Automatic Cardiac Risk Management Classification using large-context Electronic Patients Health Records

Jacopo Vitale; David Della Morte; Luca Bacco; Mario Merone; Mark de Groot; Saskia Haitjema; Leandro Pecchia; Bram van Es

cs.CLMar 10, 2026

Automatic Cardiac Risk Management Classification using large-context Electronic Patients Health Records

Authors: Jacopo Vitale, David Della Morte, Luca Bacco, Mario Merone, Mark de Groot, Saskia Haitjema, Leandro Pecchia, Bram van Es

Organizations: Research Unit of Intelligent Health Technologies, Universit`a Campus Bio-Medico di Roma, Rome, Italy. · Central Diagnostic Laboratory, University Medical Center Utrecht, Utrecht, The Netherlands.

Abstract

To overcome the limitations of manual administrative coding in geriatric Cardiovascular Risk Management, this study introduces an automated classification framework leveraging unstructured Electronic Health Records (EHRs). Using a dataset of 3,482 patients, we benchmarked three distinct modeling paradigms on longitudinal Dutch clinical narratives: classical machine learning baselines, specialized deep learning architectures optimized for large-context sequences, and general-purpose generative Large Language Models (LLMs) in a zero-shot setting. Additionally, we evaluated a late fusion strategy to integrate unstructured text with structured medication embeddings and anthropometric data. Our analysis reveals that the custom Transformer architecture outperforms both traditional methods and generative \acs{llm}s, achieving the highest F1-scores and Matthews Correlation Coefficients. These findings underscore the critical role of specialized hierarchical attention mechanisms in capturing long-range dependencies within medical texts, presenting a robust, automated alternative to manual workflows for clinical risk stratification.

Figures & tables

Characteristic	Value
Male samples	1871 (53.73%)
Female samples	1611 (46.27%)
Age (mean ± SD)	M: 74.52 ± 9.05, F: 74.40 ± 9.51
Class 0	2808 (80.61%)
Class 1	674 (19.39%)
Total number of samples	3482

Table 1: Demographic and class distribution of the dataset.

Figure 1: Graphical representation of an EHR. Highlighted items are the one used in this study, the not highlighted one are present but discarded.

Figure 2: Schematic overview of the Hierarchical Transformer classification pipeline. The architecture processes concatenated consults via BPE tokenization and hierarchical encoding, utilizing CLS Classification Token (orange arrows) or Global Average Pooling (purple arrows) strategies before late fusion with anthropometric data (or not).

Feature	HTrans (Avg)	HTrans (CLS)	ResNet	LinearSVC
Architecture Base	Hierarchical Transformer		ResConsultNet (CNN)	SVM (Linear)
Input Representation	Learned Embeddings		Learned Embeddings	TF-IDF
Aggregation	Global Avg Pool	CLS Token	Max Pooling (Inter-layer)	Hyperplane Dist.
Structural Params
Total Layers	$3$	$3$	$8$	–
Block Composition	PreNorm $\to$ Attn/FF $\to$ Add		Conv1D $\to$ BN $\to$ ReLU	–

Table 2: Model Architecture and Hyperparameter Summary . detailed breakdown of the Hierarchical Transformers (HTrans), ResNet-1D, and LinearSVC baseline. Note the distinct classification head structures and dropout rates.

Parameter	Value
Epochs	$30$
Batch Size	$12$
Learning Rate	$3\times 10^{-5}$
Weight Decay	$1\times 10^{-4}$
Optimizer Flag	Adam (AMSGrad)
Data Split Strategy

Table 3: Training & Optimization Configuration . These hyperparameters are shared across all model variations in the experimental setup.

Models	F1-Score (%)	Precision (%)	Recall (%)	MCC-Score ⊕
Dummy Classifier	31.17 (0.53)	19.36 (0.33)	79.94 (1.47)	-0.0013 (0.016)
Linear SVC △	86.13 (0.59)	82.80 (0.57)	92.78 (0.48)	0.749 (0.011)
H-transformer 1D †	89.73 (3.39)	89.73 (3.39)	91.48 (1.25)	0.718 (0.049)
H-transformer 1D ‡	91.48 (1.25)	91.48 (1.25)	89.73 (3.39)	0.646 (0.144)
ResNet1D	78.54 (2.16)	78.54 (2.16)	78.54 (2.16)	0.371 (0.102)
Zero-Shot LLM GPT-4o-mini	34.81	21.31	94.8	0.12

Table 4: Performance of Cardiovascular Risk Management prediction based on only consult texts as input. Values are shown as mean (std).

