mlsanityData to 14 Oct 2026

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

arXiv 2605.20292 ↗First posted 19 May 2026cs.LGcited by 0

Not on the map. Why?

Abstract (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.

Key facts. No citing papers in the map yet (data through 14 Oct 2026). Builds on 27 of its 53 references that are in the map. First posted on arXiv 19 May 2026.

Builds on

27 of 53 references.

References · 27

  1. 2017Attention Is All You Needcited by 60,361
  2. 2017A Unified Approach to Interpreting Model Predictionscited by 6,555
  3. 2016Recurrent Neural Networks for Multivariate Time Series with Missing Valuescited by 554
  4. 2016Self-critical Sequence Training for Image Captioningcited by 547
  5. 2019ERASER: A Benchmark to Evaluate Rationalized NLP Modelscited by 530
  6. 2017Multitask learning and benchmarking with clinical time series datacited by 290
  7. 2016Rationalizing Neural Predictionscited by 281
  8. 2021Multi-Time Attention Networks for Irregularly Sampled Time Seriescited by 238
  9. 2019Set Functions for Time Seriescited by 138
  10. 2020Learning to Faithfully Rationalize by Constructioncited by 113
  11. 2021Graph-Guided Network for Irregularly Sampled Multivariate Time Seriescited by 111
  12. 2021Self-Supervised Transformer for Sparse and Irregularly Sampled Multivariate Clinical Time-Seriescited by 96
15 more
  1. 2021Explaining Time Series Predictions with Dynamic Maskscited by 73
  2. 2023GraphCare: Enhancing Healthcare Predictions with Personalized Knowledge Graphscited by 58
  3. 2024Making Pre-trained Language Models Great on Tabular Predictioncited by 57
  4. 2024Optimized Feature Generation for Tabular Data via LLMs with Decision Tree Reasoningcited by 49
  5. 2024Reasoning-Enhanced Healthcare Predictions with Knowledge Graph Community Retrievalcited by 45
  6. 2023Encoding Time-Series Explanations through Self-Supervised Model Behavior Consistencycited by 42
  7. 2025TimeCAP: Learning to Contextualize, Augment, and Predict Time Series Events with Large Language Model Agentscited by 37
  8. 2024ClinicalBench: Can LLMs Beat Traditional ML Models in Clinical Prediction?cited by 32
  9. 2025BioClinical ModernBERT: A State-of-the-Art Long-Context Encoder for Biomedical and Clinical NLPcited by 28
  10. 2024Prompting Large Language Models for Zero-Shot Clinical Prediction with Structured Longitudinal Electronic Health Record Datacited by 22
  11. 2023DuETT: Dual Event Time Transformer for Electronic Health Recordscited by 17
  12. 2023General-Purpose Retrieval-Enhanced Medical Prediction Model Using Near-Infinite Historycited by 12
  13. 2024When Raw Data Prevails: Are Large Language Model Embeddings Effective in Numerical Data Representation for Medical Machine Learning Applications?cited by 12
  14. 2025LLM Meeting Decision Trees on Tabular Datacited by 8
  15. 2026Can we generate portable representations for clinical time series data using LLMs?cited by 4

Built on by

Not cited yet.