Inter-rater reliability of data elements from a prototype of the Paul Coverdell National Acute Stroke Registry

BMC Neurology

Table 1 Inter-rater reliability for stroke sub-type, demographic, clinical features, and pre-hospital data.

Variable (Item Number)	Dichotomous response option	N	PI*	Kappa (95% LCL)†	BI‡
Coverdell stroke sub-type (1.0)	Ischemic Stroke	104	0.24	0.61 (0.46)	-0.07
	Intracranial hemorrhage (ICH)	104	-0.86	0.93 (0.79)	0.01
	Subarachnoid hemorrhage (SAH)	104	-0.89	0.90 (0.72)	0.01
	Transient ischemic attack (TIA)	104	-0.64	0.70 (0.52)	-0.01
Gender (1.2)	Female	104	0.16	0.98 (0.94)	0.01
Race (1.3)	White	104	0.40	0.55 (0.38)	0.13
	Black or African American	104	-0.66	0.97 (0.90)	0.01
	Not documented (missing)	104	-0.76	0.07 (-0.13)	-0.11
Health Insurance status (1.7)	Medicare	104	0.43	0.83 (0.72)	0.03
	Medicaid	104	-0.86	0.50 (0.18)	- 0.03
	Self-pay	104	-0.89	0.52 (0.15)	0.01
Arrival mode (2.2)	Ambulance	104	-0.40	0.82 (0.69)	0.00
	Other	104	-0.14	0.59 (0.43)	-0.01
	Hospital transfer	104	-0.74	0.70 (0.50)	0.03
	Not documented (missing)	104	-0.72	0.40 (0.15)	-0.01
Direct admit (2.2b)	Yes	98	-0.88	0.47 (0.17)	0.04

* Prevalence Index (PI) is a measure of the true prevalence of the trait. PI is 0 when concordant responses are equally balanced between the 2 raters. A large negative PI indicates trait is rarely found, while a large positive PI indicates trait is common. Variables with extreme distributions i.e., PI < -0.90 or > 0.90 are not shown. † 95% LCL = 95% Lower Confidence Limit.
‡ Bias Index (BI) is a measure of the relative bias between the hospital and audit abstractors. BI is 0 when there is no bias. Positive BI values indicate a bias toward the hospital abstractor, negative BI values indicates a bias towards the audit abstracter. BI values > 0.10 or < -0.10 indicate substantial bias.

ISSN: 1471-2377