Trends & Insights

Key themes and future directions from R/Pharma 2025

Executive Summary

R/Pharma 2025 marked a pivotal moment in pharmaceutical R adoption, with three dominant themes emerging: AI/LLM integration, open-source transformation, and regulatory acceptance. The conference showcased 19 workshops and 30+ presentations revealing an industry in rapid transition from proprietary tools to collaborative, standards-based approaches.

Key Statistics:

🤖 8+ sessions on AI/LLM (50% increase from previous years)
📊 GSK achieved 50%+ R code across Biostatistics
✅ FDA neutrality on statistical software increasingly normalized
🌐 CDISC ARS/ARM driving automation and standardization

1. The AI/LLM Revolution 🤖

Emergence as Dominant Theme

AI and Large Language Models were the most discussed topic, appearing in:

3 dedicated workshops (ellmer basics, enterprise tooling, clinical data privacy)
8+ presentations across all sessions
Multiple vendor tools ({ellmer}, {llumen}, {mcpr}, Databot, {meRlin})

From Prototype to Production

The conversation has matured from “Can we use AI?” to “How do we deploy it safely?”

Key Developments:

1. Enterprise-Ready Frameworks

Organizations are building production AI systems:

Merck’s {llumen} - Comprehensive agentic framework used internally
- Supports multiple LLM sources (Azure, OpenAI, local)
- RAG with vector databases for documents
- Database querying (SQL, Cypher, GraphQL)
- Foundation model integration (TxGemma)
Roche’s Multi-Agent Copilot - Package-specific AI agents
- Each internal package gets its own expert agent
- LangGraph orchestration
- Integration with Cursor via MCP
A2-AI’s GxP Solutions - Validated AI workflows
- AWS Bedrock integration
- MCP server implementations
- Compliance documentation

2. Privacy-Preserving Approaches

Critical for pharma with sensitive clinical data:

On-premise LLM deployments (llama.cpp)
Query sanitization removing PII
Audit logging for compliance
Role-based access controls
{DataChat} example - Chat interface with data never leaving secure environment

3. Practical Applications

Moving beyond novelty to genuine productivity:

Data Exploration:

Natural language queries on CDISC data
Conversational interfaces for non-programmers
RAG-powered document search

Code Assistance:

Context-aware programming help
Debugging complex statistical code
Documentation generation

Report Automation:

LLM-powered table summarization
Narrative generation from results
QC log generation

Standards Emerging

Model Context Protocol (MCP):

Standardized interface for AI tools
R implementation via {mcpr}
Cross-language interoperability
Growing ecosystem (Claude Code, Cursor, VS Code)

Challenges Identified

Despite enthusiasm, several barriers remain:

⚠️ Validation complexity - How to validate non-deterministic outputs
⚠️ Cost management - API costs at scale
⚠️ Regulatory uncertainty - Limited guidance on AI in submissions
⚠️ Skills gap - Prompt engineering and AI orchestration expertise

2. Open-Source Transformation 🌐

GSK’s Landmark Achievement

Sam Warden’s presentation on GSK’s journey was a conference highlight:

Timeline:

2018-2020: Pilot programs and early adopters
2020-2022: COVID acceleration and platform adoption
2023-2024: 50%+ R code target achieved
2025: Rburst initiative for full integration

Success Factors:

Executive commitment - Clear target setting (50%+ R)
Multi-wave approach - Gradual transformation
Training investment - Bookdown courses, Resource Hub, AccelerateR
Platform deployment - Posit Workbench for infrastructure
Cultural change - Growth mindset and adaptability

Industry-Wide Movement

GSK is not alone - the tipping point has been reached:

Novartis:

Mosaic platform for ARS-driven TFL automation
Standards-based, language-agnostic approach
React UI with R backend

Roche/Genentech:

Leading {admiral}, {teal}, {gtsummary} development
autoslideR for slide automation
{crane} for pharma-specific reporting

Pfizer:

Custom R packages for RWD programming
Dual R/SAS syntax support
Quarto documentation sites

Moderna:

AI-enhanced Shiny apps for trial data
ellmer/GPT integration
Accelerating exploratory analysis

Pharmaverse Ecosystem

The collaborative open-source movement is thriving:

Key Packages:

{admiral} - Transitioning to stable maintenance (feature complete)
{teal} - Version 1.0 released (interactive apps)
{gtsummary}/{crane} - Clinical table generation
{cardinal} - Standardized TLG templates
{sdtm.oak} - SDTM programming framework

Benefits Realized:

✅ Reduced duplication across companies
✅ Shared validation burden
✅ Faster innovation cycles
✅ Transparent, auditable code

3. Regulatory Evolution ✅

FDA Software Neutrality

The 2015 FDA clarification on software neutrality is now fully operationalized:

No preference for SAS vs R vs Python
Focus on methodology and validation, not tools
R-based submissions increasingly routine

New FDA Guidance Impact

2023 Covariate Adjustment Guidance driving change:

Novartis developed {beeca} in response
ASA-BIOP collaboration on {RobinCar2}
Academic methods reaching practice faster
R enabling rapid response to guidance

Validation Maturity

Key Developments:

1. Risk-Based Approaches

{riskmetric} for package assessment
Shared metric repositories (R Validation Hub)
Automated quality scoring
Focus on critical packages

2. Acceptance-Test Driven Development (ATDD)

Plain-language tests in Quarto
Shareable with regulators
“Given-when-then” format
Brian Repko bringing JBehave concepts to R

