Campbell Kevin M, Deck Dennis, Krupski Antoinette
Washington State Division of Alcohol and Substance Abuse, PO Box 45330, Olympia, Washington 98504-5330, USA.
Health Informatics J. 2008 Mar;14(1):5-15. doi: 10.1177/1460458208088855.
The study objective was to compare the accuracy of a deterministic record linkage algorithm and two public domain software applications for record linkage (The Link King and Link Plus). The three algorithms were used to unduplicate an administrative database containing personal identifiers for over 500,000 clients. Subsequently, a random sample of linked records was submitted to four research staff for blinded clerical review. Using reviewers' decisions as the 'gold standard', sensitivity and positive predictive values (PPVs) were estimated. Optimally, sensitivity and PPVs in the mid 90s could be obtained from both The Link King and Link Plus. Sensitivity and PPVs using a basic deterministic algorithm were 79 and 98 per cent respectively. Thus the full feature set of The Link King makes it an attractive option for SAS users. Link Plus is a good choice for non-SAS users as long as necessary programming resources are available for processing record pairs identified by Link Plus.
本研究的目的是比较一种确定性记录链接算法与两款用于记录链接的公共领域软件应用程序（The Link King和Link Plus）的准确性。这三种算法被用于对一个包含50多万客户个人标识符的管理数据库进行去重。随后，将随机抽取的链接记录样本提交给四名研究人员进行盲法人工审核。以审核人员的判定作为“金标准”，估算了敏感度和阳性预测值（PPV）。理想情况下，The Link King和Link Plus都能达到95%左右的敏感度和PPV。使用基本确定性算法时，敏感度和PPV分别为79%和98%。因此，The Link King的完整功能集使其成为SAS用户的一个有吸引力的选择。只要有必要的编程资源来处理Link Plus识别出的记录对，Link Plus对于非SAS用户来说是个不错的选择。
