Conditions of Non-Unique Identifiers in Record Linkage Using Japanese Cohort Data Set

• Activity Number: 375
• Type: Contributed
• Date/Time: Tuesday, August 2, 2016 : 10:30 AM to 12:20 PM
• Sponsor: Section on Statistics in Epidemiology
• Author(s): Michikazu Nakai* and Kunihiro Nishimura and Fumiaki Nakamura and Yoshihiro Miyamoto
• Companies: National Cerebral and Cardiovascular Center and National Cerebral and Cardiovascular Center and National Cerebral and Cardiovascular Center and National Cerebral and Cardiovascular Center
• Keywords: record linkage ; probabilistic algorithm ; non-unique identifiers ; epidemiology

Abstract

Applications of unique identifiers such as name, home address and social security number to link different datasets have been commonly used and well-published. Also, the theoretical concepts of probabilistic algorithm in record linkage have been well-defined in the literature. In this presentation we investigated several variables (weight, height, waist, age, sex, smoking and alcohol habit) as non-unique identifiers using Japanese cohort dataset with three-year baseline as well as ten-year baseline to observe how effectively these identifiers can be linked on record linkage. Moreover, we modified the conditions of these identifiers and estimated the sensitivity, specificity and accuracy for comparison. As a result, we concluded that the combination of age, sex, weight and height predicted better estimation than other combinations in both men and women in case of using three-year baseline whereas the combination of age, sex and height predicted better in both men and women in case of using ten-year baseline.