BackgroundMonitoring progress with disease and injury reduction in many populations will require widespread use of verbal autopsy (VA). Multiple methods have been developed for assigning cause of death from a VA but their application is restricted by uncertainty about their reliability.MethodsWe investigated the validity of five automated VA methods for assigning cause of death: InterVA-4, Random Forest (RF), Simplified Symptom Pattern (SSP), Tariff method (Tariff), and King-Lu (KL), in addition to physician review of VA forms (PCVA), based on 12,535 cases from diverse populations for which the true cause of death had been reliably established. For adults, children, neonates and stillbirths, performance was assessed separately for individuals using sensitivity, specificity, Kappa, and chance-corrected concordance (CCC) and for populations using cause specific mortality fraction (CSMF) accuracy, with and without additional diagnostic information from prior contact with health services. A total of 500 train-test splits were used to ensure that results are robust to variation in the underlying cause of death distribution.ResultsThree automated diagnostic methods, Tariff, SSP, and RF, but not InterVA-4, performed better than physician review in all age groups, study sites, and for the majority of causes of death studied. For adults, CSMF accuracy ranged from 0.764 to 0.770, compared with 0.680 for PCVA and 0.625 for InterVA; CCC varied from 49.2% to 54.1%, compared with 42.2% for PCVA, and 23.8% for InterVA. For children, CSMF accuracy was 0.783 for Tariff, 0.678 for PCVA, and 0.520 for InterVA; CCC was 52.5% for Tariff, 44.5% for PCVA, and 30.3% for InterVA. For neonates, CSMF accuracy was 0.817 for Tariff, 0.719 for PCVA, and 0.629 for InterVA; CCC varied from 47.3% to 50.3% for the three automated methods, 29.3% for PCVA, and 19.4% for InterVA. The method with the highest sensitivity for a specific cause varied by cause.ConclusionsPhysician review of verbal autopsy questionnaires is less accurate than automated methods in determining both individual and population causes of death. Overall, Tariff performs as well or better than other methods and should be widely applied in routine mortality surveillance systems with poor cause of death certification practices.

BackgroundVerbal autopsy (VA) is an important method for obtaining cause of death information in settings without vital registration and medical certification of causes of death. An array of methods, including physician review and computer-automated methods, have been proposed and used. Choosing the best method for VA requires the appropriate metrics for assessing performance. Currently used metrics such as sensitivity, specificity, and cause-specific mortality fraction (CSMF) errors do not provide a robust basis for comparison.MethodsWe use simple simulations of populations with three causes of death to demonstrate that most metrics used in VA validation studies are extremely sensitive to the CSMF composition of the test dataset. Simulations also demonstrate that an inferior method can appear to have better performance than an alternative due strictly to the CSMF composition of the test set.ResultsVA methods need to be evaluated across a set of test datasets with widely varying CSMF compositions. We propose two metrics for assessing the performance of a proposed VA method. For assessing how well a method does at individual cause of death assignment, we recommend the average chance-corrected concordance across causes. This metric is insensitive to the CSMF composition of the test sets and corrects for the degree to which a method will get the cause correct due strictly to chance. For the evaluation of CSMF estimation, we propose CSMF accuracy. CSMF accuracy is defined as one minus the sum of all absolute CSMF errors across causes divided by the maximum total error. It is scaled from zero to one and can generalize a method's CSMF estimation capability regardless of the number of causes. Performance of a VA method for CSMF estimation by cause can be assessed by examining the relationship across test datasets between the estimated CSMF and the true CSMF.ConclusionsWith an increasing range of VA methods available, it will be critical to objectively assess their performance in assigning cause of death. Chance-corrected concordance and CSMF accuracy assessed across a large number of test datasets with widely varying CSMF composition provide a robust strategy for this assessment.