Predicting hospital readmissions in severe COPD patients using an electronic-nose.

Ana Maria Rodrigo Troyano, Jordi Giner, Diego Sousa, Jose Luis Merino, Ana Alonso, Anna Feliu, Vicente Plaza, Borja Garcia-Cosio, Alver Agusti, Oriol Sibila

Research output: Contribution to journalLiterature reviewResearchpeer-review


Introduction: Hospital readmissions are critical in the natural history of severe COPD patients. However, its prediction is complicated in clinical practice. The electronic nose is a non-invasive technology capable of distinguish volatile organic compounds (VOC) breath-prints in exhaled breath.

Objective: We aim to explore if an electronic nose can reliably predict the presence of future hospital readmissions in admitted patients with Chronic Obstructive Pulmonary Disease (COPD) and frequent severe exacerbations.

Methods: Eighty-eight hospitalized COPD patients with two ore more severe exacerbations in the previous year were included. On admission, clinical, functional, microbiological and laboratory data were recorded. During the first 24 hours of admission, exhaled breath was collected in Tedlar bags and VOCs breath-prints were detected by the commercially available electronic nose Cyranose 320®. Cross-validation accuracy was assessed using principal component reduction analysis (PCA). Patients were followed during 90 days after discharge.

Results: Seventeen patients (19%) were readmitted at 30-days, and 33 patients (37%) were readmitted at 90-days. No clinical differences on admission were observed among patients who were readmitted or not. However, VOC breath-prints were different among both groups. In predicting 30-day readmission, the accuracy was of 87.5%, sensitivity of 90% and specificity of 76.5% (AUROC 0.85, p
Conclusions: An electronic nose may identify admitted severe COPD patients at risk of 30-day and 90-day readmission.
Original languageSpanish
Pages (from-to)3854
Number of pages1
JournalEuropean Respiratory Journal
Issue number62
Publication statusPublished - 15 Sept 2018

Cite this