Keywords:-
Article Content:-
Abstract
Congestive Heart Failure (CHF) is a common and high-impact cardiovascular condition, and early risk identi- fication can support proactive clinical decision-making. We present RaX-CHF, a resource-efficient and explainable small- model pipeline that predicts the presence of CHF using structured tabular features derived from MIMIC-III ICU data. RaX-CHF integrates demographics and aggregated laboratory measure- ments at the patient level and compares classical machine learning baselines (logistic regression, Random Forest, Gradient Boosting, and XGBoost) against a locally deployed small language model (Gemma) prompted with serialized patient features. On a dataset of roughly 44,000 adult patients with 115 attributes, tree-based models achieve the best performance, with XGBoost reaching 0.858 accuracy and 0.861 sensitivity (recall for CHF). The prompted Gemma baseline achieves 0.734 accuracy, near the majority-class baseline, highlighting the continued strength of compact tabular predictors for disease presence classification. We also summarize interpretability signals from linear coefficients and tree-based feature importance, and discuss practical risks related to missingness assumptions, de-identification policy, and deployment in safety-critical settings.
References:-
References
E. Johnson, T. J. Pollard, L. Shen et al., “MIMIC-III, a freely accessible critical care database,” Scientific Data, vol. 3, 2016.
Zhu, C. Xie, Z. Wang, and H. Zhang, “RaX-Crash: A Resource Efficient and Explainable Small Model Pipeline with an Application to City Scale Injury Severity Prediction,” arXiv preprint arXiv:2512.07848, 2025. [Online]. Available: https://arxiv.org/abs/2512.07848
Z. Bao, D. Zhu, L. Jiang, S. Sheng, Z. Wang, and H. Zhang, “SLOFetch: Compressed-Hierarchical Instruction Prefetching for Cloud Microser- vices,” arXiv preprint arXiv:2511.04774, 2025. [Online]. Available: https://arxiv.org/abs/2511.04774
Zhang, S. Zhao, Z. Zhou, W. Zhang, and Y. Meng, “Domain- Specific RAG with Semantic Normalization and Contrastive Feedback for Document Question Answering,” in Proc. 2025 7th Int. Conf. Internet of Things, Automation and Artificial Intelligence (IoTAAI), 2025, pp. 750–753, doi: https://doi.org/10.1109/IoTAAI66837.2025.11213473.
Z. Zhou, C. Zhao, X. Li, H. Zhang and R. Chang, “Diverse Stacking Ensemble for Attributing LLM Outputs via Relational Reasoning,” in Proc. 2025 8th Int. Conf. on Computer Information Science and Application Technology (CISAT), 2025, pp. 1089–1092, doi:
https://doi. org/10.1109/CISAT66811.2025.11181963
P. Ponikowski, A. A. Voors, S. D. Anker et al., “2016 ESC Guidelines for the diagnosis and treatment of acute and chronic heart failure,” European Heart Journal, vol. 37, no. 27, pp. 2129–2200, 2016, doi: https://doi. org/10.1093/eurheartj/ehw128.
Shickel, P. J. Tighe, A. Bihorac, and P. Rashidi, “Deep EHR: A survey of recent advances in deep learning techniques for electronic health record analysis,” IEEE Journal of Biomedical and Health Informatics, vol. 22, no. 5, pp. 1589–1604, 2018, doi: https://doi.org/10.1109/JBHI.2017.2767063.
S. Hegselmann, A. Buendia, H. Lang, M. Agrawal, X. Jiang, and D. Sontag, “TabLLM: Few-shot classification of tabular data with large language models,” in Proc. 26th Int. Conf. on Artificial Intelligence and Statistics (AISTATS), 2023, pp. 5549–5581.