Kevin Lee, Clinvia

The rapid evolution of generative AI is reshaping how programming and data analysis are performed across industries, and the programming of Biometrics team is no exception. Vibe coding, a novel paradigm that leverages conversational AI tools for code generation, debugging, and automation, is emerging as a transformative force in the programming workflow – for example, Microsoft CEO Satya Nadella estimated in April 2025 that 20% to 30% of Microsoft’s code was generated by AI.

This presentation introduces the concept of vibe coding and explores its application in the Biometrics Department, where traditional tools like SAS, R, and Python remain foundational. The presentation will discuss how vibe coding can be integrated with established statistical coding environments for data analysis and visualization. First, the presentation will demonstrate how prompt engineering could be applied in popular AI web platforms like ChatGPT, Gemini and Claude. Secondly, the presentation will examine AI coding agents such as Cursor, Windsurf, GitHub Copilot and more where these tools can offer inline code suggestions, integration and completion with real-time collaborative coding with AI. The presentation also introduces cases where customized AI agents are developed to support data analysis, review and visualization, and even suggesting code validation steps across SAS, R, and Python environments. The presentation also examines the benefits, including enhanced productivity, democratization of complex tasks, and reduced coding overhead potential risks such as hallucinated outputs, compliance concerns, and the need for human oversight. The presentation concludes with a forward-looking view on the evolving role of statistical programmers in the era of Vibe coding. It offers a vision where vibe coding augments, not replaces, human expertise, empowering Biometrics departments to focus more on scientific insight, data-driven strategy, and cross-functional collaboration.

Ryan Lafler and Miguel Angel Bravo, Premier Analytics Consulting LLC

Machine learning (ML) continues to reshape business, technology, science, and research across all industries, with its adoption enabling systems to learn from data, automate decisions, and generate insights. This paper presents a structured roadmap through three core domains of machine learning that are increasingly adopted by organizations: supervised, unsupervised, and generative learning. Along this roadmap, readers will identify key algorithms and architectures within each domain and understand the role of parameters and hyperparameters in mitigating overfitting and underfitting. The discussion includes examples of predictive modeling on labeled data using supervised algorithms, knowledge discovery from unlabeled data using unsupervised algorithms, and the extension of these capabilities through generative learning, which enables systems to extract insights and produce new content or data representations. The paper concludes by introducing three generative architectures that define the state of AI in 2025: encoder models (BERT), decoder models (LLMs), and encoder-decoder models (T5), and describes how each supports advanced AI tasks including representation learning, language generation, natural language processing (NLP), text summarization, and translation.

Saurabh Das, Tata Consultancy

Clinical trials have increased in complexity and duration. Incorporating data sources, including wearables, electronic health records, mobile health applications, genetic sequencing, and diagnostic imaging, expand the breadth of information collected. Maintaining data quality, efficiency, and deriving actionable insights remain ongoing considerations. Artificial intelligence and machine learning along with Generative AI, are increasingly being used to improve processes related to these challenges, contributing to faster and more efficient clinical trials. These technologies are expected to have a notable impact on research and development within life sciences and healthcare. For instance, a recent McKinsey study reported that approximately 270 companies are involved in AI-driven drug discovery, while others focus on areas such as clinical trial management and reporting.

This paper outlines practical strategies for global industries to implement open-source AI/ML software, tools, and frameworks in their products and platforms. This practice can support scalability and potentially improve accuracy, efficiency, safety, and speed in clinical trials. Some key use cases in clinical research and development are:

• Monitoring Visit Reports Insights and Analytics: Natural language processing (NLP) models organize insights from unstructured reports, assisting in the identification of compliance risks and operational inefficiencies.
• Anomaly Detection: Machine learning algorithms identify outliers and protocol deviations, supporting data integrity and patient safety.
• Patient Profile Scoring: Predictive models assess patients based on adherence, risk factors, and medical history, helping facilitate targeted interventions and analysis of dropout rates and efficacy.
• Site Risk Profiling: Dashboards utilizing aggregate site metrics generate dynamic risk profiles to inform monitoring and resource allocation, increasing efficiency by 30%.
• One-Touch Statistical Reporting: Automated processes enable timely statistical report generation, reducing efforts from 6 weeks to 1 day.

