Structured Evidence Routing for Incident Risk Prediction from Multimodal Longitudinal EHRs

Animesh Agarwal; Meysam Ghaffari; Nina Fatehi; Carlos Morato

cs.AIAug 24, 2026

Structured Evidence Routing for Incident Risk Prediction from Multimodal Longitudinal EHRs

Authors: Animesh Agarwal, Meysam Ghaffari, Nina Fatehi, Carlos Morato

Organizations: Optum AI, UnitedHealth Group, Minneapolis, Minnesota, USA.

Abstract

Incident risk prediction from longitudinal electronic health records (EHRs) is challenging because relevant signals are multimodal, weak in isolation, and distributed across irregular patient histories. We propose structured evidence routing, a router-predictor-reviewer workflow that separates full-record access from disease-specific assessment. The router organizes the complete pre-index EHR into a compact summary and targeted evidence slices; the predictor uses this evidence to form an evidence-linked risk assessment, which the reviewer critiques. For comparison with supervised EHRSHOT baselines, we pair the routed evidence summaries with a supervised classifier readout. Across five 1-year incident diagnosis tasks, our method reaches the AUROC range of established supervised EHRSHOT baselines and remains competitive on AUPRC, while exposing a patient-specific evidence trail. Internal pre-readout ablations further suggest that routing, laboratory evidence, task guidance, and review each contribute to performance.

Figures & tables

Figure 1: Structured evidence routing for incident risk prediction. A record router organizes long, irregular EHR histories into a compact summary and targeted evidence slices. A review step can trigger refinement before final disease-specific prediction.

Task	Count-based	CLMBR	MoA+Cls	Ours+Cls
Hypertension	$0.73\pm 0.003$	$0.70\pm 0.003$	$0.73\pm 0.01$	$0.73\pm 0.03$
Hyperlipidemia	$0.75$	$0.70$	$0.72\pm 0.01$	$0.75\pm 0.02$
Lupus	$0.76$	$0.77\pm 0.04$	$0.82\pm 0.03$	$0.82\pm 0.04$
Pancreatic cancer	$0.89$	$0.82$	$0.84\pm 0.005$	$0.88\pm 0.03$
Acute myocardial infarction	$0.76\pm 0.01$	$0.74\pm 0.01$	$0.76\pm 0.01$	$0.77\pm 0.02$

Table 1: Supervised-readout AUROC on EHRSHOT for 1-year incident diagnosis prediction. Our method uses routed evidence summaries with a trained classifier readout, placing it in the same supervised task-label setting as count-based, CLMBR, and MoA+Cls baselines. For Ours+Cls,

\pm

denotes the bootstrap standard error estimated from 1,000 resamples; baseline uncertainties are reported as provided by prior work.

Task	Count-based	CLMBR	MoA+Cls	Ours+Cls
Hypertension	$0.40\pm 0.01$	$0.32\pm 0.01$	$0.36\pm 0.02$	$0.36\pm 0.03$
Hyperlipidemia	$0.36\pm 0.005$	$0.29$	$0.30\pm 0.01$	$0.34\pm 0.04$
Lupus	$0.14\pm 0.04$	$0.10\pm 0.01$	$0.16\pm 0.03$	$0.18\pm 0.07$
Pancreatic cancer	$0.39\pm 0.04$	$0.20$	$0.36\pm 0.01$	$0.34\pm 0.07$
Acute myocardial infarction	$0.27\pm 0.01$	$0.23\pm 0.01$	$0.28\pm 0.03$	$0.27\pm 0.04$

Table 2: Supervised-readout AUPRC on EHRSHOT for 1-year incident diagnosis prediction. For Ours+Cls,

\pm

denotes the bootstrap standard error estimated from 1,000 resamples; baseline uncertainties are reported as provided by prior work.

Configuration	Pre-readout AUROC
Full framework	$0.86\pm 0.03$
$-$ Reviewer	$0.84\pm 0.033$
$-$ Lab measurements	$0.83\pm 0.034$
$-$ Record Router	$0.83\pm 0.037$
$-$ Task guidance	$0.84\pm 0.030$

Table 3: Internal pre-readout ablation on pancreatic cancer.

Clinical Prediction

Clinical prediction uses patient records, signals, and images to estimate diagnoses, risks, outcomes, and treatment responses for care.

See all 202 papers

Learning to Select Source-Traceable Evidence for Language-Model Prediction from Irregular Clinical Time Series
May 19, 2026Kwanhyung Lee, Juhwan Choi, Jongheon Kim +5Irregular Time-Series Modeling Healthcare
Kim Jaechul Graduate School of AI, KAIST · AITRICS
Multimodal Routing for Interpretable, Robust, and Auditable Clinical Prediction
Jul 10, 2026Nikkie Hooman, Zhongjie Wu, Eric C. Larson +1Multimodal Robustness Clinical Prediction
Southern Methodist University Dallas, USA
GLoC-EHR: Evidence-Cited Clinical Reasoning over Global Context and Local EHR Events
Oct 1, 2026Chaiho Shin, Kwangsoo KimEvidence-Grounded Reasoning Clinical Outcome Prediction
Interdisciplinary Program of Medical Informatics, Seoul National University College of Medicine · Department of Transdisciplinary Medicine, Seoul National University Hospital · Center for Data Science, Healthcare AI Research Institute, Seoul National University Hospital +1

May 19, 2026J/K move · Enter open · S save

Learning to Select Source-Traceable Evidence for Language-Model Prediction from Irregular Clinical Time Series

Kwanhyung Lee, Juhwan Choi, Jongheon Kim, Joohyung Lee+4

Kim Jaechul Graduate School of AI, KAIST · AITRICS

Irregular Time-Series Modeling Healthcare Evidence Selection Clinical Outcome Prediction Longitudinal EHR Modeling

Open paper AskPDF arXiv

Numerical time-series models effectively process irregular electronic health record (EHR) trajectories, but do not expose which temporal patterns support each prediction as readable evidence. Existing text-based interfaces either serialize observations, preserving source traceability but offering limited clinical interpretation, or generate patient-level summaries that improve readability but can obscure links to source measurements. We introduce STEP-CTS (Source-Traceable Evidence for Prediction from Clinical Time-Series), which learns to select source-traceable text evidence for a language-model predictor. Multi-scale window statistics of each trajectory are verbalized as sets of deterministic threshold predicates, each set linked to its source window (e.g., "[5-7 h]: last temperature at least 38C"). An offline LLM, run once per unique predicate set without access to patient records, the prediction task, or outcome labels, attaches clinical concepts such as "fever" to their supporting predicates and abstains when none applies. Each predicate set, together with its concepts, forms an evidence unit. A learned Evidence Selector selects a fixed-size subset of the evidence units, which a pretrained clinical language-model encoder reads to make the prediction, with no patient-level text generation. Across three ICU benchmarks, STEP-CTS outperforms evaluated text-based baselines, improving AUPRC over the strongest by 5.1, 3.7, and 11.7 percentage points on P2012, MIMIC-III, and P2019, respectively, and is competitive with dedicated numerical time-series models. Ablations show that clinical concepts and learned evidence selection each contribute to predictive performance. In a blinded clinician study, the selected evidence is rated as traceable as deterministic serialization, near the ceiling of the scale, and above generated summaries on clinical interpretation.

Abstract

Clinical Prediction

Explore similar work

Learning to Select Source-Traceable Evidence for Language-Model Prediction from Irregular Clinical Time Series

Multimodal Routing for Interpretable, Robust, and Auditable Clinical Prediction

GLoC-EHR: Evidence-Cited Clinical Reasoning over Global Context and Local EHR Events