Graph-based model improves early detection of gastrointestinal cancers
GraphBEAT: Graph-based Bucketized Event Aggregation Over Time for Early Risk Stratification of Gastrointestinal Cancers
Machine LearningArtificial Intelligence
Summary
Doctors keep many records about patients over time, but these records are often messy and incomplete. The authors created GraphBEAT, a method that organizes patient data into timed groups and turns it into a graph to better predict who might get certain digestive system cancers years in advance. This method was tested on patient records from multiple places and was more accurate and faster than previous models at guessing cancer risk. Its efficiency means it could be easier to use in real hospital settings.
What this means in practice
- •For hospital data teams: Enable more accurate and faster cancer risk predictions from patient history using limited, irregular clinical data.
- •For medical software developers: Build efficient clinical decision support tools that use graph representations of patient records to identify early cancer risks.$Commercial implications: Supports development of predictive tools sold to healthcare providers for early cancer screening.
Authors
Leo Y. Li-Han, Chris Varghese, Elizabeth B. Habermann, Mark J. Truty, William R. G. Perry, Cornelius A. Thiels, Hojjat Salehinejad
Abstract
Longitudinal electronic health records (EHRs) contain complementary diagnostic and laboratory information about patient health. However, irregularity, sparsity, heterogeneity, and incompleteness in documentation complicate EHR modeling. To address the problem, we introduce GraphBEAT (Graph-based Bucketized Event Aggregation over Time), a temporal graph representation learning framework that groups clinical events into time buckets and transforms observed features into patient-specific, temporally directed graphs to enable early risk stratification for gastrointestinal cancers. Each node in an EHR graph encodes feature identity, observed value, and temporal position of clinical events. A context-aware, time-decay message-passing mechanism models transitions between successive buckets, followed by multi-head attention pooling to generate patient-level representations. GraphBEAT was trained and evaluated on multi-site longitudinal diagnosis codes and laboratory tests to predict pancreatic, liver, and intestinal cancer at 12-, 24-, and 36-month diagnostic lead times. In all experimental setups, GraphBEAT consistently outperformed baseline graph neural networks (GraphSAGE, GAT, and GCN) and sequence modeling approaches (Transformer and long short-term memory [LSTM]). GraphBEAT also required over 250x fewer floating-point operations per patient and over 20x fewer parameters than the Transformer and LSTM baselines. These results support GraphBEAT as an accurate and computationally efficient approach for cancer risk stratification from sparse, heterogeneous longitudinal health records.