ICU-acquired pneumonia: why detecting a pathogen is not the same as diagnosing disease

For decades, the diagnosis of pneumonia acquired in the intensive care unit has rested on a deceptively simple premise: find the organism, treat the organism. A new Perspective published in Nature Communications by Ignacio Martin-Loeches of St James’s Hospital in Dublin and Antoni Torres of Hospital Clínic de Barcelona argues that this premise is failing contemporary critical care. The authors propose that ICU-acquired pneumonia should be reconceived as an attribution problem, in which clinicians explicitly grade and record how plausible it is that a detected organism is actually causing the lung injury in front of them, rather than treating a positive culture as a binary confirmation of disease.

The technical core of the argument is that the diagnostic signals used at the bedside measure disparate biological processes. Respiratory deterioration reflects physiology, whereas cultures, polymerase chain reaction assays and metagenomic sequencing detect microbial material, which may represent invasive infection, airway colonization, biofilm fragments or residual nucleic acid from non-viable organisms. Surveillance definitions prioritize reproducible events for population-level monitoring. None of these signals, in isolation, confirms tissue-invasive bacterial pneumonia, which is why the syndrome remains so difficult to classify and treat consistently.

The authors also contend that the classical clinical model, ventilator-associated pneumonia, is too narrow for contemporary practice. They argue that the framework should encompass ventilated hospital-acquired pneumonia, in which pneumonia begins before intubation, and non-ventilated ICU hospital-acquired pneumonia, which arises in patients who are never intubated at all. These entities share host vulnerability, exposure to hospital-adapted pathogens and the risk of rapid deterioration, but they differ systematically in prior probability of true infection, in the quality of respiratory samples available and in the plausibility of competing explanations. Observational studies cited in the Perspective suggest that ventilated hospital-acquired pneumonia carries high mortality, in some analyses exceeding that of ventilator-associated pneumonia after adjustment for severity and comorbidity, while large electronic surveillance studies indicate that non-ventilated hospital-acquired pneumonia is common and associated with substantial inpatient mortality.

Attribution behaves differently in each syndrome, and this is where the framework becomes technically specific. In ventilator-associated pneumonia, the prior probability of parenchymal infection is comparatively well characterized, lower airway sampling is feasible and quantitative, and the principal error is over-attribution, treating airway colonization or biofilm-derived organisms as causative when atelectasis, fluid overload or acute respiratory distress syndrome explains the deterioration. In ventilated hospital-acquired pneumonia, the causal sequence is inverted because pneumonia preceded intubation, ward-level antibiotic exposure is common and often incompletely documented, and the dominant error is misclassification. In non-ventilated pneumonia, sputum quality is variable, bronchoscopy is often unsafe and portable imaging is insensitive, so the dominant error is under-attribution and delay. The same molecular result should therefore not move the estimated probability of infection by the same amount in all three settings.

The Perspective draws on an ecological view of the lower respiratory tract as a low-biomass microbial ecosystem rather than a sterile compartment. Its composition is shaped by microaspiration, mucociliary clearance, local immune surveillance, antimicrobial exposure and competition between microbial populations. Endotracheal intubation introduces a foreign surface, facilitates biofilm formation and increases microaspiration, while sedation reduces airway clearance. The best-supported determinants of lower airway community composition are antimicrobial exposure, aspiration of oropharyngeal and gut-associated organisms, and bacterial burden itself, which predicts ventilator-free days independently of pneumonia and illness severity. Endotracheal tube biofilm provides a persistent reservoir that can reseed the lower airway. The therapeutic objective that follows is not complete sterilization of the airway, which the authors describe as neither achievable nor clinically necessary, but resolution of clinically meaningful parenchymal infection while limiting toxicity and selection pressure.

Imaging occupies an equivalent position in the framework: it constrains attribution without establishing it. The portable chest radiograph performs poorly against autopsy-confirmed pneumonia, and in ventilated patients with diffuse bilateral infiltrates no radiographic sign discriminates reliably. Computed tomography adds anatomical resolution, and its principal contribution to attribution is often negative rather than positive, identifying the alternative explanation such as symmetrical dependent atelectasis with volume loss, smooth septal thickening in hydrostatic oedema, wedge-shaped peripheral opacity in pulmonary embolism or diffuse anteroposterior gradient change in ARDS. Lung ultrasound is attractive because it is bedside-available, repeatable and free of radiation, but the authors emphasize that consolidation is close to ubiquitous in dependent posterior regions of ventilated patients, that the dynamic air bronchogram is the most informative single sign rather than proof of pneumonia, and that unstructured examination is not reproducible between examiners.

