Sepsis-related mortality and ICU admissions of patients with acute infections presenting during out-of-hours at GP cooperatives: a retrospective study to identify high-risk groups

Background and aim Early recognition of severe infections is crucial for the prognosis. About half of the patients with sepsis contact an out-of-hours general practitioner (GP) cooperative before hospital admission, but identification of patients at risk for sepsis is difficult for GPs. Acute infections are common and on the other hand, the most severely ill patients often present with atypical symptoms without clear signs of infection. Apparently, the diagnosis of sepsis should also be considered in patients without clear signs and symptoms of an infection. Therefore, we aim to identify high-risk groups for adverse sepsis-related outcomes, by including a broad sample of patients. The results of the study can help to identify patients with an increased risk of sepsis at GP cooperatives. This could help GPs to tailor their management accordingly to avoid unnecessary referral but more importantly also avoid doctor’s delay that could endanger the prognosis of the patient. Design We will perform a retrospective cohort study, linking a large database from Nivel Primary Care Database (Nivel-PCD) with data from Dutch Hospital Data (DHD) and Basisregistratie Personen (BRP) via CBS remote access. Patients and setting Adult patients who had a medical assessment by a GP (clinic consultations or home visit) of an out-of-hours GP cooperative in 2017-2019 will be included in the study. In this three-year period we will include approximately 300.000 patients. Methods The primary outcome measure is “adverse sepsis-related outcome”, defined as ICU-admission due to infection within 72 hours or infection-related death within 30 days. To examine variables that are related to an adverse sepsis-related outcome we will perform a penalized logistic regression analysis. Candidate predictors are patient characteristics (e.g. age, gender and comorbidities) and information available at triage (e.g. time of contact, entry complaint and urgency category). Bootstrapping will be used to correct for optimism and model performance will be assessed in terms of discrimination and calibration.