Extrapolation refers, in clinical research and in regulatory approval, to the methodological transfer of efficacy and safety data from a studied population, dosage, route of administration or study situation to another for which only limited or no direct study data are available. The aim is to reach well-founded conclusions without having to conduct a complete, independent clinical development programme for every subgroup.
In practice, extrapolation is used particularly often when ethical or practical reasons make randomised trials difficult (e.g. in rare diseases) or when conducting large trials in special populations (e.g. paediatrics) would be disproportionate. Extrapolation is not “guessing” but requires a traceable justification, a sound evidence chain and a transparent presentation of the uncertainties.
When extrapolation is used
Typical use cases are the transfer of adult data to children, from one dosage form to another (e.g. tablet to suspension), or from one indication to a closely related indication where pathophysiology and mechanism of action are comparable. Extrapolation is also used in dose finding and pharmacometrics when models describe dose-response relationships beyond the observed range.
Extrapolation is also regulatorily relevant for biosimilars or bridging concepts, where certain data packages are to be reduced. In such cases, a “totality-of-evidence” approach is often expected, bringing together preclinical, clinical and model-based evidence.
Methodological basis: models, assumptions and uncertainty
Sound extrapolation is based on (1) a plausible mechanism of action, (2) comparable exposure at the site of action, (3) similar disease biology, and (4) consistent clinical evidence. These criteria are often understood as an evidence chain: if one link is weak, the uncertainty of the transfer increases.
Methodologically, pharmacokinetic and pharmacodynamic models, population PK, exposure-response analyses and physiologically based PK models (PBPK) are used, among others. In statistics, sensitivity analyses are used to test robustness to assumptions. It is important that models not only “fit” but are also externally validated or made plausible (e.g. by comparison with literature data or real-world evidence).
From a data integrity perspective, it is essential that model inputs (e.g. dosing times, concentrations, covariates) are complete and verifiable. Unclear missing-data rules or subsequent data corrections without an audit trail can jeopardise the acceptance of an extrapolation during an inspection. A frequent point of discussion is also the selection of suitable comparative data: where historical controls or literature data are used, their origin, inclusion and exclusion criteria, and measurement methods should be critically assessed.
Communication of uncertainty is also central: confidence intervals, scenario analyses and clear sensitivity analyses help make the consequences of alternative assumptions visible. This turns extrapolation from a mere model result into a traceable basis for decision-making in benefit-risk assessments.
Regulatory requirements in the EU and internationally
In the EU, extrapolation plays a particular role in paediatrics: within the framework of a Paediatric Investigation Plan (PIP), partial or full extrapolation of adult data can be accepted if the scientific justification is sound and supplementary data (e.g. PK in children) support the assumptions. Extrapolation may also be discussed in the context of authorisation variations or line extensions when new populations or dosages are added.
For medicinal product studies, every derivation must be transparently documented. In dossiers such as the eCTD, assumptions, modelling, data sources and uncertainties should be described in a traceable manner. International guidelines from the ICH (e.g. ICH E9 on statistical principles, ICH E6(R3) on study quality) as well as EMA guidelines provide the framework for this; Scientific Advice is also frequently recommended to align the concept in advance.
Practical relevance and typical pitfalls
For sponsors, extrapolation can reduce development time and costs, but it is only valuable if it is regulatorily accepted. The sponsor, pharmacometrics and regulatory affairs functions should therefore jointly define early on which questions may be extrapolated and where additional data are needed. For a CRO, this means that study planning, data management and statistical analyses must be set up so that the later modelling and bridging analyses are possible with high data quality.
Operationally, it is important to plan endpoints and subgroups proactively. If extrapolation targets subgroups, relevant covariates (e.g. age, weight, organ function) should be collected cleanly. Common mistakes are insufficiently justified assumptions (“children are small adults”), the conflation of extrapolation with mere interpolation within the data, or overreliance on model-based results without clinical plausibility checks. In the benefit-risk assessment, it should be clearly identifiable which part of the evidence was directly measured and which part was derived.
Regulatory references (selection)
- EU Clinical Trials Regulation (EU) No 536/2014 (CTR)
- ICH E6(R3) Guideline for Good Clinical Practice (GCP)
- ICH E9(R1) Statistical Principles (Estimands) and ICH E8(R1) General Considerations for Clinical Studies
FAQ
Is extrapolation generally permitted in authorisation procedures?
Yes, it can be accepted if the assumptions are scientifically justified and the uncertainty is appropriately addressed. The greater the “leap” between source and target, the more supporting data (e.g. PK/PD) are generally expected.
What data are typically required for paediatric extrapolation?
At a minimum, pharmacokinetic data in the target population, a plausible mechanism of action, and comparable disease biology are often required. Additionally, safety data, dose finding, and, where applicable, limited efficacy data may be necessary.
How can a CRO practically support extrapolation?
Through study-related collection of relevant covariates, robust data management processes, statistical sensitivity analyses, and the operational preparation of model-based analyses (e.g. standardised data structures, audit trail, traceability).