Pediatric extrapolation explained: evidence for children

Pediatric extrapolation is the use of evidence from a reference population, usually adults, to support a medicine's safe and effective use in children, when the disease, the drug's behavior in the body, and the response to treatment are judged similar enough across the two groups. It does not mean skipping studies in children. It means deciding, from what is already known, which questions still need pediatric data and which can reasonably be carried over.

An earlier article on this site described where extrapolation fits among the ethical rules for pediatric trials. This one goes a level deeper: how the current international framework asks the similarity question, the two main ways studies put it into practice, one published trial that used it, and where its limits sit.

What is pediatric extrapolation, in the regulators' own words?

The working definition comes from the International Council for Harmonisation of Technical Requirements for Registration of Pharmaceuticals for Human Use (ICH), which develops shared technical guidelines that regulators then adopt. Its E11(R1) addendum of April 2018 defines pediatric extrapolation as "an approach to providing evidence in support of effective and safe use of drugs in the pediatric population when it can be assumed that the course of the disease and the expected response to a medicinal product would be sufficiently similar in the pediatric [target] and reference (adult or other pediatric) population."

Two terms in that sentence carry the whole idea. The target population is the group a medicine is being evaluated for, such as children aged 10 to 17. The reference population is the group where evidence already exists, usually adults but sometimes an older pediatric age band.

The definition is reproduced in ICH E11A, Pediatric Extrapolation, the guidance that now governs how the idea is applied. The ICH Assembly endorsed E11A at its final stage, Step 4, in August 2024, and the FDA issued it as guidance for industry in December 2024.

Why is extrapolation needed at all?

Because the conventional sequence left children behind. E11A states plainly that "historically, pediatric trials have not been initiated until after adult development has been completed and/or after the drug has been approved for adults," and that enrollment into pediatric trials may then be slow because the drug is already being used in children off label, meaning outside its approved uses.

Extrapolation is now a routine part of how pediatric approvals are reached. A 2026 analysis in Clinical Pharmacology and Therapeutics by Nookala and colleagues identified 36 FDA-approved applications, covering 33 distinct drug products, that used extrapolation based on pharmacokinetics (what the body does to a drug: absorption, distribution, breakdown and clearance) to support a pediatric approval for a neuropsychiatric indication between January 2014 and July 2025. That count covers one therapeutic area, so it shows how established the approach has become rather than how often it is used overall.

What did ICH E11A change?

The most important change is conceptual. Earlier thinking sorted programs into full, partial or no extrapolation. E11A drops those labels: the guidance "does not use discrete categories (e.g., full, partial, none)," and instead describes "a continuum of similarity/dissimilarity in disease, drug pharmacology, and response to treatment" between a reference and a target population.

That sentence names the three questions every program has to answer:

• Disease. Does the condition have the same causes, course and markers in children as in the reference group?
• Drug pharmacology. Does the drug reach its target and act on it in a comparable way, allowing for differences in body size and organ maturity?
• Response to treatment. Does a given level of drug in the body produce a comparable clinical effect?

The guidance then lays out a framework in three parts: developing a pediatric extrapolation concept, then creating and executing a pediatric extrapolation plan. The concept is the written, evidence-based assessment of what is known and unknown about those three similarities. The plan is the set of studies designed to close the gaps the concept identifies. The value of the structure is that assumptions have to be stated before the studies are designed, where they can be examined.

E11A also makes clear that similarity is judged at the level of the disease subgroup, not the diagnostic label. Its own example: many causes of adult heart failure are not similar to pediatric heart failure, but heart failure due to dilated cardiomyopathy is similar between adults and children, which allows extrapolation for that subgroup.

Safety received the same treatment. The guidance notes that "historically, extrapolation of safety generally was considered unacceptable," and that understanding has evolved, so the suitability and extent of safety extrapolation are now assessed within the same concept and plan rather than ruled out in advance.

