How good is probabilistic record linkage to reconstruct reproductive histories? Results from the Aberdeen children of the 1950s study

The Children of the 1950s cohort includes 5868 females born in Aberdeen 1950–56 who were in primary schools in the city in 1962. In 2001 a postal questionnaire was sent to the cohort members resident in the UK requesting information on offspring. Probabilistic record linkage (based on surname, maiden name, initials, date of birth and postcode) was used to link the females in the cohort to birth records held by the Scottish Maternity Record System (SMR 2).We attempted to mail a total of 5540 women; 3752 (68%) returned a completed questionnaire. Of these 86% reported having had at least one birth. Linkage to SMR 2 was attempted for 5634 women, one or more maternity records were found for 3743. There were 2604 women who reported at least one birth in the questionnaire and who were linked to one or more SMR 2 records. When judged against the questionnaire information, the linkage correctly identified 4930 births and missed 601 others. These mostly occurred outside of Scotland (147) or prior to full coverage by SMR 2 (454). There were 134 births incorrectly linked to SMR 2.Probabilistic record linkage to routine maternity records applied to population-based cohort, using name, date of birth and place of residence, can have high specificity, and as such may be reliably used in epidemiological research.Probabilistic record linkage is increasingly used in health research [1,2] particularly in countries such as Canada and the United Kingdom where there is no system of unique national identity numbers that can be used for linkage purposes [3,4].Despite a sound theoretical basis for probabilistic record linkage [2], it is often regarded as a second best method in comparison to exact linkage. There is, however, only a limited literature concerned with the reliability and accuracy of probabilistic record linkage [5-9], mainly because there is rarely an independent gold-standard source of information against which the linkage can be judged. Additionally, most assessments of linka