Comparative performance of verbal autopsy methods in identifying causes of adult mortality: A case study in India.

ABSTRACT

Background & objectives: Cause of death assignment from verbal autopsy (VA) questionnaires is conventionally accomplished through physician review. However, since recently, computer softwares have been developed to assign the cause of death. The present study evaluated the performance of computer software in assigning the cause of death from the VA, as compared to physician review. Methods: VA of 600 adult deaths was conducted using open- and close-ended questionnaires in Nandpur Kalour Block of Punjab, India. Entire VA forms were used by two physicians independently to assign the cause of death using the International Statistical Classification of Diseases and Related Health Problems (ICD)-10 codes. In case of disagreement between them, reconciliation was done, and in cases of persistent disagreements finally, adjudication was done by a third physician. InterVA-4-generated causes from close-ended questionnaires were compared using Kappa statistics with causes assigned by physicians using a questionnaire having both open- and close-ended questions. At the population level, Cause-Specific Mortality Fraction (CSMF) accuracy and P-value from McNemar's paired Chi-square were calculated. CSMF accuracy indicates the absolute deviation of a set of proportions of causes of death out of the total number of deaths between the two methods. Results: The overall agreement between InterVA-4 and physician coding was 'fair' (κ=0.42; 95% confidence interval 0.38, 0.46). CSMF accuracy was found to be 0.71. The differences in proportions from the two methods were statistically different as per McNemar's paired Chi-square test for ischaemic heart diseases, liver cirrhosis and maternal deaths. Interpretation & conclusions: In comparison to physicians, assignment of causes of death by InterVA- 4 was only 'fair'. Hence, it may be appropriate to continue with physician review as the optimal option available in the current scenario.