear during review whether a table was generated from the source level or the analysis level.
Relevance for clinical trials
For analysis, biostatistics and programming must define, before database lock, which endpoints, populations, censoring rules, and repeated measures feed into which ADaM structure. Deviations between the analysis plan, TLF specifications, and data derivations are particularly noticeable in regulatory inspections because they can impair the derivation of efficacy or safety results. What matters, therefore, is an unbroken chain from the result via the program and analysis dataset back to the SDTM source.
Full-service CROs such as Mediconomics support ADaM projects through review of the analysis plan for data requirements, specifications for analysis populations and derivations, statistical programming, independent quality control, and the creation of analysis reviewer guides. At database lock, they also coordinate reconciliation between the SDTM delivery, statistical results production, and the tables, figures, and listings of the clinical study report.
Frequently Asked Questions (FAQ)
Can an ADaM dataset be created directly from raw data?
The CDISC standard expects SDTM data as the source. If additional information is required, its origin must be explained in a traceable manner in the ADaM metadata.
Why are ADaM datasets often smaller than SDTM datasets?
An analysis dataset contains the information relevant to a specific endpoint or calculation. It therefore does not need to repeat every collected observation from the SDTM level.
Does ADaM replace the Statistical Analysis Plan?
No. The plan defines the statistical methods, and ADaM implements the data structure required for them. Both artefacts must be consistent in content.
Regulatory References
- CDISC Analysis Data Model (ADaM) – defines the core principles for analysis datasets and metadata.
- CDISC ADaM Implementation Guide – describes standardised structures and variables for implementation.
- FDA Study Data Technical Conformance Guide – requires ADaM datasets to support key efficacy and safety analyses.
- CDISC Define-XML – governs the machine-readable provision of dataset metadata.
