An efficient record linkage scheme using graphical analysis for identifier error detection

Table 1 Identifiers and Record linkage operation

Start cluster id	New cluster id	NHS number	hospital number	Surname	Forename	sex	date of birth (ddmmyyyy)	frequency of occurrence
1	1	NULL	4496644	WILSON	DAVID	M	14061940	3
2	2	5170231111	NULL	WILSON	DAVID	M	01051939	1
3	3	3319004037	4118890	WILSON	DAVID	M	20011969	2
4	4	NULL	NULL	WILSON	DAVID	M	20011969	1
5	3	3319004037	NULL	WILSON	DAVID	M	20011969	2
6	6	NULL	4118890	WILSON	DAVID	M	20011969	1

An example of identifiers provided for patients with forename and surname 'David Wilson'. The details have been changed to protect patient confidentiality. Null fields indicate there was no information provided in that field.
One cycle of the record linkage is illustrated. Consider each combination of identifiers to belong to its own, discrete cluster, identified by a cluster identifier (Start cluster id). For all sets in which at least one member shares an NHS number identifier with a different set, combine these sets into a single set (New Cluster ID). The operation proceeds for all identifiers.

ISSN: 1472-6947