ARD/ARS and architecture of parameterizable reporting

[hoppanda]

interoperable approaches to analysis artifacts. Common structure for data collection, to make it easy to build tools against the data. Novo/Merck/Roche have looked at this in HTA, pharmaverse in full development

[epijim]

Can we cover this as a broad topic across reporting - so impact

• pharmaverse
• HTA
• statistics (e.g. handling outputs from simulations)

[epijim]

would suggest @orladoylenvs for where we are in trials (she built tools at Novartis using ARS/ARD) and is on the pharmaverse council

[epijim]

Also impact on AI (e.g. write CSR from an ARD, rather than a rendered table?) and/or component based authoring - e.g. a statistic appearing in 5 places within the document

[hoppanda]

Session 1: The Analysis Results Standard Revolution — From Static Tables to Living, Machine-Readable Submissions

The Core Narrative

We've spent decades rendering statistics into fixed-format tables locked in PDFs and SAS transport files. What if the "result" itself—not its rendered form—became the unit of exchange between sponsors, regulators, and HTA bodies?

Why This Matters Now

• CDISC's ARD/ARS standards are maturing (albeit slowly), moving from SAS V5 transport to JSON, Parquet, and language-agnostic formats.
• AI/LLM explosion means machine-digestible results aren't a "nice to have"—they're the prerequisite for AI-assisted regulatory review, automated CSR authoring, and component-based document assembly.
• HTA bodies increasingly demand different slices/views of the same data—parameterizable reporting from a single source of truth is the answer.
• Open-source tooling in R (e.g., pharmaverse packages, Novartis's ARS/ARD tools) is already prototyping this future.

Possible Session Structures (Options for Committee Discussion)

Option A: "The Full Stack" — End-to-End Vision

Segment	Content	Potential Speaker Profile
**Opening **	The problem statement: Why static rendered tables are a bottleneck for regulators, HTA, and AI	Senior biostatistics leader (Head of Dept level)
**Standards Deep-Dive **	ARD/ARS: What it is, where CDISC is, what's left to solve	??
Tooling in Practice	Building ARD-native pipelines in R—demo of tools (e.g., Novartis ARS/ARD implementation in pharmaverse)	Practitioner/developer (e.g., Orla Doyle or equivalent)
The AI Angle	From ARD to CSR narrative generation; component-based authoring where one statistic populates 5 document locations	??
Discussion/Q&A	Regulatory & HTA perspectives: What would reviewers do with machine-readable results?	??

Option B: "The Debate" — Is the Industry Ready?

Frame the session as a structured debate:

1. Proposition: "Within 5 years, no regulatory submission should contain a rendered table as the primary data carrier."
2. For: Efficiency gains, AI enablement, reusability, error reduction
3. Against: Regulatory inertia, validation burden, global harmonization gaps, reviewer readiness
4. Audience vote before and after

Option C: "The Collaboration Blueprint"

Focus on what open-source collaboration between sponsors and regulators could look like:

• Live demo of a prototype where a regulator queries an ARD to answer a review question
• GSK/Novartis/Roche showing converging approaches
• Call-to-action for a cross-industry working group

[orladoylenvs]

Maybe a blend of A and B could be interesting? Like the today and tomorrow perspective?
I had these points in a related issue. I think the submission angle is core here: how do we ensure concepts like ARS that enable pharma to automate and use AI also enable better submissions, review, etc..
The other angle we might want to cover is the knowledge management use case but perhaps that's already a lot to cover in one session :)

Ps I'm happy to give a demo if that's what we align on

[shannonhaughton]

I think that {gtsummary} would be a great talk for the tooling in practice section from Daniel Sjoberg. gtsummary has both ard first and ard second functionality. Daniel's {cards} and {cardx} packages are also useful tools - appreciate that he has presented on similar at main R/Pharma before though

[philbowsher]

https://www.danieldsjoberg.com/gtsummary/articles/tbl_ard-functions.html

[shannonhaughton]

From CDISC: Someone from https://www.cdisc.org/standards/cdisc-360i or https://www.cdisc.org/core? A name that came up was Bhavin Busa

[hoppanda]

Session 1: The Analysis Results Standard Revolution — From Static Tables to Living, Machine-Readable Submissions

The Core Narrative

We've spent decades rendering statistics into fixed-format tables locked in PDFs and SAS transport files. What if the "result" itself—not its rendered form—became the unit of exchange between internal stakeholders, sponsors, regulators, and HTA bodies?

Why This Matters Now

• CDISC's ARD/ARS standards are maturing (albeit slowly), moving from SAS V5 transport to JSON, Parquet, and language-agnostic formats.
• AI/LLM explosion means machine-digestible results aren't a "nice to have"—they're the prerequisite for AI-assisted regulatory review, automated CSR authoring, and component-based document assembly.
• HTA bodies increasingly demand different slices/views of the same data—parameterizable reporting from a single source of truth is the answer.
• Open-source tooling in R (e.g., pharmaverse packages, Novartis's ARS/ARD tools, {gtsummary}/{cards}/{cardx}) is already prototyping this future.

Possible Session Structures (Options for Committee Discussion)

Option A+B Hybrid: "Today and Tomorrow" — Current Reality Meets Future Vision (PREFERRED DIRECTION)

Per Orla's suggestion, blend the end-to-end vision (Option A) with the provocative forward-looking framing (Option B). Structure the session as a journey from where we are → where we're going → what's blocking us → what it enables.