3. Shiny App Validation

Litmusverse suite (Jumping Rivers)
Risk-based validation frameworks
Code quality assessment
Traceability and documentation tools

GxP-Ready AI

Pioneering work on validating AI applications:

Devin Pastoor (A2-AI): Production AI in GxP contexts
Testing strategies for non-deterministic outputs
Audit trails and logging
Version control and rollback procedures

4. Automation & Efficiency 📊

ARS/ARM as Game Changers

CDISC Analysis Results Standard driving automation:

Mosaic (Novartis):

YAML captures ARD requirements
Language-agnostic rules (metadata → R or any language)
React UI for customization
Push-button TFL generation

{cards} Ecosystem:

Analysis Results Data objects
QC becomes straightforward (compare ARDs)
LLMs can summarize language-agnostic results
{crane} and {gtsummary} building on this foundation

Report Automation

Time Savings Achieved:

autoslideR (Roche): 0.5-4 days per slide deck
DMC materials automation: Week → Day for exposure reports (AstraZeneca)
LLM-powered {gtsummary}: Submission-ready tables in minutes vs hours

Technologies Enabling This:

officer/flextable - Programmatic Word reports
Quarto - Multi-format publishing (PDF, HTML, presentations)
Template systems - Reusable structures (Cardinal)
CI/CD pipelines - Automated reanalysis on code changes

5. Advanced Analytics & Methods 📈

Bayesian Methods Maturing

Tools Reaching Production:

BayesERtools (Genentech) - Exposure-response with Stan
bmstate (Generable) - Multistate survival models
Stan debugging workshop - Making Bayesian accessible

Advantages in Pharma:

Prior information integration
Uncertainty quantification
Small sample performance
Complex model flexibility

High-Performance Computing

Polars for Clinical Data:

10-100x faster than pandas
Apache Arrow native
Lazy evaluation
Parallel processing out-of-the-box

HPC Integration:

Offloading Shiny computations to clusters
Maintaining interactive UX
Resource optimization

Machine Learning

TabPFN - Novel deep learning for tabular data:

No training required (pre-trained on priors)
Fast inference
Bayesian-like uncertainty
Promising for exploratory analysis

Synthetic Data:

{synthpop} for privacy-preserving data sharing
Software testing without PHI
Model training on synthetic patients

6. Data Quality & Validation 🔍

{pointblank} Adoption

Comprehensive data validation framework:

Use Cases:

Quick dataset understanding (scan_data())
Validation at scale (35+ tables daily)
Beautiful automated documentation
Integration with databases, Arrow, Shiny

Pharmaceutical Applications:

SDTM compliance checks
ADaM dataset verification
Cross-domain consistency
Longitudinal data integrity

Automated Traceability

DevOps Principles in Pharma:

Graticule’s approach:

Docker containers for reproducibility
CI/CD for automatic reanalysis
Version control with validated outputs
Cloud storage (AWS S3) for results

Benefits:

Clear audit trail
Reproducible analyses
Automated documentation
Integrated QC

7. Specialized Applications 🧬

Real-World Evidence

Pfizer’s RWD Programming:

Custom R package for database queries
Dual R/SAS syntax support
Shiny apps for workflow support
Quarto documentation

Niche Medical Devices

Abbott’s Synthetic Data:

{synthpop} for diagnostics
Privacy-preserving test data
Validation datasets

Oncology Innovations

PMDA Inquiries:

{cards} for rapid response (Japan regulatory)
Structured ARD approach
Accelerated turnaround

Pharmacokinetics

aNCA (Pharmaverse):

Open-source NCA software
PKNCA backend (200+ parameters)
100% test coverage
Validated against commercial tools (±0.1%)

8. Regional Insights 🌏

Asia/Pacific Innovations

Strengths:

Strong PMDA regulatory focus
Practical automation tools
Under-represented population research
Genetic diversity considerations

Notable Presentations:

SHIONOGI’s open-source culture
{meRlin} AI assistant
PMDA e-CRT automation
Vibe Coding approaches

Global Collaboration

Conference demonstrated truly global R pharma community:

Workshops from US, Europe, Asia
Shared challenges and solutions
Pharmaverse as unifying force

9. Future Directions 🔮

Short-Term (2025-2026)

AI Integration:

✅ More validated AI applications in production
✅ MCP ecosystem expansion
✅ Regulatory guidance on AI in submissions
✅ Cost-effective on-premise LLM solutions

Automation:

✅ ARS/ARM adoption across industry
✅ Push-button TFL generation becomes standard
✅ Automated slide decks and CSRs routine

Validation:

✅ Shared metric repositories operational
✅ Risk-based validation industry standard
✅ Shiny app validation frameworks mature

Medium-Term (2027-2028)

Tool Consolidation:

Language-agnostic analysis platforms
R/Python/Julia interoperability
Cloud-native pharma analytics

Regulatory:

AI-specific guidance documents
Electronic submissions with code
Real-time regulatory review platforms

Skills:

AI/LLM literacy standard for programmers
Bayesian methods more accessible
DevOps practices ubiquitous

Long-Term (2029+)

Transformative Possibilities:

AI-Designed Clinical Trials
- LLMs suggesting optimal designs
- Automated SAP generation
- Real-time adaptation
Fully Automated TFLs
- From data lock to submission package
- Human review only, no coding
- Multi-language outputs for global submissions
Real-Time Safety Monitoring
- Continuous analysis during trials
- AI-detected signals
- Predictive adverse event modeling
Personalized Trial Design
- Bayesian adaptive with real-time learning
- Biomarker-driven enrollment
- Precision medicine integration

