As artificial intelligence becomes increasingly embedded in healthcare research, the ability to generate reliable, transparent, and regulatory-grade evidence is taking on new significance. At ISPOR 2026, discussions around the future of real-world evidence highlighted both the opportunities and responsibilities associated with AI-driven analytics. In this interview with BioPharma Boardroom, Saamir Pasha explores the challenges of maintaining high-quality oncology data from electronic health records, the frameworks that support reproducibility and trust, and how generative AI is accelerating biostatistical workflows and trial emulation. He also shares insights into how regulators and industry stakeholders are adapting to advanced analytical methods while maintaining rigorous standards for scientific validity, auditability, and transparency.

Q: What are the biggest challenges in maintaining high-quality oncology real-world data derived from EHR systems?

A: Oncology data is inherently complex due to the wide range of cancer types and care pathways spanning multiple settings. As a result, data capture can vary significantly across providers. The primary challenges are ensuring consistency, completeness, and traceability back to the source. In practice, patients often move between providers and progress through multiple therapies, making it difficult to document information in a standardized way. Critical details are frequently embedded in unstructured notes, which adds additional complexity. High-quality real-world data (RWD) therefore requires strong standardization and a clear audit trail that tracks variables from source to analysis. Some clinically important information, such as comorbidities or adverse effects, may not be routinely captured in structured EHR fields, further impacting completeness. Ultimately, it comes down to data provenance—if we cannot clearly trace and defend where the data comes from, we cannot confidently interpret the results.

How does Ontada’s regulatory data quality framework strengthen the reliability and reproducibility of real-world evidence generation?

A: We apply regulatory-aligned frameworks recommended by the FDA and EMA, specifically the QCARD Initiative and Data Quality Framework (DQF), to systematically assess and monitor data quality within oncology EHR data. This starts with standardizing data into common oncology structures and elements, ensuring consistent expectations for how data is defined, captured, and organized across the industry. Reproducibility is driven by robust lineage documentation—clearly tracking where data originates and how it is transformed from bedside entry through analytic dataset creation. We also strengthen reliability through external validation against benchmarks such as the National Death Index, allowing us to compare outcomes like overall survival and ensure alignment with broader population-level data.

What opportunities do you see for generative AI to transform biostatistical workflows and trial emulation strategies?

A: The goal is to identify the optimal balance where generative AI can accelerate well-defined research questions while preserving the underlying biostatistical framework. AI reduces time spent on manual implementation steps, such as data linkage, variable derivation, and incorporating study-specific context. Traditionally, preparing analyses could take weeks, but AI can significantly shorten the time needed to generate an initial proof of concept or assess the magnitude of an association. This is particularly valuable under tight timelines. For example, AI can rapidly parse clinical trial protocols, extract key design elements, and map them directly to available data assets to support trial emulation. It can also generate tailored analytic workflows based on the structure of the underlying database.  While guidance is still required to ensure the correct analytical direction, AI enables faster cohort generation, analytic dataset creation, and execution of analyses, supported by well-documented code and structured outputs.

How can AI-driven approaches accelerate evidence generation while still ensuring scientific rigor and transparency?

A: AI accelerates evidence generation primarily through automation of well-defined analytical steps. However, it does not replace subject matter expertise or established review processes. Ensuring scientific rigor requires predefined robustness checks, including validating the plausibility of research questions and the consistency of observed results. These checks must be embedded into the workflow rather than applied retrospectively. We implement guardrails from the outset by providing clear, structured instructions—similar to onboarding a colleague—followed by iterative refinement and expert review. Maintaining auditability and reproducibility remains central, supported by transparent documentation at each step.

In your view, how will regulators and industry stakeholders adapt to increasing AI integration within oncology data analytics?

A: Regulators and industry stakeholders are increasingly supportive of advanced AI-driven methodologies, as seen in forums such as ISPOR. However, the expectation to “show the work” remains unchanged—and is likely to become more stringent. There will continue to be a strong emphasis on clearly justifying methodological choices, particularly as AI introduces more complex analytical approaches. As a result, our approach prioritizes robustness checks, expert validation, and comprehensive audit trails to ensure transparency, reproducibility, and regulatory alignment.