Estimate the deterministic record linkage specificity using indirect identifiers – birth cohort Generation XXI

Objectives Record linkage within the same dataset enables longitudinal studies with pseudonymized data using indirect identifiers including sex at birth and residence codes. However, evidence on variable combinations remains limited. This study evaluates the validity of matching algorithms using indirect identifiers within a pseudonymization framework against a gold standard. Methods Data were drawn from the Generation XXI birth cohort, including 8 647 newborns from maternity hospitals in Porto, Portugal (2005–2006). Records were pseudonymized and linked using two deterministic strategies, sex at birth, date of birth, and the first letters of the mother’s and child’s first names (strategy 1), and the same variables plus maternal education (strategy 2). Linkage quality was assessed against a gold standard using specificity, positive predictive value, and Cohen’s kappa. Results Among 8 614 eligible participants, 33 lacked sufficient information to generate an identifier. Strategy 2 outperformed Strategy 1, reducing false matches (87 vs. 439) and improving PPV (99.00% vs. 95.15%) and Cohen’s kappa (0.995 vs. 0.975), while maintaining specificity. Conclusions Combining indirect identifiers for the purpose of longitudinal studies (within the same dataset) improved record linkage accuracy in pseudonymized data, achieving higher agreement than previous approaches. These findings highlight the importance of demographic variables to reduce linkage errors while maintaining privacy protections and emphasize the need for validation when applying linkage methods to datasets.