Truace
Subscribe

LongCAF: A longitudinal context-constructive agentic framework for LLM-supported hypertension self-management

Objective Long-term hypertension self-management requires patients to make daily decisions and be cautious of when to seek medical attention. Although large language model (LLM)-based systems provide new opportunities for conversational self-management support, conventional question-answering conversational paradigms may be insufficient for providing accurate and personalized responses under conditions of fragmented patient expression, incomplete contextual reporting, and changing interaction intents. Therefore,…

By the Truvace desk1 min read1 sourcePeer-reviewedTRV-2026-1690· cite & verify
LongCAF: A longitudinal context-constructive agentic framework for LLM-supported hypertension self-management
"checking blood pressure" by jencu is licensed under CC BY 2.0. To view a copy of this license, visit https://creativecommons.org/licenses/by/2.0/.

In brief

Researchers developed LongCAF, a framework that organizes ongoing conversations between hypertension patients and a large language model into one workflow. It combines longitudinal memory, intent-aware routing, context-grounded response generation, proactive questions for missing information, and escalation of safety-critical cases to clinicians. They tested it in a 6-month real-world deployment.

The results suggest that conversational support for chronic disease can be made more personalized and clinically governed than plain question-answering. The evidence is a feasibility deployment, so it does not establish clinical outcomes such as blood pressure control, and the abstract does not say how well it generalizes beyond this setting.

Main points

  1. Context completeness per patient rose from 42.9% at baseline to 57.1% at follow-up, with 79,290 clinically relevant entries added.
  2. Non-blood-pressure domains made up 49.3% of newly acquired entries.
  3. Retrieved patient-specific context supported 69% of patient-facing outputs, and candidate-question adoption reached 84.4% among eligible interactions.
  4. Intent coordination reached a micro-F1 of 0.911, and safety escalation reached precision 0.915 and recall 0.977.

The gain

In a 6-month real-world deployment, an LLM framework built the patients' longitudinal context and escalated safety-critical interactions to clinicians with high precision and recall.

The rundown

The framework has five modules: longitudinal context memory, intent-aware interaction coordination, interaction synthesis, proactive communication, and safety escalation. These are meant to handle fragmented patient reports, incomplete context, and changing intents that conventional question-answering systems may handle poorly.

In deployment, the system added a large number of clinically relevant entries, and nearly half came from areas other than blood pressure. Many outputs drew on retrieved patient-specific context across multiple domains.

Clinician and patient evaluations suggested that responses were clinically relevant, contextually appropriate, and useful for continued self-management. The authors conclude that the deployment demonstrates feasibility.

Sources

  1. Peer-reviewedJournal of Biomedical Informatics2026-10-08

The debate