2.1 Estructura libre de riesgo
2.2.1 Metodología de estimación
An experimental medicine must pass through several phases of experimentation, and only once its safety and efficacy have been confirmed can it be approved for general use. At the time of planning a drug development programme, relevant information may already be available from routine clinical practice; clinical trials of the drug performed in related diseases or different age groups; or studies of sim- ilar medicines. The design of the development programme can then be optimised in light of this so that any new studies fill in the gaps in our existing knowledge base without replicating information. Leveraging existing data in this way is par- ticularly desirable when our aim is to develop medicines for small or vulnerable populations such as children. The European Medicines Agency (EMA) defines ex- trapolation as “. . . extending information and conclusions available from studies in one or more subgroups of the patient population (source population) . . . to make inferences for another subgroup of the population (target population). . . ”34, 35 In
many cases, we may seek to extrapolate adult efficacy data to children. Wadsworth et al.69 report the findings of a systematic review of statistical methods relevant for extrapolating efficacy and other data from adults to children. The authors identify methods originally proposed in a variety of contexts, ranging from incor- porating historical controls in new studies to evaluating the consistency of results across sites in a multi-centre trial, reflecting the wide ranging applications of ex- trapolation.
To justify the extrapolation of adult efficacy data to children, we must often make strong assumptions about the similarity of age groups in terms of disease progres- sion, response to intervention and pharmacokinetic-pharmacodynamic (PK-PD) relationships. These assumptions are made explicit in the paediatric decision tree (see FDA1 and Figure 6.1) where, in the terminology of Dunne et al.,36 judgements
about the plausibility of each aspect of similarity determine whether a ‘complete’, ‘partial’ or ‘no’ extrapolation strategy is adopted. Dunne et al.36 reviewed 370
paediatric studies submitted to the FDA between 1998 and 2008 to identify cases in which efficacy data were extrapolated: of the 166 drug products considered, 14.5% followed a complete extrapolation strategy, 68% a partial extrapolation strategy and 17.5% did not extrapolate. Sun et al.,71in an update on the review by Dunne
et al.,36 reviewed 388 paediatric studies between 2009 and 2014. The proportion of
products using partial extrapolation fell to 29%, whilst the use of no and complete extrapolation both rose to 37% and 34%, respectively. There is likely to be prior uncertainty about the plausibility of different assumptions. Hlavin et al.162 use a scepticism factor to represent uncertainty about the plausibility of a complete extrapolation approach, whereby the full weight of evidence supporting drug ef- ficacy in adults is taken to support a claim of efficacy in children. This factor could be established from historical data or expert opinion. The EMA extrapola- tion framework stipulates that data which are subsequently collected in the target population should be used to confirm extrapolation assumptions.34, 35
Since 2006, the EU paediatric regulation27 has mandated that the programme of
studies intended to support licensing of a medicine for children in the EU must follow a Paediatric Investigation Plan (PIP), which itself must be agreed ahead of time with the EMA’s Paediatric Committee (PDCO). When selecting (approv- ing) an extrapolation strategy, sponsors (regulators) must first ask themselves how plausible needed assumptions are given the data currently to hand, where
Figure 6.1: Extrapolation strategies, assumptions made and required studies, based on the FDA paediatric decision tree.1
This chapter presents a framework for using existing data to inform a decision on whether to perform a complete extrapolation of efficacy data from adults to chil- dren or a partial extrapolation instead. This decision will determine whether the sponsor will collect only PK data in children to support dose-finding, or both PK and PD data. The proposed framework requires pre-specification of a numerical criterion which PK-PD curves in adults and children must satisfy in order to be considered ‘similar’. The sponsor can then use historical data or expert opinion to quantify the prior plausibility of the stated degree of similarity. This process en- ables sponsors and regulators to define transparent success criteria which emerging data in the target population must satisfy in order to be judged as verifying the assumption.
We propose that the process of choosing between complete and partial extrapola- tion strategies should begin by performing a Bayesian random-effects meta-analysis of existing PK-PD data to derive priors for parameters representing differences between PK-PD relationships in adults and children. When studying small pop- ulations it is likely that few historical studies will be available for synthesis. The
methodological challenges associated with performing meta-analyses of few trials have been noted in Friede et al. and Turner et al.163, 164 In this setting, using a frequentist approach, we lack power to detect between-trial heterogeneity,163 while the results of a Bayesian random-effects meta-analysis are sensitive to the choice of prior for the between-trial heterogeneity parameter.163, 165 Furthermore, ‘external
biases’51 may be inherent in the existing data if there are differences between the
source and target populations, for example, if existing data are measurements on adults and adolescents but our question is whether PK-PD relationships in adults and children aged 2-11 years are similar. This data scenario often arises in practice because drug development in adults and children is typically staggered, starting in adults first. Furthermore, older adolescents are also often recruited into adult tri- als in therapeutic areas such as epilepsy155, 156, 166, 167 and asthma.168, 169 To derive
prior distributions for key parameters accounting for external biases, existing data may be down-weighted according to either a pre-specified weight (see, for example, Ibrahim and Chen;72Tan et al.;170 Rietbergen et al.87) or a dynamic weight reflect- ing their commensurability with new data collected in the target population.72–74 The challenges of dynamic downweighting are noted in Galwey.171 Alternatively, we can model the external biases and either define empirical priors51, 172 or priors
elicited from expert opinion51 on the bias parameters. We adopt the latter ap-
proach here.
To make things consistent, throughout this chapter we illustrate the proposed ex- trapolation framework with applications to anti-epileptic drug (AED) development in mind. In this setting, there is broad agreement about the acceptability of extrap- olating efficacy data in adults with partial-onset seizures (POS) to older children with POS, although there is some uncertainty about what age we can extrapolate down to.43–45, 70 This chapter proceeds as follows. In Section 6.3 we introduce
extrapolation of the efficacy data. Section 6.4 describes a scheme for eliciting expert opinion on external biases that may be inherent in the existing data. In Sections 6.5 and 6.6, we describe the simulation study used to evaluate properties of our framework in a range of scenarios before concluding in Section 6.7 with a discussion.