Contingency tables.

Background
Previous work has identified that malarial retinopathy has diagnostic value in paediatric cerebral malaria (CM). To improve our understanding of malarial retinopathy as a predictor of cerebral parasite sequestration in paediatric CM we reviewed data from the Blantyre autopsy study, to test the hypothesis that malarial retinopathy is an accurate predictor of cerebral parasite sequestration in an autopsy cohort.
Methods and findings
We performed a retrospective analysis of data collected from a consecutive series of patients presenting to the Pediatric Research Ward at Queen Elizabeth Central Hospital in Blantyre, Malawi between 1996 and 2010. We determined the diagnostic accuracy of malarial retinopathy as a predictor of cerebral parasite sequestration in a cohort of children with fatal CM. Of 84 children included in the study, 65 met the World Health Organization clinical diagnostic criteria for CM during life. Eighteen (28%) of 65 did not have evidence of cerebral parasite sequestration at autopsy and 17 had an alternative cause of death. Malarial retinopathy had a sensitivity of 89.4% (95% CI [77.6%, 95.6%]) and specificity of 73.0% (95% CI [57.2%, 84.8%]) to predict cerebral parasite sequestration. In a subset of patients with graded retinal assessments, this was improved to 94.3% (95% CI [81.7%, 98.7%]) and 88.0% (95% CI [70.4%, 96.2%]) by reclassifying patients in whom the only retinal sign was 1–5 haemorrhages in a single eye as retinopathy negative. This study is limited by its retrospective nature and the inherent selection bias associated with autopsy studies.
Conclusions
Malarial retinopathy remains the most specific point-of-care test for CM in endemic areas. Its specificity may be improved, without sacrificing sensitivity, by reclassifying patients in whom the only retinal sign is fewer than 5 haemorrhages in a single eye as malarial retinopathy negative. A management algorithm is proposed for integration of malarial retinopathy into clinical care in both well-resourced and resource-limited environments.