This report presents data from a study carried out in three African countries to assess the validity of verbal autopsies--based on information about symptoms and signs observed antemortem by relatives or associates of deceased individuals--for determining the causes of institutional maternal death. The validity of the verbal autopsy was assessed for each cause of death; and for groups of "direct" and "indirect" maternal causes, by comparing the verbal autopsy diagnoses with the reference diagnoses and calculating their sensitivity, specificity, and positive predictive value. Verbal autopsies were found to be highly specific (98 percent specificity for all causes of maternal death) but not very sensitive (< or = 60 percent sensitivity for all causes except ante/postpartum hemorrhage). Verbal autopsy estimates of cause-specific mortality were comparable to expected values for most of the causes. The study shows that certain direct causes of hospital-based maternal mortality can be determined by means of verbal autopsies with a reasonable level of confidence.
PIP: This study shows that certain direct causes of hospital-based maternal mortality can be validly determined by verbal autopsies. Data were obtained during 1993-95 from a rural district hospital in Kilombero District, Tanzania; a rural teaching hospital in Oromiya Region, Ethiopia; and a rural district hospital in Bawku District, Ghana. Hospital deliveries averaged 1200-1500/year. Maternal deaths averaged 20-30/year. The study population included all adults aged over 15 years dying at ages 15-49 years in the specified hospitals and who lived within 60 km. Data also included hospital records and death certificates. Findings among physicians indicate that direct maternal causes (DMCs), including abortion, had a sensitivity of 82% and a specificity of 93%. For indirect maternal causes (IMCs), the specificity was 97% and the sensitivity was 38%. The positive predictive value (PPV) was 70% for DMCs and 67% for IMCs. Sensitivity was the lowest measure of reliability for all causes. Sensitivity was higher than 60% for all DMCs, with the exception of eclampsia (40%), and lower than 50% for common IMCs. IMCs had a specificity over 98%. The PPV was under 60% for most IMCs and DMCs, except obstructed labor (80%), abortion (64%), and hepatitis (100%). Findings using the algorithm showed lower specificities (93% for DMCs and IMCs). Sensitivity was 60% for DMCs and 68% for IMCs. There was reasonable agreement between physician diagnoses and algorithms. Individual misclassifications of causes were higher in algorithm-based verbal autopsies. False nonmaternal causes were greater among algorithm-based verbal autopsy diagnoses.