Track papers across labs, topics, and trends.
Search papers, labs, and topics across Lattice.
This study introduces ClinTraceBench, a benchmark consisting of 385 verified dialogues derived from MIMIC-IV, designed to evaluate the effectiveness of various history representation strategies for clinical reasoning in LLM assistants. The authors assess eight different strategies across four model backbones, revealing that compressed representations significantly impair the ability to track multi-visit trends and make cross-patient comparisons. Key findings indicate that while larger models generally perform better, the choice of representation can dramatically affect outcomes, challenging the assumption that bigger models always yield superior results.
Compressed history representations can lead to a staggering 62.7 percentage point drop in reasoning accuracy for clinical LLMs, challenging the assumption that larger models are always better.
Clinical LLM assistants must reason over multi-visit patient trajectories, yet whether the compact history representations used to scale them---retrieval, structured timelines, LLM summaries, agentic memory---preserve the longitudinal signal clinical reasoning needs has not been measured. We introduce ClinTraceBench: 385 MIMIC-IV-derived verified dialogues with event-ID provenance, a nine-task taxonomy (T1--T9), and L0--L4 deterministic + L5 human-audit validation (98.92\% agreement). We evaluate eight history representation strategies---a no-context floor, \textit{last-visit-only}, \textit{full-context}, BGE-M3 \textit{dense-retrieval}, two compression schemes, and two agentic-memory systems (\textit{Mem0}, \textit{A-Mem})---across four backbones (DeepSeek-V3, GPT-4o-mini, Haiku~4.5, Sonnet~4.6) on 6{,}271 questions: 32 cells, 200{,}672 predictions. Four findings: (SP4) a controlled T3 injection probe isolates compression-induced \textit{relation} loss---with the attribution sentence present \textit{before} construction, \textit{Mem0}, \textit{A-Mem} and \textit{llm-summary} still recover only 0--5.3\% of the injected positives; (SP1) compressed strategies pay an aggregation tax on multi-visit trends and cross-patient comparisons; (SP2) the blind-to-full gap spans $+29.8$~pp (GPT-4o-mini) to $+62.7$~pp (Haiku); (SP3) abstention scales non-monotonically with context length. On the Pareto frontier Haiku dominates Sonnet under \textit{full-context} (\$25.76 vs.\ \$106.21), inverting the ``biggest backbone wins'' heuristic.