CARE-RAG: Clinically Aligned Risk-Calibrated Evidence-Aware Retrieval-Augmented Generation for Safe Medical Question Answering
International Journal of Scientific Research in Engineering and Management · 2026-04 · 10 (4)
DOI: 10.55041/ijsrem62213 ↗Kurz gefasst
Fachartikel in der Zeitschrift International Journal of Scientific Research in Engineering and Management, 2026-04. Verfasst von Prudvi Saisaran Ponduru Prudvi Saisaran Ponduru.
Abstract (Originalsprache)
Medical large language models can improve access to clinical information, but high-stakes deployment is limited by unsupported claims, retrieval noise, poor citation support, and unsafe over-answering under uncertainty. This paper proposes CARERAG, a Clinically Aligned Risk-Calibrated Evidence-Aware Retrieval-Augmented Generation framework for medical question answering and clinical decision support. The proposed framework treats answerability control as the central bottleneck in high-stakes medical RAG. It combines risk-tier classification, hybrid retrieval, evidence graph construction, facet coverage scoring, claim-level entailment and contradiction verification, and risk-tiered conformal calibration. The key contribution is an end-to-end decision policy that can answer, redact unsupported claims, or abstain and escalate when evidence is insufficient. The paper presents mathematical formulation, system architecture, algorithms, evaluation protocol, safety governance, reproducibility checklist, and limitations. The manuscript is written as a framework and protocol paper: it does not fabricate clinical trial outcomes, but it specifies a rigorous pathway for empirical validation on public biomedical QA datasets and governed clinical review. The resulting design provides a scholarly foundation for safer, more transparent, and more accountable medical RAG systems.
Zitieren
APA
Ponduru, P. S. P. P. S. (2026). CARE-RAG: Clinically Aligned Risk-Calibrated Evidence-Aware Retrieval-Augmented Generation for Safe Medical Question Answering. International Journal of Scientific Research in Engineering and Management, 10(4). https://doi.org/10.55041/ijsrem62213
Vancouver
Ponduru PSPPS. CARE-RAG: Clinically Aligned Risk-Calibrated Evidence-Aware Retrieval-Augmented Generation for Safe Medical Question Answering. International Journal of Scientific Research in Engineering and Management. 2026;10(4). doi:10.55041/ijsrem62213
BibTeX
@article{Ponduru2026,
author = {Ponduru, Prudvi Saisaran Ponduru Prudvi Saisaran},
title = {CARE-RAG: Clinically Aligned Risk-Calibrated Evidence-Aware Retrieval-Augmented Generation for Safe Medical Question Answering},
journal = {International Journal of Scientific Research in Engineering and Management},
year = {2026},
volume = {10},
number = {4},
doi = {10.55041/ijsrem62213}
}
EUVIMED