How is it done? Two study approaches

Exposure matching. Exposure is the amount of active drug the body actually experiences over time, and exposure-response is the relationship between that amount and the effect. When there is strong evidence that the disease is similar and that a given exposure produces a similar response in both populations, E11A says targeting the exposures known to be effective in the reference population "may be reasonable." Its examples are infectious diseases and partial onset seizures. In practice, modeling and simulation guide the first pediatric doses, allometric scaling adjusts for how clearance and distribution change with body weight, and a pharmacokinetic study in children confirms that the target exposure is reached.

Borrowing adult data into a smaller pediatric trial. When response similarity is less certain, a pediatric efficacy trial is still needed, but it can be made feasible by formally borrowing information from adult data. A 2026 paper by FDA statisticians, Crackel and colleagues in Therapeutic Innovation and Regulatory Science, works through this for type 2 diabetes in children. They note that variability in treatment effect is much larger in children with type 2 diabetes than in adults, so a pediatric trial powered to stand entirely on its own would need many more participants than can realistically be recruited.

Their approach uses Bayesian statistics, in which prior information is expressed as a probability distribution (the prior) and combined with new trial data. The paper sets out four steps:

1. Identify the external data that can be leveraged.
2. Prespecify the model parameters.
3. Assess the operating characteristics, meaning how the design behaves across plausible scenarios, including how often it would wrongly declare success.
4. Prespecify the weights and the maximum amount of borrowing.

The authors use a mixture prior, which combines an informative component built from adult data with a less informative one. Its practical effect is that the analysis borrows less when the pediatric results disagree with the adult data, a safeguard against the populations turning out to be less similar than assumed.

A worked case: the DINAMO trial

The DINAMO trial tested empagliflozin and linagliptin, each against placebo, in young people with type 2 diabetes. Its primary outcome was the change in HbA1c, a blood measure reflecting average blood glucose over the preceding months, over 26 weeks.

A 2025 paper by Sailer and colleagues describes how the trial handled a problem common in pediatric research: variability in the primary outcome was greater than expected, which threatened the trial's statistical power. The team had prespecified a Pharmacometrics Enhanced Bayesian Borrowing analysis. Previously fitted pharmacokinetic and exposure-response models, based on historical adult and pediatric data, were used to simulate participant data and build the informative part of a robust mixture prior. External experts and representatives of the FDA advised on the prior's effective sample size (how many patients' worth of information the prior represents) and on the weight given to its informative component.

The reported result is modest and specific. The Bayesian analyses for each drug produced estimates consistent with the trial's own results, and sensitivity analyses were run across a full range of alternative weights. The abstract does not report effect sizes, so none are given here. Readers should also know that several of the paper's authors list Boehringer Ingelheim as their affiliation.

The methodological lesson is the one the FDA paper also stresses: borrowing is defensible when its sources, weights and limits are fixed in advance and tested, not chosen after the results are seen.

What extrapolation cannot settle

Response similarity remains the weakest link. The 2026 Clinical Pharmacology and Therapeutics analysis states that "uncertainties in treatment response and drug tolerance similarity remain a fundamental concern," and that while acute treatment responses are often studied, longer-term effects on disease progression and neurodevelopment in children are less well characterized.

The youngest children are the hardest case. E11A notes that extrapolation to neonates may be challenging because of rapid physiologic change and organ maturation, even though its general principles still apply.

Borrowing is a judgment with a dial. A mixture prior limits the damage when adult and pediatric data disagree, but it does not remove the underlying assumption. A careful reader of any borrowing analysis should look for the stated maximum borrowing, the prior's weight, and the operating characteristics, and should treat their absence as a reason for caution.

Similarity is argued, not observed. When an extrapolation concept is wrong, the resulting confidence is misplaced rather than merely imprecise. The strength of the E11A framework is that it requires the argument to be written down where others can check it.

Where this sits in pediatric health research

Extrapolation is best understood as a discipline for deciding what evidence a child still needs, not as a shortcut around collecting it. It matters most in exactly the settings where pediatric trials are hardest to run: small eligible populations, high variability, and outcomes that unfold over years. Related work on how pediatric studies are designed and interpreted appears in the pediatric health and GI research program on this site, and in the peer-reviewed publications listed here.