10. Challenges & Barriers ⚠️

Technical Challenges

Validation Complexity:

Non-deterministic AI outputs
Rapidly evolving tools
Keeping pace with innovation

Integration Issues:

Legacy system compatibility
Data silos
IT security constraints

Organizational Challenges

Cultural Resistance:

“SAS is what we’ve always used”
Fear of change
Risk aversion in regulated environment

Resource Constraints:

Training investment required
Dual maintenance (SAS + R) during transition
Validation documentation burden

Regulatory Challenges

Uncertainty:

Limited AI/LLM guidance
Evolving expectations
Regional variations (FDA vs EMA vs PMDA)

Documentation:

What level of detail for AI systems?
How to handle model updates?
Third-party API dependencies

11. Success Patterns 🎯

What Works

Based on successful implementations presented:

1. Executive Sponsorship

Clear targets (GSK’s 50%)
Resource commitment
Long-term vision

2. Gradual Transformation

Multi-wave approach
Pilot programs
Learn and adapt

3. Support Infrastructure

Training programs (not just one-time)
Help desks and office hours
Documentation and resources

4. Community Building

Internal R user groups
External pharmaverse participation
Shared learning culture

5. Platform Investment

Posit Workbench
Version control (Git/GitHub)
CI/CD pipelines
Cloud infrastructure

6. Standards Adoption

CDISC ARS/ARM
Pharmaverse conventions
Style guides and linters

12. Actionable Insights 💡

For Organizations Starting R Adoption

Start with low-risk projects (exploratory analysis, visualizations)
Invest in training early and continuously
Build internal community (lunch & learns, Slack channels)
Adopt pharmaverse packages (don’t reinvent)
Establish validation framework from day one
Use Posit products (Workbench, Connect, Package Manager)

For Organizations Scaling R

Embrace automation (ARS/ARM, template systems)
Implement CI/CD for reproducibility
Explore AI carefully (privacy first, validate thoroughly)
Contribute to pharmaverse (share internal packages)
Formalize support model (beyond training)
Plan SAS sunset (don’t maintain dual indefinitely)

For R Professionals

Learn AI/LLM basics (will be essential skill)
Master one reporting framework (officer/flextable or Quarto)
Understand validation (not just coding)
Contribute to open source (career differentiator)
Stay current (R/Pharma, webinars, blogs)
Network (pharmaverse Slack, conferences)

13. The Bigger Picture 🌍

R/Pharma 2025 in Context

This conference revealed an industry at an inflection point:

From:

Proprietary, siloed tools
Manual, repetitive coding
Conservative, risk-averse culture
Isolated company solutions

To:

Open-source, collaborative development
Automated, standards-driven workflows
Innovative, evidence-based adoption
Shared industry platforms

Why This Matters

For Patients:

Faster drug development
More rigorous analysis
Transparent, reproducible science
Precision medicine enablement

For Industry:

Reduced costs
Faster time to market
Better talent recruitment
Competitive advantage

For Science:

Reproducibility crisis addressed
Method innovation accelerated
Global collaboration enabled
Democratized advanced analytics

14. How R/Pharma Has Evolved 📈

Conference Evolution (2018-2025)

Based on publicly available information and this year’s content, here’s how the conference themes have shifted:

2018-2020: Foundation Years

Key Themes:

📦 Package development basics - Building pharma packages
📊 Basic Shiny apps - Interactive visualizations
🔄 SAS vs R debates - “Should we switch?”
📝 Simple reporting - Basic TFLs

Sentiment: Cautious optimism, early adopters sharing success stories

Notable:

Pharmaverse not yet established
Validation was major concern
FDA software neutrality clarification (2015) still fresh

2021-2022: Acceleration Phase

Key Themes:

🚀 COVID acceleration - Remote work drove R adoption
🤝 Pharmaverse launch - Collaborative ecosystem born
✅ Validation frameworks - {riskmetric}, R Validation Hub
📦 Production packages - {admiral}, {teal} gaining traction

Sentiment: Growing confidence, momentum building

Milestones:

Major pharma (Roche, Novartis) sharing production solutions
Regulatory submissions with R increasing
Training programs formalized (GSK example)

2023-2024: Mainstream Adoption

Key Themes:

📊 CDISC integration - Analysis Results Standard (ARS/ARM)
🔧 Automation focus - Template-based TFL generation
📈 Advanced analytics - Bayesian methods, ML
🌐 Multi-language - R + Python workflows

Sentiment: R is mainstream, focus shifts to optimization

Evidence:

Multiple talks on automation (not just feasibility)
Validation becoming standardized, less controversial
Advanced topics (multistate models, Stan) gaining space

2025: AI Revolution Year

Dominant Themes:

🤖 AI/LLM explosion - 8+ sessions (50% increase)
🏢 Enterprise transformation - GSK’s 50%+ achievement
✅ Regulatory maturity - GxP-ready AI, validated Shiny
🚀 Production at scale - From “can we?” to “how fast?”

Shift in Discourse:

2018: “Is R viable for pharma?”
2022: “How do we migrate from SAS?”
2025: “How do we integrate AI with R?”

