Study Protocol: A retrospective observational analysis of patients treated for hospital-acquired pneumonia.

<ns3:p> Background Hospital-acquired pneumonia (HAP) is an important complication of hospital admission, with both high incidence and consequences for patients. However, our understanding of causative organisms and prognostic factors is limited. Although ventilator-associated pneumonia (VAP,) an important subset of HAP,has been - extensively investigated, less is known about non-ventilated cases, leading to calls for focused research in this group. This retrospective observational cohort study aims to define a population of patients treated as HAP by comparing ventilated and non-ventilated cases. It aims to clarify how often a microbiological diagnosis is reached, what organisms are frequently identified, and whether this has a relevant impact on the outcomes. The relative impact of positive radiographic changes among patients treated for HAP will also be assessed. Methods Data will be obtained from the Health Data Research UK acute care hub, ‘. ’PIONEER’ Cases meeting the coding criteria or a clinical surveillance definition of HAP over a 5-year period will be extracted. Demographic, clinical, and microbiological variables will be analysed initially descriptively, and subsequently, with multiple logistic regression analysis to investigate factors affecting microbiological diagnosis. Key outcome variables are in-hospital, 30-day and 1 year mortality, as well as all-cause readmissions within 1 year. Secondary outcomes include nosocomial infections, such as <ns3:italic>C. difficile</ns3:italic> . Kaplan-Meier curves and a Cox proportional hazards regression model will be used to investigate outcomes and compare subgroups. A key comparison is between those in whom a putative pathogen is identified and those treated entirely empirically. For this purpose, we will also compare outcomes using an inverse probability of treatment weighting analysis. Additionally, we will explore identifying consolidation in chest imaging reports using natural language processing to allow consideration of the relative impact this may have on mortality and readmission rates. </ns3:p>