Synthetic real-world data reshapes oncology trial decision-making

Key Takeaways

• Frontiers in Oncology (Yap et al., 2022) reviews how real-world data can support external control groups in oncology drug development.
• Nature Medicine (Orcutt et al., 2025) uses machine learning-based trial emulations to test whether oncology RCT results generalize to real-world patients.
• FDA’s RWE program frames when real-world evidence may inform regulatory decisions, without treating every RWD analysis as substantial evidence.
• For BD and medical teams, the decision gate is protocol quality and transportability—not the mere presence of a synthetic or external arm.

Why do oncology trials look to synthetic and real-world controls?

Randomized controls remain the default for causal inference, but some oncology settings face ethical, feasibility, or sample-size limits. In those cases, sponsors evaluate external or synthetic control strategies built from electronic health records, registries, or claims.

Yap et al., Frontiers in Oncology (PMC8771908), discuss applying real-world data to external control groups in oncology clinical trial drug development. The review stresses matching, confounding control, and endpoint definitions as make-or-break design choices.

What does trial emulation say about generalizability?

Orcutt, Chen, Mamtani, Long, and Parikh (Nature Medicine, 2025) evaluate generalizability of oncology trial results to real-world patients using machine learning-based trial emulations. The study title itself defines the method: emulate trial eligibility and treatment contrasts in observational data, then compare emulated outcomes with published RCT results.

For decision-makers, a successful emulation increases confidence that a trial’s effect estimate may transport. A failed emulation is equally informative: it flags populations or practice patterns where the RCT may not apply.

How do regulators frame real-world evidence?

The FDA Real-World Evidence program page explains that RWE can support regulatory decisions in defined circumstances and points to guidance on using RWD/RWE for drugs and biologics. That framing matters for Advancing Clinical Trials and Decision-Making teams because it separates exploratory HEOR analyses from evidence packages intended for labeling or approval support.

• Pre-specify objectives, estimands, and bias diagnostics.
• Document data provenance and fitness-for-purpose.
• Do not treat “synthetic control” as a synonym for “regulatory-grade evidence.”

Where synthetic controls still fail diligence

Common failure modes include immortal-time bias, incomplete biomarker capture, differential follow-up, and soft endpoints that clinics code inconsistently. Synthetic cohorts can also overfit historical outcomes if generative models are tuned after seeing trial results. Those risks are why Yap et al. and Orcutt et al. emphasize design transparency rather than automation alone.

What remains unproven

Neither paper claims that synthetic RWD can replace randomized evidence across all oncology indications. Commercial “synthetic control” vendor claims that lack peer-reviewed methods or regulator-facing protocols should be excluded from diligence conclusions until primary methods are available.

Related NovaPharma coverage

• Data harmonization and AI in clinical research
• EMA and African regulators on Ebola clinical trials
• Oncology disease hub

Frequently Asked Questions

Primary Sources

1. PMC — Application of Real-World Data to External Control Groups in Oncology (Yap et al.)
2. Nature Medicine — Evaluating generalizability of oncology trial results (Orcutt et al.)
3. FDA — Real-World Evidence program