C-PPDSS: A Comprehensive Patient Personal Data Sovereignty System for Real-Time PII Redaction in Healthcare LLM Workflows

Counts differ because each database indexes a different set of publications. We treat OpenAlex as the canonical count; Google Scholar is not shown (no API, and crawling it violates its ToS).

Health surveillance in modern healthcare increasingly relies on AI, raising concerns about the exposure of patients' identifiable information (PII) in automated workflows. This paper presents the Comprehensive Patient Personal Data Sovereignty System (CPPDSS), a privacy-preserving framework designed to enforce real-time PII redaction during outbound interactions with large language models. C-PPDSS employs a fine-tuned Custom PIIRANHA model, based on Microsoft's DeBERTa-v2 architecture, to intercept and sanitize HTTP POST requests containing sensitive healthcare data before they reach external LLM services. Evaluated on a held-out test set of 20,927 annotated examples, Custom PIIRANHA achieved 99.51% token-level accuracy and a micro-averaged F1 score of 95.45%, outperforming a fine-tuned Distilled BERT baseline across all major metrics. Quantitative benchmarking demonstrated a mean inference latency of 229.66 ms per request, a model memory footprint of 305.7 MB, and a sustained throughput of 4.35 requests per second on CPU-only hardware, confirming suitability for real-time deployment on modest intranet infrastructure. By securing patient information at the network level, C-PPDSS bridges a gap left by previous static-document-focused systems, enabling safe AI integration in healthcare workflows. Its architecture ensures reproducibility and feasibility for deployment in hospital intranets or standard commodity hardware setups.