Temporally Phenotyping GLP-1RA Case Reports with Large Language Models: A Textual Time Series Corpus and Risk Modeling

Sayantan Kumar; Jeremy C. Weiss

cs.CLMar 12, 2026

Temporally Phenotyping GLP-1RA Case Reports with Large Language Models: A Textual Time Series Corpus and Risk Modeling

Authors: Sayantan Kumar, Jeremy C. Weiss

Organizations: National Library of Medicine, Bethesda, Maryland, USA

Abstract

Type 2 diabetes case reports describe complex clinical courses, but their timelines are often expressed in language that is difficult to reuse in longitudinal modeling. To address this gap, we developed a textual time-series corpus of 136 PubMed Open Access single-patient case reports involving glucagon-like peptide 1 receptor agonists, with clinical events associated with their most probable reference times. We evaluated automated LLM timeline extraction against gold-standard timelines annotated by clinical domain experts, assessing how well systems recovered clinical events and their timings. The best-performing LLM produced high event coverage (GPT5; 0.871) and reliable temporal sequencing across symptoms (GPT5; 0.843), diagnoses, treatments, laboratory tests, and outcomes. As a downstream demonstration, time-to-event analyses in diabetes suggested lower risk of respiratory sequelae among GLP-1 users versus non-users (HR=0.259, p<0.05), consistent with prior reports of improved respiratory outcomes. Temporal annotations and code will be released upon acceptance.

Figures & tables

Figure 1: Left: Example case report (top) with text-ordered event-time tuples (bottom). Clinical events and temporal cues are marked in green and underline respectively. Right: Overview of our pipeline. Left panel: filtering the PMOA corpus to identify case reports of patients administered GLP1-RA medications. Middle panel: textual time series generation for each case report via LLM prompting and the creation of structured feature subsets. Right panel: Evaluation process, including text event matching, time discrepancy analysis (log- time CDF, AULTC), temporal order concordance and time-to-onset survival modeling of selected outcomes.

Figure 2: a Distribution of time series lengths (timesteps) across the dataset. b Most frequently occurring events across all case reports.

Figure 3: Frequency and prevalence patterns of UMLS-normalized diagnoses in PMOA-TTS. ( a ) Top 20 diagnoses by frequency, reported using canonical UMLS names. ( b ) Prevalence of broad disease categories in PMOA-TTS compared with published U.S. adult baseline estimates, highlighting systematic differences between case-report cohorts and general-population distributions.

Figure 4: Sensitivity analysis of clinical textual time series (TTS) quality across event-matching thresholds. Performance is summarized as concordance (ordering agreement) and AULTC (timestamp accuracy) plotted against event match rate for comparisons to Annotator 1 (top row) and Annotator 2 (bottom row). Solid circle (

\bullet

) represents threshold of 0.1, with ticks (—) indicating 0.01 increments of the threshold in [0.01, 0.50]. Curves closer to the upper-right indicate models that recover a larger fraction of expert events while preserving temporal order and producing smaller timestamp discrepancies.

Figure 5: Time-to-onset survival modeling using GLP-1RA textual time series. Left : age/sex-adjusted event-free survival curves from Cox proportional hazards models for cardiovascular, respiratory, and kidney outcomes (treatment/control: diabetes patients with/without GLP medication exposure). Shaded bands denote uncertainty for the adjusted curves. Right : corresponding adjusted hazard ratios (95% CI, p-value) and the keyword lexicons used to identify outcomes from timeline events.

Clinical Information Extraction

Clinical Information Extraction turns clinical notes, reports, and patient narratives into structured, searchable facts and relations.

See all 106 papers

GlyRAG: Context-Aware Retrieval-Augmented Framework for Blood Glucose Forecasting
Jan 8, 2026Shovito Barua Soumma, Hassan GhasemzadehTime Series Forecasting Blood Glucose Forecasting
College of Health Solutions, Arizona State University, Phoenix, AZ 85004, USA · School of Computing and Augmented Intelligence, Arizona State University, Tempe, AZ 85281, USA
Hindsight Bias in Clinical Temporal Reasoning: How Future Data Exposure Affects Large Language Model Judgment
Sep 11, 2026Misaki Matsuura, Sayantan Kumar, Ojas Kadam +1Temporal Reasoning in Language Models Clinical Reasoning
Case School of Engineering, Case Western Reserve University, USA · National Library of Medicine, National Institutes of Health, USA · Rice University University, USA
It's All in the Way You Say It: The Role of Information Representation in LLM-Based Glycemic-Event Prediction
Sep 8, 2026Andrea Apicella, Pasquale Arpaia, Matteo Orefice +2LLM Prompting Blood Glucose Forecasting
Department of Information Engineering, Electrical Engineering, and Applied Mathematics (DIEM), University of Salerno, Fisciano (SA), Italy · Department of Electrical Engineering and Information Technology, University of Naples Federico II, Naples, Italy

Jan 8, 2026J/K move · Enter open · S save

GlyRAG: Context-Aware Retrieval-Augmented Framework for Blood Glucose Forecasting

Shovito Barua Soumma, Hassan Ghasemzadeh

College of Health Solutions, Arizona State University, Phoenix, AZ 85004, USA · School of Computing and Augmented Intelligence, Arizona State University, Tempe, AZ 85281, USA

Time Series Forecasting Blood Glucose Forecasting LLM-Based Time Series Forecasting Retrieval-Augmented Time Series Forecasting Wearable Health Monitoring Continuous Glucose Monitoring

Open paper AskPDF arXiv

Accurate blood glucose forecasting using continuous glucose monitoring (CGM) data can support the early prediction of dysglycemic risk. However, current neural-network-based forecasting models treat CGM data as a purely numerical sequence without integrating the contextual information contained in CGM signal morphology. Recently, large language models (LLMs) have shown promise for time-series forecasting, yet their role as agentic context extractors in diabetes care remains largely unexplored. In this study, we bridge glucose forecasting and LLM-based contextualization by developing GlyRAG, a context-aware, retrieval-augmented forecasting framework that uses an LLM as a contextualization agent to summarize glucose morphology directly from a timed CGM window. The generated CGM-only narrative is embedded and fused with patch-based glucose representations, while a retrieval module incorporates similar historical training episodes through cross-attention. We evaluate GlyRAG on the OhioT1DM and AZT1D datasets for 5-, 30-, and 60-minute forecasting horizons. Compared with strong CGM-only baselines, GPT-4 GlyRAG significantly improves long-horizon root mean square error (RMSE) over PatchTST on both datasets. For example, RMSE decreases from 13.8 to 10.6 at 30 minutes and from 23.1 to 20.2 at 60 minutes on OhioT1DM. LLaMA 3.1 produces smaller but significant long-horizon gains, suggesting that the contextualization pipeline is not limited to GPT-4. Clinical error-grid analyses further show that approximately 85 percent of predictions fall within the clinically acceptable Clarke Error Grid Zones A and B. These results suggest that CGM-derived linguistic context and case-based retrieval can improve long-horizon glucose forecasting without requiring additional sensing modalities.

Abstract

Clinical Information Extraction

Explore similar work

GlyRAG: Context-Aware Retrieval-Augmented Framework for Blood Glucose Forecasting

Hindsight Bias in Clinical Temporal Reasoning: How Future Data Exposure Affects Large Language Model Judgment

It's All in the Way You Say It: The Role of Information Representation in LLM-Based Glycemic-Event Prediction