What’s New in 2025:

AI Dominance
- 2018-2022: Barely mentioned
- 2023: Experimental projects
- 2024: Pilot implementations
- 2025: Production systems ({llumen}, multi-agent copilots, enterprise frameworks)
Regulatory Confidence
- 2018-2020: “Will FDA accept this?”
- 2021-2022: “Here’s how we validated”
- 2023-2024: “Validation frameworks established”
- 2025: “Validated AI in GxP” - previously unthinkable
Speed of Innovation
- 2018: Yearly package updates
- 2022: Quarterly releases (pharmaverse)
- 2025: Real-time AI integration - tools released months ago
Collaboration Level
- 2018: Individual company solutions
- 2020: Pharmaverse begins
- 2023: Cross-company working groups
- 2025: Shared AI infrastructure (MCP, metric repos)

Topic Frequency Evolution

Topic	2020	2022	2024	2025
Package Development	🔥🔥🔥	🔥🔥	🔥	🔥
Validation	🔥🔥🔥	🔥🔥🔥	🔥🔥	🔥
Shiny Apps	🔥🔥	🔥🔥🔥	🔥🔥	🔥🔥
Clinical Reporting	🔥🔥	🔥🔥	🔥🔥🔥	🔥🔥🔥
Bayesian Methods	🔥	🔥	🔥🔥	🔥🔥
AI/LLM	-	-	🔥	🔥🔥🔥🔥
Automation	🔥	🔥🔥	🔥🔥	🔥🔥🔥
Python Integration	🔥	🔥	🔥🔥	🔥🔥

Interpretation:

Traditional topics (package dev, validation) declining as they become solved problems
AI/LLM emerged from nowhere to dominate the conversation
Automation shifted from “nice to have” to strategic imperative
Python integration normalized (not R vs Python, but R and Python)

What Changed Between 2024 and 2025?

Major Shifts:

{ellmer} Package Launch (2024)
- Made LLM integration trivial in R
- Multiple workshops/presentations built on it
- Enterprise adoption accelerated
Claude/ChatGPT API Maturity
- Function calling standardized
- MCP protocol emerged
- Privacy-preserving deployments viable
GSK Milestone
- 50%+ R code achieved
- Proof that full transformation possible
- Other pharma following suit
CDISC ARS Adoption
- Analysis Results Standard gaining traction
- Mosaic (Novartis) and others automating TFLs
- Machine-readable results enabling AI
Posit Product Evolution
- Positron IDE launched (VS Code + R)
- Better Python support
- AI assistants integrated

What Stayed Constant:

✅ Pharmaverse continues thriving
✅ Validation remains priority (but less controversial)
✅ Collaboration over competition
✅ Open-source commitment

Predictions for R/Pharma 2026

Based on the trajectory, expect:

Likely Topics:

“AI in Submissions” - First AI-powered analysis in regulatory filing
“Real-time Clinical Trial Analysis” - Adaptive designs with AI
“Synthetic Patient Generation” - Privacy-preserving AI for trials
“Quantum ML for Drug Discovery” - Cutting edge
“Fully Automated Study Reports” - From data lock to CSR in hours

Maturing Topics:

LLM integration will be assumed (not taught from scratch)
Validation frameworks standardized across industry
Python-R seamless interop (not separate tracks)

Declining Topics:

“Intro to R packages” (basics widely known)
“Why leave SAS?” (decision already made)
“Can we validate R?” (settled question)

Bold Prediction:

R/Pharma 2026 will feature the first fully AI-generated clinical study report accepted by a regulatory agency. The discussion will shift from “Is this possible?” to “What should humans still review?”

15. Conclusion

R/Pharma 2025 demonstrated that open-source pharmaceutical analytics is not the future it’s the present. With AI integration, regulatory acceptance, and enterprise adoption converging, the question is no longer “Should we adopt R?” but “How fast can we transform?”

The conference showcased:

✅ Production-ready AI systems
✅ Validated, GxP-compliant workflows
✅ Major pharma success stories (GSK 50%+)
✅ Thriving ecosystem (pharmaverse)
✅ Regulatory confidence (FDA neutrality)

The momentum is undeniable. The future is open source. The time is now.

Stay Connected

R/Pharma website: rinpharma.com
Pharmaverse: pharmaverse.org
Posit: posit.co
R Validation Hub: pharmar.org

Analysis compiled from R/Pharma 2025 Conference materials | Last updated: November 2025