In conclusion, the integration of artificial intelligence and machine learning into clinical trial operations marks a transformative period for the industry, promising significant advances in speed, accuracy, and safety. Looking ahead, the emergence of Generative AI—capable of synthesizing complex datasets, generating clinical documentation, and even simulating patient populations—offers new horizons for protocol design, data analysis, and regulatory submissions. Meanwhile, Agentic AI systems, distinguished by their autonomous decision-making and adaptive learning, have the potential to orchestrate end-to-end trial activities, proactively identify risks, and optimize resource allocation without constant human intervention. Together, these innovations stand to redefine the boundaries of clinical research, enabling more personalized, efficient, and reliable trials. As open-source platforms and collaborative frameworks mature, the adoption of Generative AI and Agentic AI will not only enhance data-driven insights but also foster a more agile and responsive clinical trial ecosystem. Embracing these technologies will be essential for organizations seeking to stay at the forefront of life sciences and deliver transformative benefits to global healthcare.

Shivani Gupta and Anna Yaggi, Clymb Clinical

The industry is steadily advancing toward TFL automation, yet automating analysis results has remained a persistent challenge. With the release of the Analysis Results Standard (ARS) by CDISC in April 2024, we now have an opportunity to build automation tools on a common standard. ARS metadata is structured and machine-readable, making it highly compatible with Gen-AI. By combining ARS metadata with study documents such as the Statistical Analysis Plan (SAP) and TFL mock shells, Gen-AI can generate complete programming code in R and SAS, ready for integration and customization with your company standards.

This presentation introduces siera, an open-source R package that ingests ARS metadata and generates ready-to-run, transparent, and modifiable R code to create Analysis Results Datasets (ARDs). Unlike black-box solutions, siera produces fully inspectable code, giving users both flexibility and control. Together, ARS, siera, and Gen-AI can bring us closer to true end-to-end TFL automation powerful, efficient, and standards-driven

Weishan Song, Vertex

The U.S. Food and Drug Administration (FDA) requires a Bioresearch Monitoring (BIMO) package for pivotal studies to support the planning and conduct of clinical site inspections. A key component is the subject-level data line listings by site, which provide site-specific subject data underpinning safety and efficacy assessments. These listings are often produced by manually programming outputs from SDTM/ADaM datasets for each site and each listing, a resource-intensive process that can introduce variability and misalignment with the clinical study report (CSR).

We developed an R tool that automatically generates the subject-level data line listings by site using an output-to-output approach. The tool ingests finalized CSR listings, extracts content, page layout, and formatting, and reproduces the required BIMO listings for every site without reprogramming listing specifications. This preserves the CSR’s structure, maintains consistency between CSR and BIMO outputs, and standardizes presentation.
By automating and formalizing listing generation, the tool meets regulatory requirements and standards, reduces manual effort, simplifies quality control, and improves traceability from CSR to BIMO deliverables.

Kevin Lee, Clinvia

The pharmaceutical industry stands at the beginning of a transformative era, with Generative AI (e.g., ChatGPT) revolutionizing clinical trial development. It will explore the integration of Generative AI in Biometrics, highlighting its potential to streamline workflows, redefine clinical trial development, and lead innovation.

The presentation will start with the introduction of Generative AI and its wide-ranging applications in Biometrics, such as information query, codes generation (e.g., SAS, R & Python), codes conversion (e.g. SAS to R/Python), document generation (e.g., SAP, CSR), data analysis, data visualization, patient profiling and many more. It will also explore the tools, systems, processes and people that are reshaping clinical trials with Gen AI integration. Looking toward the future, the presentation will evaluate the lasting impact of Generative AI on Biometrics department. By strategically adopting these cutting-edge technologies, Biometrics teams can dramatically enhance operational efficiency, optimize trial outcomes, and expedite regulatory approval processes. It will culminate in a forward-looking exploration of how Biometrics teams can evolve into “super Biometrics teams”—leveraging Generative AI to achieve unprecedented levels of innovation, precision, and effectiveness in clinical trial development.

Peng Zhang, CIMS Global

The preparation of Clinical Study Reports (CSRs) for regulatory submission often requires medical writers to manually integrate tables, figures, and listings (TFLs) into narrative documents. This process can be time-consuming and prone to human error. To address this challenge, we leverage open-source solutions including {quarto} and {shiny} for medical writers to generate the report. By interacting with the study outputs, medical writers can efficiently produce customizable contents in the desired format. In this presentation, we will demonstrate the workflow, illustrate feasibility through practical examples, and discuss key development practices.

