Innovative Use of Genomics and AI in Respiratory Diagnostics Enhances Stewardship

A genomics–AI platform that pairs a host biomarker with a large language model improved pathogen detection in a study; investigators estimated that, if available at admission, it could have reduced inappropriate antibiotic use by more than 80%.

In an observational study of critically ill adults, the combined classifier reached an accuracy of 96% in an independent validation cohort and outperformed clinician admission diagnosis on key performance metrics. Clinically, this approach could enable earlier, more precise bedside treatment decisions for patients with acute lower respiratory failure.

The study was an observational analysis conducted in a critical care setting that evaluated diagnostic accuracy and downstream antibiotic-decision impact in adults with suspected lower respiratory tract infection. Specifically, the authors report an independent validation accuracy of 96% accuracy for the combined classifier versus lower accuracy for clinician admission diagnosis, and they assessed the model’s potential to change empiric prescribing. These results carry high validation confidence and represent a clear improvement over typical culture-based diagnostic yields.

The genomic component supplies a host-derived transcriptomic signal while the AI synthesizes clinical text and biomarker data to improve classification. The genomic measure detects host-response patterns consistent with infection, and the large language model analyzes electronic medical record text to contextualize clinical features. Together, they distinguish infection from noninfectious respiratory failure more reliably than either modality alone, yielding faster, more specific diagnostic information to support targeted therapy.

Stewardship gains are substantial: the investigators estimated that availability of the combined approach at admission could reduce inappropriate antibiotic use by more than 80%, primarily by limiting empiric broad-spectrum regimens when the classifier indicates a noninfectious cause. Greater diagnostic certainty allows clinicians to shorten or withhold empiric courses—translating into fewer drug-related adverse events and lower selection pressure for resistance. Operational consequences could include shorter antibiotic courses, fewer complications, and reduced antimicrobial-resistance pressure.

Prospective validation studies and targeted workflow pilots are the next practical steps to translate diagnostic performance into improved patient outcomes and measurable stewardship gains.

Key Takeaways:

• Combining a host transcriptomic biomarker with a large language model achieved high diagnostic accuracy (reported 96% in validation) and can substantially reduce unnecessary antibiotics.
• The primary beneficiaries are ICU and inpatient lower respiratory infection care pathways, where diagnostic uncertainty is highest.
• Implementation requires laboratory capacity, EHR integration, regulatory validation, and clinician engagement before widespread adoption.