Segment	Content	Potential Speaker Profile
Opening: The Problem & The Promise	Why static rendered tables are a bottleneck for regulators, HTA, and AI. Frame the "today vs. tomorrow" tension.	Senior biostatistics leader (Head of Dept level)
Standards Deep-Dive	ARD/ARS: What it is, where CDISC is, what's left to solve. Connection to CDISC 360i and CDISC CORE initiatives.	CDISC representative — Bhavin Busa (suggested name, from CDISC 360i/CORE)
Tooling in Practice	Building ARD-native pipelines in R — demo of tools. {gtsummary} with ARD-first and ARD-second functionality; {cards} and {cardx} packages. Novartis ARS/ARD implementation in pharmaverse.	Orla Doyle (confirmed willing to demo) and/or Daniel Sjoberg ({gtsummary}/{cards}/{cardx}) — Note: Daniel has presented on similar at main R/Pharma before; consider whether this is sufficiently differentiated or if he covers the "submission angle" specifically
The Submission & Review Angle	How do concepts like ARS that enable pharma to automate and use AI also enable better submissions, review, and regulatory interaction? What would a reviewer actually do with machine-readable results?	Regulatory perspective speaker (FDA/EMA) or cross-industry submission expert
The AI Angle	From ARD to CSR narrative generation; component-based authoring where one statistic populates 5 document locations	? Orla
Provocative Close + Discussion	Debate prompt: "Within 5 years, no regulatory submission should contain a rendered table as the primary data carrier." Audience pulse check. Open Q&A.	Moderator / all panelists

Key Narrative Arc for the Hybrid

1. Today: We have tools ({gtsummary}, pharmaverse) that already produce ARDs — the building blocks exist
2. The Standard: CDISC is formalizing this (ARS, 360i, CORE) — but progress is slow and adoption is uneven
3. Tomorrow — Submission: ARD-native submissions enable regulators to query results directly, not reverse-engineer PDFs
4. Tomorrow — AI: Machine-readable results feed CSR authoring, component-based documents, AI-assisted review
5. The Provocation: Are we ready? Is the industry? Are regulators?

Option C: "The Collaboration Blueprint" (RETAINED AS ALTERNATIVE)

Focus on what open-source collaboration between sponsors and regulators could look like:

• Live demo of a prototype where a regulator queries an ARD to answer a review question
• GSK/Novartis/Roche showing converging approaches
• Call-to-action for a cross-industry working group

Scoping Decision: Knowledge Management Use Case

Orla raised the knowledge management angle (e.g., reusing ARD-structured results across studies, building searchable result repositories, enabling cross-study AI queries). This is a compelling extension. To discuss if we squeeze this in somewhere, replace a proposal or round table topic in the afternoon as an continuation?

Options:

• Include briefly as a "what else this enables" teaser in the closing segment
• Exclude and flag as a potential standalone talk or future session topic
• Include as a full segment if we have time and the right speaker

Committee to decide on scope.

Key Questions for Committee Discussion

1. Hybrid A+B direction: Does the "Today and Tomorrow" arc work? Any concerns about trying to cover too much ground?
2. Submission angle as core: Orla suggested that the submission/review enablement should be central, not just an afterthought. Do we all agree that this is the differentiating angle vs. a "tooling" talk?
3. Daniel Sjoberg: He's presented on {gtsummary} at R/Pharma before. Not sure about the details.
4. Bhavin Busa / CDISC: Need to confirm availability and willingness. Who can reach out?
5. Knowledge management: how to position?
6. Provocation level: The debate framing ("no rendered tables in 5 years") — do we use this as the session's through-line or just a closing discussion prompt?

Confirmed / Potential Speaker Status

Name	Role / Contribution	Status
Orla Doyle	Demo of ARS/ARD tooling in practice (Novartis/pharmaverse)	✅ Willing to present
Daniel Sjoberg	{gtsummary}/{cards}/{cardx} — ARD-first/second functionality	🟡 Suggested — need to assess differentiation from prior talks and confirm interest
Bhavin Busa	CDISC standards perspective (360i / CORE)	🟡 Suggested — need outreach
Senior biostatistics leader	Opening framing / problem statement	🔴 TBD — need to identify
??	??	🔴 TBD — need to identify

[philbowsher]

Nice view of TLG packages supporting ARD here https://pharmaverse.org/e2eclinical/tlg/

[hoppanda]

HTA variation of ARD/ARS: https://github.com/[hta-pharma/ramnog](https://github.com/hta-pharma/ramnog)

[hoppanda]

Suggestion:

Orla: Today + Standard and Tooling in Regulatory
Greg: Standard and Tooling in HTA/Market Access + Tomorrow
Together: Provocation + Discussion

[orladoylenvs]

Some ideas to discuss:
The automation pull: pharma has strong internal reasons to move toward structured, machine-readable outputs. Tooling pharmaverse is already available.
The standard as formalisation: ARS isn't a regulator-driven mandate, it's industry formalising its own direction.
The submission process: machine-readable TFLs could enable / evolve review: direct querying, reproducibility, AI-assisted analysis.
Is there a consumption hurdle: but regulators receiving richer data isn't the same as regulators being able to use it. What does it actually take on their side? Can AI simplify this hurdle?
The shared question: how do we design a transition that creates real value for both sides, rather than transferring burden in the name of progress?

[orladoylenvs]

Revised ideas with feedback from the team that providing an overview of ARS, how it is structure and used and general demystification would be highly value.

• ARS intro and examples (15 mins)
• demo of ARS-driven automation: Clinical Study Report use case(10 mins)
• use case for market access (10 mins)
• Discussion prompts, Q&A and wrap up (10 mins)