5arXiv cs.CL (Computation and Language)·
MOSAIC: Multi-LLM agentic framework for clinical severity phenotyping in EHR data
MOSAIC is a two-phase agentic LLM framework for disease severity phenotyping applied to type 2 diabetes, evaluated on a synthetic EHR cohort of up to 4,886 patients. The system incorporates domains absent from traditional algorithmic comparators—including glycaemic staging, beta-cell function, and social determinants of health—and shows open-weight models matching proprietary pipelines (weighted kappa 0.773). Agentic classification diverged meaningfully from deterministic rule execution of the same rubric (kappa 0.428), suggesting genuine reasoning beyond fixed rules. The work provides early evidence that agentic LLM systems can generate clinically meaningful severity phenotypes from structured EHR data.
Related guides (1)
Related events (8)
Large Cancer Assistant (LCA): Model-agnostic orchestration framework for oncology clinical decision support
Researchers propose the Large Cancer Assistant (LCA), a model-agnostic orchestration framework for multimodal AI in oncology that decouples data ingestion, clinical routing, and AI inference via a 7-tuple architecture grounded in 'Algorithmic Impermeability.' The system uses Geometric Deep Learning to standardize multimodal patient data and outputs a Standardized Intermediate Payload (SIP) to isolate AI execution from hospital IT infrastructure. A proof-of-concept validated orchestration logic across four scenarios, demonstrating invariant routing during model swaps and 100% recall on failure-safety under injected data anomalies. The framework targets EMR interoperability and modular deployment of heterogeneous oncology AI models.
MedRLM: Recursive multimodal agent framework for long-context clinical decision support
MedRLM is a proposed framework for clinical decision support that uses recursive multi-agent reasoning over heterogeneous patient data including EHRs, medical images, physiological sensor streams, and clinical guidelines. Rather than single-step prompting, it decomposes patient cases into an inspectable external environment coordinated by specialized agents, with a Clinical Evidence Graph Memory and sensor-triggered deeper reasoning. The paper outlines an evaluation design using public and credentialed clinical datasets spanning radiology, ECG, ICU time series, and referral outcomes. The work targets a gap between static medical QA benchmarks and real-world longitudinal clinical workflows.
Agentic CLEAR: Automating Multi-Level Evaluation of LLM Agents
Agentic CLEAR is an automatic evaluation framework for LLM-based agentic systems that analyzes behavior at three granularity levels: system, trace, and node. Unlike existing tools that rely on static error taxonomies or focus only on observability, it dynamically generates textual insights and integrates above the observability layer with an accessible UI. Experiments across four benchmarks and seven agentic settings demonstrate strong alignment with human-annotated errors and predictive accuracy for task success rates.
Risk-Aware Hybrid Selective Classification for HIV Suspicion Identification in Spanish Clinical Notes
This paper proposes a hybrid selective classification framework for clinical NLP that explicitly handles both aleatoric and epistemic uncertainty to avoid overconfident predictions in medical triage settings. The system combines Mondrian conformal prediction with a Multi-Centroid Mahalanobis Distance veto, evaluated on HIV suspicion identification in Spanish clinical notes. The authors demonstrate that standard uncertainty metrics and baseline classifiers suffer coverage collapse under strict reliability constraints, while their dual-verification approach isolates a trustworthy operational domain. The work critiques inflated benchmark metrics that arise from forcing deterministic classification on inherently ambiguous clinical instances.
ClinEnv: Interactive Multi-Stage Long-Horizon EHR Benchmark for Clinical Agent Evaluation
ClinEnv is a new interactive benchmark that evaluates LLMs as attending physicians over real inpatient admissions using a Longitudinal Inpatient Simulation paradigm. Each case is decomposed into sequential decision stages where models must query four specialized agents before committing to medications, procedures, and diagnoses. Across seven evaluated models, the best achieves only 0.31 decision F1, with a sharp gap between diagnosis recovery (0.51 F1) and management actions (0.17 F1). The benchmark uniquely measures information-acquisition process quality alongside outcome quality, exposing a gap invisible to static or outcome-only evaluations.
Persona-conditioned LLM support for people who use drugs reveals tension between generic empathy and clinical alignment
Researchers present a proof-of-concept study using Latent Profile Analysis on Reddit data to identify four self-stigma personas among people who use drugs, then train classifiers to detect these personas from posting history (macro-F1 = 0.74 at 30 posts). Persona-matched LLM responses achieved targeted behavioral shifts, but clinical expert raters preferred the generic empathy of persona-neutral baselines. The core finding is a misalignment: holistic empathy judgments and clinically-aligned response design can pull in opposite directions, suggesting current evaluation rubrics for LLM-based mental health support are inadequate.
LLM-guided MAP-Elites evolution improves medical decision pipelines at inference time
Researchers propose using LLM-guided MAP-Elites evolutionary search as an inference-time alternative to fine-tuning for adapting LLMs to clinical workflows, formulating triage, consultation, and image classification as evolutionary searches over executable artifacts. Across three medical settings, evolved programs substantially outperform manually designed baselines: triage accuracy improves from 77.3% to 87.1% and emergency recall from 0.60 to 0.97, with gains also shown on MIMIC-ESI, iCRAFTMD, and PneumoniaMNIST. The approach works across Llama-3, Qwen-3.5, and Gemma-4 backbones and produces interpretable program-level mechanisms rather than superficial prompt changes.
ClaMPAPP: Hybrid LLM-ML system uses language models as interfaces for pediatric appendicitis diagnosis
Researchers introduce ClaMPAPP, a hybrid clinical decision support system that uses an LLM solely for structured feature extraction from free-text clinical notes, then passes validated features to an XGBoost classifier for final diagnosis. Evaluated on two independent German pediatric appendicitis cohorts, ClaMPAPP outperformed end-to-end LLM baselines on diagnostic performance and showed greater robustness to narrative reordering. The work formalizes an 'LLM-as-interface, ML-as-predictor' design pattern that separates natural-language usability from predictive inference, offering a more auditable pathway for clinical AI.