POLARIS

Rationale & methods

A shared observatory.
Multiple research questions.

Standardized prospective data to study ARDS phenotypes, care practices, and outcomes across healthcare systems.

The POLARIS observatory

POLARIS is an international research network built around an ongoing repository of de-identified clinical data. Adult ICUs contribute routinely collected information using shared definitions and assessment timepoints.

The repository follows the 2023 Global Definition of ARDS, including eligible patients receiving non-invasive respiratory support. It supports collaborative observational research and evaluation of ARDS care practices.

Our first prospective study

The initial phenotyping study evaluates whether inflammatory phenotypes assigned with AI-CLARITY remain informative across patients and settings. It examines their association with 28-day mortality, their transportability, and differences in associations with early intravenous corticosteroids within a prespecified causal framework.

From local care to shared research

Participating centers collect data through their national or regional coordination. These teams oversee harmonization, quality review, and de-identification before secure transfer to the central repository at BIDMC under the applicable agreements.

Assessments use routine information at baseline, days 3 and 7, and day 28 for vital status. The repository does not require additional clinical tests or assign treatment.

Future collaborative projects

Investigators can propose studies on phenotypes, prognosis, treatment-response heterogeneity, and clinical practices. Each project is reviewed through the POLARIS governance process, with data access defined for the approved research question.

Biology and clinical classification

AI-CLARITY estimates inflammatory phenotypes from routine measurements. This clinical estimate is a surrogate for the biological phenotype. Prognostic separation does not establish biological agreement or demonstrate that phenotype-guided treatment improves outcomes. Biomarker agreement is an exploratory objective where measurements are available.

AI-CLARITY

AI-CLARITY uses the original, frozen XGBoost model and 18 routinely available clinical and laboratory variables. XGBoost combines decision trees to estimate the probability of an inflammatory phenotype. The model has been validated in independent cohorts.

Scientific sources

  1. Pensier J, Schaefer MS, Talmor D. Phenotyping ARDS: A pragmatic path forward with the POLARIS international cohort. Intensive Care Med. 2026. doi: 10.1007/s00134-026-08614-z.
  2. Pensier J, Fosset M, Paschold BS, von Wedel D, Redaelli S, Braeuer BLP, et al. Temporal stability of phenotypes of acute respiratory distress syndrome: clinical implications for early corticosteroid therapy and mortality. Intensive Care Med. 2025;51(10):1784-96. doi: 10.1007/s00134-025-08089-4.
  3. van Amstel RBE, Reilly JP, Bos LDJ. Open-source clinical classifiers for ARDS: democratizing precision or diluting biology? Intensive Care Med. 2025;51(11):2123-5. doi: 10.1007/s00134-025-08169-5.
  4. Pensier J, Fosset M, Paschold BS, von Wedel D, Sharif M, Braeuer BLP, et al. Effects of inflammatory phenotypes in acute respiratory distress syndrome on mortality and partitioning of lung and chest wall mechanics in patients in the USA and Canada: a retrospective cohort study. Lancet Respir Med. 2026;14(7):609-19. doi: 10.1016/S2213-2600(26)00045-7.
  5. Matthay MA, Arabi Y, Arroliga AC, Bernard G, Bersten AD, Brochard LJ, et al. A new global definition of acute respiratory distress syndrome. Am J Respir Crit Care Med. 2024;209(1):37-47. doi: 10.1164/rccm.202303-0558WS.

For questions about study participation and site activation, contact the coordinating team.