BACKGROUND. Ventilator-associated pneumonia (VAP) is the most common intensive care unit (ICU)-acquired infection, yet accurate diagnosis remains difficult, leading to overuse of antibiotics. Low concentrations of IL-1 and IL-8 in bronchoalveolar lavage (BAL) fluid have been validated as effective markers for excluding VAP. VAP-Rapid-2 sought to determine whether a rapid IL-1/IL-8 test could effectively and safely improve antibiotic stewardship in patients with clinically suspected VAP. METHODS. VAP-Rapid-2 was a multi-centre randomised controlled trial in 24 intensive care units in the United Kingdom. Patients with suspected VAP in 24 ICUs were randomised 1:1, via a web-based randomisation service, to a “biomarker-based recommendation on antibiotics” or “routine use of antibiotics”. Clinicians were blinded to randomisation for an initial period until the biomarker results were reported. Bronchoalveolar lavage (BAL) was performed in all patients, with concentrations of IL-1 and IL-8 rapidly determined in BAL fluid from patients randomised to the “biomarker-based antibiotic recommendation” arm. If concentrations were below a previously validated cut-off, clinicians were advised that VAP was unlikely and to consider discontinuing antibiotics. Patients in the “routine use of antibiotics” arm received antibiotics according to usual practice. Microbiology was performed on BAL fluid from all patients and VAP was confirmed by ≥104 colony forming units per mL of BAL fluid. The primary outcome measure was the distribution of antibiotic-free days (AFD) in the seven days following BAL. Data were analysed on an intention-to-treat basis, with an additional per-protocol analysis that excluded patients randomised to the intervention arm who defaulted to routine use of antibiotics due to failure to return an adequate biomarker result. An embedded process evaluation assessed factors influencing trial adoption, recruitment and decision making. This study is registered as an International Standard Randomised Controlled Trial, number ISRCTN: 65937227, and with ClinicalTrials.gov: NCT01972425. FINDINGS. The trial randomised the target number of 210 patients with suspected VAP. One patient was withdrawn by the clinical team prior to bronchoscopy and so was excluded from the intention-to-treat analysis (‘biomarker-guided recommendation on antibiotics’, n=103; ‘routine use of antibiotics’, n=106). Antibiotics were being used at baseline in 83 (n=103, 80·6%) patients in the intervention arm and 87 (n=106, 82·1%) patients in the control arm. The negative likelihood ratio of the IL-1/IL-8 test was 0·09. There was no difference in the frequency distribution of AFD at day seven (p=0·58). There were no between-group differences in AFD at day 14 (HR 1·13 [95% CI 0·83-1·54]), in AFD at day 28 (HR 1·01 [95% CI 0·73-1·40]) or in mortality at 28 days (OR 1·52 [95% CI 0·78-2·98]). BAL was associated with a small and transient increase in oxygen requirements. Established prescribing practices, reluctance for BAL, and dependence on a chain of trial-related procedures emerged as factors impairing trial processes. INTERPRETATION. Antibiotic use remains high in patients with suspected VAP. Antibiotic stewardship was not improved by a rapid, highly sensitive rule-out test. Prescribing culture, rather than poor test performance, may explain the lack of effect.