Wednesday, Aug 6: 2:00 PM - 3:50 PM
4190
Contributed Papers
Music City Center
Room: CC-102B
Main Sponsor
Biopharmaceutical Section
Presentations
For a biomarker to be at least a "level 3 surrogate" that is "reasonably likely to predict clinical benefit for a specific disease and class of interventions" it must be either a mediator on the causal pathway between treatment and response, or else be causally downstream of such a mediator. We investigate causal mediation analysis as an approach to statistically infer potential mediation effects of biomarkers. Steps involve graphically stating the causal structure using DAGs, formulating estimands of interest and using statistical methods to derive estimates. However, longitudinal clinical data are commonplace and causal estimation of such data is notoriously challenging, standard statistical methods might not provide appropriate target estimates. Thus, we also explore methods to account for time-varying confounding in mediation analysis, one such method discussed provides a reasonable approximation by "Landmarking" the biomarker process at a particular timepoint t, and modeling the clinical outcome data after time t. We aim to outline fundamental ideas of causal mediation analysis and delineate a potential framework for its use in clinical development.
Keywords
biomarkers
causal inference
mediation
estimands
The minimal clinically important difference (MCID) concept recognizes the limitations of statistical significance in determining practical relevance for patients. One popular methodology for determining the MCID is the anchor-based approach. However, the theoretical properties and robustness of the methodology are not fully understood. To address the gap, we conducted a simulation study to explore the performance of anchor-based methods across a range of values for clinical outcome assessment (COA) variance, placebo effects, anchor measurement noise, and confounding. Our findings revealed that some scenarios exhibited bias exceeding 50%. This bias can be in either direction, meaning the MCID estimate is not always conservative. We observed an increase in bias when the calculated MCID deviated significantly from the mean COA. In the COA variation scenario, the bias may also be more pronounced when the standard deviation of the COA is small relative to the calculated MCID. Finally, confounding effects are more likely when the COA represents only one of several factors influencing the anchor. We conclude by discussing strategies for identifying and mitigating these biases.
Keywords
COA - clinical outcome assessment
MCID - minimal clinically important difference
Anchor-based method
Bias
Placebo effect
Confounding
A surrogate endpoint is a biomarker or intermediate outcome used instead of a direct clinical endpoint to predict drug benefit. It serves as a substitute for a primary endpoint, offering advantages when measured earlier or more conveniently. Before its use in scientific conclusions, qualification is required. A valid surrogate must meet two associations: I-Association (linking the surrogate and true endpoints, e.g., disease response and overall survival) and T-Association (linking treatment effects on both, e.g., odds ratio and hazard ratio). While I-Association is commonly evaluated, T-Association is often overlooked. This study proposes methods to assess T-Association, assuming treatment effects on both endpoints follow a bivariate normal distribution. The key evaluation metric is the correlation coefficient, estimated via maximum likelihood, restricted maximum likelihood, and a Bayesian approach. Simulated and real-world data assess bias, standard error, and coverage probability. This method will support future FDA Accelerated Approval drugs.
Keywords
Surrogate endpoint
Bivariate normal
maximum likelihood
restricted maximum likelihood
Bayesian
Predicting survival remains a critical challenge for many diseases where more traditional statistical models often rely on baseline demographics or disease characteristics. Studying the trajectory of specific biomarkers is important for understanding the dynamic of disease progression and clinical outcome that could predict the overall survival (OS). This project investigated the performance of survival prediction using the trajectory of a longitudinal biomarker (serum protein electrophoresis, SPEP), which captures the dynamic of disease progression over time. Using area under the curve (AUC) and prediction error (PE), model performance was evaluated for several joint models and Cox models for overall survival that considered observed or Bayesian derived clinical data. Using simulated and real data from a Multiple Myeloma study, our findings indicate that joint model incorporating the trajectory of SPEP data improves the predictions for OS.
Keywords
Longitudinal biomarker
Survival prediction
Joint Model
Oncology clinical trials
Companion diagnostics (CDx) are essential in precision oncology, guiding targeted therapies based on biomarker status. During biomarker discovery stage, research-use-only assay and early-phase clinical trial data are often used to identify a potential predictive biomarker. However, the non-fully validated assay and single-arm design of Phase I trials with limited sample size in multiple indications pose significant challenges.
Real-world evidence (RWE) can help address these challenges by providing insights into biomarker expression within target populations, evaluating patient demographics and treatment patterns, and estimating biomarker prevalence. Additionally, RWE facilitates the construction of synthetic control arms, enhancing the robustness of predictive biomarker assessments and supporting CDx assay development.
This presentation will showcase current applications of RWE in biomarker discovery and CDx development. Furthermore, we will discuss the potential applications for developing AI-based approach to predictive biomarker value using real-world available genomic data.
Keywords
Real-World Evidence
Biomarker Discovery
Companion Diagnostics (CDx)
Oncology
Per the International Council on Harmonisation clinical (ICH E14) guidance in 2005, all new chemical entities (NCEs) with systemic exposure should undergo an evaluation as to whether they cause an effect on the heart rate QTc before marketing, as pronounced QTc prolongation in susceptible patients may lead to a potentially lethal ventricular arrhythmia called torsades de pointes (TdP). Therefore, the cardiodynamic ECG assessment is performed, among which the by-time point analysis and concentration-QTc analysis are the two main analyses, based on mixed models for repeated measures of ECGs and concentrations. Since ICH E14 Questions & Answers (R3) was released in 2015, the latter analysis has served as an alternative to the former analysis as the primary basis for decisions to classify the risk of a drug, which has been the primary analysis for most ECG evaluation in early clinical trials. Model description, model assumptions, sample size determination, and goodness-of-fit will be introduced for the concentration-QTc analysis with application to case studies with healthy subjects or patients at this presentation.
Keywords
cardiodynamic ECG assessment
concentration-QTc analysis
by-time point analysis
Precision medicine is an evolving area in the medical field and rely on biomarkers to make patient enrichment decisions, thereby providing drug development direction. A traditional statistical approach is to find the cut-off that leads to the minimum p-value of the interaction between the biomarker dichotomized at that cut-off and treatment. Such an approach does not incorporate clinical significance and the biomarker is not evaluated on a continuous scale. We are proposing to evaluate the biomarker in a continuous manner from a predicted risk standpoint, based on the model that includes the interaction between the biomarker and treatment. The predicted risk can be graphically displayed to explain the relationship between the outcome and biomarker, whereby suggesting a cut-off for biomarker positive/negative groups. We borrow ideas from the TreatmentSelection approach and extend it to account for covariates via G-computation. Other features include biomarker comparisons using net gain summary measures and calibration to assess the model fit. The PRIME approach is flexible in the type of outcome and covariates considered. A R package is available and examples will be demonstrated.
Keywords
biomarker
precision medicine
patient enrichment
prediction