Research map / Time series / Foundation models for time series
Forecasting Clinical Risk from Textual Time Series: Structuring Narratives for Temporal AI in Healthcare
Shahriar Noroozizadeh, Sayantan Kumar, Jeremy C. Weiss
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| use encoder models instead of decoder LLMs (ablation) | event occurrence forecasting · T2S2 F1 at 24h (DeepSeek-R1) | F1: 0.482 -> 0.576 | Improved |
| fine-tune an MLP head instead of prompting an LLM (ablation) | Llama-3.3-70B forecasting · T2S2 (Llama annotations) F1 at 24h | F1: 0.460 -> 0.653 | Improved |
| MLP-head fine-tuning instead of fine-tuned masking (ablation) | ModernBERT-base · T2S2 (DeepSeek-R1) F1 at 1w | F1: 0.697 -> 0.877 | Improved |
| fine-tuned masking instead of MLP head (ablation) | ModernBERT-base · T2S2 (Llama annotations) event ordering | c-index: 0.594 -> 0.632 | Improved |
| add biomedical pretraining (BioClinical-ModernBERT) (ablation) | ModernBERT-base encoder · T2S2 (DeepSeek-R1) event ordering | c-index: 0.646 -> 0.677 | Improved |
| add biomedical pretraining (BioClinical-ModernBERT) (ablation) | ModernBERT-base encoder · T2S2 (DeepSeek-R1) F1 at 24h | F1: 0.576 -> 0.559 | No clear effect |
| add biomedical pretraining (BioClinical-ModernBERT) (ablation) | ModernBERT-base encoder · sepsis-100 F1 at 24h | F1: 0.607 -> 0.635 | Improved |
| use biomedical-pretrained decoder (MediPhi-PubMed) vs general decoder (ablation) | LLM-MLP head · T2S2 (DeepSeek-R1) F1 at 24h | F1: 0.482 -> 0.268 | Hurt |
| exclude PubMed from pretraining (RedPajama-INCITE-7B) (ablation) | decoder LLM-MLP head · T2S2 event ordering | c-index: 0.618 vs 0.618-0.632 | No clear effect |
| use decoder LLMs instead of encoders (ablation) | survival analysis · T2S2 / sepsis-100 survival | time-dependent c-index: 0.64 vs 0.60 at 0h; RedPajama 0.76 at 168h (sepsis-100) | Improved |
| increase observation window to 168h for survival (ablation) | survival models · T2S2 survival | time-dependent c-index: DeepSeek-R1-Llama-70B 0.64 -> 0.59 | No clear effect |
| zero-shot masking with biomedical encoder (ablation) | BioClinical-ModernBERT-base · T2S2 F1 at 1h | F1: 0.246 F1 vs near-zero for BERT/RoBERTa | Improved |
Closest papers by meaning
- XTSFormer: Cross-Temporal-Scale Transformer for Irregular-Time Event Prediction in Clinical Applications (AAAI 2025 · similarity 0.88)
- Risk Horizons: Structured Hypothesis Spaces for Longitudinal Clinical Prediction (NeurIPS 2026 · similarity 0.87)
- One Loss to Rule Them All: Marked Time-to-Event for Structured EHR Foundation Models (NeurIPS 2026 · similarity 0.87)
- JoLT: Jointly Learned Representations of Language and Time-Series for Clinical Time-Series Interpretation (Student Abstract) (AAAI 2024 · similarity 0.86)
- Context Clues: Evaluating Long Context Models for Clinical Prediction Tasks on EHR Data (ICLR 2025 · similarity 0.86)
- MIRA: Medical Time Series Foundation Model for Real-World Health Data (NeurIPS 2025 · similarity 0.86)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