Models	F1 (%)	Precision (%)	Recall (%)	MCC ⊕
Dummy Classifier	50.76 (0.0)	50.75 (0.0)	50.78 (0.0)	0.015 (0.0)
Linear SVM △	86.30 (0.56)	82.97 (0.54)	92.87 (0.43)	0.752 (0.010)
H-transformer1D †	92.48 (0.67)	92.48 (0.67)	92.48 (0.67)	0.758 (0.024)
H-transformer1D ‡	91.02 (1.11)	91.02 (1.11)	91.02 (1.11)	0.730 (0.028)
ResNet1D	85.48 (0.69)	85.48 (0.69)	85.48 (0.69)	0.565 (0.023)

Table 5: Performance of Cardiovascular Risk Management prediction based on consult texts and anthropometric measures as input. Values are shown as mean (std).

Explore similar work

LLMs for Cardiovascular Risk Prediction from Structured Clinical Data
Apr 21, 2026Jeba Maliha, Md Rafiul KabirMedical LLMs Clinical Risk Prediction
Central Michigan University, Mount Pleasant, MI, USA
CLIC: Contextual Language-Informed Cardiac Pathology Classification
May 18, 2026Giovani D. Lucafo, Rafael da Costa Silva, João Lucas Luz Lima Sarcinelli +2Electrocardiogram Classification Multimodal Learning
Institute of Mathematical and Computer Sciences Universidade de S˜ao Paulo S˜ao Carlos, S˜ao Paulo, Brazil
Narrative Feature or Structured Feature? A Study of Large Language Models to Identify Cancer Patients at Risk of Heart Failure
Mar 18, 2024Ziyi Chen, Mengyuan Zhang, Mustafa Mohammed Ahmed +4Healthcare Oncology
Department of Health Outcomes and Biomedical Informatics, College of Medicine, University of Florida, Gainesville, FL, USA · Division of Cardiovascular Medicine, Department of Medicine, College of Medicine, University of Florida, Gainesville, FL, USA · Division of Hematology & Oncology, Department of Medicine, College of Medicine, University of Florida, Gainesville, FL, USA

Apr 21, 2026J/K move · Enter open · S save

LLMs for Cardiovascular Risk Prediction from Structured Clinical Data

Jeba Maliha, Md Rafiul Kabir

Central Michigan University, Mount Pleasant, MI, USA

Medical LLMs Clinical Risk Prediction

Open paper AskPDF arXiv

Coronary artery disease (CAD) remains one of the leading causes of death globally, highlighting the need for reliable predictive systems to support early diagnosis and risk assessment. While traditional machine learning models perform well on structured clinical data, large language models (LLMs) present new possibilities to interpret medical information expressed in natural language. In this work, we develop a hybrid framework that bridges structured clinical data and natural-language representations for CAD prediction. Using a publicly available dataset of 1,190 patient records with 11 clinical attributes, structured variables are converted into interpretable feature representations and synthetic clinical narratives using LLMs. A validation pipeline performs reverse extraction of clinical variables and computes a consistency score with the original records, achieving an average fidelity of 94.61%. We then evaluate four conventional machine learning models and compare their performance with LLM-based classification under zero-shot and few-shot prompting settings. We use two LLMs here, GPT and Gemini. Experimental results show that Random Forest achieves the highest accuracy. Despite this advantage, LLM-based classification remains beneficial in real-world clinical settings. This is because LLMs operate directly on natural language patient descriptions, meaning that sensitive numerical patient data such as exact lab values, blood pressure readings, and diagnostic codes are kept private. Findings suggest that combining structured clinical data with LLM-generated narratives can enable new directions for hybrid clinical prediction systems.

Abstract

Explore similar work

LLMs for Cardiovascular Risk Prediction from Structured Clinical Data

CLIC: Contextual Language-Informed Cardiac Pathology Classification

Narrative Feature or Structured Feature? A Study of Large Language Models to Identify Cancer Patients at Risk of Heart Failure