Research map / Graph learning / Graph self-supervised learning
Learning the Graphical Structure of Electronic Health Records with Graph Convolutional Transformer
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 |
|---|---|---|---|
| Combine conditional-probability guidance with masked, learned self-attention | Transformer for synthetic graph reconstruction · Synthetic graph reconstruction test | AUCPR: 0.5752 (0.0015) -> 0.5965 (0.0031) | Helped |
| Use conditional-probability guidance and masked, learned self-attention instead of a shallow bag-of-features model | Synthetic diagnosis-treatment classification · Synthetic diagnosis-treatment classification test | AUCPR: 0.8555 (0.0206) -> 0.8671 (0.0247) | Helped |
| Use learned structure without access to the true adjacency matrix | GCT on synthetic masked diagnosis prediction · Synthetic masked diagnosis code prediction test | accuracy: 0.2834 (0.0065) -> 0.2179 (0.0071) | Hurt |
| Combine conditional-probability guidance with learned attention | GCN_P on eICU masked diagnosis prediction · eICU masked diagnosis code prediction test | accuracy: 0.7432 (0.0086) -> 0.7704 (0.0039) | Helped |
| Learn latent encounter structure with GCT rather than using a deep feedforward bag-of-features model | eICU readmission prediction · eICU readmission prediction test | AUCPR: 0.5050 (0.0126) -> 0.5244 (0.0142) | Helped |
| Learn latent encounter structure with GCT rather than using a deep feedforward bag-of-features model | eICU mortality prediction · eICU mortality prediction test | AUCPR: 0.5924 (0.0121) -> 0.5992 (0.0223) | Helped |
| Use conditional probabilities instead of unguided Transformer attention | Synthetic structure learning · Synthetic graph reconstruction, diagnosis-treatment classification, and masked diagnosis prediction | KL divergence to true adjacency: 19.6268 (2.9114) -> 7.6490 (0.0476); 14.3178 (0.2084) -> 8.0363 (0.0305); 15.1837 (0.8646) -> 8.9648 (0.1944) | Helped |
| Use latent encounter structure for mortality prediction | GCT versus Deep on eICU · eICU mortality prediction test | AUROC: 0.9092 (0.0060) -> 0.9120 (0.0048) | Mixed |
Closest papers by meaning
- One Loss to Rule Them All: Marked Time-to-Event for Structured EHR Foundation Models (NeurIPS 2026 · similarity 0.82)
- SMART: Towards Pre-trained Missing-Aware Model for Patient Health Status Prediction (NeurIPS 2024 · similarity 0.82)
- GraphCare: Enhancing Healthcare Predictions with Personalized Knowledge Graphs (ICLR 2024 · similarity 0.82)
- Advancing Medical Multimodal Learning and Data Generation with Diffusion Model and LLM (AAAI 2025 · similarity 0.82)
- Medformer: A Multi-Granularity Patching Transformer for Medical Time-Series Classification (NeurIPS 2024 · similarity 0.82)
- Risk Horizons: Structured Hypothesis Spaces for Longitudinal Clinical Prediction (NeurIPS 2026 · similarity 0.81)
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.