Assessing proxy indicators of record linkage bias in the establishment of a cohort from a Portuguese community-based screening network

Background Record linkage is important for integrating data into longitudinal epidemiological research, particularly in community-based screening programs where pseudonymized identifiers protect participant privacy. However, linkage errors and missing data in the identification codes used for record linkage can introduce bias and affect study validity. Methods We analysed data from the Portuguese Community-Based Screening Network, covering 188,585 HIV, hepatitis B/C, and syphilis testing sessions (2016–2023). Participants were categorized into three age groups ≤25, 26–49, and ≥50 years. A pseudonymized code, based on sex assigned at birth, date of birth, and initials of the participant’s and their mother’s first names, enabled record linkage. We compared characteristics of testing sessions available and not available for record linkage, identified coding errors, and calculated Shannon entropy to assess randomness in identifier components. Meta-analysis and meta-regression were conducted to evaluate factors associated with linkage participation and coding errors, while crude linear regression was used to analyse entropy. Results Overall, 92.6% of sessions were available for record linkage. Participation was highest among older individuals (97.1%) and lowest among younger participants (87.1%). The northern region exhibited the lowest participation (61.9%). Sex assignment errors occurred in 2.3% of cases, particularly among transgender participants, while date of birth errors (0.2%) reflected invalid or implausible values. Entropy scores varied across testing services, indicating inconsistencies in code generation, with regional differences confirmed by crude linear regression (North: −0.92, 95% CI = −1.74, −0.09). Conclusion Record linkage in community-based screening networks is feasible, though participation and data quality vary by region and population subgroup. Errors in key identifiers and inconsistent coding practices highlight the need for standardized protocols. Entropy-based scoring is a useful tool for monitoring identifier quality and identifying potential linkage problems, thereby supporting data integration.