ObjectiveSplinter haemorrhages are an examination finding that has classically been associated with infective endocarditis (IE), but are not included in current diagnostic algorithms. Splinter haemorrhages have not been evaluated as a diagnostic tool using modern definitions of IE. We determined their sensitivity and specificity in patients with suspected IE and investigated their inclusion in the Duke criteria.MethodsThis is a retrospective diagnostic accuracy study using data from 1119 patients with suspected IE referred to the IE service. Patients were categorised according to the Duke criteria, the current diagnostic gold standard, into Duke ‘rejected’, ‘possible’ or ‘definite’ groups. Definite cases (n=451) served as the true positives and rejected cases (n=486) as the true negatives against which splinter haemorrhages were compared. Duke possible cases (n=182) were used the assess the clinical impact of adding splinter haemorrhages to the Duke criteria.ResultsIn clinically suspected cases of IE and using the Duke criteria as the gold standard comparator, splinter haemorrhages had a sensitivity of 26% (95% CI 22 to 31) (119 out of 451) and a specificity of 83% (95% CI 79 to 86) (403 out of 486). Inclusion of splinter haemorrhages as a minor vascular phenomenon in the Duke criteria would result in a reclassification of 12% of cases from Duke rejected to possible and 13% from Duke possible to definite.ConclusionSplinter haemorrhages are an insensitive tool in the diagnosis of IE, but their high specificity indicates they do have clinical value in patients with suspected infection. Their inclusion in the Duke criteria as a minor vascular criterion reduces diagnostic uncertainty for some Duke possible cases, while increasing it for a similar proportion of Duke rejected cases.