The short version
- ADaM is built around traceability: a reviewer should be able to follow an analysis result back through the analysis data to the tabulation data.
- Traceability has two parts: metadata that explains how a value was derived, and data that point to the records it came from.
- Pick three numbers from your key tables each week and trace them yourself. If it takes you more than a few minutes, it will take a reviewer longer.
What the standard asks
Traceability is one of the fundamental principles of ADaM: analysis datasets and their metadata should make clear how each analysis value was obtained from the data it came from [1]. The implementation guide distinguishes two levels. Metadata traceability is the description of the derivation in the define.xml and the reviewer's guide. Data point traceability points from a specific analysis record to the specific source records, for example through the source domain, variable and sequence variables SRCDOM, SRCVAR and SRCSEQ, or by keeping SDTM sequence numbers in the analysis dataset [2].
FDA's Technical Conformance Guide asks for analysis data that can be traced to the tabulation data they were derived from, and singles out traceability as a particular problem when legacy data are converted to the standards after the fact [3].
The five-step test
Take one number from a key results table, for example the count of responders at week 12 in one arm. Then:
- Output to program. Can you identify the program and the analysis dataset that produced it, from the output itself or its footnotes?
- Output to analysis records. Can you select exactly the records that make up that number, using the population flag, parameter, visit and analysis flags named in the define?
- Analysis record to derivation. For one of those records, does the define or reviewer's guide explain how the analysis value was derived, in words a reviewer can follow?
- Analysis record to tabulation record. Can you find the SDTM record or records it came from, using source variables, sequence numbers or a documented key?
- Tabulation record to collection. Does the annotated CRF, or the external data documentation, show where that value was collected?
Each step that needs a programmer to explain it is a question a reviewer will ask in writing.
Where it usually breaks
| Break | Typical cause | Fix |
|---|---|---|
| Step 2 | The table applies a selection that no flag records | Add the flag in ADaM, not in the table program |
| Step 3 | Method says "see program" | Write the derivation in words in the define |
| Step 4 | Values derived from several records, with no pointer | Keep source sequence numbers or create intermediate records |
| Step 5 | External vendor data with no documentation | Describe the source and transfer in the reviewer's guide |
A synthetic example
The fictional ZEPHYR-101 team traces three numbers at its dry run. The responder count passes all five steps. A laboratory shift table fails at step 2, because the table program excludes unscheduled visits with no flag in the dataset. A time-to-event table fails at step 4: the censoring date is taken from whichever of four dates is latest, with no record of which one. Two new variables close both gaps in an afternoon.
Do this today
- Choose three numbers from your primary efficacy and key safety tables and run the five-step test on each.
- Move any record selection that lives only in a table program into an analysis flag.
- Replace every "see program" method in your define with a derivation a reviewer can read.
Check yourself
One question
Show the answer
Answer 2. SRCDOM, SRCVAR and SRCSEQ identify the source domain, variable and sequence number of the record a value came from. Subject and visit narrow the search, but they do not identify the record.
Apply it
Turn it into a plan
The brief says what to check. The Navigator and the Handbook turn it into dated, owned actions for your milestone, in your browser, with no study data.
Sources
References
- CDISC. Analysis Data Model (ADaM) version 2.1. www.cdisc.org ↗
- CDISC. Analysis Data Model Implementation Guide (ADaMIG) version 1.3. www.cdisc.org ↗
- U.S. Food and Drug Administration. Study Data Technical Conformance Guide: Technical Specifications Document. www.fda.gov ↗
Standards and guidance change. Confirm the version in force for your study, sponsor and region before relying on any reference. Examples are synthetic and illustrative. This brief is generalized professional guidance, not regulatory advice.
Next brief · Thursday 10:00 ET · Capacity Math