Table 2 Measures of agreement between registration data and verbal autopsy diagnoses, and implied validity* of registered mortality fractions by cause in the Disease Surveillance Points system in rural China, 2002.

Category	Deaths (n)		κ	Sensitivity*	PPV*	Change in cause‐specific mortality fraction (%)
	In registration data	From VA re‐investigation
Drowning	48	44	0.91	95.5	87.5	8.3
Road traffic accidents	81	85	0.86	84.7	88.9	−4.9
Liver cancer	92	86	0.79	82.6	77.2	6.5
Diabetes	27	29	0.78	75.9	81.5	−7.4
Oesophageal cancer	45	52	0.76	71.2	82.2	−15.6
Stomach cancer	91	89	0.73	75.3	73.6	2.2
Suicide	64	63	0.73	74.6	73.4	1.6
Perinatal conditions	36	36	0.72	72.2	72.2	0.0
Colorectal cancers	29	24	0.71	79.2	65.5	17.2
Cerebrovascular diseases	482	497	0.69	73.6	75.9	−3.1
Lung cancer	88	79	0.68	73.4	65.9	10.2
Liver disease	43	54	0.57	51.9	65.1	−25.6
Falls	28	44	0.52	43.2	67.9	−57.1
Other external causes	55	51	0.46	49.0	45.5	7.3
Pneumonia	32	40	0.44	40.0	50.0	−25.0
Ischaemic heart disease	169	128	0.42	53.1	40.2	24.3
COPD	252	297	0.41	44.1	52.0	−17.9
Tuberculosis	34	50	0.35	30.0	44.1	−47.1
Hypertensive diseases	59	50	0.33	38.0	32.2	15.3
Poisoning	14	25	0.30	24.0	42.9	−78.6
Ill‐defined conditions	137	229	0.28	26.64	44.53	−67.2
Other cardiac disease	224	73	0.05	19.2	6.3	67.4
All other causes	352	357				−1.4
Total deaths	2482	2482

COPD, chronic obstructive pulmonary disease; PPV, positive predictive value; VA, verbal autopsy.

A positive change in cause‐specific mortality fractions suggests over‐reporting by the routine registration system, a negative change suggests under‐reporting.

*To measure validity, the VA diagnoses were used as “proxy” reference diagnoses, hence the use of the qualification term “implied”.