© The Author(s) 2020. Background: Under-coding of dementia during hospitalisation results in an inability to identify all patients with dementia using hospital administrative data. Clinical coding can be viewed as a proxy for management; therefore, under-coding indicates dementia was not considered in the patient’s management. While under-coding of dementia is well established, there is sparse evidence on whether dementia is coded in subsequent hospitalisations among patients with a known diagnosis. Objective: (a) To describe patterns of dementia coding over 5 years after a first-coded (i.e. index) admission for dementia; (b) to identify factors associated with clinical coding of dementia; and (c) to identify patient subgroups at risk of not being coded to inform future interventions to improve hospital identification and management of dementia. Method: Retrospective study of longitudinal hospital data from 1 July 2006 to 30 June 2015 for 7919 patients hospitalised during the 5 years’ post-index admission for dementia in a regional local health district of New South Wales, Australia. Results: Dementia was coded in 63.9% of admissions in the 12 months following index admission for dementia; this decreased to 53.7% after 5 years. Patients were 20% more likely to have dementia actively managed when it co-occurred with delirium. Under-coding varied across conditions, with dementia more likely to be coded in admissions for falls and pneumonitis, and less likely for heart failure, pneumonia and urinary tract infection (UTI). Conclusion: The frequency with which dementia was not coded highlights opportunities to improve identification and management of dementia through dementia-specific care, enhanced clinical protocols, and interventions focused around heart failure, pneumonia and UTI admissions.