Background: High failure rates in Phase III trials are a major challenge in clinical research. Often, promising treatment effects in Phase II cannot be reproduced in the following Phase III study. This lack of reproducibility might be related to the two main differences between Phase II and Phase III trials: Phase II being conducted in a limited number of (selected) regions or sites – opposed to global phase III trials – and Phase III only being conducted after observing promising results in phase II (go/no-go decision rules). Both aspects can be sources of overoptimism/bias. Suitable methods for bias adjustment are needed. Additive and multiplicative adjustment was proposed for go/no-go related bias, and the approximate-Bayesian-computation (ABC) approaches were used for addressing different forms of selection bias. The objective was to define a quantitative approach for bias adjustment due to selection (region and go/no-go) for the Phase II treatment effect estimates and corresponding probability of success (PoS) estimates for phase III.

Method: The go/no-go decision after Phase II is based on observed treatment effects leading to a bias between the observed and true effect. In contrast to that, region selection in Phase II is performed before any patient data is collected. Therefore, to describe characteristics of the different sources of bias and to compare properties of the different effect estimates, an additional type of bias is introduced: the “true effect bias” between the selected regions’ true effect and the true overall effect (composed of all potential regions for phase III). In a simulation study, the different adjustment methods are examined according to the level of bias reduction in treatment effects and estimated PoS for three bias scenarios: Only decision rule, only region selection or both.

Results: Each of the adjustment methods correct for overestimation. Adjusted treatment effect estimates are generally preferable to non-adjusted estimates. While the additive and multiplicative adjustment methods are appropriate in specific combinations of the true overall effect, the number of events, the region selection, and/or the decision rule, the ABC-adjusted treatment effect estimates perform best across all combinations. The adjusted PoS shows less consistency over simulation scenarios.

Discussion: Stricter go/no-go rules increase bias but reduce (false) decisions to go to Phase III. In contrast to that, more extreme region selection increases bias and go decisions at the same time. Therefore, in drug development processes, the risk of a region bias should be addressed. ABC methods can adjust for region bias and seem useful for practice.

The two-trials rule requires "at least two adequate and well-controlled studies, each convincing on its own, to establish effectiveness". This is usually employed by requiring two significant pivotal trials and is the standard requirement by regulators before new drugs are approved. However, drug applications are often based on more than two trials and some alternatives and generalizations have been recently proposed to properly deal with this case, among them the harmonic mean chi-squared test (Held, 2020, doi: 10.1111/rssc.12410) and the 2-of-3 rule (Rosenkranz, 2022, doi: 10.1007/s43441-022-00471-4). Both can be even extended to more than 3 pivotal trials. I will compare the different approaches in terms of their statistical properties, with a focus on Type-I error rate, project power and expected sample sizes. Type-I error rates will be considered either under the intersection null hypothesis, where all studies are assumed to have a null effect, or the union null hypothesis, where only some of the studies have a null effect. The applicability of the different methods to combine evidence from clinical trials in conditional or accelerated approval will also be discussed.

Introduction: Formal proof of efficacy of a new drug requires that in a prospective clinical trial, superiority towards a placebo or an established standard is demonstrated. Traditionally one primary endpoint is specified, but several diseases (e.g., Alzheimer’s disease, cancer) exist where treatment can be based on the assessment of multiple primary endpoints. With two (or more) co-primary endpoints, significant superiority must be shown for both to claim study success. No adjustment of the study-wise type-1-error is needed in this scenario, but the sample size usually needs to be increased to maintain the pre-defined power. Recently, also trials that use the so-called dual primary endpoint concept have been proposed. Here, a Bonferroni split is applied to guarantee that the study-wise type-1-error is controlled at the pre-specified level. Study success is claimed as soon as superiority for at least one of the primary endpoints is demonstrated. This approach is sometimes also called the at-least-one concept. In contrast to co-primary endpoints, it is not covered in the European guideline on multiplicity because study success can be declared as soon as superiority for one primary endpoint is demonstrated, even if the other indicates a deterioration. This is obviously illogical and not in line with practical decision-making and thus may lead to post-hoc discussions on overall study success if results for one of the endpoints are ambiguous.

