Considerable research suggests that regression models with random effects can be used to establish a solid paradigm for the construction of the mathematics and statistics of personalized medicine research and practice, especially in the treatment of chronic diseases. The fact that generalized linear mixed models (GLMMs) have concepts that allow describing patient populations as a whole (the fixed effects) and, simultaneously, concepts that allow describing patients as individuals (the random effects) suggests that these models contain the key ideas for providing personalized medicine with a rigorous mathematical language. Underlying this is the belief that the variability of a random coefficient is not just a mathematical artifact to control for patients heterogeneity, but also the result of real variation in the biological and environmental factors that have made humans develop as individuals. Moreover, solid empirical and theoretical work by the Sheiner School of Pharmacology and others shows that a combination of mixed models with empirical Bayesian feedback (EBF) can be employed successfully in pharmacotherapy individualization and that EBF is well anchored in standard decision theory. Thus, both biological and mathematical arguments support the development of methodological instruments for personalized medicine based on GLMMs.
The objective of this half-day course is to introduce the main ideas of generalized linear mixed models, making emphasis on the interpretations from a personalized medicine viewpoint. Pharmacological applications taken from the extensive professional experience of the instructor will be shown. These applications will include: 1) methods to measure the individual benefit of medical or behavioral treatments; 2) analyses of bioequivalence studies; 3) the study of drug-drug interactions with patient samples, including the examination of the inducing or inhibiting effects of comedications; and 4) the utilization of mixed models in drug dosage individualization. Initially a historical account will be presented that will show how the original idea that random effects models are the key to developing personalized medicine can be traced back to pharmacological and genetic research developed in the second half of the past century. Examples of data analyses with the SAS and Stata computer packages will be shown.
Some of the applications will be taken from the following publications authored or coauthored by the instructor. A more complete reference list will be provided in the course class notes:
1. Diaz, F.J. Measuring the Individual Benefit of a Medical or Behavioral Treatment Using Generalized Linear Mixed-Effects Models. Stat Med 2016; 35:4077-4092.
2. Diaz FJ, Yeh H-W and de Leon J. Role of statistical random-effects linear models in personalized medicine. Curr Pharmacognomics Person Med 2012; 10:22-32.
3. Diaz FJ and de Leon J. The mathematics of drug dose individualization should be built with random effects linear models. Ther Drug Monit 2013; 35:276-277.
4. Diaz FJ, Rivera TE, Josiassen RC, et al. Individualizing drug dosage by using a random intercept linear model. Stat Med 2007; 26:2052-2073.
5. Diaz FJ, Cogollo M, Spina E, et al. Drug dosage individualization based on a random-effects linear model. J Biopharm Stat 2012; 22:463-484.
6. Diaz FJ, Berg MJ, Krebill R, et al. Random-effects linear modeling and sample size tables for two special cross-over designs of average bioequivalence studies: the 4-period, 2-sequence, 2-formulation and 6-period, 3-sequence, 3-formulation designs. Clin Pharmacokinet 2013; 52: 1033-1043.
7. Diaz FJ, Eap CB, Ansermot N, et al. Can valproic acid be an inducer of clozapine metabolism? Pharmacopsychiatry 2014; 47:89-96.
8. Botts S, Diaz FJ, Santoro V, et al. Estimating the effects of co-medications on plasma olanzapine concentrations by using a mixed model. Prog Neuro-Psychoph 2008; 32:1453-1458.
9. Diaz, F.J., Santoro, V., Spina, E., Cogollo, M., Rivera, T.E., Botts, S., de Leon, J. (2008). “Estimating the size of the effects of co-medications on plasma clozapine concentrations using a model that controls for clozapine doses and confounding variables.” Pharmacopsychiatry, Vol. 41; pp. 81-91.
Statisticians are extremely effective at analyzing data, performing simulations, and generating pages upon pages of analysis results. Despite their analytical prowess, however, statisticians continue to struggle to communicate the story hidden within the data to their colleagues. First and foremost, with the high cost of conducting translational clinical research, it is common to collect as much data as possible on as many endpoints as possible. This phenomenon is further reinforced due to our limited understanding of biological mechanisms and pathways, including the potential genomic underpinnings of a disease or treatment response. For example, we may have a clear understanding for how a novel therapy induces an efficacious response, but there is typically limited knowledge into the downstream effects of the drug to other body systems. A second challenge to communication lies in the increased use of sensitivity analyses to assess the consistency and robustness of results to varying assumptions. Given the volume of data to review and the variety of analyses to perform, it should come as no surprise that clear insight is often out of reach. In this environment, the traditional means of data summary – tables and listings – are ineffective for gaining insight; visualization is the key to effective communication for the modern statistician. Ben Shneiderman stated that “the purpose of visualization is insight.” Therefore, the goal of this short course is to describe data visualization techniques to aid in the understanding and communication of results from applications in clinical trials and genomics research.
Futility analyses (FA) are increasingly utilized in clinical trials. FA involves interim evaluation of the trial’s primary hypothesis to determine if there is a low probability of a positive result with trial continuation, or if the desired clinically meaningful effects can already be ruled out with reasonable confidence. FA can improve resource efficiency by the halting of trials with ineffective interventions and enabling sponsors to redirect efforts to more promising pursuits. FA also have ethical advantages in that fewer trial participants may be exposed to ineffective and possibly toxic interventions, and public health advantages in that trial results may be conveyed to the medical community in a more timely fashion.
FA should be carefully planned during trial design and described in the protocol as there are important statistical and operational consequences. Concerns include the control of statistical error rates and the concern for operational bias resulting from interim evaluations. There are varied and expanding statistical tools available for FA. Challenging questions arise during trial design regarding how FA should be conducted, a threshold at which futility would be established, and when futility should be assessed. Non-constancy of effect size and familiar limitations of accruing interim data can raise further challenges.
