Maintenance Notice

Due to necessary scheduled maintenance, the JMIR Publications website will be unavailable from Wednesday, July 01, 2020 at 8:00 PM to 10:00 PM EST. We apologize in advance for any inconvenience this may cause you.

Who will be affected?

Accepted for/Published in: Journal of Medical Internet Research

Date Submitted: Oct 12, 2025
Date Accepted: May 8, 2026

The final, peer-reviewed published version of this preprint can be found here:

Human-in-the-Loop as a Safety Guardrail: Clinical Accountability in the Large Language Model Era

Zablah I, Molina Y, Garcia-Loureiro A

Human-in-the-Loop as a Safety Guardrail: Clinical Accountability in the Large Language Model Era

J Med Internet Res 2026;28:e85726

DOI: 10.2196/85726

PMID: 42313658

Human-in-the-Loop as a Safety Guardrail: Clinical Accountability in the LLM Era

  • Isaac Zablah; 
  • Yolly Molina; 
  • Antonio Garcia-Loureiro

ABSTRACT

We talk about Zhang et al.'s review of LLMs in healthcare and stress that how well they work in practice is what matters most. There are still some big problems: 2–10 second latencies, a lot of VRAM needed, and the ability to handle many users at once. We suggest standard metrics (operations per diagnosis, energy per inference, cost-effectiveness) and ways to improve performance, such as quantization/pruning, edge deployment, and hybrid architectures. In an initial benchmarking, approximately 14 billion-parameter medical models attained 85–90% of GPT-4’s accuracy utilizing roughly 15% of the resources. We think that HPC and biomedical informatics should be aligned for fair and efficient use.


 Citation

Please cite as:

Zablah I, Molina Y, Garcia-Loureiro A

Human-in-the-Loop as a Safety Guardrail: Clinical Accountability in the Large Language Model Era

J Med Internet Res 2026;28:e85726

DOI: 10.2196/85726

PMID: 42313658

Download PDF


Request queued. Please wait while the file is being generated. It may take some time.

© The authors. All rights reserved. This is a privileged document currently under peer-review/community review (or an accepted/rejected manuscript). Authors have provided JMIR Publications with an exclusive license to publish this preprint on it's website for review and ahead-of-print citation purposes only. While the final peer-reviewed paper may be licensed under a cc-by license on publication, at this stage authors and publisher expressively prohibit redistribution of this draft paper other than for review purposes.