Key Insights from PHUSE APAC Connect 2026

Automating Character Length Audits to Optimize SUPPVAR Integration
Presenter: Ram Mohan Konda, Manager - SAS Programming

One of the most common SDTM challenges is managing character variables that exceed the 200-character threshold, a risk that can lead to truncation, compliance issues and regulatory queries.

In this session, Ram Mohan Konda presented a reusable SAS macro framework designed to automatically scan entire dataset libraries, compute maximum character lengths, flag variables exceeding thresholds, and generate audit-ready Excel summaries. The solution not only eliminated manual review time but also ensured consistent SUPPVAR handling across studies.

The key takeaway was clear: with parameter-driven automation, what once required hours of review can now be executed in under a minute with improved accuracy and full traceability.

Q: How can automation streamline SUPPVAR handling and SDTM review processes in SAS?
A: Automating SUPPVAR compliance and character length audits relies on rule based SAS macros that detect long values, move overflow text to SUPPQUAL domains, and generate structured, timestamped reports. Reusable validation frameworks embedded in batch pipelines reduce manual review effort, improve SDTM data quality, and accelerate preparation of regulatory ready clinical programming deliverables.

Q: How to prepare CDISC compliant regulatory submission datasets?
A: Preparing CDISC compliant SDTM datasets involves mapping raw clinical data into standardized structures, applying controlled terminology, and ensuring all variables comply with SDTM rules. Automated checks—including character length audits and value based validations—help identify issues early and ensure no truncation or data loss.

Q: How to reduce manual review time in clinical programming?
A: Manual review time is reduced by implementing automated, rule based SAS frameworks that scan all datasets, profile variable metadata, and flag anomalies without requiring variable by variable inspection. Using parameter driven macros, batch processes, and automated reporting removes repetitive review tasks and minimizes human error. This leads to faster turnaround, improved consistency, and near real time validation across large study libraries.

Q: What are the best practices for automated character length audits in SAS?
A: Best practices include dynamically identifying all character variables across a library, computing true maximum lengths using length(strip()), and comparing them against a CDISC aligned threshold (commonly 200 characters). Any values exceeding the limit should be automatically flagged, logged, and prepared for SUPPQUAL integration. Generating timestamped, multi sheet Excel reports ensure auditability, traceability and submission ready documentation.

Precision Timing: Implementing Epoch Calculations with R for Clinical Data Integrity
Presenters: Ram Mohan Konda, Manager - SAS Programming
Sheik Ibrahim, Technical Lead - SAS Programming

Accurate epoch assignment is foundational for treatment classification, safety interpretation and regulatory submissions, yet partial ISO dates, overlapping ranges and inconsistent boundaries often create complexity.

This session introduced an automated R-based framework (m_epoch) designed to standardize partial dates, align observations with subject element boundaries and apply configurable tie-breaking logic when multiple matches occur. The solution ensures reproducibility, handles edge cases and strengthens regulatory-grade dataset preparation.

The discussion resonated strongly with attendees navigating increasingly complex trial designs and multi-epoch data structures.

Q: How to calculate study epochs in R for partial ISO dates?
A: Calculating study epochs in R when working with partial ISO-8601 dates such as YYYY or YYYY-MM requires careful temporal standardization and overlap logic. The process begins by parsing partial dates into standardized comparable formats and joining the observation dataset with the SE (Subject Elements) domain containing epoch start and end boundaries. Conditional logic is then applied to determine whether the observation falls within a valid epoch range, even when only year or month-level precision is available. In cases of multiple matches, configurable tie-breaking rules such as EARLY, LATE, TRT_EARLY, or TRT_LATE ensure consistent selection. An automated R function that validates inputs, handles edge cases and produces reproducible outputs strengthens data integrity and ensures regulatory-grade epoch assignment.

RWE and Data Visualization
Presenter: Rahul Manohar Somavanshi, Associate manager- R Programming

Navitas Life Sciences was also represented in the RWE and Data Visualization stream, where Rahul Manohar Somavanshi delivered a session titled “Leveraging large language models for missing data imputation and interpretation of RWE outputs.”

Real-world data (RWD) frequently contain substantial missingness, which can compromise the validity and interpretability of real-world evidence (RWE) analyses. Traditional statistical imputation methods perform well under restrictive assumptions but struggle with complex, nonlinear, and context-dependent clinical data. In this study, we evaluated the performance of a Large Language Model (LLM; Claude Sonnet 4.5) for imputing missing clinical variables and interpreting RWE outputs using the MIMIC-IV intensive care database. LLM-based imputation was compared with Multiple Imputation by Chained Equations (MICE) and MissForest across varying levels of missingness and sample sizes. For continuous variables, the LLM demonstrated accuracy comparable to MissForest and greater robustness than MICE, particularly at higher missingness levels, while categorical imputation performance varied across methods without a consistent best performer. Bias remained low across methods, though variability increased with extreme missingness. Few-shot prompting provided minimal benefit over zero-shot prompting, indicating stable model behavior without extensive prompt engineering. Additionally, the LLM generated reasonably accurate and clinically appropriate summaries of baseline tables and survival analyses, as validated by expert review. These findings suggest that LLMs can complement existing statistical approaches by improving both data completeness and interpretability in RWD-based clinical research.

The presentation formed part of the broader dialogue at PHUSE APAC Connect 2026 around emerging technologies and their evolving role in real world evidence analytics.

Poster Presentation
Creating Clinical Trial Data Package for Regulatory Submission

Presenters: Anbarasan Durairajan, Associate Manager - SAS Programming
Pradip Sivaramakrishnan, Associate Director - SAS Programming

The presentation outlined considerations involved in preparing structured, submission-ready clinical trial data packages aligned with CDISC standards and regulatory expectations.

Hot off the heels of this inaugural event, we are continuing the action and are excited to be heading to Austin, Texas, for PHUSE US Connect 2026.

Both our Senior Director of Clinical Reporting, Sid Kumar, and our Associate Director, Clinical Reporting, Jeff Dickinson, will be taking to the podium to present as part of the Scripts, Macros, and Automation stream and the RWE and Data Visualization stream, respectively.

Also in attendance will be our EVP, Data Sciences, Neeraj Malhotra. Stay tuned for more information on our involvement to follow soon!

Schedule a meet with our on-site team NOW!