Statistical Significance, Clinical Meaningfulness, and Regulatory Consideration for Rare Disease Drugs' Approval - Examples Sharing
Chair: Yeh-Fong Chen (FDA) and Huan Wang (FDA)
Speaker: Hengrui Sun, PhD (FDA)
Title: Finding right therapies under a public health emergency – A story from Ebola drug development
Abstract: Ebola virus disease is a rare and often deadly illness caused by the infection with ebolaviruses that are found primarily in sub-Saharan Africa. Infected patients normally start with general symptoms followed by decreased liver and kidney functions, and even unexplained internal and external bleeding. Mortality rate for the infected patients could range between 20% to 90%. Since Ebola was first identified, intermittent outbreaks occurred with the largest and the most serious outbreak in West Africa from 2013 to 2016. To respond to this public health crisis, scientists started the Partnership for Research on Ebola Virus in Liberia II (PREVAIL II) trial to study ZMapp for treating Ebola. PREVAIL II was a randomized, controlled trial with an adaptive design. However, the enrollment for PREVAIL II stopped prematurely due to the waning of Ebola epidemic. This made it difficult to reach a definitive conclusion about the efficacy of ZMapp. Two years later, when another Ebola outbreak occurred in the Democratic Republic of Congo, PALM trial, which had a platform design, randomized patients to ZMapp and two other investigational agents initially, and then allowed a fourth investigational agent arm to enter on a later date. Two of the investigational agents showed superiority to ZMapp and were later approved by the FDA.
During this presentation, the speaker will discuss some interesting design features of PREVAIL II and PALM in the context of a major public health emergency as well as under a rare disease setting, such as feasibility of conducting a randomized controlled trial, utilizing master protocols, Bayesian design and analyses, and interim stopping criteria for efficacy and futility. The PALM trial will also be perused in light of the newly issued FDA draft guidance of Master Protocols for Drug and Biological Product Development.
Speaker: Qing Liu, PhD (Quantitative & Regulatory Medical Science, LLC), Lisa Jiao, PhD (Mandos, LLC.)
Title: On Survival Analysis with Zero or Very Few Events for Patients Receiving a New Treatment
Abstract: Survival analysis usually relies on time-to-event data arising from a counting process for which the information is based on the number of events. We are interested in medical research to compare the survival of patients receiving a new treatment with the survival of patients receiving a control (e.g., placebo or standard-of-care). The most often used methods of inference are derived from the Cox proportional hazards model. These event-based methods run into serious problems when patients receiving the new treatment have zero or very few events (e.g., 1 or 2 events). We propose to measure survival of an individual patient via a binary outcome variable of being alive or died at the last-follow-up time. We develop a many-to-1 matched analysis by which the last-follow-up time for each patient receiving the new treatment is matched by only patients receiving the control whose survival status (alive or died) at the last-follow-up time of the treated patient being matched can be determined. Thus, conditional on the matched patients, the survival (alive or died) of the treated patient being matched is a binary outcome variable for which the probability of being alive is the observed survival rate of the matched control patients. For the vector of binary survival outcomes of all treated patients where matched controls exist, we consider the test statistic based on an inverse probability weighted mean of binary outcomes. Since the binary outcomes are independent, conditional on the observed survival rates of the matched controls, the exact null distribution of the test statistic can easily be obtained by MonteCarlo simulations. Because the patients receiving the controls are used for each treated patient in the many-to-1 matched procedure, their survival rates corresponding to individual treated patients are not independent. Therefore, a bootstrap method is used to quantify the variability of the inverse probability weighted survival rate for the control group. This leads to a two-sample test procedure to compare the inverse probability weighted survival rates between the treatment and control groups. The many-to-1 procedure can easily be extended to allow matching based on an index time to eliminate immortal time bias, a form of selection bias first recognized by Gail (1972) by which patients need to survive long enough to receive a treatment of interest whereas there is no wait time for patients receiving a control. The immortal time bias arises very often in comparative analysis involving natural history controls. Immortal time bias is also present in randomized controlled trials for settings where patients are randomized to a treatment group and a control group. However, for patients randomized to the treatment group, the treatment of interest for individual patients is not available at the time of randomization, rather each patient has to go through a selection process of meeting certain outcome dependent criteria as an integral part of the treatment of interest. The presence of this outcome dependent selection process has the duo consequences of elimination of patients who do not meet the required criteria from the intentto-treat population and enriching patients who receive the treatment of interest with favorable prognosis for treatment outcomes. Examples include but are not limited to scenarios where the treatment of interest 1) is not immediately available at the time of randomization (e.g., CART-T for cancer), 2) is contingent upon patient being successfully complete a pre-treatment procedure (e.g., surgery) during an initial run-in period, 3) requires a predicative biomarker of individual patients reaches a certain predetermined level (e.g., CD4 counts in HIV treatment), and 4) is individualized according to efficacy and safety comes for patients have successfully complete an initial treatment (e.g., individualized maintenance therapy following the results of an induction therapy in cancer). The proposed methodology is applicable to other many clinical trial settings where a current event (e.g., disease progression) needs to be confirmed by future data. This violation of the counting process has so far been ignored and the validity of statistical inference with existing statistical methods is largely unknown. The proposed methodology also addresses issues in certain applications where censoring is informative and potentially differential between the treatment and control groups.
Speaker: Xiaofei Wang, PhD (Duke University)
Title: Challenges and Statistical Issues in Externally Controlled Trials for Drug Evaluation in Rare Diseases
Abstract: The utilization of historical control data in ongoing clinical trials, aimed at enhancing the efficiency of drug evaluation in rare diseases, has garnered considerable attention and support. This talk delves into the challenges and statistical intricacies inherent in externally controlled trials, with a focus on the unique constraints posed by small sample sizes in rare disease clinical trials. The discussion encompasses established approaches such as regression-based models, propensity score matching, inverse propensity weighting, calibration weighting, and their augmented estimators. Novel methods accommodating hidden confounders will also be discussed. The identifiability assumptions of these methods and their operating characteristics, such as bias, standard deviation (SD), mean squared error (MSE), power, and the issue of type I error rate, in clinical trials of small sample sizes will be discussed. The statistical issues will be illustrated with standard clinical trials with external historical controls, as well as multi-stage adaptive trials with external historical controls and enrichment of biomarker-defined subgroups.
Speaker: Xiaoyu Cai, PhD (FDA), Huan Wang, PhD (FDA) and Yeh-Fong Chen, PhD (FDA)
Title: Case Sharing for Assessing Collective Evidence from Hematological Trial Review Experience
Abstract: To establish effectiveness of a new drug, the FDA guidance for industry Providing Clinical Evidence of Effectiveness for Human Drug and Biological Products recommends at least two adequate and well-controlled studies, each convincing on its own, to be conducted. Considering unmet medical need, the FDA can grant flexibility within the limits imposed by the congressional scheme, broadly interpreting the statutory requirements to the extent possible where the data on a particular drug were convincing. In special situations, two following alternative options can be adopted:
• One adequate and well-controlled large multicenter or otherwise particularly persuasive trial that is the functional equivalent of two trials, or
• One adequate and well-controlled clinical investigation supported by confirmatory evidence.
For rare diseases, in particular, conducting two adequate and well-controlled studies or a single adequate and well-controlled large multicenter study may not be feasible due to limited sample size. On the other hand, some drugs may need to be approved only for a subpopulation of a disease and the requirement may be different and case by case.
In this presentation, a couple of case examples regarding challenges and unique consideration for efficacy evaluation from rare hematological disease trials will be shared and thoroughly discussed for learning purpose.
