Objective Coding of health data has implications both on quality of care and allocation of resources. In the past coding quality has been checked by manual reviews of hospital records, but this approach has limited applicability. We analyse longitudinal administrative hospital data to detect changes over time in coding practice and quality. Methods We obtained 25 million episodes of coded admission data over a period of several years from 180 hospitals around Australia and New Zealand. We define indicators of ICD coding quality by looking at the frequency of failure to properly code Acute vs. Chronic respiratory failure and Ischemic vs. Haemorrhagic stroke. We apply these indicators to 6 financial years of data. Key Messages 1. There has been a slow decrease over time of the proportion of strokes that are coded as â€œUnspecifiedâ€ type while we find that coding of stroke became more descriptive of the type of stroke. 2. There has been a sharp increase in the past few years in the proportion of respiratory failures that are coded as neither Acute nor Chronic.