--- title: "Trends & Insights" subtitle: "Key themes and future directions from R/Pharma 2025" toc: true --- ## Executive Summary R/Pharma 2025 marked a pivotal moment in pharmaceutical R adoption, with three dominant themes emerging: **AI/LLM integration**, **open-source transformation**, and **regulatory acceptance**. The conference showcased 19 workshops and 30+ presentations revealing an industry in rapid transition from proprietary tools to collaborative, standards-based approaches. **Key Statistics:** - 🤖 **8+ sessions** on AI/LLM (50% increase from previous years) - 📊 **GSK achieved 50%+ R code** across Biostatistics - ✅ **FDA neutrality** on statistical software increasingly normalized - 🌐 **CDISC ARS/ARM** driving automation and standardization ------------------------------------------------------------------------ ## 1. The AI/LLM Revolution 🤖 ### Emergence as Dominant Theme AI and Large Language Models were the most discussed topic, appearing in: - **3 dedicated workshops** (ellmer basics, enterprise tooling, clinical data privacy) - **8+ presentations** across all sessions - **Multiple vendor tools** ({ellmer}, {llumen}, {mcpr}, Databot, {meRlin}) ### From Prototype to Production The conversation has matured from "Can we use AI?" to "How do we deploy it safely?" **Key Developments:** #### 1. **Enterprise-Ready Frameworks** Organizations are building production AI systems: - **Merck's {llumen}** - Comprehensive agentic framework used internally - Supports multiple LLM sources (Azure, OpenAI, local) - RAG with vector databases for documents - Database querying (SQL, Cypher, GraphQL) - Foundation model integration (TxGemma) - **Roche's Multi-Agent Copilot** - Package-specific AI agents - Each internal package gets its own expert agent - LangGraph orchestration - Integration with Cursor via MCP - **A2-AI's GxP Solutions** - Validated AI workflows - AWS Bedrock integration - MCP server implementations - Compliance documentation #### 2. **Privacy-Preserving Approaches** Critical for pharma with sensitive clinical data: - **On-premise LLM deployments** (llama.cpp) - **Query sanitization** removing PII - **Audit logging** for compliance - **Role-based access** controls - **{DataChat} example** - Chat interface with data never leaving secure environment #### 3. **Practical Applications** Moving beyond novelty to genuine productivity: **Data Exploration:** - Natural language queries on CDISC data - Conversational interfaces for non-programmers - RAG-powered document search **Code Assistance:** - Context-aware programming help - Debugging complex statistical code - Documentation generation **Report Automation:** - LLM-powered table summarization - Narrative generation from results - QC log generation ### Standards Emerging **Model Context Protocol (MCP):** - Standardized interface for AI tools - R implementation via {mcpr} - Cross-language interoperability - Growing ecosystem (Claude Code, Cursor, VS Code) ### Challenges Identified Despite enthusiasm, several barriers remain: - ⚠️ **Validation complexity** - How to validate non-deterministic outputs - ⚠️ **Cost management** - API costs at scale - ⚠️ **Regulatory uncertainty** - Limited guidance on AI in submissions - ⚠️ **Skills gap** - Prompt engineering and AI orchestration expertise ------------------------------------------------------------------------ ## 2. Open-Source Transformation 🌐 ### GSK's Landmark Achievement **Sam Warden's** presentation on GSK's journey was a conference highlight: **Timeline:** - **2018-2020:** Pilot programs and early adopters - **2020-2022:** COVID acceleration and platform adoption - **2023-2024:** 50%+ R code target **achieved** - **2025:** Rburst initiative for full integration **Success Factors:** 1. **Executive commitment** - Clear target setting (50%+ R) 2. **Multi-wave approach** - Gradual transformation 3. **Training investment** - Bookdown courses, Resource Hub, AccelerateR 4. **Platform deployment** - Posit Workbench for infrastructure 5. **Cultural change** - Growth mindset and adaptability ### Industry-Wide Movement GSK is not alone - the tipping point has been reached: **Novartis:** - Mosaic platform for ARS-driven TFL automation - Standards-based, language-agnostic approach - React UI with R backend **Roche/Genentech:** - Leading {admiral}, {teal}, {gtsummary} development - autoslideR for slide automation - {crane} for pharma-specific reporting **Pfizer:** - Custom R packages for RWD programming - Dual R/SAS syntax support - Quarto documentation sites **Moderna:** - AI-enhanced Shiny apps for trial data - ellmer/GPT integration - Accelerating exploratory analysis ### Pharmaverse Ecosystem The collaborative open-source movement is thriving: **Key Packages:** - **{admiral}** - Transitioning to stable maintenance (feature complete) - **{teal}** - Version 1.0 released (interactive apps) - **{gtsummary}/{crane}** - Clinical table generation - **{cardinal}** - Standardized TLG templates - **{sdtm.oak}** - SDTM programming framework **Benefits Realized:** - ✅ Reduced duplication across companies - ✅ Shared validation burden - ✅ Faster innovation cycles - ✅ Transparent, auditable code ------------------------------------------------------------------------ ## 3. Regulatory Evolution ✅ ### FDA Software Neutrality The 2015 FDA clarification on software neutrality is now **fully operationalized**: - No preference for SAS vs R vs Python - Focus on **methodology** and **validation**, not tools - R-based submissions increasingly routine ### New FDA Guidance Impact **2023 Covariate Adjustment Guidance** driving change: - Novartis developed {beeca} in response - ASA-BIOP collaboration on {RobinCar2} - Academic methods reaching practice faster - R enabling rapid response to guidance ### Validation Maturity **Key Developments:** #### 1. **Risk-Based Approaches** - {riskmetric} for package assessment - Shared metric