Design-Ignoring versus Design-Respecting World Models for Epidemiology

2026-09-24

AI Summary

This study addresses the intervention estimation bias in existing epidemiological world models arising from the neglect of research design. By formalizing study design as model constraints, this work proposes a novel “design-aligned evaluation” paradigm and validates it through techniques including latent world interface modeling, action mechanism constraints, and cluster-randomized trial resampling. Results demonstrate that factual reconstruction alone cannot guarantee valid causal inference. In contrast, design-respecting models maintain stable intervention estimates across varying sampling intensities, significantly outperforming design-agnostic baselines. This research provides critical methodological support for constructing causally reliable epidemiological world models.

Abstract

World models for epidemiology learn from records shaped by study designs, including assignment, sampling, measurement, and related processes. A model may therefore reconstruct observed trajectories while learning an intervention contrast that depends on how records were collected. We formalize study design as constraints on a world model latent-world interface, action mechanism, observation likelihood, and target readout. This distinguishes design-respecting models, which encode these constraints, from design-ignoring models, which fit selected records without relating assignment and observation to the intended intervention question. Using a large scale cluster-randomized test-negative trial, we hold the latent structure and fitting settings fixed and compare a design-ignoring case-count model with a design-respecting test-negative observation model. Both models achieve comparable factual reconstruction. Yet across 500 paired resampling experiments at different relative sampling intensity, median contrast changes are substantial for the design-ignoring model, versus merely marginal for the design-respecting model. The same qualitative separation holds when sampling also varies across clusters. Thus, factual reconstruction alone does not establish design alignment; testing whether fitted contrasts preserve design-implied observation-process invariances provides a sharper evaluation.

Problem

• epidemiology
• world models
• study design
• intervention contrast
• observational bias

Innovation

• World Models
• Epidemiology
• Study Design
• Test-Negative Trial
• Observation-Process Invariance