Benefit Risk Assessment: Methodology

Organizers: Richard Payne (Lilly), Yeh-Fong Chen (FDA)
Chair: Yeh-Fong Chen (FDA)
Vice Chair: Richard Payne (Lilly)

Saurabh Mukhopadhyay (Abbvie)
Lan Huang (FDA)
Qian Helen Li (BMS)


Title: Benefit-Risk Assessment Using Bayesian Discrete Choice Experiment
Speaker: Saurabh Mukhopadhyay (AbbVie)

Early assessment of benefit-risk balance of a treatment is very important in a drug development program to understand the medicinal product’s utility in a population of interest. Benefit-risk assessment about treatments however are often complex and involve tradeoffs between multiple, sometimes conflicting, assessments of benefits and risks. Discrete choice experiments are used in health outcomes research to assess tradeoffs in preferences, but they often impose a high cognitive burden to assess multiple attributes and the requirement for a large pool of respondents.
A novel Bayesian framework that borrows strength from respondents will be discussed that allows to conduct a discrete choice experiments with only a limited number of respondents. Furthermore, will discuss how this framework only requires respondents to choose from a few pairs of profiles to state their preferences, thus drastically reducing the cognitive burden. Specifically, a hierarchical Bayes benefit-risk (HBBR) model and an associated discrete choice experiment will be leveraged in which, by design, each respondent needs to evaluate only a fraction of all choice pairs; thus, respondents would not become fatigued from a long questionnaire. In addition to making the survey task operationally efficient as described above, will discuss and illustrate with a pilot experiment in an oncology setting how this framework is also expected to produce very high-quality preference data. Ultimately, patients are the most important voice in the benefit–risk balance. Therefore, using a simulated data and an augmented model it will be further shown how to incorporate patients’ characteristics to obtain a more precise estimate of benefit-risk preferences. An R-package developed and available on CRAN for this purpose will be utilized for implementation of the HBBR model.

Title: Benefit-Risk Assessment for Binary Diagnostic Tests using Frequentist and Bayesian approaches
 Speaker: Lan Huang (FDA)

In diagnostic device evaluation, it is important to have an integrated Benefit-risk (BR) assessment for safety and effectiveness, which is not same as the assessment for drugs and therapeutic devices. Correct diagnosis does not lead to direct clinical outcome such as longer survival, release of symptoms, tumor shrinkage, etc.; but leads to the proper treatment in time while incorrect diagnosis may result in serious consequences of unnecessary tests and wrong treatments. Some common measures used in evaluating the accuracy of a diagnostic device include sensitivity, specificity, positive and negative predictive values (PPV and NPV), positive and negative likelihood ratios (PLR and NLR). Here, we will discuss a new BR measure by incorporating information from true positive and true negative cases (correct diagnosis/benefit), and false positive and false negative cases (incorrect diagnosis/risk) for facilitating the necessary decision making. Three decision rules are considered depending on the purpose of the clinical study. Different statistical models will be used for estimating the BR measure for data obtained from a clinical study enrolling all-comers or using enrichment design strategy with known positive cases and negative cases. One can derive the point and interval estimators for the BR measure using frequentist’s methods. In addition, the Bayesian point estimate and credible interval for the BR measure can also be derived.  For illustration, the proposed methods are applied to two examples with diagnostic devices for cancer screening and for preterm delivery evaluation.

Title: Dependency of Efficacy and Safety Responders – A New Metric for Risk-Benefit Assessment
Speaker:  Qian Helen Li (BMS)

The risk-benefit profile of a treatment plays important role in patients’ treatment decision. Many statistical methods have been developed to quantify the risk-benefit profiles. The approaches include multicriteria decision analysis, Quality-adjusted Time Without Symptoms and Toxicity, and etc. Those methods help understand certain aspects of the risk-benefit relationship of a treatment, may not answer all questions. It is often desirable to understand the correlation of certain efficacy and safety responders. For example, if patients who are efficacy responders are also experiencing safety events. Are such occurrences correlated incidences or random coincidence? In this presentation, we propose a simple statistical metric that can quantify the relative dependency of the efficacy and safety events. This metric is easy to interpret and understand. The properties of this metric are evaluated for bias, variability and sample size requirement using simulations.