Fundamental Principles of Clinical Trials
As a highly regulated industry, biomedical research must meet strict requirements that are encapsulated in the form of regulations, guidelines, and best practices. The principles underlying these requirements are principles of good science in general and concern the advancement of public health through state-of-the-science research, while safeguarding the rights and safety of study participants. In this section, we provide an overview of the statistical principles underpinning clinical trials intended for registration. Relevant literature on other aspects of medical research, including protection of trial subjects, investigator and sponsor responsibilities, quality assurance, and other operational requirements may be found elsewhere (see, e.g., International Conference of Harmonization 1996; World Health Organization 2002). Although clinical trials are typically classified into phases I to IV, differing mainly in scope, objective, and methodology, they all share certain fundamental principles in terms of prespecification of relevant aspects of the trial in a protocol or Statistical Analysis Plan (SAP), data processing and analytical approaches, subject safety protection, and other operational procedures. Further, they all require the institution of quality assurance processes to guarantee the integrity of the trial and the accompanying results. It is also a requirement that results of phases I-IV trials be appropriately summarized and interpreted in a study report,
highlighting the strengths and weaknesses of the trial findings, and their contributions to advance medical science and public health.
Phase I trials are primarily concerned with the assessment of the safety, tolerability, and pharmacokinetic (PK) profiles of the investigational agent using healthy volunteers. PK studies are especially conducted to assess how a drug is absorbed, distributed, metabolized, and excreted in a human body. In general, a key objective of Phase I research is the identification of the optimal doses for use in subsequent Phase II studies. The determination of a Maximum Tolerated Dose (MTD) with acceptable Dose Limiting Toxicity (DLT) is typically achieved using either a rule-based or a modelbased design. An example of the former is the conventional 3 + 3 cohort expansion design, in which dose escalation or de-escalation decision is made with reference to the toxicity observed on the current group of three study participants assigned to a dose. Specifically, the dose is lowered if the current dose is considered toxic; otherwise, the current dose is maintained or escalated to the next level. On the other hand, a model-based approach, such as the Continual Reassessment Method (CRM), involves fitting a dosetoxicity curve to estimate the MTD. Estimation of parameters associated with the curve may be performed either via the Maximum Likelihood (ML) or Bayesian framework. Although Phase I trials are relatively less complex than other phases, they can still pose certain operational and analytical challenges. Further, while the primary ethical consideration is to ensure that participants are not exposed to unsafe levels of the study drug, this assessment still involves a difficult benefit-risk judgment. For example, in trials involving cancer immunotherapy, the traditional Phase I paradigms appear to be challenged, especially with respect to determination of patient eligibility, as well as the MTD and PK profiles of the drug (Postel-Vinay et al. 2016).
In contrast, Phase II trials are typically conducted in a sample of the target patient population and aim at establishing preliminary evidence of efficacy and safety, and, in some cases, the optimal dose ranges. Phase II trials are often planned as two subphases (Phase IIA and IIB), depending on the complexity of the objectives and disease area. For efficiency reasons, Phase IIA trials may employ a single arm and a historical control group. This approach, while attractive, requires care to ensure minimization of bias associated with confounding factors and heterogeneity (Lara and Redman 2012). Phase IIB trials are intended to provide more definitive evidence to inform decisions to proceed into Phase III. As a result, a Phase IIB trial usually involves randomization and multiple arms, with a suitable control group. In oncology, Phase II trials are customarily conducted using flexible, multistage designs, with prespecified minimax criteria (Kramar et al. 1996). Notably, while in a typical optimal two-stage design the goal is to minimize the expected sample size under the null hypothesis, a minimax design targets minimization of the maximum sample size.
Phase III trials, generally conducted using randomized, double-blind designs, are intended to provide relatively more definitive results relating to the short- and long-term safety and efficacy of the drug under investigation. Phase III trials are often referred to as confirmatory trials and are the critical trials in the regulatory approval process. Accordingly, these studies tend to be larger than the corresponding Phase I or II trials, and may pose additional operational and analytical complexities, depending on the number of participating sites, nature of study endpoints, and study procedures. When they are conducted in multiple centers, the trials require well-established processes to ensure consistency of study conduct across the various sites. In particular, to ascertain the interpretability of data pooled from the various centers, certain measures may need to be put in place, including periodic training of site personnel and establishment of committees charged with providing guidance for data collection, study endpoint evaluation, or other aspects of the conduct of the study.
