multi-agent-llm-pipelines-multimodal-medical-referral-real-inpatient-records CLM-0613For multi-agent LLM pipelines on multimodal medical referral from real inpatient records, requiring agents to exchange typed logical edges (causal cascade, mutual reinforcement, exclusion, conditional context) in a shared reasoning graph instead of free-form dialogue raises referral F1, and the causal-chain edge type carries most of that gain.
Capability: Getting more from several models than from one
Observed on
952 real inpatient cases over 37 departments, four modality agents plus an expert arbitration agent, DeepSeek-V3 backbone, zero-shot, four-option multiple choice; gains saturate at.
Sources
- Validated by: Referral F1 on 952 real inpatient cases: 95.3% vs 91.4% for MMA and 92.6% for ToR multi-agent baselines and 89.7% for GPT-5; replacing the graph with natural-language output cost 5.1 F1, removing inter-agent collaboration 6.1, removing the knowledge base 2.3; removing the cascade relation alone dropped F1 to 90.6%, more than any other relation type. Measured on one self-constructed dataset from a single hospital, one backbone for all baselines, with module and relation-type ablations and a human-scored reasoning rubric; [truncated]
Disagreeing is the most useful thing you can do here. Nothing on this page waits for a reviewer: the counts above move when evidence lands, and every source stays attached with who brought it.
Related claims
- For medical multiple-choice QA with a multi-agent pipeline, mining natural-language lessons from its own past trajectories (both correct and failed) into a reusable library gives a larger accuracy gain than the multi-agent role decomposition itself, and the gain comes mostly from the most abstract tier: removing principle-level entries costs more than removing pattern- or case-level entries.Improving its own scaffolding from failures · reviewed by AI
- On knowledge, math and medical QA benchmarks with 8B open backbones (Llama-3.1-8B-Instruct, Qwen3-8B), the accuracy advantage of multi-agent debate pipelines over a single agent can be matched by a single-turn single agent whose hidden state is steered by a query-composed mixture of role vectors, at roughly a twentieth of the token cost, so the benefit of these pipelines comes from multi-perspective specialization rather than from inter-agent text exchange.Getting more from several models than from one · reviewed by AI
- In multi-agent LLM systems where one agent starts with an erroneous shared belief, broadcasting every agent's full context at every step raises the rate of false assertions above doing no synchronization at all, because the error propagates to agents that were previously correct — and the harm appears only in tasks where one wrong fact cascades across semantically linked dimensions (destination to airport to weather), not where agent contexts are largely orthogonal.Stating false facts confidently · reviewed by AI
- For LLM agents that must compose and then execute domain tool workflows, training separate adapters for the two roles splits the gains cleanly: supervised fine-tuning of the planner on reference tool sequences mainly improves workflow-order fidelity, while node-level RL on argument-match rewards mainly improves tool-argument grounding and final-answer quality, and combining both beats either on a 9B backbone.Planning a sequence of actions before acting · reviewed by AI
- When automatically rewriting the prompts of a multi-agent LLM system from its failures, locating the responsible agent by injecting a hint into one agent at a time and using that agent's hint-corrected intermediate output as the supervision signal produces prompt updates that generalize better than critique derived only from the system-level output.Improving its own scaffolding from failures · reviewed by AI
Notes
Drafted from the paper by a model and filed unreviewed. Visible here so it can be read, not because anyone has vouched for it: it does not move any technique's standing and does not count toward the internal scorecard. Drafted confidence: low. Falsifier as drafted: A matched comparison in which free-form multi-agent dialogue with the same agents, knowledge base and round budget reaches the same referral F1, or in which removing the cascade relation costs no more than removing any other relation type.