Generation of Drug-Induced Cardiac Reactions towards Virtual Clinical Trials

TL;DR

Cardiac drug development suffers from high clinical trial failure due to efficacy and safety concerns. The authors introduce the Drug-Aware Diffusion Model (DADM), which combines an Ordinary Differential Equation based External Physical Knowledge (EPK) with a Denoising Diffusion Probabilistic Model, uses Dynamic Cross-Attention (DCA) to adapt EPK during denoising, and extends ControlNet into Clinical Information ControlNet (CICN) to condition on demographic and drug data for individualized ECG synthesis. On two public ECG datasets covering eight drug regimens, DADM achieves superior fidelity and more accurate simulated drug-induced ECG changes than eight state-of-the-art generators, and improves downstream drug-effect classification. The approach enables safer, cost-effective virtual clinical trials and has potential to accelerate cardiac drug development by enabling realistic, subject-specific ECG simulations that reflect post-dose pharmacodynamics.

Abstract

Clinical trials remain critical in cardiac drug development but face high failure rates due to efficacy limitations and safety risks, incurring substantial costs. In-silico trial methodologies, particularly generative models simulating drug-induced electrocardiogram (ECG) alterations, offer a potential solution to mitigate these challenges. While existing models show progress in ECG synthesis, their constrained fidelity and inability to characterize individual-specific pharmacological response patterns fundamentally limit clinical translatability. To address these issues, we propose a novel Drug-Aware Diffusion Model (DADM). Specifically, we construct a set of ordinary differential equations to provide external physical knowledge (EPK) of the realistic ECG morphology. The EPK is used to adaptively constrain the morphology of the generated ECGs through a dynamic cross-attention (DCA) mechanism. Furthermore, we propose an extension of ControlNet to incorporate demographic and drug data, simulating individual drug reactions. Compared to the other eight state-of-the-art (SOTA) ECG generative models: 1) Quantitative and expert evaluation demonstrate that DADM generates ECGs with superior fidelity; 2) Comparative results on two real-world databases covering 8 types of drug regimens verify that DADM can more accurately simulate drug-induced changes in ECGs, improving the accuracy by at least 5.79% and recall by 8%. In addition, the ECGs generated by DADM can also enhance model performance in downstream drug-effect classification tasks.

Figures (9)

• Figure 1: Physical clinical trials v.s. virtual clinical trials. In realistic scenarios, we aim to train the DADM for virtual trials with limited physical trial data, thereby avoiding financial and safety risks associated with large-scale clinical trials.
• Figure 2: (a) Workflow of integrating EPK; (b) Structure of RBDCA. $\mathbf{x}_{ode}$ represents the external physical knowledge. The output of the previous residual block is the input of the next one. Each block is also connected through a skip connection; (c) Structure of DCA. $\mathbf{y}_{inter}$ and $\mathbf{A}$ represent the output of the Bi-DilConv and the cross-attention map, respectively.
• Figure 3: The architecture of CICN.
• Figure 4: (a) Ablation study of attention mechanisms; (b) Changing trend of $\alpha$ over time-steps.
• Figure 5: Visualization of generated ECGs with and without EPK. For clarity in comparison, we use the electrocardiograms of three leads: II, aVR, and V3.