Rapid molecular diagnostics, including syndromic PCR panels and metagenomic sequencing, have shifted diagnostic uncertainty earlier in the pathway by identifying pathogens and resistance determinants within clinically actionable timeframes, even after antibiotic initiation. Yet the authors warn that analytical sensitivity must not be mistaken for diagnostic specificity. In the ventilated airway, molecular assays detect nucleic acids within a disturbed ecosystem, and a positive signal may represent invasive infection but can just as easily reflect low-abundance colonization, oral contamination, biofilm or residual DNA from suppressed organisms. Bacterial burden is described as the essential metric for interpretation: high molecular burden is persuasive when it corresponds to a high bacterial burden, whereas low-abundance detection is frequently misleading, particularly after antibiotic exposure when molecular tests may remain positive by detecting nucleic acid from non-viable organisms.

The dominant clinical risk of high-sensitivity testing, the authors argue, is not missed infection but antibiotic overuse, and this risk is structural rather than incidental. Four features drive it. Syndromic panels report organisms that culture would never have reported, because they detect below the threshold at which growth is called significant and because they detect non-viable material. They are frequently polymicrobial, inviting coverage of three or four organisms when only one, or none, is causative. Resistance determinants are reported as genotype rather than phenotype, exerting a strong pull toward pre-emptive escalation. Most importantly, clinician response is asymmetric: a positive result reliably prompts action, whereas a negative result rarely prompts equivalent de-escalation. Trial evidence is consistent with this asymmetry, since in the largest randomized evaluation of multiplex bacterial PCR of bronchoalveolar lavage the demonstrated gain was a reduction in the duration of inappropriate therapy, not an increase in the proportion of patients treated.

Host immune competence modifies the meaning of both positive and negative microbiological tests, and the authors argue that an attribution statement omitting it is incomplete. Critically ill patients are immunologically heterogeneous: some exhibit hyperinflammatory responses with cytokine-mediated tissue injury, while others develop critical illness-associated immune dysfunction characterized by persistent lymphopenia, impaired antigen presentation, reduced monocyte HLA-DR expression, T cell exhaustion and reduced microbial clearance. Experimental and human data show that after resolution of a primary inflammatory insult, alveolar macrophages undergo an epigenetic programme of tolerogenic training and remain functionally impaired for weeks. Timing of pneumonia onset therefore serves as a proxy for host immune state: in late-onset disease, a moderate microbial signal in a patient with an impaired host response should be given more weight than the same signal early in the ICU course, because the probability of progression from containment to invasion is higher.

Therapeutically, the framework makes attribution confidence the explicit first term in antimicrobial decision-making. Four actions are available at any point, start, continue, narrow or stop, and each maps onto a combination of attribution confidence and clinical stability rather than onto microbiology alone. The practical test is that the framework should sometimes produce different decisions for identical microbiological reports: an endotracheal aspirate growing Pseudomonas aeruginosa in a stable patient whose oxygenation is improving is a reason to stop, whereas the same organism at the same susceptibility in a deteriorating patient with new dense consolidation and rising vasopressor requirement is a reason to treat and optimize exposure. A formal 48-to-72-hour reassessment revisits three questions in order, whether ICU-acquired pneumonia remains the most plausible explanation, which detected organism is most likely causal, and what antimicrobial exposure is now required. The authors call for prospective studies to test structured recording of attribution confidence, for trials to distinguish the three syndromes, and for evaluation to extend beyond mortality to antibiotic-free days, toxicity, recurrence and ecological effects, so that high-sensitivity diagnostics add value only when they improve decisions without converting detection into automatic treatment.

Subject of Research: Causal attribution of ICU-acquired pneumonia integrating clinical trajectory, imaging, microbial burden and host immune competence

Article Title: ICU-acquired pneumonia beyond pathogen detection

Article References: Martin-Loeches, I., & Torres, A. (2026). ICU-acquired pneumonia beyond pathogen detection. Nature Communications, 17(1), Article 10875. https://doi.org/10.1038/s41467-026-78901-2

Image Credits: AI Generated

DOI: 10.1038/s41467-026-78901-2

Keywords: ICU-acquired pneumonia, ventilator-associated pneumonia, molecular diagnostics, antimicrobial stewardship, host immune competence, critical care, ICU-acquired, pneumonia, beyond, pathogen, detection, scientific research