Skip to main content
. 2022 May 14;42(8):999–1009. doi: 10.1177/0272989X221098699

Figure 1.

Figure 1

Measures of linkage performance displayed for different match weight thresholds for a subset of compared records with valid National Health Service (NHS) numbers. All links with a match weight above the threshold were considered valid; however, some ambulance records matched to more than 1 hospital record (or vice versa). In this case, only the greatest weighted link was retained and only if its weight was twice (or more) of the weight of the next greatest weighted link. For each probabilistically linked pair, both the ambulance record and the hospital record may have themselves been deterministically linked to other records. Before proceeding, it was necessary to ensure that these wider pools of linked records did not conflict with one another; that is, we needed to ensure there was no conflict in the transitivity due to the probabilistic linkage. If a conflict was detected (e.g., conflicting NHS numbers or date of births), the probabilistic link was disregarded.