repositories (R Validation Hub) - Automated quality scoring - Focus on critical packages #### 2. **Acceptance-Test Driven Development (ATDD)** - Plain-language tests in Quarto - Shareable with regulators - "Given-when-then" format - Brian Repko bringing JBehave concepts to R #### 3. **Shiny App Validation** - Litmusverse suite (Jumping Rivers) - Risk-based validation frameworks - Code quality assessment - Traceability and documentation tools ### GxP-Ready AI Pioneering work on validating AI applications: - **Devin Pastoor (A2-AI):** Production AI in GxP contexts - Testing strategies for non-deterministic outputs - Audit trails and logging - Version control and rollback procedures ------------------------------------------------------------------------ ## 4. Automation & Efficiency 📊 ### ARS/ARM as Game Changers **CDISC Analysis Results Standard** driving automation: **Mosaic (Novartis):** - YAML captures ARD requirements - Language-agnostic rules (metadata → R or any language) - React UI for customization - Push-button TFL generation **{cards} Ecosystem:** - Analysis Results Data objects - QC becomes straightforward (compare ARDs) - LLMs can summarize language-agnostic results - {crane} and {gtsummary} building on this foundation ### Report Automation **Time Savings Achieved:** - **autoslideR (Roche):** 0.5-4 days per slide deck - **DMC materials automation:** Week → Day for exposure reports (AstraZeneca) - **LLM-powered {gtsummary}:** Submission-ready tables in minutes vs hours **Technologies Enabling This:** 1. **officer/flextable** - Programmatic Word reports 2. **Quarto** - Multi-format publishing (PDF, HTML, presentations) 3. **Template systems** - Reusable structures (Cardinal) 4. **CI/CD pipelines** - Automated reanalysis on code changes ------------------------------------------------------------------------ ## 5. Advanced Analytics & Methods 📈 ### Bayesian Methods Maturing **Tools Reaching Production:** - **BayesERtools** (Genentech) - Exposure-response with Stan - **bmstate** (Generable) - Multistate survival models - **Stan debugging workshop** - Making Bayesian accessible **Advantages in Pharma:** - Prior information integration - Uncertainty quantification - Small sample performance - Complex model flexibility ### High-Performance Computing **Polars for Clinical Data:** - 10-100x faster than pandas - Apache Arrow native - Lazy evaluation - Parallel processing out-of-the-box **HPC Integration:** - Offloading Shiny computations to clusters - Maintaining interactive UX - Resource optimization ### Machine Learning **TabPFN** - Novel deep learning for tabular data: - No training required (pre-trained on priors) - Fast inference - Bayesian-like uncertainty - Promising for exploratory analysis **Synthetic Data:** - {synthpop} for privacy-preserving data sharing - Software testing without PHI - Model training on synthetic patients ------------------------------------------------------------------------ ## 6. Data Quality & Validation 🔍 ### {pointblank} Adoption Comprehensive data validation framework: **Use Cases:** - Quick dataset understanding (`scan_data()`) - Validation at scale (35+ tables daily) - Beautiful automated documentation - Integration with databases, Arrow, Shiny **Pharmaceutical Applications:** - SDTM compliance checks - ADaM dataset verification - Cross-domain consistency - Longitudinal data integrity ### Automated Traceability **DevOps Principles in Pharma:** Graticule's approach: - Docker containers for reproducibility - CI/CD for automatic reanalysis - Version control with validated outputs - Cloud storage (AWS S3) for results **Benefits:** - Clear audit trail - Reproducible analyses - Automated documentation - Integrated QC ------------------------------------------------------------------------ ## 7. Specialized Applications 🧬 ### Real-World Evidence **Pfizer's RWD Programming:** - Custom R package for database queries - Dual R/SAS syntax support - Shiny apps for workflow support - Quarto documentation ### Niche Medical Devices **Abbott's Synthetic Data:** - {synthpop} for diagnostics - Privacy-preserving test data - Validation datasets ### Oncology Innovations **PMDA Inquiries:** - {cards} for rapid response (Japan regulatory) - Structured ARD approach - Accelerated turnaround ### Pharmacokinetics **aNCA (Pharmaverse):** - Open-source NCA software - PKNCA backend (200+ parameters) - 100% test coverage - Validated against commercial tools (±0.1%) ------------------------------------------------------------------------ ## 8. Regional Insights 🌏 ### Asia/Pacific Innovations **Strengths:** - Strong PMDA regulatory focus - Practical automation tools - Under-represented population research - Genetic diversity considerations **Notable Presentations:** - SHIONOGI's open-source culture - {meRlin} AI assistant - PMDA e-CRT automation - Vibe Coding approaches ### Global Collaboration Conference demonstrated truly global R pharma community: - Workshops from US, Europe, Asia - Shared challenges and solutions - Pharmaverse as unifying force ------------------------------------------------------------------------ ## 9. Future Directions 🔮 ### Short-Term (2025-2026) **AI Integration:** - ✅ More validated AI applications in production - ✅ MCP ecosystem expansion - ✅ Regulatory guidance on AI in submissions - ✅ Cost-effective on-premise LLM solutions **Automation:** - ✅ ARS/ARM adoption across industry - ✅ Push-button TFL generation becomes standard - ✅ Automated slide decks and CSRs routine **Validation:** - ✅ Shared metric repositories operational - ✅ Risk-based validation industry standard - ✅ Shiny app validation frameworks mature ### Medium-Term (2027-2028) **Tool Consolidation:** - Language-agnostic analysis platforms - R/Python/Julia interoperability - Cloud-native pharma analytics **Regulatory:** - AI-specific guidance documents - Electronic submissions with code - Real-time