Understanding the ADaM Framework for Data Analysis (Practical Guide for CROs and Sponsors)

While SDTM organizes information “as it is reported/collected” for review, ADaM organizes it “as it is analyzed.” In practice, ADaM translates the intent of the protocol and the Statistical Analysis Plan (SAP) into structures that make it easier to execute analyses without reinterpretation, with standardized derivations and keys that allow traceability back to the origin when someone asks: “Where does this number come from?”

Moreover, ADaM is not a “decorative” standard for the submission package: it is a central component of modern data-driven review. CDISC explicitly states that ADaM is one of the required standards for data submission to the FDA (USA) and the PMDA (Japan).

The “building blocks” of ADaM: ADSL, BDS and OCCDS

The ADaM Implementation Guide (ADaMIG) v1.3 (published by CDISC) defines structures and conventions for building consistent analysis datasets. In particular, it describes two standard structures:

• ADSL (Subject-Level Analysis Dataset): one record per subject with key variables (analysis populations, treatment, reference dates, stratification factors, etc.).
• BDS (Basic Data Structure): a “long” structure parameter-oriented (e.g., changes in continuous variables, repeated assessments by visit/time, etc.).

As a complement, the guide highlights a third structure, OCCDS (Occurrence Data Structure), documented in a specific IG, which is very useful when analysis focuses on events/occurrences (e.g., onset of an adverse event with analysis criteria).

Practical principles that make a difference

• Bidirectional traceability: a reviewer must be able to go from a result to the dataset and from there to the tabulated source. ADaM is specifically designed to preserve this chain of evidence.
• Clear metadata (Define-XML) and narrative (ADRG): datasets without documentation are a black box. The FDA technical guidance refers to definition files and the reviewer’s guide as key elements of the submission package to facilitate review and reproducibility.
• Standardized derivations: baseline, changes, visit windows, imputations, or selection rules must be defined and applied consistently, minimizing “interpretations.” This is part of the spirit of implementation guides and the technical conformance required in submissions.

Practical guide: 6 steps to implement ADaM without surprises

1. Align protocol, SAP, and ADaM from the design stage. Before programming, identify primary/secondary objectives, populations, and data handling rules so that the dataset is “born” ready for analysis.
2. Define a “traceability map.” Specify which SDTM domains feed each ADaM variable and what derivations are applied.
3. Build ADSL as the backbone. ADSL establishes populations and temporal references; if it is well built, the rest of the datasets fit together. ADaMIG v1.3 treats it as a central standard structure.
4. Choose the appropriate structure for each analysis. Repeated measures and parameters → BDS; occurrence-based analyses → OCCDS; always with consistent naming and conventions.
5. Document and validate with a regulatory mindset. Generate Define-XML and ADRG, and apply compliance and consistency checks.
6. Prepare a “reproducible” package. The final goal is for a third party to replicate results with minimal friction: datasets, metadata, reviewer’s guide, and, when applicable, supporting programs within the submission framework.

In summary

ADaM is the piece that transforms clinical data into analyzable, traceable evidence ready for review. Proper adoption means less ambiguity, more efficient analyses, and a stronger regulatory submission. Above all, it means that study results can be defended transparently: from the final figure back to the original data point.

At Sermes, we understand ADaM as a quality system for analysis, not as a final deliverable. By working consistently with a common ADaM format, we have been able to implement automations that make us more efficient without compromising traceability or rigor. We work to ensure that standardization is not a burden, but an accelerator: we combine statistical expertise and technological tools to design ADaM databases ready for analysis, with robust metadata and reproducible TLFs.

If you want us to help you become more efficient without compromising the highest data quality standards, at Sermes we are your trusted partner: we support you throughout the entire journey. In addition, we accompany companies throughout their path, helping ensure that all their studies maintain a homogeneous, solid, and coherent structure, ready for review and for submission to regulatory authorities.