Joe Madden, SAS

In the world of data analysis, two mighty warriors stand tall: R, the shining star for statistical open-source freedom, and SAS, the seasoned veteran. Join us as we dive into the epic showdown between these two data titans. Will R’s quirky charm and endless packages win the day, or will SAS’s polished prowess and enterprise support reign supreme? Expect a rollercoaster of code, a sprinkle of statistical magic, and a whole lot of nerdiness as we explore the strengths, quirks, and the things we obsess over for these beloved tools. Whether you’re an R enthusiast, a SAS devotee, or just here for the data drama, this presentation promises to be a data duel for the ages!

Anuja Das, Biogen

R Validation Hub is a cross-industry collaboration to support the adoption of R within a biopharmaceutical regulatory setting through appropriate tools and resources that leverage the open source, collaborative nature of the language. Using R in submissions to healthcare regulators often requires documentation showing that the quality of the programming packages used was adequately assessed. This can pose a challenge in R where many of the commonly used tools are open source. Through this presentation, we will highlight the R Validation Hub’s risk assessment framework for R packages that has been utilized by key pharma companies across the industry. We also showcase the products our working groups have developed including the {riskmetric} R package that evaluates the risk of an R package using a specified set of metrics and validation criteria, and the {riskassessment} app that augments the utility of the {riskmetric} package within a Shiny app front end. Lastly, we will illustrate a prototype of a technical framework to maintain a ‘repository’ of R packages with accompanying evidence of their quality and the assessment criteria. All of our work is designed to facilitate the use of R within a highly regulated space and ease the burden of using R packages within a validated environment.

Xing Chen and Weijie Zhang, Moderna

Artificial intelligence (AI) is transforming drug development by accelerating how clinical and statistical teams interact with data. This talk presents an AI agent framework designed to deliver rapid, flexible analysis with clinical data. The agent leverages a multi-LLM architecture to interpret natural language questions and translate them into actionable workflows. Statistical modeling and visualization are executed natively in R, ensuring alignment with common toolsets for clinical data analyses while maintaining trackability and reproducibility for high quality outputs. This approach enables automatic construction of graph databases, dynamic retrieval of records, generation of reproducible statistical models, and automated production of publication-quality plots. By bridging AI-driven reasoning with R-based analytics, the system empowers clinical teams to obtain insights in minutes rather than days, enhancing decision-making in drug development pipelines.

Bo Wang, Novartis

The Generic Safety App is a customizable tool that streamlines interactive safety data review in clinical trials. Initially introduced as a pilot, it has now completed full validation and is available as a scalable, production-ready solution.
Trial teams can quickly configure trial-specific instances without coding, supported by training and user guides. This ease of setup reduces deployment time while ensuring consistent, transparent safety analyses. Validation under the Scientific Software Validation Working Practice ensures compliance, and subsequent use in new trials requires no additional validation. During its pilot phase, the Generic Safety App supported over 20 Data Monitoring Committee (DMC) meetings and, after successful validation in Q4 2024, has been implemented in more than 10 standardized deployments. These deployments span multiple Disease Units, including Immunology, Oncology, and Cardio-Renal-Metabolism, with expansions to additional areas such as Neuroscience. By advancing from pilot to fully validated platform, the Generic Safety App illustrates how standardized, scalable tools can enhance collaboration, support external partners, and promote innovative approaches to safety data review – enabling faster, more informed decision-making in clinical development.

Sean Sinnott, Posit

This talk will provide an overview of some recent Pharma updates from Posit. In particular, the talk will focus on Workbench, and a new feature: Audited Workbench Jobs. Audited Jobs enables R and Python code to run with cryptographically signed outputs and environment details for verifiable, reproducible results, crucial for Pharma use cases. This talk will highlight how Workbench computes a digital signature of the job’s output, verifying the integrity and authenticity of the data that runs in the Workbench job. Presentation will also discuss how Audited Workbench jobs capture information about the environment used to run the job, including R or Python versions as well as how to customize auditing by including user-defined information in the audit record. Audited Workbench Jobs and other Posit updates will be highlighted in the talk.