Acute respiratory distress syndrome (ARDS) is a life-threatening lung injury that affects an estimated 3 million people worldwide each year, and accounts for around 10% of intensive care unit admissions. Community-acquired pneumonia (CAP) is among its most common triggers, and clinicians are often left with a window of only two to six hours to identify patients at risk before the condition can be confirmed.
At ERS Congress 2026 in Barcelona, Vahid Esmaeili, Volv Global’s Data Science and Digital Health Director, will present new research tailored to a research question defined with CSL Behring, addressing that gap using machine learning applied to a large-scale US claims database, together with independent validation on intensive care data and expert clinical review.
What the research demonstrates
Flagging risk ahead of diagnosis
The model was trained on de-identified US claims data covering 341,697 patient records, learning to recognise the clinical patterns that precede ARDS. In retrospective testing, it remained predictive up to five days before the ARDS diagnosis code appeared, or at the time of CAP diagnosis – offering a materially wider window than the two to six hours clinicians typically have today.
Validating against independent clinical judgement
The model’s outputs were reviewed independently by three ARDS specialists in the US, UK and Germany, who confirmed that the flagged clinical phenotypes matched established ARDS pathophysiology. Their assessment showed 95% agreement with the model’s top flagged high-risk cases – a strong concordance between the model’s output and expert clinical judgement.
Why this matters
This work sits within inFlow, one of Volv Global’s solutions for prognostic modelling and outcome prediction. The underlying methodology is built to recognise disease-specific patterns in real-world data, and ARDS is one proof point of an approach designed to apply wherever a disease leaves a distinct signature in the data – subject, in every case, to its own prospective validation.
About the study
- Disease: Acute respiratory distress syndrome (ARDS) following community-acquired pneumonia (CAP)
- Data sources: Komodo Health US administrative claims database; MIMIC-IV intensive care dataset (independent validation)
- Cohort: 341,697 patient records (2016–2023)
- Independent clinical validation: three ARDS specialists, US, UK and Germany
- Presented at: Poster PA 2100, Session 167, Poster Session 5 – Severe respiratory infections, chest trauma, airways and resuscitation, ERS Congress 2026, Barcelona, Sunday 6 September 2026
- Collaboration: Volv Global and CSL Behring
All findings are based on retrospective analysis of claims and electronic health record data. Prospective validation has not been conducted. The model is intended to support clinical decision-making and research, and does not diagnose ARDS or replace clinician judgement.
Links
- Event page: Volv Global at ERS Congress 2026
- Press release: Volv Global Model Flags Early ARDS Risk in Pneumonia [link to come]
- ERS Congress 2026 details
- Volv Global inFlow – Predicting patient outcomes
Download
The full poster is available to download below.
