Determining causes of death through an abridged verbal autopsy tool: a pilot test conducted among vulnerable communities in Kolkata, India

Users often regard the 2022 WHO verbal autopsy (VA) instrument as lengthy and time-consuming to implement, emphasising the need for a more concise and practical tool, particularly for use in unstable or resource-limited settings. This study aimed to develop, pilot test and validate an abridged version of the WHO VA instrument capable of identifying probable causes of neonatal (0-27 days), child (1 month-11 years) and adult (12-49 years) deaths in vulnerable populations.

An abridged VA instrument based on the expert algorithm method was developed and applied to deaths recorded between 2021 and 2022 in three tertiary general hospitals and one tertiary paediatric hospital in Kolkata, West Bengal. VA interviews were conducted between July 2024 and January 2025. Causes of death were assigned using two approaches: (1) underlying medical record diagnoses and (2) a hierarchical computer-based VA algorithm. Diagnostic metrics were used to validate algorithmic causes against medical diagnoses.

Among 270 deaths (90 per age group), 95.5% yielded a VA diagnosis. Among neonates, the leading causes were sepsis (34.4%: 83.8% VA sensitivity, 94.9% VA specificity), IPRE (25.6%: 69.6%, 97.0%) and pneumonia (16.7%: 100%, 98.7%). Among children, the two most common medical diagnoses were pneumonia (53.3%: 100%, 50%) and other infections including sepsis without a known source (25.6%: 21.7%, 98.5%). Among adults, the two most common medical diagnostic categories were acute febrile illness (38.5%: 80.9%, 98.6%) and cardiovascular diseases (38.5%: 100%, 98.6%). The accuracy of the new VA instrument compared with the medical diagnosis was 81.1% for neonates, 71.1% for children and 87.8% for adults, with an overall accuracy of 80%.

The abridged VA instrument demonstrated satisfactory accuracy in identifying the probable causes of death across all age groups in the study populations. This instrument holds potential for enhancing cause-of-death reporting in resource-limited settings.