Enhancing the accuracy of register-based metrics: Comparing methods for handling overlapping psychiatric register entries in Finnish healthcare registers

Abstract

Objectives: Healthcare registers are invaluable resources for research. Partly overlapping register entries and preliminary diagnoses may introduce bias. We compare various methods to address this issue and provide fully reproducible open-source R scripts. Methods: We used all Finnish healthcare registers 1969–2020, including inpatient, outpatient and primary care. Four distinct models were formulated based on previous reports to identify actual admissions, discharges, and discharge diagnoses. We calculated the annual number of treatment events and patients, and the median length of hospital stay (LOS). We compared these metrics to non-processed data. Additionally, we analyzed the lifetime number of individuals with registered mental disorders. Results: Overall, 2,130,468 individuals had a registered medical contact related to mental disorders. After processing, the annual number of inpatient episodes decreased by 5.85%–10.87% and LOS increased by up to 3 days (27.27%) in years 2011–2020. The number of individuals with lifetime diagnoses reduced by more than 1 percent point (pp) in two categories: schizophrenia spectrum (3.69–3.81pp) and organic mental disorders (1.2–1.27pp). Conclusions: The methods employed in pre-processing register data significantly impact the number of inpatient episodes and LOS. Regarding lifetime incidence of mental disorders, schizophrenia spectrum disorders require a particular focus on data pre-processing.