With the ever-rising cost of drug development, prolonged development time, and dwindling number of new medicines achieving marketing authorization, there has been increased focus on novel approaches to clinical development and trial design. This has particularly been the centerpiece of the Food and Drug Administrations (FDA) Critical Path Initiative, launched with the aim of improving efficiency and reducing attrition rates. Proposed approaches under that framework include use of historical data, modeling and simulation, and trial designs, such as the seamless Phase II/ III trials, having greater degrees of flexibility than those employed traditionally. Although these novel approaches have tremendous potential to generate reliable information in a speedy manner, their implementation may require substantial quantitative work and operational efforts. Consequently, while there are cases of successful applications, routine use of the approaches has not yet been fully realized. In the traditional paradigm, the efficacy of the drug in the targeted population along with the manner in which it can be safely administered should be established at the end of Phase III. The drug developer then submits the entire body of research in a New Drug Application (NDA) to the regulatory agencies.
After a drug is approved and marketed, Phase IV trials are conducted to gather additional data on safety or to understand further the therapeutic value or alternative treatment strategies of the drug. In some instances, postmarketing studies may be conducted for a new indication or label enhancement. To achieve such objectives, which may also include claims about a new dosage or strength of the drug, or the way the drug is manufactured, the drug developer may file the so-called supplemental new drug application or sNDA.
When the goal is to obtain data about a drugs effectiveness, as used in the real world, or to evaluate resource utilization, non-interventional (N1) studies are carried out in a naturalistic setting. In such studies the drug is typically prescribed in accordance with the approved label, and the healthcare providers perform only procedures required in routine practice. Thus, the effectiveness of a drug evaluated through N1 studies can complement the efficacy and safety data gathered in the restricted randomized controlled trials (RCTs) conducted for preapproval review by regulatory bodies. Other options include Phase IV studies that are intended for Post-Marketing Surveillance (PMS) purposes, per regulatory requirement, and disease registries, which involve patients with common characteristics and collect ongoing data over time on selected outcomes of interest. These prospective observational studies can provide valuable data on various aspects of the drug, including safety and effectiveness, but require caution in the interpretation of the results. A major drawback of such studies is the potential for bias emanating from lack of randomization. In certain situations, large simple trials (LSTs) or pragmatic trials may be conducted as a hybrid between an RCT and N1 study (Maclure 2009; Patsopoulos 2011; Roehr 2013; Mentz 2016).
When it is desired to enhance administrative efficiency or minimize investigator bias, cluster randomization trials, in which the randomization unit is not the individual subject, but groups defined by suitable criteria, e.g., clinics or communities, have been proposed as a viable option (Donner and Klar 2004). From a statistical perspective, a major drawback of such studies is the loss in statistical precision relative to the corresponding RCT with the same number of subjects. Further, the analysis of data from such studies requires caution, since the techniques will need to take into account the intracluster correlation.
To ensure that the safety of study participants is protected and that the trial achieves the desired outcome, the sponsor should seek input from all applicable stakeholders, including patient advocates, key opinion leaders, and drug regulatory bodies. In addition, reference should be made to relevant guidance documents issued by regulatory agencies, especially when considering nontraditional trial designs or novel analytical approaches. In certain situations, it may also be essential to establish Data Monitoring Committees (DMCs) with a mandate to periodically assess the safety and scientific validity and integrity of clinical trials, or to make appropriate recommendations for changes in trial design or duration (Ellenberg et al. 2002; US FDA 2006).
In the rest of this chapter, we highlight some of the general statistical considerations that should be taken into account in the design, analysis, and reporting of clinical trials. In subsequent chapters, detailed descriptions of key statistical concepts will be provided, with particular reference to their importance in regulatory review and approval.