This study evaluated the diagnostic accuracy of a verbal autopsy (VA) tool in ascertaining the causes of stillbirths and neonatal deaths in rural Ghana and was nested within a community-based maternal vitamin A supplementation trial (ObaapaVitA trial). All stillbirths and neonatal deaths between 1 January 2003 and 30 June 2004 were prospectively included. Community VAs were carried out within 6 months of death and were classified with a primary cause of death by three experienced paediatricans. The reference standard diagnosis was obtained by the study paediatrician in 4 district hospitals in the study area.
There were 20 317 deliveries, 661 stillbirths and 590 neonatal deaths with a VA diagnosis in the study population. A total of 311 stillbirths and 191 neonatal deaths had both a VA and a hospital reference standard diagnosis. The VA performed poorly for stillbirth diagnoses such as congenital abnormalities and maternal haemorrhage. Accuracy was higher for intrapartum obstetric complications and antepartum maternal disease. For neonatal deaths, sensitivity was >60% for all major causes; specificity was 76% for birth asphyxia but >85% for prematurity and infection. Overall, VA diagnostic accuracy was higher than expected in this rural African setting. Our classification system was based on the expected public health importance of the individual causes of death, differing implications for intervention and the ability to distinguish between the individual causes in low-resource settings. We believe this system was easier to use than traditional approaches and resulted in high precision and accuracy. However, further simplifications are needed to allow use of the World Health Organisation VA in routine child health programmes. The diagnostic accuracy of the VA tool should also be assessed in other regions and in multicentre studies.