regulatory review platforms **Skills:** - AI/LLM literacy standard for programmers - Bayesian methods more accessible - DevOps practices ubiquitous ### Long-Term (2029+) **Transformative Possibilities:** 1. **AI-Designed Clinical Trials** - LLMs suggesting optimal designs - Automated SAP generation - Real-time adaptation 2. **Fully Automated TFLs** - From data lock to submission package - Human review only, no coding - Multi-language outputs for global submissions 3. **Real-Time Safety Monitoring** - Continuous analysis during trials - AI-detected signals - Predictive adverse event modeling 4. **Personalized Trial Design** - Bayesian adaptive with real-time learning - Biomarker-driven enrollment - Precision medicine integration ------------------------------------------------------------------------ ## 10. Challenges & Barriers ⚠️ ### Technical Challenges **Validation Complexity:** - Non-deterministic AI outputs - Rapidly evolving tools - Keeping pace with innovation **Integration Issues:** - Legacy system compatibility - Data silos - IT security constraints ### Organizational Challenges **Cultural Resistance:** - "SAS is what we've always used" - Fear of change - Risk aversion in regulated environment **Resource Constraints:** - Training investment required - Dual maintenance (SAS + R) during transition - Validation documentation burden ### Regulatory Challenges **Uncertainty:** - Limited AI/LLM guidance - Evolving expectations - Regional variations (FDA vs EMA vs PMDA) **Documentation:** - What level of detail for AI systems? - How to handle model updates? - Third-party API dependencies ------------------------------------------------------------------------ ## 11. Success Patterns 🎯 ### What Works Based on successful implementations presented: **1. Executive Sponsorship** - Clear targets (GSK's 50%) - Resource commitment - Long-term vision **2. Gradual Transformation** - Multi-wave approach - Pilot programs - Learn and adapt **3. Support Infrastructure** - Training programs (not just one-time) - Help desks and office hours - Documentation and resources **4. Community Building** - Internal R user groups - External pharmaverse participation - Shared learning culture **5. Platform Investment** - Posit Workbench - Version control (Git/GitHub) - CI/CD pipelines - Cloud infrastructure **6. Standards Adoption** - CDISC ARS/ARM - Pharmaverse conventions - Style guides and linters ------------------------------------------------------------------------ ## 12. Actionable Insights 💡 ### For Organizations Starting R Adoption 1. **Start with low-risk projects** (exploratory analysis, visualizations) 2. **Invest in training** early and continuously 3. **Build internal community** (lunch & learns, Slack channels) 4. **Adopt pharmaverse packages** (don't reinvent) 5. **Establish validation framework** from day one 6. **Use Posit products** (Workbench, Connect, Package Manager) ### For Organizations Scaling R 1. **Embrace automation** (ARS/ARM, template systems) 2. **Implement CI/CD** for reproducibility 3. **Explore AI carefully** (privacy first, validate thoroughly) 4. **Contribute to pharmaverse** (share internal packages) 5. **Formalize support model** (beyond training) 6. **Plan SAS sunset** (don't maintain dual indefinitely) ### For R Professionals 1. **Learn AI/LLM basics** (will be essential skill) 2. **Master one reporting framework** (officer/flextable or Quarto) 3. **Understand validation** (not just coding) 4. **Contribute to open source** (career differentiator) 5. **Stay current** (R/Pharma, webinars, blogs) 6. **Network** (pharmaverse Slack, conferences) ------------------------------------------------------------------------ ## 13. The Bigger Picture 🌍 ### R/Pharma 2025 in Context This conference revealed an industry at an **inflection point**: **From:** - Proprietary, siloed tools - Manual, repetitive coding - Conservative, risk-averse culture - Isolated company solutions **To:** - Open-source, collaborative development - Automated, standards-driven workflows - Innovative, evidence-based adoption - Shared industry platforms ### Why This Matters **For Patients:** - Faster drug development - More rigorous analysis - Transparent, reproducible science - Precision medicine enablement **For Industry:** - Reduced costs - Faster time to market - Better talent recruitment - Competitive advantage **For Science:** - Reproducibility crisis addressed - Method innovation accelerated - Global collaboration enabled - Democratized advanced analytics ------------------------------------------------------------------------ ## 14. How R/Pharma Has Evolved 📈 ### Conference Evolution (2018-2025) Based on publicly available information and this year's content, here's how the conference themes have shifted: #### **2018-2020: Foundation Years** **Key Themes:** - 📦 **Package development basics** - Building pharma packages - 📊 **Basic Shiny apps** - Interactive visualizations - 🔄 **SAS vs R debates** - "Should we switch?" - 📝 **Simple reporting** - Basic TFLs **Sentiment:** Cautious optimism, early adopters sharing success stories **Notable:** - Pharmaverse not yet established - Validation was major concern - FDA software neutrality clarification (2015) still fresh ------------------------------------------------------------------------ #### **2021-2022: Acceleration Phase** **Key Themes:** - 🚀 **COVID acceleration** - Remote work drove R adoption - 🤝 **Pharmaverse launch** - Collaborative ecosystem born - ✅ **Validation frameworks** - {riskmetric}, R Validation Hub - 📦 **Production packages** - {admiral}, {teal} gaining traction **Sentiment:** Growing confidence, momentum building **Milestones:** - Major pharma (Roche, Novartis) sharing production solutions - Regulatory submissions with R increasing - Training programs formalized (GSK example) ------------------------------------------------------------------------ #### **2023-2024: