Rituparna Datta

Background: Accurate clinical outcome prediction using electronic health records (EHRs) is crucial for patient care and resource allocation. EHRs include both structured data and rich, unstructured clinical notes. However, prior machine learning methods struggle with the multimodality, long context of notes, and severe class imbalance in clinical tasks. Objective: This study aimed to introduce and evaluate KAMELEON (Knowledge-Augmented Multimodal EHR Learning for Outcome Prediction), a unified, 2-stage hybrid framework that integrates diverse EHR modalities and external biomedical knowledge to enhance clinical risk prediction. Methods: This study used the publicly available, deidentified Medical Information Mart for Intensive Care-III dataset, which includes structured and unstructured data for over 40,000 intensive care unit patients. The 2 tasks studied were 30-day readmission (403/10,031, 4% positive rate) and in-hospital mortality prediction (2423/17,903, 13% positive rate). Train-test splits were patient-disjoint (80:20). Performance was evaluated against general and medical large language models (LLMs) and structured baselines. Key metrics included the area under the receiver operating characteristic curve (AUROC), area under the precision-recall curve (AUPRC), and macro F1-score. Results: The KAMELEON framework consistently outperformed all existing baselines. 30-day readmission: the KAMELEON-balanced random forests model achieved an AUROC of 0.85 and a sensitivity (recall) of 0.79. Ablation analysis shows the critical role of the LLM-generated reasoning, with its removal causing the AUROC to drop from 0.85 to 0.7 and sensitivity to fall by over 80%. In-hospital mortality: the KAMELEON-extreme gradient boosting model achieved an AUROC of 0.92 and an AUPRC of 0.650. Unstructured-only models showed limited ability to discern mortality, with AUROC values near chance (around 0.51-0.53). Conclusions: To our knowledge, KAMELEON represents one of the first systematic frameworks to enhance LLMs for health care prediction through graph-guided knowledge retrieval combined with structured machine learning. The framework demonstrates superior performance across both prediction tasks, highlighting the synergistic value of combining diverse data modalities and LLM reasoning for robust clinical risk estimation.

BACKGROUND Accurate clinical outcome prediction using electronic health records (EHRs) is crucial for patient care and resource allocation. EHRs include both structured data and rich, unstructured clinical notes. However, prior machine learning methods struggle with the multimodality, long context of notes, and severe class imbalance in clinical tasks. OBJECTIVE This study aimed to introduce and evaluate KAMELEON (Knowledge-Augmented Multimodal EHR Learning for Outcome Prediction), a unified, 2-stage hybrid framework that integrates diverse EHR modalities and external biomedical knowledge to enhance clinical risk prediction. METHODS This study used the publicly available, deidentified Medical Information Mart for Intensive Care-III dataset, which includes structured and unstructured data for over 40,000 intensive care unit patients. The 2 tasks studied were 30-day readmission (403/10,031, 4% positive rate) and in-hospital mortality prediction (2423/17,903, 13% positive rate). Train-test splits were patient-disjoint (80:20). Performance was evaluated against general and medical large language models (LLMs) and structured baselines. Key metrics included the area under the receiver operating characteristic curve (AUROC), area under the precision-recall curve (AUPRC), and macro F1-score. RESULTS The KAMELEON framework consistently outperformed all existing baselines. 30-day readmission: the KAMELEON-balanced random forests model achieved an AUROC of 0.85 and a sensitivity (recall) of 0.79. Ablation analysis shows the critical role of the LLM-generated reasoning, with its removal causing the AUROC to drop from 0.85 to 0.7 and sensitivity to fall by over 80%. In-hospital mortality: the KAMELEON-extreme gradient boosting model achieved an AUROC of 0.92 and an AUPRC of 0.650. Unstructured-only models showed limited ability to discern mortality, with AUROC values near chance (around 0.51-0.53). CONCLUSIONS To our knowledge, KAMELEON represents one of the first systematic frameworks to enhance LLMs for health care prediction through graph-guided knowledge retrieval combined with structured machine learning. The framework demonstrates superior performance across both prediction tasks, highlighting the synergistic value of combining diverse data modalities and LLM reasoning for robust clinical risk estimation.

Growing vaccine hesitancy is contributing to the decline in immunization rates for highly contagious, vaccine-preventable childhood diseases. Therefore, there has been a significant interest in understanding how hesitancy is spreading at higher spatio-temporal resolutions, enabling more targeted interventions. Motivated by this, we study the problem of prediction of vaccine hesitancy at the ZIP Code level, referred to as the VaxHesitancy problem. A significant challenge for this problem is the lack of high-resolution data that indicates hesitancy. Here, we develop a hybrid VaxHesSTL framework that combines a Graph Neural Network (GNN) and a Recurrent Neural Network (RNN) to address the VaxHesitancy problem. The GNN uses a ZIP Code-level network to capture spatial signals from neighboring areas, while the RNN models the temporal dynamics present in the data. We train and evaluate VaxHesSTL using a large dataset, namely the All-Payer Claims Databases (APCD), for Virginia, consisting of insurance claims from over five million individuals for six years. We find that an aggregated contact network or graph, developed from a detailed activity-based population network, plays an important role in the performance of VaxHesSTL, compared to graph models based solely on spatial proximity. Experiments demonstrate that VaxHesSTL outperforms a range of state-of-the-art baselines, which rely solely on historical time series data without accounting for spatial relationships. Since hesitancy data at higher spatial resolution is often unavailable or hard to get, we incorporate an active learning approach with our VaxHesSTL framework to optimize the training set without compromising the prediction performance. We find that hesitancy data for only 18% of ZIP Codes selected by active learning allows us to forecast hesitancy for all the ZIP Codes in the Virginia.