Accuracy of record linkage software in merging dental administrative data sets

Abstract

Objective: To determine the accuracy of record matching using "Link King" software that uses an ordinal score for the certainty that linked records are valid matches.

Methods: We linked records in North Carolina Medicaid files to public health surveillance files using Link King matching software. We selected a stratified random sample of 230 of 45,295 linked records and 50 of 35,119 non-linked surveillance records, then manually reviewed the records. Sensitivity (Sn) and Specificity (Sp) were calculated based on each cut point of the Link King score, using manual review as the gold standard.

Results: The Sn increased from 0.837 (95% CI: 0.785, 0.892) to 0.935 (0.879, 0.994) and Sp decreased from 0.893 (0.816, 0.976) to 0.865 (0.790, 0.947) as cut points were varied to widen the scope of declared matches. With a goal of both Sn and Sp being large, accuracy was best when linked record pairs from the top three levels of certainty were included in the match.

Conclusions: This study found that a publicly available software program accurately merged Medicaid and surveillance data. The Link King could be useful to researchers in merging administrative databases.

Figure 1

The data merging process resulted in 56% of the NCSoDC records being linked to a NC Medicaid record. Numbers of linked record pairs (N_i; i=1,2,3,4,5) varied widely across certainty levels. We defined non-linked record pairs as certainty level 6 (N=35,119). Analysis weights were equal to the inverse of the fraction sampled (n_i/N_i) for each certainty level.