Augmenting clinical trials with external information - why, when and how? (Part 1)

Why use external information?

Randomised Clinical Trials (RCTs) are the gold standard for evidence and underpin submissions to regulatory agencies and payers. However, in some cases conducting a fully powered RCT is not feasible, for example in rare diseases, paediatrics and small biomarker-defined populations. In other cases, there might be substantial existing evidence, such as on patient outcomes when treated with placebo that could help improve the evidence provided by a trial.

External information is increasingly available in practice. Sources include previous trials, disease registries, and routinely collected healthcare data, such as the Clinical Practice Research Datalink in the UK.

How can using external information help?

Statistical approaches such as Bayesian dynamic borrowing or synthetic control methods can allow a clinical trial to utilise this external information to:

• Replace a concurrent control arm with an external comparison group when randomisation is infeasible or inappropriate;
• Supplement a control arm, perhaps allowing the trial randomisation to favour the intervention arm;
• Give additional information to inform estimating the intervention effect in subgroups or in a broader patient population than was enrolled in the trial.

The potential gains are numerous: use of external information can allow smaller trials that are more feasible to do and may give better evidence.

When can we use external information?

However, we need to be careful before routinely using this approach in practice. To be able to trust results when borrowing information, the external data must meet certain requirements. If external data and concurrent trial data are fundamentally different then analyses will give misleading and incorrect results. It is particularly important that the external data collects data on outcomes and patient characteristics in a high-quality way so that trial inclusion/exclusion criteria can be applied and any differences between trial participants and external patients adjusted for.

Although it’s impossible to rule out all sources of bias when utilising external data, we can seek to minimise bias. For the best results, it’s critical to plan carefully when designing the trial; leaving it as an afterthought or to rescue a trial that is failing to meet its recruitment targets is dangerous: it opens up strong possibilities for critique by regulatory agencies and payers. Steps that help reduce this critique include: careful consideration of whether potential sources of data reach the quality required; specifying a robust analysis plan with sensitivity analyses; and conducting statistical simulation studies to investigate which scenarios lead to meaningful bias.

An important feature of some Bayesian approaches is that the amount of external information incorporated can depend on its compatibility with the new trial data. If the findings are consistent, more information can be borrowed; if they disagree, borrowing can be reduced. This provides a degree of protection against misleading conclusions when external evidence is not fully applicable.

A real-world example: learning from adult trials to evaluate a treatment in children

An example that showcases the benefits of using external evidence comes from a trial of nintedanib, a treatment for fibrosing interstitial lung disease. Six large previous randomised trials had demonstrated that nintedanib slowed deterioration in lung function in adults. However, establishing efficacy in children was much more challenging because the condition is thankfully rare in this population.

The InPedILD trial[1] recruited 39 children and adolescents from 21 countries. Although the trial results suggested a potential benefit from nintedanib, the uncertainty was substantial.

A prespecified analysis explored whether evidence from six previous trials in adults could strengthen the conclusions from InPedILD[2].

Using a Bayesian approach, they combined the paediatric trial results with information about treatment effects in adults. Importantly, they did not assume that the treatment would necessarily work equally well in both populations. Instead, they used expert opinion to determine the initial relevance of the adult evidence and a statistical model designed to reduce borrowing if the adult and paediatric results were inconsistent.

The probability that nintedanib improved the lung-function outcome increased from approximately 70% based on the paediatric evidence alone to 95.5% when incorporating adult evidence. This exceeded the prespecified 90% threshold for concluding a positive treatment effect.

This illustrates how external information can substantially strengthen conclusions from a small trial, provided there is a convincing rationale for borrowing and appropriate safeguards against bias.

Conclusion

Borrowing external information offers exciting opportunities to make clinical trials more efficient, particularly where recruiting sufficient participants is challenging and substantial relevant evidence already exists.

However, the benefits are not automatic. The availability of external data does not necessarily mean that it should be used. Careful consideration of its relevance, quality and potential biases is essential, alongside rigorous statistical evaluation of the proposed approach.

By considering external borrowing early in the trial design process, researchers can identify opportunities to improve efficiency while maintaining the robustness and credibility of the evidence generated.

Better Trials

Better Trials, a strategic partner of Evidax, is a specialist clinical trial design initiative based at Newcastle University, bringing together expertise in innovative statistical methods and clinical trial design.

We support pharmaceutical and biotechnology companies in exploring how innovative approaches can make their clinical development programmes more efficient and informative. For trials considering the use of external evidence, this includes evaluating the suitability of available data, selecting appropriate methods for borrowing information, and using statistical simulations to quantify potential benefits and risks before a trial begins.

Our expertise also includes adaptive and platform trial designs and other innovative approaches that can help make clinical trials faster and more efficient.

References

1. Deterding R, Young LR, DeBoer EM, Warburton D, Cunningham S, Schwerk N, et al. Nintedanib in children and adolescents with fibrosing interstitial lung diseases. Eur Respir J. 2023;61:2201512. https://doi.org/10.1183/13993003.01512-2022.

2. Maher TM, Brown KK, Cunningham S, DeBoer EM, Deterding R, Fiorino EK, et al. Estimating the effect of nintedanib on forced vital capacity in children and adolescents with fibrosing interstitial lung disease using a Bayesian dynamic borrowing approach. Pediatr Pulmonol. 2024;59:1038–46. https://doi.org/10.1002/ppul.26882.