MIMIC-EXT-TS: A Discrete Clinical Time-Series Dataset for ICU Patients

ACL ARR 2026 January Submission7541 Authors

Keywords: rag, agent, llm

Abstract:

Early detection and timely treatment are critical in medicine. For example, surgical excision of skin lesions can cure early-stage skin cancer, but once metastasis occurs, even the most advanced therapies often fail. In this work, we introduce MIMIC-EXT-TS, a large-scale dataset with an over 11 million clinical event, timestamp pairs from over 267k free-text discharge summaries. It is the first discrete time series clinical events dataset of ICU patients. It significantly reduces the storage time and presents important clinical events and related timestamp in structured way.
To achieve the dataset, we propose an end-to-end RAG model with an LLM agent for temporal reasoning.
The integrity check confirms over 94\% of events are can be traced back to source clinical note with mean token overlap around 91\%. All important events, such as diagnosis ICD codes are captured. We further validate the effectiveness of our dataset in downstream tasks, and fine-tuning LLMs, such as Qwen and MedGemma, for medical question answering tasks on MedMCQA, MMLU, and PubMedQA dataset.

Paper Type: Long

Research Area: AI/LLM Agents

Research Area Keywords: nlp, rag, llm

Contribution Types: Data resources

Languages Studied: english

Submission Number: 7541