Intervention Granularity Matters: Coherent Treatment Bundles in Counterfactual Simulation with Clinical World Models

Fangzhou Wang; Yixuan Yang; Camilla Balzarotti; Rishikesan Kamaleswaran

cs.LGSep 18, 2026

Intervention Granularity Matters: Coherent Treatment Bundles in Counterfactual Simulation with Clinical World Models

Authors: Fangzhou Wang, Yixuan Yang, Camilla Balzarotti, Rishikesan Kamaleswaran

Organizations: Duke University

Abstract

Counterfactual simulation with a clinical world model means fixing a patient's history, changing the treatment, and reading off the predicted response. Doing so requires deciding what counts as one intervention. In clinical settings, interventions are documented as bundles: a co-occurrence audit of 945,707 patient-hours from MIMIC-IV shows groups of components, such as every parameter of a dialysis circuit, that never appear apart, so an edit that changes one component on its own describes an hour that never occurs in the data. We hypothesize that the granularity at which an intervention is edited changes how a world model responds, and test this with Clin-JEPA, a latent world model of patient trajectories conditioned on hourly treatment text. At 1,019 documented onsets of invasive ventilation, we keep the patient's history and other treatments fixed and compare editing one ventilator setting with editing the complete configuration recorded for a real patient with the most similar recent trajectory. The complete bundle moves the predicted next state further than any single setting, consistently across all five settings, and the difference remains after accounting for how much each edit changes the model's input. Intervention granularity therefore materially affects the response of a clinical world model: single-component edits may understate treatment sensitivity, and bundle-aware editing may offer a better-supported basis for counterfactual treatment simulation.

Figures & tables

Figure 1: Bundle-consistent editing at a ventilation onset. The 24-hour history is fixed. At the onset hour, the ventilation description is replaced by one setting or by the complete configuration recorded for the reference patient, the real patient with the most similar trajectory; other treatments are kept. The frozen world model predicts the next state under each text, and the response

d_{z}

is the relative distance from the prediction under the recorded text. Inset: a coherent bundle is a configuration observed in real care; a component on its own is not.

Setting	$n$	Single setting	Coherent bundle	Bundle $-$ single [95% CI]	Adjusted	Adj. / single
FiO 2	918	0.0309	0.0487	$+0.0178$ [0.0157, 0.0201]	$+0.0083$	27%
PEEP	967	0.0303	0.0481	$+0.0178$ [0.0157, 0.0199]	$+0.0088$	29%
Set respiratory rate	685	0.0307	0.0553	$+0.0246$ [0.0220, 0.0274]	$+0.0121$	39%
Set tidal volume	677	0.0417	0.0556	$+0.0140$ [0.0112, 0.0167]	$+0.0057$	14%
Ventilator mode	338	0.0415	0.0656	$+0.0241$ [0.0202, 0.0281]	$+0.0253$	61%

Table 1: The complete bundle produces a larger next-hour response than any single setting. Mean next-hour response

d_{z}

under the single-setting and the bundle condition, on the targets whose reference documents the setting in the row. Bundle

-

single: mean paired difference with its bootstrap 95% confidence interval. Adjusted: the bundle effect

\beta_{b}

at equal edit magnitude. Adj. / single: the adjusted effect as a fraction of the single-setting mean, for scale only.

Figure 2: Immediate representation responses to ventilation intervention construction. (a) Mean factual-relative next-state latent response $d_{z}$ for single-component (blue) and coherent-bundle (green) edits on paired targets; $n$ is the number of paired target instances for each setting. (b) Relationship between action-embedding perturbation magnitude $d_{a}$ and $d_{z}$ ; curves are descriptive quadratic fits (overall Pearson $r=0.21$ ). (c) Action-distance-adjusted bundle-minus-single differences in $d_{z}$ with 95% target-level bootstrap confidence intervals. Percentages give the adjusted difference relative to the corresponding mean single-component $d_{z}$ . (Appendix D )

Appendix figures & tables2 assets

Supplementary material from the paper’s appendix.

