IntroductionStudies based on health claims data (HCD) have been increasingly adopted in medical research for their strengths in large sample size and abundant information, and the Taiwan National Health Insurance Research Database (NHIRD) has been widely used in medical research across disciplines, including dementia. How the diagnostic codes are applied to define the diseases/conditions of interest is pivotal in HCD-related research, but the consensus on the issue that diagnostic codes most appropriately define dementias in the NHIRD is lacking. The objectives of this scoping review are (1) to investigate the relevant characteristics in the published reports targeting dementias based on the NHIRD, and (2) to address the diversity by a case study.Methods and analysisThis scoping review protocol follows the methodological framework of the Joanna Briggs Institute Reviewer’s Manual and the guidance of the Preferred Reporting Items for Systematic Reviews and Meta-Analyses extension for scoping reviews. The review will be performed between 1 March and 31 December 2022 in five stages, including identifying the relevant studies, developing search strategies, individually screening and selecting evidence, collecting and extracting data, and summarising and reporting the results. The electronic databases of MEDLINE, EMBASE, CENTRAL, CINAHL, and PsycINFO, Airiti Library Academic Database, the National Health Insurance Administration’s repository, and Taiwan Government Research Bulletin will be searched. We will perform narrative syntheses of the results to address research questions and will analyse the prevalence across the included individual studies as a case study.Ethics and disseminationOur scoping review is a review of the published reports and ethical approval is not required. The results will provide a panorama of the dementia studies based on the NHIRD. We will disseminate our findings through peer-reviewed journals and conferences, and share with stakeholders by distributing the summaries in social media and emails.