Bench evaluation of six non-invasive ventilation home ventilators: comparison with an ICU ventilator and unsupervised clustering
Joris Pensier, Mathieu Capdevila, Dany Jaffuel, Abdelkebir Sabil, Fabrice Galia, Albert Prades, Aurélie Vonarb, Julien Boudjemaa, Audrey De Jong, Samir JaberIntroduction
Chronic hypercapnic respiratory failure often necessitates non-invasive ventilation (NIV) at home. Our study aimed to assess the static and dynamic performance of six modern NIV home ventilators and one intensive care unit (ICU) ventilator and to identify performance clusters among the devices.
Methods
A two-compartment lung model was connected to seven NIV ventilators (Sefam Stan, Philips A40, Philips DreamStation, Resmed Lumis 150, Löwenstein PrismaVent 30C, Löwenstein PrismaVent 40 and BellaVista 1000) in pressure-support mode. Static and dynamic (triggering and pressurisation) performances were assessed through three distinct clinical phenotypes and four levels of unintentional leak. Clustering analysis was performed using K-means.
Results
For each of the seven ventilators, 144 conditions were tested, and a total of 3024 cycles were analysed. Static and dynamic performances were good to excellent across home ventilators, significantly higher than the ICU ventilator. Clustering analysis identified three performance clusters. Cluster 1 (Sefam Stan and Philips A40) showed significantly more precise accuracy of inspiratory pressure than Cluster 2 (Philips DreamStation, Resmed Lumis 150, Löwenstein PrismaVent 30C and Löwenstein PrismaVent 40) and Cluster 3 (BellaVista 1000): mean error=4.3%±5.1% versus 8.5%±6.7% versus 10.6%±14.7% respectively, p<0.001. For the triggering delay, Cluster 1 displayed shorter delays than Cluster 2 and Cluster 3 (41±5 ms vs 58±11 ms vs 67±13 ms, respectively, p<0.001). For the pressurisation delay, Cluster 1 displayed shorter delays than Cluster 2 and Cluster 3 (42±6 ms vs 64±14 ms vs 87±14 ms, respectively, p<0.001). For the pressure-time product at 300 ms, Cluster 1 displayed higher area under the curve for the first 300 ms than Cluster 2 and Cluster 3 (2.1±1.1 cmH2O /s vs 1.6±0.8 cmH2O/s vs 1.3±1.0 cmH2O/s, respectively, p<0.001). Continuous unintentional leaks did not modify the pressurisation performances in Cluster 1 but altered them in Clusters 2 and 3.
Conclusion
The six NIV home ventilators demonstrated superior performance compared with the tested ICU ventilator. The ventilators of Cluster 1 were identified as top performers in clustering analysis and compensated for unintentional continuous leaks.