Proactive Anomaly Screen for Multiple Endpoints Using Bayesian Latent Class Modeling: A k-Step Ahead Approach

TL;DR

This work tackles data quality in clinical trials by introducing a proactive anomaly-screening framework that combines a Bayesian latent class model with risk-adjusted factors to monitor multiple efficacy endpoints. By modeling endpoints jointly through a Dirichlet process mixture, it enables individualized $k$-step ahead predictions and posterior predictive checks, implemented via grid-based credible regions and two region-identification algorithms. Simulation and real-data analyses (EASI with IGA and SCORAD) demonstrate high predictive coverage and effective detection of anomalous patterns, supporting real-time alerts within electronic data capture systems. The approach provides a flexible, extensible tool for risk-based monitoring that can be integrated with existing workflows to improve data quality and patient safety in trials.

Abstract

In clinical trials, ensuring the quality and validity of data for downstream analysis and results is paramount, thus necessitating thorough data monitoring. This typically involves employing edit checks and manual queries during data collection. Edit checks consist of straightforward schemes programmed into relational databases, though they lack the capacity to assess data intelligently. In contrast, manual queries are initiated by data managers who manually scrutinize the collected data, identifying discrepancies needing clarification or correction. Manual queries pose significant challenges, particularly when dealing with large-scale data in late-phase clinical trials. Moreover, they are reactive rather than predictive, meaning they address issues after they arise rather than preemptively preventing errors. In this paper, we propose a joint model for multiple endpoints, focusing on primary and key secondary measures, using a Bayesian latent class approach. This model incorporates adjustments for risk monitoring factors, enabling proactive, $k$-step ahead, detection of conflicting or anomalous patterns within the data. Furthermore, we develop individualized dynamic predictions at consecutive time-points to identify potential anomalous values based on observed data. This analysis can be integrated into electronic data capture systems to provide objective alerts to stakeholders. We present simulation results and demonstrate effectiveness of this approach with real-world data.

Figures (14)

• Figure 1: An Illustrative Example: Deviating Patterns between multiple endpoints of Change from baseline in EASI ("Chg in EASI" denoted in red) and Change from baseline in IGA ("Chg in IGA" denoted in blue).
• Figure 2: Clinical data management workflow with proposal of embedded workflow (in blue and orange).
• Figure 3: Simulation Scenarios. Left: observed trajectories for the first simulation data; Right: mean trajectory. In each figure, left is placebo and right is treatment.
• Figure 4: EASI and IGA: Classification (number of subjects) of $570$ subjects from best configuration dahl2006model: 1(12), 2(15), 6(157), 10(8), 11(4), 12(146), 14(19), 17(38), 18(39), 23(77), 25(4), 28(45), 29(6). Top: Mean Trajectory; Bottom: Individual Trajectory. Note that for the bottom of individual trajectories, it is faceted by treatment for easier visualization, and training was based on blinded data.
• Figure 5: Prediction Scenario 1 (EASI and IGA): $80\%$ credible regions for Scenario 1 Selected Subject $\#1$ at Week 2 conditional on baseline. The $80\%$ credible regions are identified with X over the regions. Left: Algorithm 1 (Branching out); Right: Algorithm 2: HDR. Baseline (EASI:21, IGA:3); True value at Week 2(EASI:-10, IGA:0). The observation at Week 2 is contained in the credible regions for both algorithms.