In the past, most phase III randomised clinical trials tested whether a new treatment was better than the current standard of care treatment in terms of a specific objective, called the primary outcome. The primary outcome was, for example, time to death, disease progression, or relapse within a certain time-window.

In recent years, more and more potential novel treatments are not expected to do better than existing ones in terms of these primary outcomes, but only to do equally well. This is often considered enough to warrant their use. This could be because they have secondary advantages, for example reduced costs, reduced toxicity or less burden to patients, or simply because they provide an additional option that clinicians could use in situations where the standard treatment was not available.

The goal of non-inferiority trials is to test whether a new treatment is similar in effectiveness to an existing one. This is done by collecting evidence that the new treatment is not worse than the current one by a substantial amount or more. This is referred to as the non-inferiority margin.

Non-inferiority trials often require a large number of participants to show that the new treatment has the current standard of care, in terms of the primary outcome. The primary outcome might be an event, e.g. death or disease progression. Therefore, it is important to guarantee that these trials are carefully designed, anticipating any issue that could affect their efficiency and success.

Our work focuses on making the design of non-inferiority trials resilient to unexpected differences between the observed risk of the primary outcome event and what had been expected. For example, when the event risk in patients getting the current treatment turns out to be very different in a trial from what was expected, standard methods can lead to reduced statistical power (a lower probability of the trial giving a correct positive result) or lack of interpretability (a technically “positive” result that is not interpreted as such by most readers).

We also created the ‘non-inferiority frontier’. This is a system that defines the best possible non-inferiority margin for each possible value of event risk in the control treatment, rather than just at the expected level.

We have proposed ways of choosing a non-inferiority frontier and using it to analyse the trial when the primary outcome is binary, for example, yes/no or death/survival. These methods have already been implemented in three of our HIV trials: D3, Breather+ and LATA.

We are extending this work to the time-to-event framework. In this case, it has implications in terms of the choice of measure to quantify the effect of the treatment. Most trials use the hazard ratio measure, which can be difficult to interpret and is not always the most efficient measure to look at. We showed that using difference in, i.e. the average survival time in a pre-defined time window, can often be more powerful. We are implementing this in our PATCH trial in prostate cancer.

The results from non-inferiority trials can be difficult to interpret for non-specialists, and the meaning of the non-inferiority margin is often not well understood. We have improved the interpretability of trial results by using Bayesian statistics to present the results in a more intuitive way. Our ACCEPT (ACceptability Curve Estimation using Probability Above Threshold) analyses provide the probability that a treatment is better or worse by different amounts when compared to an existing treatment. We have also developed guidelines on how to design a non-inferiority trial when planning to use Bayesian methods, particularly with uncommon conditions and rare events.

By improving the efficiency and resilience of non-inferiority trials, we are making sure our research can deliver results requiring fewer participants and benefitting the wider patient population more quickly, maximising the probability that each non-inferiority trial we design gives meaningful results.

**Publications:**

- Handling an uncertain control group event risk in non-inferiority trials: non-inferiority frontiers and the power-stabilising transformation (2020)
- The Smooth Away From Expected (SAFE) non-inferiority frontier: theory and implementation with an application to the D3 trial (2023)
- A comparison of different population-level summary measures for randomised trials with time-to-event outcomes, with a focus on non-inferiority trials (2023)
- Improving clinical trial interpretation with ACCEPT analyses (2022)
- Practical approaches to Bayesian sample size determination in non-inferiority trials with binary outcomes (2023)