Intervention-Conditioned Counterfactual Human Digital Twin Generator
Abstract
Chronic diseases evolve through long, heterogeneous trajectories shaped by interacting mediators, confounders, and treatment histories. Estimating long-term patient trajectories under alternative interventions is essential for in silico clinical trial design. However, most existing models overlook irregular visit intervals and cross-mediator dependencies, both of which are crucial for reliable long-term longitudinal prediction. Moreover, these models typically focus on outcome prediction rather than generating complete patient-state trajectories. To address these challenges, we propose CHDT, an intervention-conditioned Human Digital Twin generator under standard identification assumptions for multimodal longitudinal trajectory estimation. Its main empirical advantage arises from explicit modeling of irregular visit intervals, while cross-mediator refinement provides additional gains in selected intervention-conditioned settings. Across real-world ADNI and simulator-generated ALPACA cohorts, CHDT shows improved long-horizon patient-state fidelity relative to competing baselines. Ablation results indicate that irregular-time conditioning provides the primary empirical gain, and cross-mediator refinement provides additional improvements.
Related
- Neural-DISCO: Source-Conditioned Counterfactual Editing of Neural Population Activity
- Warburg metabolism in tumor-conditioned macrophages promotes metastasis in human pancreatic ductal adenocarcinoma
- A generator-matrix model quantifies the limited contribution of measured biomarkers to human mortality acceleration
- A study protocol evaluating the effectiveness of the older adult’s digital health literacy (ODDHeaL) intervention
- Alternative digital secondary education for marginalized, out-of-school youth in low- and middle-income countries: Protocol for a scoping review of intervention models, outcomes, and implementation challenges
BibTeX
@misc{farzi2026interventionconditioned,
title = {{Intervention-Conditioned Counterfactual Human Digital Twin Generator}},
author = {Saeed Farzi and Sofia Michopoulou and Christopher Kipps and Rahman Attar},
howpublished = {arXiv},
year = {2026},
month = oct,
doi = {10.48550/arxiv.2610.40664},
eprint = {2610.40664},
archivePrefix = {arXiv},
url = {https://arxiv.org/abs/2610.40664},
note = {CC BY 4.0},
}
RIS
TY - UNPB
TI - Intervention-Conditioned Counterfactual Human Digital Twin Generator
AU - Saeed Farzi
AU - Sofia Michopoulou
AU - Christopher Kipps
AU - Rahman Attar
PY - 2026
DA - 2026/10/09
PB - arXiv
DO - 10.48550/arxiv.2610.40664
UR - https://arxiv.org/abs/2610.40664
AB - Chronic diseases evolve through long, heterogeneous trajectories shaped by interacting mediators, confounders, and treatment histories. Estimating long-term patient trajectories under alternative interventions is essential for in silico clinical trial design. However, most existing models overlook irregular visit intervals and cross-mediator dependencies, both of which are crucial for reliable long-term longitudinal prediction. Moreover, these models typically focus on outcome prediction rather than generating complete patient-state trajectories. To address these challenges, we propose CHDT, an intervention-conditioned Human Digital Twin generator under standard identification assumptions for multimodal longitudinal trajectory estimation. Its main empirical advantage arises from explicit modeling of irregular visit intervals, while cross-mediator refinement provides additional gains in selected intervention-conditioned settings. Across real-world ADNI and simulator-generated ALPACA cohorts, CHDT shows improved long-horizon patient-state fidelity relative to competing baselines. Ablation results indicate that irregular-time conditioning provides the primary empirical gain, and cross-mediator refinement provides additional improvements.
ER -