Generalisability of ARDS biological subphenotype models in Asians: an international, multicentre, prospective biomarker study.

PURPOSE: Subphenotype classifiers for acute respiratory distress syndrome (ARDS) dichotomise patients into hyperinflammatory versus hypoinflammatory subgroups. These models demonstrated prognostic and predictive values but were developed primarily in Caucasian populations. Generalisability of these models in Asian patients, who experience worse clinical outcomes, has not been established. We aimed to profile host responses in Asian patients with ARDS and evaluate the generalisability of established classifiers in this understudied population compared with a Caucasian cohort.

METHODS

We prospectively enrolled patients with ARDS from medical intensive care units in Beijing, China, and Pittsburgh, Pennsylvania, USA. In the Beijing cohort, 37 protein biomarkers were measured, with 10 overlapping biomarkers measured in the Pittsburgh cohort. Six established subphenotype models were assessed for generalisability and intermodel agreement. Sensitivity analyses, including latent class analysis, were conducted to explore biological heterogeneity within Asians.

RESULTS

Between 2011 and 2020, a total of 356 patients with ARDS (83% meeting the Berlin Definition; the rest on high-flow nasal cannula (HFNC) meeting the New Global Definition) were enrolled across Beijing (97% Han Asian) and Pittsburgh (90% Caucasian) sites, with comparable baseline hypoxaemia severity but disparate outcome. While the proportion of hyperinflammatory versus hypoinflammatory subphenotypes was predicted to be overall similar across different cohorts per each model, we observed poor intermodel agreement. We observed heightened inflammation in Berlin patients with ARDS compared with HFNC-ARDS within our Asian cohort.

CONCLUSION

Established subphenotype classifiers demonstrated similar distribution of subphenotypes in Asian patients with ARDS. However, poor intermodel agreement highlights the need for further investigation into model variability with models coming closer to bedside implementation.

TRIAL REGISTRATION NUMBER: NCT02975908.