BackgroundEarly recognition of sepsis is challenging, and diagnostic criteria have changed repeatedly. We assessed the robustness of sepsis-3 criteria in intensive care unit (ICU) patients.MethodsWe studied the apparent incidence and associated mortality of sepsis-3 among patients who were prospectively enrolled in the Molecular Diagnosis and Risk Stratification of Sepsis (MARS) cohort in the Netherlands, and explored the effects of minor variations in the precise definition and timing of diagnostic criteria for organ failure.ResultsAmong 1081 patients with suspected infection upon ICU admission, 648 (60%) were considered to have sepsis according to prospective adjudication in the MARS study, whereas 976 (90%) met sepsis-3 criteria, yielding only 64% agreement at the individual patient level. Among 501 subjects developing ICU-acquired infection, these rates were 270 (54%) and 260 (52%), respectively (yielding 58% agreement). Hospital mortality was 234 (36%) vs 277 (28%) for those meeting MARS-sepsis or sepsis-3 criteria upon presentation (p < 0.001), and 121 (45%) vs 103 (40%) for those having sepsis onset in the ICU (p < 0.001). Minor variations in timing and interpretation of organ failure criteria had a considerable effect on the apparent prevalence of sepsis-3, which ranged from 68 to 96% among those with infection at admission, and from 22 to 99% among ICU-acquired cases.ConclusionThe sepsis-3 definition lacks robustness as well as discriminatory ability, since nearly all patients presenting to ICU with suspected infection fulfill its criteria. These should therefore be specified in greater detail, and applied more consistently, during future sepsis studies.Trial registrationThe MARS study is registered at ClinicalTrials.gov (identifier NCT 01905033).