Equivalence trials for biosimilars
A biosimilar's clinical trials set out to show that it is similar to an approved reference product, not that it is better than a comparator or placebo. That question shapes how each trial is designed and analysed. Pharmacokinetic similarity is tested against a fixed interval, a three-arm study usually bridges the reference products authorised in different regions, the margin for the confirmatory trial is derived from the reference's effect against placebo, and the analysis set and the handling of missing data work differently from a superiority trial.
Similarity rather than superiority
Biosimilar development sets out to show that the proposed biosimilar is similar to an already approved reference product, whereas new-drug development often has to prove superiority over an active comparator or placebo. This difference affects the statistical aspects of its trials, including the choice of study design, endpoints and analyses. Biosimilar development also puts a greater focus on justifying the margin and on imputing missing efficacy data.
An equivalence design, which has an upper and a lower comparability margin, is generally preferred to a non-inferiority design, which has only one. A non-inferiority design may still be considered if it is appropriately justified. In phase III comparative trials, the equivalence design uses a 90% or 95% confidence interval and establishes that the biosimilar is neither superior nor inferior to the reference.
The pharmacokinetic similarity study
The pharmacokinetic/pharmacodynamic similarity study aims to show bioequivalence between the proposed biosimilar and the authorised reference, meaning no clinically meaningful difference in pharmacokinetics, pharmacodynamics or both. Statistically, pharmacokinetic similarity is shown if the 90% confidence interval for the ratio of the geometric means of test to reference falls entirely within a predefined margin of 0.80 to 1.25 for each parameter. The parameters are typically the area under the curve to infinity (AUCinf), the maximum measured concentration (Cmax) and the area under the curve to the last quantifiable concentration (AUClast). The method is equivalent to two one-sided tests at the 5% level.
Regulatory guidelines suggest this standard margin for all pharmacokinetic parameters, and it accounts for an acceptable difference in systemic exposure of up to 20% between treatments. Each parameter is log-transformed and analysed by analysis of variance, with estimates computed for each treatment comparison. For a crossover study, the model includes sequence, treatment and period as fixed effects and subject nested within sequence as a random effect. For a parallel-group study, the model should include treatment as its only fixed effect. Stratification factors used in randomisation and other important baseline characteristics may also be added as covariates if clinically justified. Three example intervals show how the result is read: an interval wholly inside 0.80–1.25 meets equivalence, one partly outside it does not, and one wholly inside that does not contain the equality point of 1 meets equivalence but needs additional explanation of why it misses 1.
A biosimilar may need a different confidence interval from a generic's to demonstrate similarity in exposure, and the choice needs to be discussed and justified. A generic's comparison focuses on absorption, whereas for a biosimilar a potential difference is of interest in both the absorption and the elimination phase.
By 2017 the FDA had given specific guidance on clinical trial design for biosimilars and interchangeables. Under that guidance, a crossover design should be used for a biosimilar with a half-life shorter than five days, a rapid pharmacodynamic response and a low incidence of immunogenicity. A parallel-group design should be used for one with a long half-life and a higher immunogenicity that affects pharmacodynamics and pharmacokinetics after multiple doses. To judge pharmacological similarity, the FDA suggested the same 90% confidence interval with limits of 80% to 125%, applied to the ratio of the geometric means of the pharmacodynamic as well as the pharmacokinetic parameters.
For most products and indications no pharmacodynamic marker exists. Where one does, its margin depends heavily on the marker chosen, so it has to be defined for each compound and agreed with the health authorities, following the same principles as the efficacy margin of the confirmatory study. If a sensitive pharmacodynamic marker is available, efficacy can also be assessed in the similarity study and may not have to be established in the confirmatory study, which would then focus on safety and immunogenicity only. In any case, as of 2019 the European Medicines Agency (EMA) required at least one year of safety data from the confirmatory study.
Three arms for the regional references
Different health authorities approve medicines in different regions, so the authorised reference can differ by region, as with a US-licensed and an EU-authorised version. The pharmacokinetic/pharmacodynamic similarity study therefore usually has three arms, the proposed biosimilar, the EU reference and the US reference, randomised 1:1:1, commonly in a three-period crossover with washouts. The three arms give three comparisons: biosimilar against EU reference, biosimilar against US reference, and EU reference against US reference. The EU-approved and the US-approved reference products are not equivalent by default, and the equivalence margins and confidence interval requirements may differ between regions.
The comparison of the two references is the pharmacokinetic bridge. Together with an analytical bridge, such as structural and functional data comparing all three products, it can form the basis for justifying the relevance of data from in vivo non-clinical or clinical studies that compare the biosimilar with a reference authorised in a different region, for example using EU reference data for an FDA submission. This potentially reduces cost and development time by including only one reference product in the animal studies or the confirmatory study.
Three comparisons on up to three primary parameters, AUCinf, Cmax and AUClast, make as many as nine comparisons. Each carries a 5% false-positive rate for a one-sided directional hypothesis, so counter-measures are needed against an inflated rate of false-positive conclusions. Where the comparisons on several primary parameters cover different aspects of the drug effect, all of them have to succeed for the study to be conclusive. One option is a hierarchical testing strategy, which ranks the comparisons and tests each one only if the comparison ranked above it succeeded, and which needs no adjustment of the significance level for each comparison. Multiple comparisons also affect power. Studies are often powered at 80% overall, so each comparison should be powered higher, for example at 96%, to keep the overall power at 80%.
