Physiological subphenotype categorization model in acute respiratory distress syndrome: An Approach to Personalised Medicine
Date
2025
Type:
Thesis
item.page.extent
188 p.+ appendices
item.page.accessRights
Privado
item.contributor.advisor
ORCID:
Journal Title
Journal ISSN
Volume Title
Publisher
Universidad del Desarrollo. Facultad de Medicina
item.page.isbn
item.page.issn
item.page.issne
item.page.doiurl
item.page.other
item.page.references
Abstract
Acute Respiratory Distress Syndrome (ARDS) is a life-threatening condition characterized by non-cardiogenic pulmonary edema resulting from increased alveolar–capillary membrane permeability. It accounts for approximately 10% of all intensive care unit (ICU) admissions worldwide, affecting an estimated 750,000 patients annually in Europe and causing more than 400,000 deaths each year in the United States, with mortality rates approaching 40–45%. Its incidence increased markedly during the H1N1 influenza and SARS-CoV-2 pandemics, placing unprecedented strain on critical care resources globally. Survivors often experience long-term sequelae, including reduced exercise capacity, persistent respiratory and psychological impairments, diminished quality of life, and substantial healthcare utilization.
Despite decades of research and numerous clinical trials, few effective therapies have been established for ARDS. The condition’s complexity, driven by diverse etiologies, risk factors, and pathophysiological mechanisms, generates marked clinical heterogeneity. This variability complicates diagnosis, limits the delivery of individualized treatment, and contributes to the high rate of failure in randomized clinical trials. Current definitions, including the 2012 Berlin criteria and the proposed 2023 updates, risk further broadening inclusion criteria, potentially amplifying heterogeneity.
A promising strategy to address this challenge is the identification of distinct ARDS subphenotypes capable of capturing the individual physiological complexity of each case. Historically, classification has relied on simple physiological metrics such as the ratio of arterial oxygen partial pressure to inspired oxygen fraction (PaO₂/FiO₂) or respiratory system compliance (Crs), which fail to fully capture the complex interaction between mechanical ventilation, gas exchange, and respiratory mechanics, as well as the variability of these interactions across individual patients.
In this doctoral project, machine learning techniques, specifically Gaussian Mixture Models and XGBoost, were applied to develop a multidimensional,
physiology-based subphenotyping model. The model was trained using a dataset from Chile, externally validated in the Amsterdam UMC database, and
subsequently applied to randomized controlled trial datasets.
This approach consistently identified two reproducible physiological subphenotypes. The Restrictive subphenotype was characterized by severe ventilatory impairment, higher dead space, poorer oxygenation, and worse clinical outcomes. The Efficient subphenotype displayed improved ventilatory mechanics, better gas exchange, and lower mortality. External validation confirmed the stability of these patterns across diverse populations and clinical settings, supporting their potential to refine ARDS stratification and guide personalized medicine approaches.
Description
Thesis submitted to the Faculty of Medicine of the Universidad del Desarrollo in partial fulfillment of the requirements for the degree of Doctor in Science and innovation in Medicine
item.page.coverage.spatial
Santiago
item.page.sponsorship
Citation
Keywords
090036S, Acute Respiratory Distress Syndrome, Intensive care unit, Personalised medicine, Physiological subphenotypes