Sunday, Aug 3: 2:00 PM - 3:50 PM
4009
Contributed Papers
Music City Center
Room: CC-209C
Main Sponsor
Biopharmaceutical Section
Presentations
Remote safety monitoring using biosensor devices presents advantages compared to traditional clinical approach for physiological parameters (e.g. blood pressure and heart rate). It operates remotely and do not need frequent site visits, thus significantly reduces burden on patients and clinical sites. However, physiological parameters are highly impacted by physical activity (PA), which makes the variability of physiological parameters in remote setting larger than that in clinical visit. The higher variability results in less precision in safety endpoint estimation. We developed a spline-based model to quantify the individualized relationship between heart rate (HR) and PA, and conducted adjustment of HR to remove the impact of PA. This method is applied on a biosensor data which contains simultaneous high-frequency data of HR and PA. Compared with traditional parametric model, our model has better fitting performance. The adjusted HR has much less influence of PA and shows reduced variability. This adjustment approach can help future clinical trials when applying biosensor device and is easy to implement.
Keywords
biosensor data
spline method
high frequency
adjustment for physical activity
Biomarkers are increasingly important in diagnosing and treating cancer. However, due to instruments' inability to measure biomarker signals below a certain threshold, such observations are left censored and classified as Below the Level of Detection (BLoD). Incorporating data from these biomarkers into analysis presents significant challenges. Most of the literature focuses on estimating the location parameter and evaluating their efficiency. However, interest often lies in comparing biomarkers between groups, such as cases and controls. Typically, one would apply an estimation method and then use appropriate techniques to compare the groups. The evaluation of the performance of these methods in the context of comparing two groups has not been performed. In our analysis, we perform a comparison study of common method used for data subject to a Level of Detection (DL) and assess their efficacy regarding a case-control study framework for the comparison of two groups. We also assess the robustness of the test procedures when the underlying distribution is not lognormal. We examine the usefulness of these test procedures by applying them to a dataset from St. Jude on IL-6.
Keywords
Censored data
Below the Limit of Detection
Biomarkers
Lognormal
Robustness
Interleukin 6
Predictive biomarkers have the potential to increase the benefit to patients and sponsors. Patient benefit from predictive biomarker is that they can receive more optimal treatment. For sponsors successful biomarker development can improve probability of product success and better differentiate the product if it is approved. However, biomarker development can add cost, complexity, and time to drug development. Additionally, biomarkers will reduce the market size, because they are usually approved only in a subset of population.
In this presentation we will first evaluate financial value of developing a biomarker in the context of one individual program. In practice, investment to develop a biomarker will usually compete with investments in other programs within the portfolio, and decision should be made in the portfolio context. The second part of this presentation will evaluate biomarker development in the context of portfolio, taking into account all decision parameters.
Keywords
biomarker development
optimal decision
probability of success
differentiation
cost and complexity
market size
Placebo response in chronic pain clinical trials presents a significant challenge to demonstrating treatment efficacy, often obscuring potential therapeutic benefits of investigational analgesics. High placebo response rates in chronic pain studies are attributed to multiple factors, including heightened expectation effects, trial design elements, and participant characteristics. These factors encompass conditioning from previous treatment experiences, intensive contact with healthcare providers, regression to the mean, natural symptom fluctuation, and psychological factors. This work examines how implementing a placebo lead-in design can enhance assay sensitivity by identifying and stratifying participants based on their initial placebo response patterns. The placebo lead-in approach involves an initial blind placebo phase before randomization, based on pre-specified criteria, participants are then stratified into "placebo responders" and "non-responders" before randomization to active treatment or control arms. This methodology helps identify participants whose responses may confound treatment effects and enables more balanced allocation of placebo responder across treatment arms
Keywords
Placebo lead-in, enrichment design, signal detection, chronic pain clinical trial
First Author
Yan Dong, Eli Lilly and Company
Presenting Author
Yan Dong, Eli Lilly and Company
Pharmaceutical research that involves bounded bioanalytical data often requires the calculation of statistical intervals to establish quality specifications and evaluate the integrity of pharmaceutical products. An illustrative technique is size exclusion chromatography (SEC), a widely used bioanalytical method for determining drug product purity by quantifying the proportion of monomers. To ensure the quality of these products, statistical intervals are developed to assess whether a significant proportion of production batches meets a high standard for the percentage of monomer content. Many traditional statistical interval derivations often rely on the assumption of a normal distribution. However, such assumption can lead to unreliable results and potentially inaccurate evaluations of product quality. To effectively tackle these challenges, it is crucial to employ alternative statistical methods that are tailored for bounded data. In this study, we survey and compare three inferential methods for developing statistical intervals based on the Kumaraswamy distribution. We illustrate the benefits of our proposed methods using simulations and a real data example.
Keywords
Fiducial-based tolerance interval
Fiducial prediction interval
Specification
Kumaraswamy distribution
The function of cytokine interactions in initiating and advancing pancreatic cancer was studied with GeoMx® dsp spatial sequencing data using PDAC FFPE samples from 15 patients in up to 5 regions of interest (ROIs) of one tumor each for two cell types: CAFs vs ductal. Usually, each ROI was exposed on two of the eight slides used. One goal is to study within-subject (across ROIs in each subject) and between-subject correlations among the expression levels between cytokines of interest and other genes. The correlations could be strong between the changes in two variables within a patient and/or among the average expression levels between patients. Challenges arise from crossed and nested effects among ROIs, slides and patients. We applied methods proposed by Bland and Altman: within-subject correlations were derived from ANCOVA with fixed effects for the subject by slide interaction, and between-subject correlations used weighted averages weighting by number of ROIs. A false discovery rate criterion was used to identify correlations. The identified correlations aligned well with previous hypotheses. This work was in part funded by NIH-P01 and the Salk Cancer Center Support Grant.
Keywords
Spatial genomics
Spatial sequencing
Gene expression
Pancreatic cancer
Co-Author(s)
Yuan Sui, Salk Institute for Biological Studies, La Jolla, California, United States
Tony Hunter, Salk Institute for Biological Studies, La Jolla, California, United States
Charles Berry
First Author
Minya Pu
Presenting Author
Minya Pu
Gene expression change has emerged as an important biomarker for evaluating treatment effects in immunology and inflammatory (I&I) indications. Historically, gene expression data have been primarily utilized as exploratory endpoints in many I&I studies. However, recent advancements have positioned gene expression data as key study endpoints (primary or secondary) with gene signatures in biomarker-driven mechanistic studies. Unlike traditional biomarkers with single-valued measurements, gene expression data are high-dimensional, often exceeding 20,000 variables. The implementation of gene expression data to select, validate, and evaluate gene signatures presents unique statistical challenges when used as key study endpoints. We address the following critical considerations for implementing gene expression endpoints in clinical study design:
1. Sample Size Calculation for Gene Signature Detection
2. Validation of Gene Signature in an Independent Sample
3. Evaluation of Gene Signature in a Different Cohort
Keywords
Clinical trial design
sample size calculation
Gene expression
Gene signature