Data Monitoring Committees play a pivotal role in futility evaluation. Ensuring DMC access to appropriate data, ensuring DMC member understanding of futility methodologies, and thoughtful and efficient DMC reports describing FA are important for optimal recommendations. In this course, we will describe current practices and recent advances in methodological approaches and procedural issues, and illustrate with examples and case studies. We describe what FA are, why they are conducted, where and when they should be considered, and how they should be methodologically and operationally conducted.
Drug development has rapidly been globalized. Multi-regional clinical trial (MRCT) for regulatory submission has widely been conducted in the ICH and non-ICH regions. Regulatory agencies currently face challenges in evaluating data from MRCTs for drug approval. In order to harmonize points to consider in planning/designing MRCTs and minimize conflicting opinions, an ICH working group was established in late 2014 to create an international guideline for MRCT (ICH E17).
In September 2016, the US FDA announced the draft guidance entitled ‘‘E17 General Principles for Planning and Design of Multi-Regional Clinical Trials’’. The draft guidance describes general principles for planning and designing multi-regional clinical trials (MRCT). MRCTs conducted according to the guidance will investigate treatment effects in overall populations with multiple ethnic factors (intrinsic and extrinsic factors as described in the ICH guidance entitled ‘‘E5 Ethnic Factors in the Acceptability of Foreign Clinical Data’’). This half-day short course will (1) review regulatory history in ICH and non-ICH regions regarding the MRCT; (2) describe the key contents in the draft ICH E17 guidance; (3) discuss the statistical methodologies in designing MRCTs; and (4) illustrate relevant concepts using case studies.
Bayesian adaptive trial designs have drawn a tremendous amount of attention from industry, academia and government, and are increasingly used in practice. These designs have great potential to improve clinical trial ethics and increase the success rate and efficiency of clinical trials. However, due the newness of such designs, practitioners are less familiar with these methods, especially how to use them in practice. This course will introduce novel Bayesian adaptive designs, with a special focus on immunotherapy and drug combination trials, and illustrate the methodologies with real-world examples. More important, the course will provide a step-by-step tutorial to show attendees how to use R and other freely available software programs to design real-world clinical trials, thereby giving attendees a hands-on experience.
This short course will provide an exposition on health measurement scales – specifically, on patient-reported outcomes – based on the recently published book "Patient-Reported Outcomes: Measurement, Implementation and Interpretation" (Cappelleri et al., Chapman & Hall/CRC Press, December 2013). Some key elements in the development of a patient-reported outcome (PRO) instrument will be noted. Highlighted here will be the importance of the conceptual framework used to depict the relationship between items in a PRO instrument and the concepts measured by it. The core topics of validity and reliability will be discussed. Validity, which is assessed in several ways, provides the evidence and extent that the PRO measure taps into the concept that it is purported to measure in a particular setting. Reliability of a PRO measure involves its consistency or reproducibility as assessed by internal consistency and test-retest reliability. Anchor-based and distributed-based approaches to interpret PRO results will be elucidated in order to make these results useful and meaningful. Illustrations will be provided mainly through real-life published examples and also through selected simulated examples using SAS. Exploratory factor analysis and confirmatory factor analysis, mediation modeling, item response theory, longitudinal analysis, and missing data will be among the topics considered if time permits.
Precision medicine has paved the way for a new era of delivering tailored treatments to patients according to their biological profiles. In combination with innovative clinical design, this has presented drug developers unprecedented opportunities to engage novel thinking to accelerate drug discovery. In the first part of this course, step-by-step introductions to basic biology and genetics will be presented, and is followed by overviews of cutting edge technologies such as microarray and next generation sequencing technologies that have been widely used to generate omics data. Built on the basic knowledge of biology and omics data, key concepts of precision medicine studies and strategies of how in practice this novel approach can be applied to drug discovery will be discussed. In addition, statistical considerations and challenges posed in omics data such as data normalization, statistical modeling and interpretation will also be discussed. Examples are case studies from the instructors’ work and from medical literature. The second part of this course will cover design considerations in modern drug development for precision medicine. Different classical and adaptive design options including platform trial designs will be introduced with case studies. In addition, related statistical theories and analysis strategies will be covered . No prerequisite knowledge needed.
Clinical trials in the regulatory environment specify a primary outcome variable to avoid problems of multiplicity. A single outcome measurement is often insufficient to understand the effect of a drug, however. In particular, various things may happen that make the outcome variable unobservable, irrelevant, or nonexistent, or that change its interpretation. The outcome in such cases should be considered to be multivariate: either no such event occurs and the outcome is the value of the primary variable, or an event occurs and the outcome is the ensemble of the fact, the time, and the nature of the event, the observations before the event, and possibly further observations after the event.
In this respect trials are very different from sample surveys. In surveys the problem is not the existence or interpretability of the variable in question but the simple failure to ascertain it. There is no doubt that the value that would have been ascertained is the relevant quantity for analysis, and if it is not ascertained it must be estimated. Methods like those used to deal with missing data in surveys are commonly applied in clinical trials, with disastrous results, requiring implausible interpretations of what “would have” happened under different conditions.
We will discuss ways of defining effects that respect the multivariate nature of the outcome. These effects are of at least five kinds:
1. Actual values notwithstanding intercurrent events. 2. Transformed outcomes taking intercurrent events into account. 3. Values under hypothetical conditions. Careful attention will be given to what hypothetical conditions can yield estimable and interpretable effects and what conditions can’t. 4. Values in a subset without intercurrent events. The difference between this and “completers” or “per protocol” analysis will be carefully explained. 5. Values before an intercurrent event. We will consider when these reasonably represent a benefit to the patient and when they do not.