The confirmatory study's indication and endpoint
The confirmatory study aims to show that the biosimilar and the reference have no clinically meaningful differences in efficacy, safety and immunogenicity. To detect any product-related difference that exists, the products have to be compared in the most sensitive model, meaning indication plus endpoint, and under study conditions in a homogeneous population. The indication chosen is not necessarily the one in which the product is most often used. Similar efficacy and safety need not be shown in every approved indication, and the confirmatory study instead runs in the most sensitive indication only. Running the study in a sensitive indication is an important part of the concept of biosimilarity, so the chosen indication needs to be justified. Which indication is the most sensitive, and which population within it, is usually agreed upfront with the health authorities before the comparative efficacy and safety study runs.
The same principle of the most sensitive setting governs the choice of the primary endpoint. The endpoint should be objective and show a clear treatment effect, and a continuous endpoint can often detect differences more sensitively than a binary one. In oncology, for example, overall survival and progression-free survival are important endpoints for establishing patient benefit from a new anticancer drug. For assessing similarity they may be neither feasible nor sensitive enough, because they may be influenced by factors unrelated to any difference between the products, such as tumour burden, performance status and previous lines of treatment. Overall response rate or percentage change in tumour mass from baseline may be used instead.
The equivalence margin
The confirmatory study does not aim to demonstrate patient benefit, which was already established for the reference, so it needs no placebo arm. Instead, an indirect comparison with placebo should be made through the equivalence margin. A systematic review identifies the studies relevant to comparing the reference with placebo in the indication, and these studies can be used to estimate the reference's effect size against placebo with its confidence interval. The planned biosimilar trial will also give an estimate of the treatment effect with a confidence interval. Combining the two intervals gives an indirect interval for the biosimilar against placebo, and the biosimilar is shown superior to placebo if the lower bound of that interval is above zero.
The margin defines the equivalence criteria and drives the sample size, so it has to be chosen carefully and agreed with the health authorities. It rests on statistical as well as clinical considerations: statistical significance concerns whether the observed result could occur by chance alone, and clinical significance whether it has important clinical, research or public health relevance. The comparability margins must be justified on clinical relevance, so that the margin is the largest difference in efficacy that would not matter in clinical practice. The margin is derived from the reference's past performance against placebo, often in the pivotal studies behind the reference's approvals, so that the biosimilar keeps an agreed proportion of the reference's effect size, usually 50% or more. The effect size is estimated as the lower bound of the 95% confidence interval for the difference between reference and placebo, and a meta-analysis is performed where several data sources exist.
Efficacy equivalence against the margin
Efficacy equivalence is shown when the confidence interval for the difference between the products on the primary endpoint lies entirely within the predefined margin. As of 2019, the FDA used a 90% interval and the EMA a 95% interval. In one example, the primary endpoint is PASI75, the proportion of patients achieving a 75% reduction in the Psoriasis Area and Severity Index, and the difference between the biosimilar and the EU reference is estimated as a risk difference against a margin of −18.0% to +18.0%. A risk difference of −2.5% with a 95% confidence interval of −10.0% to 5.1%, or any other interval entirely inside the margin, would then demonstrate equivalent efficacy.
The primary analysis set
The intent-to-treat set is the first choice for the primary analysis of a superiority or non-inferiority trial, whereas a biosimilar equivalence trial usually uses the per-protocol set. The per-protocol set should be used for the primary analysis of the primary endpoint, and the full analysis set for a secondary analysis. The per-protocol set is the cleanest population for avoiding a bias towards equivalence from the effect distortion that protocol deviations and imputed missing data cause. Efficacy data from patients with major protocol deviations may not show the product's own effect and are likely influenced by other factors. When such factors distort the results in both arms, any potential differences between the products' actual effects become harder to detect, which biases the comparison towards equivalence. Additional analyses on the intent-to-treat set are always required.
Missing data
The occurrence and imputation of missing data need special consideration in an equivalence comparison, so as not to bias the results towards equivalence. Imputing both arms with the same rule may lead to equivalence being falsely concluded from the imputation rather than from a similar therapeutic effect. For example, imputing every missing value as a non-responder would reduce the treatment effect in both arms and so reduce the difference between them. To counter this, the main analysis is usually based on the per-protocol set, which excludes patients with missing data. Sensitivity analyses should test how robust its conclusion is under different missing-data scenarios. One option is to impute the reference's missing data as responders and the biosimilar's as non-responders, and vice versa, as extreme-case scenarios. Alternatively, a tipping-point analysis can show the possible impact of missing data and which missing-data scenarios would tip the analysis so that it no longer demonstrates equivalent efficacy.
Switching designs for interchangeability
For an interchangeable, the Biologics Price Competition and Innovation (BPCI) Act requires the sponsor to show that the risk in safety or diminished efficacy of alternating or switching is no greater than the risk of using the reference product without switching. By 2017 the FDA therefore suggested a switching design, with at least three switches between test and reference products in the experimental arm. As of 2019, the phase III confirmatory study usually included switching between treatment groups. Because of the US concept of interchangeability, biosimilar studies in a global submission dossier then often had several study periods, with at least the originator group split into two, typically after the primary endpoint, in a so-called switching design. The interchangeability concept did not then apply in other regions, such as the EU.
Concepts carried over from new-drug development
As more guidelines on planning biosimilar trials become available, medical writers need to work closely with statisticians to decide which concepts from new-drug development can be applied to biosimilar development and which aspects need a different approach.