Appendix

Criterion	Target hours	Stays
Raw transition (no InvasiveVent at $t$ , InvasiveVent at $t+1$ )	3,553	2,903
6-h washout	3,442	2,877
+ 24-h history + 12-h future	1,127	805
+ persistent $\geq$ 2 h	1,125	803
12-h washout	3,404	2,869
+ 24-h history	1,125	809

Appendix

Table 2: Cohort flow for invasive-ventilation onset targets in the test split. The primary analysis uses the 12-hour washout (bold).

	Full window ( $t-23$ to $t$ )	Masked window ( $t-23$ to $t-6$ )
Targets matched	1,089 / 1,089	1,089 / 1,089
Mean similarity of best match	0.929	0.934
5th percentile of similarity	0.894	0.899
Distinct references selected	518	541
Reuse per reference: median / 95th pct. / max	1 / 5 / 12	1 / 5 / 15
Bundles with $\geq 3$ of 5 settings	88.2%	88.3%

Appendix

Table 3: Reference matching with the full 24-hour window versus the window with the last six hours before onset masked. Similarity: mean hourly cosine similarity of the best match.

Causal Intervention

Causal intervention studies how deliberate changes to variables or mechanisms reveal cause and effect and guide prediction, discovery, and decisions.

See all 89 papers

Same Predictions, Different Harms: Causal Auditing of Patient World Models
Oct 4, 2026Yicheng Qi, Xiyi XiongStructural Causal Models Causal World Models
Imperial College London, London, United Kingdom
Medical world models: representing medical states, modelling clinical dynamics and guiding intervention policies
Jun 15, 2026Ke Liu, Mengxuan Li, Yanyi Bao +4World Model Learning Clinical Decision Support
College of Computer Science, Zhejiang University, Hangzhou, China. · School of Medicine, Zhejiang University, China. · Department of Biomedical Informatics, Harvard University, USA.
Compact but Moving: Intervention-Relevant Geometry in Recurrent World Models
Sep 18, 2026Yuming Chen, Yang LiuWorld Models World Model Learning
College of Intelligent Robotics and Advanced Manufacturing, Fudan University Shanghai, China

Oct 4, 2026J/K move · Enter open · S save

Same Predictions, Different Harms: Causal Auditing of Patient World Models

Yicheng Qi, Xiyi Xiong

Imperial College London, London, United Kingdom

Structural Causal Models Causal World Models Partial Identification

Open paper AskPDF arXiv

Patient world models used for clinical trial simulation can agree on transition kernels and arm-specific risks, yet disagree on the fraction of patients harmed by switching treatment---the counterfactual quantity that matters for intervention-aware reasoning. We audit this reliability gap in a two-stage shared-response SCM: a categorical intermediate health state is followed by common terminal care. Under independent stages, the sharp harm interval has closed-form endpoints for at most three intermediate states, with an exactness boundary at four states. Declared dependence and response-mismatch budgets yield calibrated outer bounds when stage independence or complete mediation is relaxed; in a symmetric three-state model the entire sensitivity frontier is sharp,

[0,\min\{1/2,1/3+(ρ+δ)/2\}]

, and shows exactly how budgets erase the gain over endpoint-only bounds. Two eight-variable response LPs propagate interventional uncertainty for finite-sample audits. Exact witnesses verify attainability. On public clinical simulators (EpiCare; sepsis), native configurations show little resolved stage dependence and no additional joint-compatibility gain over pairwise transport---honest negative results for reliability claims. All experiments are locally reproducible; guarantees remain conditional on the stated causal model. The results provide a concrete protocol for deciding when a patient world model is safe to trust for counterfactual harm.

Abstract

Causal Intervention

Explore similar work

Same Predictions, Different Harms: Causal Auditing of Patient World Models

Medical world models: representing medical states, modelling clinical dynamics and guiding intervention policies

Compact but Moving: Intervention-Relevant Geometry in Recurrent World Models