Performance of four computer-coded verbal autopsy methods for cause of death assignment compared with physician coding on 24,000 deaths in low- and middle-income countries

Table 3 Positive predictive values of computer-coded verbal autopsy methods versus physician-coded verbal autopsy reference standards

	Test cases	Open-source random forest		Open-source tariff method		InterVA-4		Average for top cause (%)
Dataset		Top (%)	Top 3 (%)	Top (%)	Top 3 (%)	Top (%)	Top 3 (%)
China	400	35	57	36	70	N/A	N/A	36
Institute for Health Metrics and Evaluation	400	33	55	34	53	N/A	N/A	34
Million Death Study	6100	58	82	52	76	42^a	63^a	51
Agincourt	2900	45	77	42	69	42	58	43
Matlab	1600	49	74	52	74	48	64	50
Average		44	69	43	68	44	62

Top cause represents accuracy of the CCVA method’s most probable cause matching the cause assigned by PCVA; Top 3 represents whether CCVA’s three most probable causes contain the cause assigned by PCVA. Averages calculated across CCVA methods only use results for the top cause. ^aThe Million Death Study dataset used for InterVA-4 contained a sample of 552 cases, in which we extracted additional InterVA-4 indicators from the narratives.

ISSN: 1741-7015