The Automated Coding of Causes of Death in the Netherlands

Background: The production of cause-of-death statistics requires the coding of an underlying cause of death from death certificates. To date, more and more countries switch from manual to automated coding. Such a change of method can cause a change in frequency of major causes of death in statistics. Therefore we coded a dataset both manually and automatically in order to study differences between these two methods for producing cause-of-death statistics. Methods: We performed a bridge (double) coding study. A death certificate was coded by medical coders (manual) and also by IRIS, free software for automated coding of causes of death, independently of each other. For 86 930 death certificates, we could compare ICD-10 codes for the underlying cause of death. We calculated a Comparability Ratio (CR) and a Perfect Compatibility Percentage (PCP). A CR indicates the expected change in the frequency of occurrence of a cause of death when changing from manual to automated coding (reproducibility). A PCP indicates the (perfect) agreement between medical coders and IRIS on coding the underlying cause of death (validity). Results: Of the double coded death certificates (n= 86 900), 75 per cent showed exactly the same underlying cause of death (ICD-10, four digits). On the three digit level of the ICD-10 code, the overall agreement between manual and automated coding was 84 per cent and on ICD-10 chapter level the agreement was 89 per cent. Agreement differed by ICD-10 chapter. Compared to manual coding, IRIS selected significant more infectious diseases (47 per cent), endocrine disorders (16 per cent), mental disorders (32 per cent) and diseases of the nervous system (18 per cent) as underlying cause of death; IRIS selected significant less diseases of the respiratory system (22 per cent), the digestive system (15 per cent), the skin (30 per cent), the genitourinary tract (22 per cent) and symptoms or signs (10 per cent) as underlying cause of death. Conclusions: A change from manual to automated coding causes (large) changes in the frequency of occurrence of major causes of death. In general, an automated coding system prefers degenerative disorders above infectious diseases as cause of death. Users of death statistics should be aware of this when studying trends in time or regional variations of causes of death.