Point-of-contact Interactive Record Linkage (PIRL): A software tool to prospectively link demographic surveillance and health facility data

<ns4:p>Linking a health and demographic surveillance system (HDSS) to data from a health facility that serves the HDSS population generates a research infrastructure for directly observed data on access to and utilization of health facility services. Many HDSS sites, however, are in areas that lack unique national identifiers or suffer from data quality issues, such as incomplete records, spelling errors, and name and residence changes, all of which complicate record linkage approaches when applied retrospectively. We developed Point-of-contact Interactive Record Linkage (PIRL) software that is used to prospectively link health records from a local health facility to an HDSS in rural Tanzania. This prospective approach to record linkage is carried out in the presence of the individual whose records are being linked, which has the advantage that any uncertainty surrounding their identity can be resolved during a brief interaction, whereby extraneous information (e.g., household membership) can be referred to as an additional criterion to adjudicate between multiple potential matches. Our software uses a probabilistic record linkage algorithm based on the Fellegi-Sunter model to search and rank potential matches in the HDSS data source. Key advantages of this software are its ability to perform multiple searches for the same individual and save patient-specific notes that are retrieved during subsequent clinic visits. A search on the HDSS database (n=110,000) takes less than 15 seconds to complete. Excluding time spent obtaining written consent, the median duration of time we spend with each patient is six minutes. In this setting, a purely automated retrospective approach to record linkage would have only correctly identified about half of the true matches and resulted in high linkage errors; therefore highlighting immediate benefit of conducting interactive record linkage using the PIRL software.</ns4:p>