POLARIS cohort tests whether ARDS phenotypes can reach every ICU

Acute respiratory distress syndrome, or ARDS, has never been a single disease. It is a clinical syndrome that gathers, under one operational definition, patients with different underlying injuries, host responses, lung physiology, imaging patterns, and clinical trajectories. This heterogeneity is widely regarded as a principal reason why so many clinical trials in ARDS have returned neutral results and why treatment effects have appeared inconsistent across studies. If biologically and physiologically distinct subgroups of patients respond differently to the same therapy, then aligning treatments with the mechanisms most likely to influence outcomes becomes a compelling strategy. The rationale for phenotyping ARDS rests on precisely this logic: clinically and biologically meaningful phenotypes may help match therapies to the patients whose disease biology they actually target. The outstanding challenge, as a new commentary in Intensive Care Medicine by Joris Pensier, Maximilian S. Schaefer, and Daniel Talmor argues, is translating phenotypes into tools that can guide treatment decisions worldwide, well beyond highly resourced research environments.

Phenotyping is one component of precision medicine, and it complements the mechanistic and biological groundwork needed to understand disease and identify therapeutic targets. But statistical performance alone cannot determine whether a phenotype will influence clinical practice. According to the framework laid out in the commentary, a clinically useful ARDS phenotype should satisfy three complementary principles. First, it should be mechanistically grounded, capturing meaningful differences in biology, physiology, respiratory mechanics, or response to injury rather than simply separating patients by prognosis. Because ARDS is heterogeneous across multiple dimensions, no single phenotype is likely to guide every therapeutic decision; complementary biological and physiological domains will probably need to be integrated to achieve true precision medicine. Second, phenotypes should be decision-oriented. Their value lies in informing defined scientific or clinical questions: which mechanism to investigate, which therapy to initiate or discontinue, which ventilatory strategy to apply, which patients to enrich in clinical trials, or which disease trajectory to anticipate. Third, phenotypes must be generalizable, because ARDS is managed across the full range of intensive care unit and resource settings, and classification systems that depend on specialized assays, research personnel, or sophisticated computational infrastructure may advance discovery while struggling to improve routine care.

This third principle is where many promising phenotyping approaches have faltered. Scalability and prospective implementation, the commentary contends, should be considered from the outset rather than after discovery has been completed. A classification system that can only be applied in a handful of academic centers may retain high scientific value while its clinical impact remains limited. If phenotypes cannot be measured wherever patients with ARDS are actually treated, the gap between discovery and delivery persists. The commentary therefore weighs two complementary strategies against these requirements: biomarker-based inflammatory phenotyping, which offers high biological fidelity but carries operational friction, and clinical classifiers built from routinely available variables, which scale across intensive care units but require rigorous prospective validation. The POLARIS international prospective cohort is designed as the bridge between the two, testing prognostic validity, decision relevance, and transportability across diverse resource settings.

Inflammatory ARDS phenotypes provide the clearest test case for these principles. Hyperinflammatory and hypoinflammatory phenotypes are biologically plausible, reflecting distinct systemic inflammatory responses that may inform the search for new therapeutic targets. In parallel, retrospective analyses have demonstrated heterogeneous treatment effects for fluid management, simvastatin, and corticosteroids, suggesting that inflammatory phenotypes may help stratify responses to therapies that already exist. More recently, parsimonious biomarker panels incorporating interleukin-6, soluble tumor necrosis factor receptor-1, and bicarbonate have enabled prospective inflammatory classification, moving the field beyond retrospective discovery. Two multicenter prospective studies, PHIND and SPARC, used near-patient biomarker assays and represent important milestones in translating inflammatory phenotyping from retrospective analysis toward real-time clinical implementation. Together, they establish that inflammatory phenotypes can be assigned prospectively at the bedside, at least under favorable conditions.

Yet these same studies expose the field’s principal challenge: feasibility is not the same as scalability. PHIND enrolled approximately one patient per center every three months, with only one hyperinflammatory patient identified per center every sixteen months. In SPARC, successful real-time phenotyping required dedicated laboratory workflows and progressive implementation before approximately 80 percent of patients could be classified in the final study phase. Even within motivated academic centers, nearly one in five patients remained un-phenotyped. Biomarker-based phenotyping therefore remains the biologically most faithful approach for enrichment trials targeting inflammatory pathways, but this fidelity comes at the cost of assay availability, laboratory infrastructure, turnaround time, expense, and unequal access across healthcare systems. If phenotypes cannot be measured beyond specialized centers, their scientific value may remain high while their clinical impact remains limited, and precision medicine in ARDS risks becoming a privilege of well-equipped institutions rather than a standard of care.