Methods: We investigate the situation with two primary endpoints where it may be sufficient to demonstrate superiority in at least one of them. In line with Röhmel et al., we argue that interpretational issues should be discussed upfront and examine several additional constraints to the dual primary endpoint concept to assure a statistically and clinically consistent strategy for decision-making. Our principal idea is first to require a certain “minimal requirement” for all primary endpoints before addressing the superiority hypothesis. Specifically, either the treatment effect estimate must be on the right side of zero, a positive trend is required for the primary endpoints, or non-inferiority to a pre-defined margin has to be demonstrated. Additionally, superiority to the control has to be shown for at least one of the primary endpoints. We performed a simulation study to examine our approach. The main intention was to compare our decision strategy to the dual primary endpoint concept regarding increased costs (in terms of power (and sample size)). Simulation scenarios included various treatment effects and two correlations for three different sample sizes.

Results/Conclusion: Our simulations illustrate that if the treatment effects are as planned, additional constraints to the dual primary endpoint concept lead to statistically and clinically consistent decisions on study success and improve interpretation with limited costs in terms of sample size. Moreover, our approach allows flexible modeling of the minimum requirements for all endpoints and leads back to the co-primary endpoint concept reflected in the European multiplicity guideline. In summary, our work emphasizes that the more aspects are discussed and pre-defined at the planning stage of a clinical trial, the better the certainty and interpretability of its results.

Many clinical variables such as biomarkers can change in a patient after initiation of a therapeutic treatment and can help identify patients who benefit most from the treatment. In a randomized controlled trial (RCT), often such post-randomization variables are only measured in the experimental treatment arm, and may contain a significant amount of missing data. Such a situation can be addressed by a principal stratum estimand strategy (ICH E9 addendum).

We present a novel weighted imputation regression approach (WRI) for this setting with missing data. The methodology and assumption required to yield valid causal inference are discussed. The good performance of WRI is demonstrated via a simulation study and its application to a large number of clinical trials to investigate treatment effect differences among patients developping or not anti-drug antibodies (ADA; the intercurrent event) to a therapeutic treatment.

Malaria is an infectious disease transmitted by Anopheles mosquitoes (the ‘vector’) and still kills more than half a million people globally each year. Reducing the human-vector contact (‘vector control’) by insecticide-treated nets and indoor residual spraying is the most effective public health measure to control malaria. While the death toll was almost halved since 2000, the global progress in malaria control has drastically slowed down since about 2015, presumably to a large extent due to gaps in vector control, such as outdoor transmission. Therefore, new vector control tools addressing outdoor transmission and settings difficult to reach with nets or spraying are urgently needed.

We implemented a randomized double-blind placebo-controlled field study in Ahero, western Kenya, to evaluate a transfluthrin-based spatial repellent (‘push’ intervention), an odour-baited trap (‘pull’ intervention), and the combined ‘push-pull’ package. The primary outcomes were outdoor and indoor human-vector contact, measured by human landing catches and and light-traps catches, respectively. We analysed the mosquito count data with Bayesian hierarchical, regression type models, with inference by Hamiltonian Monte Carlo (stan). After extensive model selection by leave-one-out cross-validation, we averaged two models jointly accounting for all possible dependencies by experimental design. As Bayesian data analysis isn’t yet common in this field, the publication puts emphasis on carefully explaining the modelling framework.

Mosquito count data is typically very variable and highly dependent on the small-scale local environment and weather conditions. Still, similar studies usually only report interval estimates of the intervention effect with respect to an average situation (average house and week), and thus do not cover the variability of the intervention effect across different houses and weeks. We advocate for a more robust intervention assessment and therefore present a complementary arbitrary house-week analysis with interval estimates (highest density credible intervals) reflecting the intervention effect to be expected for an unknown house and week in the field.

We could demonstrate a strong protective efficacy of the spatial repellent against indoor biting but failed to achieve any protection from outdoor biting malaria vectors, by either of the three intervention strategies. The reason for adding an attractive trap to the spatial repellent was to avoid diverting bites to others, and in fact, our results indicate that repelling vectors from the indoor space resulted in an increased biting in the outdoor space. Hence there remains an urgent need to further develop and evaluate odour-baited attract-and-kill approaches that can be effectively combined with spatial repellents for a push-pull intervention for malaria control.