ChronoMedicalWorld: A Medical World Model for Learning Patient Trajectories from Longitudinal Care Data

arXiv 2026 55 method, application

TLDR

A medical world model for predicting patient trajectories from EHR data, using action-conditioned latent dynamics and evaluated on CKD eGFR forecasting.

Reasoning

The paper introduces a novel world model framework tailored to longitudinal clinical data, with a strong empirical evaluation on a real-world CKD cohort and a clear improvement over a GPT baseline. However, the abstract focuses on a single disease and does not discuss generalization to other conditions or limitations of the approach.

Read-first score

Read-first score 55, weighted from topical fit, citation, graph, method, reproducibility, and recency signals. Original total remains 45.

Recency 6%

100

Uses a gentle age decay so recent papers surface without erasing older foundations. 2026

Citation impact 18%

83.1

Uses OpenAlex-shaped citation metadata as a bibliometric attention signal, separate from paper quality. citation_normalized_percentile=0.83051346

Topical relevance 29%

64.3

Uses existing LLM keyword relevance scores normalized to 0-100. world model,world simulator,generative world model,interactive world model,video world model,world dynamics prediction,model-based reinforcement learning world model

Methodology quality 18%

Screens visible abstract and analysis fields for experiment, dataset, baseline, metric, and limitation evidence. markers=baseline

Reproducibility 18%

Screens links and visible text for paper, code, dataset, artifact, and repository signals. pdf=True; code=False; dataset=False; markers=code

Citation velocity 12%

Citation velocity estimates citations per publication-year to reduce old-paper bias. velocity=0.00

Field roles

FoundationFrontierBridge

Rank sensitivity

Stability: volatile; rank range: 340.

Keyword Scores

world model

world dynamics prediction

generative world model

interactive world model

world simulator

model-based reinforcement learning world model

video world model

Deep Analysis

Innovations

Action-conditioned latent world-model framework for learning patient trajectories from longitudinal care data
Joint-embedding state encoder with wide action encoder that admits structured intervention indicators and free-text communication embeddings
Six-term objective combining next-observation supervision, next-latent prediction, SIGReg latent regularisation, and three physiology-aware shape priors (slope, continuity, large-jump penalty)
Closed-loop rollout-prefix protocol that matches training to deployment, optimizing against multi-step error
Framework is not CKD-specific; applicable to any chronic condition cast as periodic clinical state interleaved with structured and conversational interventions

Methodology

CMWM couples a joint-embedding state encoder with a wide action encoder (structured indicators and free-text embeddings) and trains a recurrent latent transition module under a six-term objective. A closed-loop rollout-prefix protocol aligns training with deployment. The CKD instantiation is evaluated on a 2,232-patient nephrology cohort using MAE and RMSE against a tuned GPT-5.5 structured-prompting baseline.

Key Results

On the CKD case study, CMWM achieves a dynamic-50% history rollout test MAE of 7.384 and RMSE of 10.256, outperforming the GPT-5.5 baseline (7.964 MAE, 11.069 RMSE) by -7.28% and -7.35%, with gains dominated by the dialogue portion of patient–health-coach communication.

Resources

Paper