Background There is currently no universally accepted classification of childhood retinal haemorrhages.

Aim To measure the inter- and intra-observer agreement of clinical classifications of retinal haemorrhages in children.

Methods Four examiners (two consultant ophthalmologists and two other clinicians) were shown 142 retinal haemorrhages on 31 RetCam photographs. The retinal haemorrhages were from children with accidental or abusive head injury, or other encephalopathies, and included retinal haemorrhages of different ages. Specified haemorrhages were initially classified by each examiner according to their clinical understanding. Altogether, 26 haemorrhages were re-presented to test intra-observer consistency. Examiners then agreed a common description for each haemorrhage type and five categories were described (vitreous, pre-retinal, nerve fibre layer, intra-retinal/sub-retinal or indeterminate) and the study repeated.

Results There was ‘fair agreement’ initially (Fleiss' unweighted κ=0.219) and, with the agreed classification, slight improvement (0.356). Intra-observer agreement marginally improved on re-test. The two consultant ophthalmologists showed ‘fair’ agreement on both occasions (paired κ statistic). The other rater pair improved from ‘fair’ to ‘substantial’ agreement with the new classification.

Conclusions The classification of retinal haemorrhage in children by appearance alone shows only fair agreement between examiners. Clinicians who are not consultant ophthalmologists appear to benefit from the new succinct classification.