Data cleaning and management protocols for linked perinatal research data: a good practice example from the Smoking MUMS (Maternal Use of Medications and Safety) Study

Background: Data cleaning is an important quality assurance in data linkage research studies. This paper presents the data cleaning and preparation process for a large-scale cross-jurisdictional Australian study (the Smoking MUMS Study) to evaluate the utilisation and safety of smoking cessation pharmacotherapies during pregnancy. Methods: Perinatal records for all deliveries (2003-2012) in the States of New South Wales (NSW) and Western Australia were linked to State-based data collections including hospital separation, emergency department and death data (mothers and babies) and congenital defect notifications (babies in NSW) by State-based data linkage units. A national data linkage unit linked pharmaceutical dispensing data for the mothers. All linkages were probabilistic. Twenty two steps assessed the uniqueness of records and consistency of items within and across data sources, resolved discrepancies in the linkages between units, and identified women having records in both States. Results: State-based linkages yielded a cohort of 783,471 mothers and 1,232,440 babies. Likely false positive links relating to 3703 mothers were identified. Corrections of baby's date of birth and age, and parity were made for 43,578 records while 1996 records were flagged as duplicates. Checks for the uniqueness of the matches between State and national linkages detected 3404 ID clusters, suggestive of missed links in the State linkages, and identified 1986 women who had records in both States. Conclusions: Analysis of content data can identify inaccurate links that cannot be detected by data linkage units that have access to personal identifiers only. Perinatal researchers are encouraged to adopt the methods presented to ensure quality and consistency among studies using linked administrative data.