Mainstream Adoption** **Key Themes:** - 📊 **CDISC integration** - Analysis Results Standard (ARS/ARM) - 🔧 **Automation focus** - Template-based TFL generation - 📈 **Advanced analytics** - Bayesian methods, ML - 🌐 **Multi-language** - R + Python workflows **Sentiment:** R is mainstream, focus shifts to optimization **Evidence:** - Multiple talks on automation (not just feasibility) - Validation becoming standardized, less controversial - Advanced topics (multistate models, Stan) gaining space ------------------------------------------------------------------------ #### **2025: AI Revolution Year** **Dominant Themes:** - 🤖 **AI/LLM explosion** - 8+ sessions (50% increase) - 🏢 **Enterprise transformation** - GSK's 50%+ achievement - ✅ **Regulatory maturity** - GxP-ready AI, validated Shiny - 🚀 **Production at scale** - From "can we?" to "how fast?" **Shift in Discourse:** - **2018:** "Is R viable for pharma?" - **2022:** "How do we migrate from SAS?" - **2025:** "How do we integrate AI with R?" **What's New in 2025:** 1. **AI Dominance** - 2018-2022: Barely mentioned - 2023: Experimental projects - 2024: Pilot implementations - **2025: Production systems** ({llumen}, multi-agent copilots, enterprise frameworks) 2. **Regulatory Confidence** - 2018-2020: "Will FDA accept this?" - 2021-2022: "Here's how we validated" - 2023-2024: "Validation frameworks established" - **2025: "Validated AI in GxP"** - previously unthinkable 3. **Speed of Innovation** - 2018: Yearly package updates - 2022: Quarterly releases (pharmaverse) - **2025: Real-time AI integration** - tools released months ago 4. **Collaboration Level** - 2018: Individual company solutions - 2020: Pharmaverse begins - 2023: Cross-company working groups - **2025: Shared AI infrastructure** (MCP, metric repos) ------------------------------------------------------------------------ ### Topic Frequency Evolution | Topic | 2020 | 2022 | 2024 | **2025** | |-------------------------|--------|--------|--------|--------------| | **Package Development** | 🔥🔥🔥 | 🔥🔥 | 🔥 | 🔥 | | **Validation** | 🔥🔥🔥 | 🔥🔥🔥 | 🔥🔥 | 🔥 | | **Shiny Apps** | 🔥🔥 | 🔥🔥🔥 | 🔥🔥 | 🔥🔥 | | **Clinical Reporting** | 🔥🔥 | 🔥🔥 | 🔥🔥🔥 | 🔥🔥🔥 | | **Bayesian Methods** | 🔥 | 🔥 | 🔥🔥 | 🔥🔥 | | **AI/LLM** | \- | \- | 🔥 | **🔥🔥🔥🔥** | | **Automation** | 🔥 | 🔥🔥 | 🔥🔥 | **🔥🔥🔥** | | **Python Integration** | 🔥 | 🔥 | 🔥🔥 | **🔥🔥** | **Interpretation:** - Traditional topics (package dev, validation) declining as they become **solved problems** - AI/LLM emerged from nowhere to **dominate** the conversation - Automation shifted from "nice to have" to **strategic imperative** - Python integration normalized (not R vs Python, but R **and** Python) ------------------------------------------------------------------------ ### What Changed Between 2024 and 2025? **Major Shifts:** 1. **{ellmer} Package Launch** (2024) - Made LLM integration trivial in R - Multiple workshops/presentations built on it - Enterprise adoption accelerated 2. **Claude/ChatGPT API Maturity** - Function calling standardized - MCP protocol emerged - Privacy-preserving deployments viable 3. **GSK Milestone** - 50%+ R code achieved - Proof that full transformation possible - Other pharma following suit 4. **CDISC ARS Adoption** - Analysis Results Standard gaining traction - Mosaic (Novartis) and others automating TFLs - Machine-readable results enabling AI 5. **Posit Product Evolution** - Positron IDE launched (VS Code + R) - Better Python support - AI assistants integrated **What Stayed Constant:** - ✅ Pharmaverse continues thriving - ✅ Validation remains priority (but less controversial) - ✅ Collaboration over competition - ✅ Open-source commitment ------------------------------------------------------------------------ ### Predictions for R/Pharma 2026 Based on the trajectory, expect: **Likely Topics:** 1. **"AI in Submissions"** - First AI-powered analysis in regulatory filing 2. **"Real-time Clinical Trial Analysis"** - Adaptive designs with AI 3. **"Synthetic Patient Generation"** - Privacy-preserving AI for trials 4. **"Quantum ML for Drug Discovery"** - Cutting edge 5. **"Fully Automated Study Reports"** - From data lock to CSR in hours **Maturing Topics:** - LLM integration will be **assumed** (not taught from scratch) - Validation frameworks **standardized** across industry - Python-R **seamless** interop (not separate tracks) **Declining Topics:** - "Intro to R packages" (basics widely known) - "Why leave SAS?" (decision already made) - "Can we validate R?" (settled question) **Bold Prediction:** R/Pharma 2026 will feature the **first fully AI-generated clinical study report** accepted by a regulatory agency. The discussion will shift from "Is this possible?" to "What should humans still review?" ------------------------------------------------------------------------ ## 15. Conclusion R/Pharma 2025 demonstrated that open-source pharmaceutical analytics is not the future **it's the present**. With AI integration, regulatory acceptance, and enterprise adoption converging, the question is no longer "Should we adopt R?" but "How fast can we transform?" The conference showcased: - ✅ Production-ready AI systems - ✅ Validated, GxP-compliant workflows - ✅ Major pharma success stories (GSK 50%+) - ✅ Thriving ecosystem (pharmaverse) - ✅ Regulatory confidence (FDA neutrality) **The momentum is undeniable. The future is open source. The time is now.** ------------------------------------------------------------------------ ::: callout-tip ## Stay Connected - **R/Pharma website:** [rinpharma.com](https://rinpharma.com) - **Pharmaverse:** [pharmaverse.org](https://pharmaverse.org) - **Posit:** [posit.co](https://posit.co) - **R Validation Hub:** [pharmar.org](https://www.pharmar.org) ::: ------------------------------------------------------------------------ *Analysis compiled from R/Pharma 2025 Conference materials \| Last updated: November 2025*