Lung ultrasound vs chest radiography for pneumonia diagnosis in critically ill patients in the intensive care unit: a systematic review and Bayesian network diagnostic meta-analysis

Background: Diagnosing pneumonia in intensive care unit (ICU) patients is challenging. Chest X-ray (CXR) is widely used, whereas lung ultrasound (LUS) is increasingly applied at the bedside. Their comparative accuracy vs computed tomography (CT), the reference standard, is uncertain. Methods: We performed a systematic review and Bayesian hierarchical network meta-analysis of prospective diagnostic accuracy studies. PubMed, CINAHL, Scopus, and Web of Science were searched to May 15, 2025. Eligible studies enrolled ICU adults with LUS and/or CXR vs CT. Exclusions were retrospective designs, case series, non-ICU settings, and studies without CT. Risk of bias was assessed with QUADAS-C. Primary outcomes were sensitivity, specificity, log diagnostic odds ratio (logDOR), and hierarchical summary receiver operating characteristic [HSROC] area under the curve (AUC). Bayesian models accounted for heterogeneity. Sensitivity, calibration, and meta-regressions (posterior lung zones and sample size) were conducted. This study is registered with PROSPERO, CRD420251081853. Findings: Of 3080 records, nine studies with 746 patients were included. Risk of bias was generally low for patient selection and flow/timing but high for index test blinding. Grading of Recommendations, Assessment, Development and Evaluation [GRADE] certainty was moderate for LUS and low for CXR. LUS showed higher sensitivity (0.93, 95% credible interval [CrI] 0.91–0.95) than CXR (0.65, 95% CrI 0.62–0.68), comparable specificity (0.83, 95% CrI 0.81–0.85 vs 0.81, 95% CrI 0.79–0.83), higher accuracy (AUC 0.88, 95% CrI 0.77–0.95 vs 0.76, 95% CrI 0.60–0.88), and greater logDOR (4.17, 95% CrI 2.53–6.64 vs 1.43, 95% CrI 0.22–2.86). Between-study heterogeneity was moderate to substantial. For LUS, τ was 0.99 (95% CrI 0.23–3.04) for sensitivity, 0.68 (95% CrI 0.18–2.06) for specificity, 1.66 (95% CrI 1.08–2.84) for logDOR, and 0.086 (95% CrI 0.046–0.156) for AUC. For CXR, τ was 0.89 (95% CrI 0.25–2.43) for sensitivity, 0.26 (95% CrI 0.05–1.05) for specificity, 2.32 (95% CrI 1.39–4.77) for logDOR, and 0.175 (95% CrI 0.117–0.231) for AUC. Interpretation: LUS shows higher sensitivity and accuracy than CXR in ICU pneumonia, particularly when dorsal zones are assessed. Moderate specificity, operator dependence, and bias require caution in interpreting the results. Randomised trials should determine whether broader LUS use improves antibiotic stewardship and ICU outcomes. Funding: There was no funding source for this study. Open access publication fees are supported by institutional funds from the University of Udine.