PURPOSE: Prognostic studies derived from samples of patients managed in tertiary hospitals are subject to referral bias. We aimed to characterise this bias using the example of infective endocarditis (IE). METHODS: We analysed data from a French population-based cohort which included 497 patients with IE in 2008. Patients were admitted directly to a tertiary hospital (group T), or admitted to a non-tertiary hospital and referred to a tertiary hospital (group NTT) or not (group NT). We compared patients' characteristics and survival rates and prognostic factors between groups. RESULTS: Compared to group T (N=291), NTT patients (N=144) were more often males (81.3% vs 72.5%, p=0.046), injection drug users (9.7% vs 4.5%, p=0.033), had more frequent surgical indications (78.5% vs 64.3%, p=0.003). Compared to group NT (N=62), NTT patients were more often males (81.3% vs 67.7%, p=0.034) and had surgical indications more often (78.5% vs 19.4%, p<0.001). One-year survival was higher in (NTT+T) patients than in NT patients (73.0% vs 56.1%, p=0.01). Prognostic factors as well as HR estimates varied across groups. CONCLUSION: When derived from samples mixing patients admitted directly and those referred to tertiary hospitals, validity of characteristics description, survival estimates, and HRs is threatened by referral bias.Read less <