Background Surrogate outcomes are measures of treatment effect that can be used to predict treatment effect on the true outcome of interest. Surrogates are valued as they can be used in place of true outcomes to reduce the length, size, or intrusiveness of a clinical trial. However, validation of surrogacy is a conceptually complicated area and much theoretical and practical statistical development has been conducted in recent years. Methods A systematic review was conducted to identify which surrogate evaluation approach was best suited to be extended to ordinal outcomes. I extended a foremost approach to the case where the surrogate, the true clinical outcome, or both are ordinal outcomes. This extension investigated surrogacy at both the trial and individual levels; trial level surrogacy was based on a two stage method. The extension was developed through large simulation studies and used to investigate whether deep venous thromboembolism (DVT) was a surrogate for the ongoing measure of death and disability the Oxford Handicap Scale (OHS), using data from the stroke trial CLOTS3. CLOTS3 was a large multi-centre randomised clinical trial which investigated whether intermittent pneumatic compression (IPC) applied to the legs reduced the occurrence of deep venous thromboembolism (DVT) in stroke clinical trial patients. Results The systematic review identified the information theory approach as the most intuitively and practically worthwhile approach to surrogacy evaluation. I extended this approach to: a binary surrogate and ordinal true outcome (the binary-ordinal setting); the ordinal-binary and the ordinal-ordinal settings. The simulation studies showed that the approach worked well in most scenarios tested. However, trial level surrogacy was impacted by loss of efficiency due to the use of the two stage method. Bias imposed at the trial level by separation of discrete outcomes was effectively dealt with using a penalised likelihood method. The information theory approach for ordinal outcomes identified no surrogate that would predict treatment effect of IPC on the true outcome OHS measured at six months in the stroke trial CLOTS3.

Accelerating regulatory science has become an urgent task due to the ongoing emergence of novel and highly complex diseases, such as coronaviruses and mpox, and the growing real-world evidence and precision medicine paradigms. In research and development, achieving this goal requires the formulation and implementation of innovative clinical trial designs and statistical approaches. Further efforts are needed to devise robust strategies toward expediting the evaluation of drugs and other medical products, which will help to quicken the delivery of safe and effective treatments to individuals in need. One pioneering concept that has greatly contributed to speeding up product assessment is the application of pre-specified adaptations in clinical trials. Adaptive techniques (like early dose or treatment selection and sample size recalibration) are based on participant data collected during the trial, where investigators/sponsors are able to learn from the study population and adjust trial characteristics accordingly without diminishing the integrity of the study. However, adaptations must be employed cautiously; it is essential to ensure that they are both clinically reasonable and statistically sound (in the sense that they will not inflate the nominal type I error rate). The evolution of seamless trial designs has also propelled drug/medical product evaluation. For instance, seamless trials that combine a phase II study (e.g., using a biomarker or short-term intermediate endpoint for early dose or treatment selection) and a phase III study (e.g., using a definitive endpoint for confirming efficacy) have become very popular in practice due to their flexibility and shorter duration compared to running separate studies. In phase II/III seamless trials, a correlation is usually pre-specified between the intermediate and definitive endpoints in order to bridge the phase II and phase III studies and accomplish the overall aims of the intended trial. However, this can be difficult to quantify, especially when the endpoints are non-continuous. An inappropriate or unsubstantiated enumeration of this correlation can yield flawed results and thus misguide approval and labeling decisions, leading to potentially serious consequences. To avert this issue, a three-stage phase II/III transitional seamless trial design was constructed that does not demand prior knowledge of the correlation between the intermediate and definitive endpoints. The design also allows for interim sample size re-estimation according to the accrued intermediate endpoint data. The utility and validity of the design, with and without sample size adaptation, were considered for different types of endpoints. For each, theoretical proofs establish that the design can control the nominal type I error rate while still reaching the targeted power, and simulations reinforce these findings. Therefore, the design exhibits promise in precipitating the assessment of experimental interventions in a wide range of therapeutic areas. This dissertation is organized as follows. First, Chapter 1 presents (1) an introduction to the drug/medical product evaluation process, (2) an overview of adaptive techniques used in clinical trials, (3) a description of seamless trial designs, including a real-world example to demonstrate how they can be applied in different data situations, (4) the research goals, and (5) a display of the proposed three-stage phase II/III transitional seamless trial design and its workflow. Chapters 2-5 lay out the models and procedures for the proposed design when using a continuous intermediate endpoint and a continuous definitive endpoint, a binary intermediate endpoint and a continuous definitive endpoint, and a binary intermediate endpoint and a time-to-event definitive endpoint in the presence and absence of censored observations. This is proceeded by statistical justification and supporting simulations for each procedure. As some authors remain skeptical toward the use of adaptations in clinical trials, Chapter 6 counters an argument posed in the literature. In closing, Chapter 7 summarizes the work and its limitations and provides several additional considerations toward enhancing the proposed design and its generalizability.

Platform trials evaluating multiple treatment arms against a shared control are an efficient alternative to multiple two-arm trials. Motivated by a randomized clinical trial of the effectiveness of two neuroprotection devices during aortic valve surgery against a control, this dissertation addresses two open questions in the optimal design of these trials. First, to explore whether multiplicity adjustments are necessary in a platform design, simulation studies evaluating the operating characteristics of platform designs relative to independent two-arm trials were conducted. Under the global null hypothesis, relative to a set of two-arm trials, we found that platform trials have slightly lower familywise error; however, conditional error rates for an experimental treatment being declared effective given another was declared effective are above the nominal alpha-level. Adjusting for multiplicity reduces familywise error, but has little impact on conditional error. These studies show that multiplicity adjustments are unnecessary in platform trials of unrelated treatments. Second, to determine the optimal approach for comparing delayed entry arms to the shared control, five methods for incorporating historical controls into two-arm trials were applied to the analyses of simulated open platform trials and compared to pooling all controls. We found that when response rates are constant, pooling yields the lowest error and most precise, unbiased estimates. However, if drift occurs, pooling results in type I error inflation or deflation depending on the direction of drift, as well as biased estimates. Although superior to naive pooling, none of the alternatives explored guarantee error control or unbiased estimates in the presence of drift. Thus, only concurrent controls should be used as comparators in the primary analysis of confirmatory studies. Finally, these findings were applied to assess the design and analysis of the neuroprotection trial.

<p>Allocation concealment is the process of implementing the randomization sequence in a manner that prevents foreknowledge of upcoming group assignments. It protects against preferential enrolment of study participants, which could disrupt the prognostic balance that randomization aims to create in the first place. Envelopes are one method perceived by clinical trial authorities to adequately conceal allocation, despite evidence suggesting otherwise. We do not believe that envelopes are adequate, and we wanted to know the extent to which our sentiment resonated within the clinical trials community. We administered an internet-based survey to a random sample (n=1,926) of corresponding authors of recently published randomized clinical trials (RCTs). We sent non-responders up to two e-mail reminders starting from two weeks after the original invitation. We received 490 complete surveys (25.4% response rate) after collecting data for seven weeks. Most participants (61%) preferred central randomization to conceal allocation, yet a majority (64%) also accepted that envelopes are adequate. After they were shown examples that suggested envelopes’ vulnerability, 11% of participants shifted their preference away from envelopes and 38% of participants became less accepting of envelopes. Compared to their initial ratings and after they were shown the examples, significantly more participants (69%) preferred central randomization (p</p> / Master of Science (MSc)