Factors Influencing Self-Reported Health in Puerto Rican Populations: A Systematic Review.

Large language models (LLMs) are moving into clinical practice faster than local governance can adapt. They generate fluent medical text, summarise evidence and propose actions. This creates a tempting “digital kerbside consult”. Yet fluency is not safety. In high-stakes care, the key question is whether a system remains guideline-concordant and stable when case complexity or input format changes [–]. Respiratory medicine is a stringent test-bed for LLMs. Acute decisions require rapid triage, risk stratification and multi-step management. Many actions are mandatory. Even small omissions can shift outcomes [–]. Functional respiratory diagnostics add a second layer. The interpretation of pulmonary function tests (PFTs) relies on fixed thresholds, pattern hierarchies and quality control. Errors cluster in mixed patterns and in borderline obstruction or restriction [–]. Imaging adds a third layer. Chest radiograph (CXR) interpretation depends on subtle visual patterns and clear anatomic localisation. Generalist LLMs are not trained for that. Multimodal, chest-tuned models can reduce hallucination, but still require oversight [–]. Three domains map directly to respiratory workflows and expose distinct error modes.