Comment
Real-time subphenotyping in acute respiratory distress syndrome: a step towards precision medicine
Denise Battaglini, Marcus J Schultz, Patricia R M Rocco
The Lancet Respiratory Medicine, Available online 23 March 2026 https://doi.org/10.1016/S2213-2600(26)00054-8
Acute respiratory distress syndrome (ARDS) has long been defined by pragmatic clinical criteria that favour sensitivity over biological specificity. This strategy has enabled early recognition and standardised enrolment in clinical trials, but at the cost of grouping patients with profoundly different inflammatory states, physiological disturbances, and disease trajectories.1 Many randomised trials have therefore tested therapies across biologically heterogeneous populations, implicitly assuming homogeneity when it is unlikely to exist.2 The persistent failure of pharmacological and physiological interventions is now increasingly viewed not as proof of inefficacy, but as a limitation of the syndrome-based framework itself.
Over the past decade, reproducible inflammatory subphenotypes have challenged this paradigm. A hyperinflammatory subphenotype, present in a minority of patients, is characterised by elevated circulating inflammatory mediators, greater haemodynamic instability, more frequent sepsis, and higher mortality. In contrast, a hypoinflammatory subphenotype accounts for most cases and is associated with more favourable outcomes than the hyperinflammatory subphenotype.3 Retrospective analyses have suggested differential responses to ventilatory strategies, fluid management, and selected pharmacological therapies. However, these observations have largely emerged post-hoc and have therefore not informed patient selection or treatment allocation at the bedside.4 The absence of tools for real-time biological classification has remained a major barrier.5
Against this background, the PHIND study,6 reported by Kiran Reddy and colleagues in The Lancet Respiratory Medicine, represents a methodological advance. In a large, multicentre prospective cohort, the investigators show that inflammatory ARDS subphenotypes can be identified in real time using near-patient biomarker measurement. The study does not propose a new biological construct nor claim therapeutic benefit. Rather, it shows that subphenotypes previously derived from retrospective latent class analyses can be prospectively assigned during routine intensive care practice. Using rapid measurements of IL-6, TNFR1, and arterial bicarbonate, classification was achievable without reliance on central laboratories or complex post-hoc modelling.6
This ability to classify patients at the bedside directly addresses a key limitation of earlier work. Previous subphenotyping approaches have depended on retrospective population-level models applied to curated datasets, limiting clinical applicability.6, 7 In PHIND, approximately one in five patients were prospectively classified as having a hyperinflammatory subphenotype, consistent with previous estimates, and this group had substantially higher 60-day mortality than those classified as having the hypoinflammatory subphenotype. Clinicians were masked to subphenotype assignment, reducing the likelihood that management decisions influenced outcomes. Although not designed to assess treatment effect modification, the study shows that prospective biological stratification retains prognostic relevance in practice.6
Several practical considerations warrant attention. Although only 4–5% of patients could not be classified owing to technical issues, this result still implies that roughly one in twenty patients would remain without biological assignment, an important concern if subphenotype-directed interventions were to benefit one group while adversely affecting another. These failures also occurred under study conditions, raising questions about feasibility in everyday practice. In addition, the timing of classification is not reported, despite its importance for early decision making. Whether inflammatory subphenotypes remain stable over time also remains uncertain.8
PHIND should not be interpreted as evidence that subphenotype-guided management improves outcomes. The study is observational, and causal inferences regarding therapy cannot be drawn.9 Identifying biological subphenotypes is a necessary but insufficient step towards precision medicine; benefit depends on whether such stratification can guide interventions that improve patient-centred outcomes.10
Nonetheless, PHIND removes a crucial technical barrier that has constrained progress in ARDS research. Although biologically stratified trials have long been advocated, their implementation has been limited by the absence of practical real-time identification tools. By showing that prospective subphenotype assignment is achievable across multiple intensive care units, PHIND provides a foundation for a new generation of trial designs. The remaining challenge is no longer technical, but strategic: how best to integrate biological stratification into interventional studies without compromising feasibility or methodological rigour.6
Several implications follow, and timing is likely to matter. The prevalence of the hyperinflammatory subphenotype was lower in PHIND than in previously published cohorts and might reflect later sampling and inclusion of patients receiving high-flow oxygen therapy. A lower prevalence of the hyperinflammatory subphenotype due to later sampling is consistent with evidence that inflammatory states in ARDS evolve rapidly early in the disease course and aligns with efforts to refine ARDS definitions towards earlier and more dynamic frameworks. Binary classification might also oversimplify a biologically continuous process. Subphenotype probability appears to be graded rather than dichotomous, suggesting that rigid thresholds might be suboptimal for some interventions.
Beyond ARDS, PHIND offers a pragmatic example of how biomarker-based stratification can move from retrospective discovery to prospective bedside application using tools compatible with routine clinical practice.6 Its relevance lies not in immediate changes to management, but in what it enables: the prospective testing of biologically informed trial designs. Whether such approaches can ultimately improve outcomes in ARDS is now a question that can be addressed.
