Tuesday, Aug 5: 2:00 PM - 3:50 PM
4135
Contributed Papers
Music City Center
Room: CC-201A
Main Sponsor
Section on Risk Analysis
Presentations
Clinical trials and postmarket surveillance data both provide valuable information for detecting adverse events (AEs) for drugs. Clinical trials provide estimates for incidence rate but often have insufficient sample size to detect rare adverse events. Post market surveillance data can accumulate more reports for adverse effects, but the data quality is often uneven and cannot provide direct estimate of incidence rate.
Using PD-1 inhibitors as an example, we compared the safety signal in clinical trials and postmarket surveillance. The meta-analysis for clinical trials were based on a sparse Bayesian mixed-effect model. Correspondingly, postmarket reports of adverse events were tabulated from the FDA Adverse Event Reporting System (FAERS) and a sparse Bayesian model was constructed. Finally, a Bayesian model was constructed using prior distributions based on the clinical trial data to guide the inference in the analysis of postmarket surveillance data.
Signals detected in clinical trials data have a high probability to be confirmed in the postmarket surveillance. Integrating clinical trial safety data and postmarket databases appear to be promising in enhancing signals.
Keywords
adverse event
postmarket surveillance
Bayesian model
clinical trial
First Author
Dong Wang, FDA National Center for Toxicological Research (NCTR)
Presenting Author
Dong Wang, FDA National Center for Toxicological Research (NCTR)
A probability method to estimate cancer risk for asymptomatic individuals for the rest of life was developed based on one's current age and screening history
using the disease progressive model. The risk is a function of the transition probability density from the disease-free to the preclinical state, the sojourn time in the
preclinical state, and the screening sensitivity if one had a screening history with negative results. The method can be applied to any chronic disease.
As an example, the method was applied to estimate women's breast cancer risk using parameters estimated from the Health Insurance Plan of Greater New York (HIP) under two scenarios: with and without a screening history, and obtain some meaningful results.
Keywords
incidence
risk
sensitivity
sojourn time
transition probability density
Incorporating historical data into the analysis of treatment effects for rare diseases has gained increasing attention. However, determining the appropriate level of congruence between historical and current data remains a significant challenge. In this work, we introduce a novel Bayesian p-value-based congruence measure to quantify heterogeneity between historical and current control data. We investigate its asymptotic properties under both congruent and incongruent scenarios and develop the pointwise predictive density calibrated-power prior (PPD-CPP) to dynamically leverage historical information. The PPD-CPP framework provides a flexible approach, allowing the power parameter to be modeled as either a fixed scalar or a random variable and enabling the assignment of unique power parameters to individual observations. Through extensive numerical studies with normal endpoints, we demonstrate that our method effectively borrows information from congruent sources while appropriately discarding incongruent data.
Keywords
Bayesian p-value
Calibrated power prior
Congruence measure
Dynamic historical borrowing
Humans are exposed to myriad pollutants, so it is important to identify the most sensitive health effects at low exposure levels. We developed a weighted mixed effects quantile regression approach to determine relative endpoint responsiveness among different organ systems. The lowest observed effect level (LOEL) is the lowest dosage associated with a significant change in an endpoint, so quantiles of LOELs for an organ system were comparable measures for endpoint responsiveness. The no observed effect level (NOEL) is the highest dosage not associated with a significant change, so LOELs close to their respective NOELs are better estimates of a toxicological response. Thus, weighted quantile regression of LOELs against organ system, with weights determined by NOEL presence and magnitude and random effects accounting for variation due to the same laboratory reporting multiple LOELs, present a novel approach to rank endpoint responsiveness. Ad hoc analyses of the estimated effects determined if sensitivities were significantly different between systems. We applied this method on a toxicological database of endpoints measured after exposure to polychlorinated biphenyl (PCB) mixtures.
Keywords
Systematic review
Quantile regression
Toxicology
PCB
This paper introduces a novel measure to quantify the directional dependence of extreme events between two variables. We propose a new approach to capture the asymmetric tail dependence. By studying the conditional tail expectation and of the rank transformed variables, we quantify the behaviour of one variable when the other is extreme. The effectiveness of the approach is demonstrated through an extensive simulation framework, and the theoretical asymptotic behaviour of the estimator is investigated. We apply this method to environmental and financial data, such as wind speed and temperature extremes, and stock-bond market extremes. Our results show the strong asymmetric nature of extreme events.
Keywords
Directional Dependence
Tail Dependence
Copula
Tail Expectation
Anomalies in the stock market, such as price manipulation, speculative bubbles or unusual capital flows, can lead to market destabilization. Traditional analysis methods, based on statistical models, often fail to keep up with the dynamic and complex nature of modern financial markets. The study aims to evaluate the effectiveness of artificial intelligence (AI) methods in detecting and forecasting stock market anomalies. AI can significantly improve the ability to monitor markets and provide early warnings of threats. The study will apply selected AI techniques to detect and analyze unusual market behavior.
Modern stock markets are characterized by a large number of market participants who generate a large number of transactions and a large amount of information. This situation means that traditional methods of analysis, based, for example, on econometric models and simple financial indicators, are often unable to identify risks associated with stock market anomalies on time. In this respect, artificial intelligence tools can be helpful. The study used data on selected companies listed on stock exchanges in the Baltic countries. The data covered the years 2007-2024.
Keywords
AI
econometric models
stock market anomalies
stock market indices
This study presents a systematic machine learning (ML) framework to classify loan applicants into creditworthy and non-creditworthy categories. The proposed methodology encompasses a comprehensive data preprocessing pipeline, including the imputation of missing values and feature selection using the Random Forest algorithm to identify key predictive variables. To address potential multicollinearity issues, a variance inflation factor (VIF) analysis is conducted to ensure model robustness and interpretability. Additionally, to mitigate class imbalance within the dataset, the Synthetic Minority Over-sampling Technique (SMOTE) is applied, enhancing the representativeness of the training data. Three ML models are subsequently trained and rigorously evaluated based on performance metrics. The results demonstrate the efficacy of the proposed approach in improving credit risk classification, providing financial institutions with a data-driven framework to enhance decision-making processes, optimize resource allocation, and support strategic lending initiatives. These findings underscore the transformative potential of predictive analytics in advancing financial risk management practices.
Keywords
Machine Learning (ML)
Creditworthy
Classification
Credit Risk
Financial Institutions
Financial Risk Management