This implementation gap motivates a complementary strategy. Rather than replacing biomarker-based phenotyping, clinical classifiers seek to infer inflammatory phenotypes using routinely collected clinical data. Machine-learning models based on vital signs, laboratory values, and ventilatory variables can be embedded within electronic health records and deployed without specialized biomarker platforms. An open-access artificial intelligence classifier exemplifies this approach by assigning inflammatory phenotypes from routinely available variables while permitting independent external validation by any interested center. In the architecture proposed by the commentary, biomarker models define the biological structure of inflammatory ARDS, while clinical classifiers determine whether those biological insights can be translated into routine care at a global scale. The two approaches are presented as complementary layers of the same enterprise rather than competitors.

Clinical classifiers, however, introduce their own challenges. Routine clinical variables may reflect institutional practice as much as underlying biology. Vasopressor use, ventilator settings, laboratory testing frequency, and patterns of missing data vary across hospitals, creating the potential for model drift across geography, time, and resource settings. A classifier trained in one healthcare system may behave differently in another, not because the biology of ARDS has changed but because the clinical variables encode different local practices. Consequently, clinical classifiers require the same rigorous prospective validation as biomarker-based approaches, including external validation in centers that did not contribute to model development. Without this validation, the apparent scalability of clinical classifiers could mask systematic errors that only become apparent after deployment across heterogeneous healthcare environments.

Logically, the next step must be to determine whether prospective phenotyping can be implemented equitably across the full spectrum of ARDS care. The POLARIS cohort, short for Prospective Observatory of Lung Injury and ARDS for Patient and Disease Phenotyping, was designed with this objective in mind. Building on the proof of concept established by PHIND and SPARC, POLARIS emphasizes scalability by enrolling patients across academic, community, and private hospitals in both high- and low-resource settings. Its inclusion criteria follow the 2023 Global Definition of ARDS, ensuring that patients identified using SpO2/FiO2 ratios, high-flow nasal oxygen, or lung ultrasound are included. These are precisely the settings where biomarker platforms are least available, which makes the cohort’s design a deliberate test of whether phenotyping can function where conventional laboratory infrastructure cannot reach.

POLARIS is structured to address three questions. First, do inflammatory phenotypes, prospectively assigned based on routine clinical parameters, predict 28-day mortality? Second, can they identify heterogeneous responses to early intravenous corticosteroids within a pre-specified causal framework, testing whether clinical phenotypes retain the decision relevance that retrospective treatment-effect analyses have suggested? Third, do the phenotypes remain reproducible across diverse healthcare systems and resource environments, the transportability question that determines whether any phenotype can serve as a global clinical tool rather than a research artifact? Beyond these questions, the commentary emphasizes that thorough assessment of ARDS heterogeneity will need to integrate complementary alveolar, morphological, and physiological dimensions, account for temporal transitions through repeated assessment over the disease course, and undergo regulatory evaluation before phenotypes can inform approved clinical decision-making.

The future of ARDS phenotyping, the commentary concludes, is unlikely to belong exclusively to biomarkers or to clinical classifiers. Biomarkers reveal the biological architecture of disease; clinical classifiers determine whether those biological insights can reach every patient with ARDS treated in an intensive care unit anywhere on the planet. Progress will require a continuous learning framework in which biological discovery refines clinical classifiers and large-scale implementation tests their robustness across diverse healthcare settings. POLARIS was designed to provide that bridge. By enabling any ICU to prospectively phenotype patients using routinely available clinical data while contributing standardized outcomes, it offers an opportunity to determine whether inflammatory phenotypes can move beyond discovery cohorts and become a practical foundation for equitable precision medicine in ARDS. Whether the field’s biological insights can survive the journey from specialized laboratories to the bedside in every resource setting will now be tested prospectively, at international scale.

Subject of Research: Clinical phenotyping of acute respiratory distress syndrome and the POLARIS international prospective cohort

Article Title: Phenotyping ARDS: A pragmatic path forward with the POLARIS international cohort

Article References: Pensier, J., Schaefer, M. S., & Talmor, D. (2026). Phenotyping ARDS: A pragmatic path forward with the POLARIS international cohort. Intensive Care Medicine. https://doi.org/10.1007/s00134-026-08614-z

Image Credits: AI Generated

DOI: 10.1007/s00134-026-08614-z

Keywords: Phenotyping, ARDS, pragmatic, path, forward, POLARIS, international, cohort, scientific research