# Structuring clinical trial data for integrity and reuse
A patient in a Phase II oncology trial takes their dose on a Tuesday morning. A nurse records the time, the tumor is measured on a CT scan three weeks later, and a lab flags an elevated liver enzyme. By the time this reaches the FDA, that single event has been transformed into a standardized, audit-trailed row of data that a reviewer in Silver Spring can trace back to the source. That transformation is the invisible engineering behind every drug approval.
This lesson walks through how clinical trial data gets structured for two goals that often pull against each other: integrity (you can trust it and prove where it came from) and reuse (you can pool it, re-analyze it, and submit it without rebuilding everything).
Clinical trials generate messy, high-stakes data. A single Phase II trial might involve 100 to 300 patients across dozens of sites, each collecting hundreds of data points per patient over months.
If every trial invented its own format, the FDA would need to relearn the data for every submission. So since 2016, the FDA has required standardized data formats for most new drug and biologic applications. That standard is CDISC (Clinical Data Interchange Standards Consortium), a non-profit that maintains the data models the industry uses.
You can read the FDA's own requirements in the Study Data Technical Conformance Guide.
Think of these as two layers.
SDTM (Study Data Tabulation Model) is the *organized raw record*. It captures what happened, close to how it was collected, in a standard structure. Minimal interpretation.
ADaM (Analysis Data Model) is the *analysis-ready layer*. It is derived from SDTM and built specifically so a statistician can run the trial's planned analyses, and so a reviewer can reproduce them.
The rule that ties them together: ADaM must be traceable back to SDTM, and SDTM must be traceable back to the raw source (the case report form or the electronic health record). This chain is the backbone of data integrity.
SDTM organizes data into domains, each a two-letter code representing a category of information. A few you will see constantly in an oncology trial:
Each domain is a table. Each row is one observation for one patient. Here is a simplified slice of an AE domain:
STUDYID USUBJID AETERM AESTDTC AESEV AEREL
ONC201 ONC201-0042 NAUSEA 2026-03-14 MODERATE PROBABLE
ONC201 ONC201-0042 ALT INCREASED 2026-03-21 SEVERE PROBABLE
ONC201 ONC201-0117 FATIGUE 2026-03-09 MILD POSSIBLENote the conventions. USUBJID is a unique subject identifier that stays constant across every domain, so you can link this patient's adverse events to their labs and their tumor measurements. AESTDTC is a date in ISO 8601 format (YYYY-MM-DD), a standard that removes ambiguity between US and European date orders. AEREL captures the investigator's judgment on whether the drug likely caused the event.
These are not arbitrary column names. They are defined in the CDISC controlled terminology, so "SEVERE" means the same thing in every trial from every sponsor.
ADaM datasets are built for a specific question. The most common structures are:
The key ADaM concept is the derived variable. Suppose the trial's primary endpoint is Overall Response Rate (the percentage of patients whose tumors shrank by a defined amount). ADaM will contain a variable like AVAL (analysis value) and flags marking which records feed the primary analysis.
Critically, ADaM documents *how* each derived value was calculated. If a responder flag was set, the dataset and its documentation show the underlying tumor measurements from SDTM that justified it. A reviewer can follow the thread from the headline result back to the CT scan.
Integrity is not just clean data. It is *provable* data. Two mechanisms matter.
Every submission includes a machine-readable metadatametadataDonnées sur les données, informations décrivant le contexte, la structure, la provenance et les caractéristiques d'un asset de données (auteur, date, format, source, définition). file called define.xml. Think of it as the data dictionary and mapmapUsing software to automate repetitive marketing tasks and campaigns, enabling personalisation at scale across channels like email, web, and social.Voir la définition complète →. It lists every dataset, every variable, its origin (collected? derived?), and the derivation logic. Reviewers open it first. A weak or inconsistent define.xml is one of the fastest ways to draw regulatory questions.
Data integrity in trials is governed by the ALCOA+ principles. The original ALCOA acronym means data should be Attributable, Legible, Contemporaneous, Original, and Accurate. The "plus" adds Complete, Consistent, Enduring, and Available.
In practice, this is enforced by the electronic systems that capture the data. Under 21 CFR Part 11 (the FDA regulation governing electronic records and signatures), the EDC (Electronic Data Capture) system must log who entered or changed each value, when, and why. If that nurse corrects the dose time from 9:00 to 9:15, the system keeps both values and the reason. Nothing is overwritten silently.
🎬 [VIDEO: "CDISC Standards Explained" — https://www.youtube.com/results?search_query=CDISC+SDTM+ADaM+explained — a plain-language overview of how SDTM and ADaM fit together in a submission]
Standardization is not only about pleasing regulators. It is what makes data reusable.
Because every trial uses the same domains and terminology, a sponsor can pool data across multiple studies to analyze safety signals across a whole drug program. A rare adverse event that appears twice in one trial might appear twenty times when three trials are combined, which changes the safety picture.
Standard structure also feeds:
The payoff is compounding. Structure the data once, correctly, and it serves the submission, the safety database, and future science.
Vérification des acquis
1. The lesson frames clinical trial data structuring as balancing two goals that often pull against each other. What are these two goals?
2. Based on the distinction drawn in the lesson, why is SDTM described as the 'organized raw record' while ADaM is the 'analysis-ready layer'?
3. A company builds a custom, proprietary data format for its Phase II trial submission to the FDA. Based on the lesson's reasoning, what is the core problem with this approach?
4. Select ALL correct answers about the role of data integrity in clinical trial data structuring.
Sélectionnez toutes les réponses correctes.
5. Select ALL correct answers about why structuring clinical trial data is challenging and important.
Sélectionnez toutes les réponses correctes.
Even experienced teams stumble. A few recurring problems worth recognizing:
Traceability breaks. An ADaM value is derived, but nobody can explain from which SDTM records. This is the single most common source of FDA data queries.
Terminology drift. A site records "high blood pressure" while the standard term is "hypertension." Left uncorrected, these fragment the data and hide patterns.
Late standardization. Teams sometimes collect data in a custom format and try to convert it to SDTM at the very end. This is expensive and error-prone. Mature organizations design the collection forms with SDTM in mind from day one, a practice sometimes called "standards-first" design.
Weak define.xml. If the metadatametadataDonnées sur les données, informations décrivant le contexte, la structure, la provenance et les caractéristiques d'un asset de données (auteur, date, format, source, définition). does not match the actual datasets, reviewers lose trust quickly, and lost trust means more scrutiny of everything.
You do not need to write SDTM to lead in this space. But you should be able to ask the right questions:
Those four questions separate teams that treat data as a compliance chore from teams that treat it as a durable asset.