Impact of automated coding system implementation based on ICD-10 on mortality statistics in Taiwan: A bridge coding study

Objectives: To assess the effects of implementing an automated coding system and the International Classification of Disease (ICD) 10th Revision (ICD-10) on mortality statistics in Taiwan. Methods: We used a new automated coding system based on ICD-10 (automatic ICD-10) to assign the underlying cause of death (UCOD) for all deaths occurring in 2007. We then compared the number of deaths according to automatic ICD-10 with that coded manually by coders according to ICD-9 (manual ICD-9). The comparability ratio (CR, automatic ICD-10/manual ICD-9) for a specific cause of death (COD) category was calculated. We assessed the impact on two mortality tabulation lists: the main (10 leading) tabulation list and the 113-category tabulation list used by the US National Center for Health Statistics. Results: The three leading CODs were the same according to both methods: malignant neoplasm, heart diseases, and cerebrovascular diseases. However, compared with manual ICD-9, automatic ICD-10 recorded 2,544 more deaths for malignant neoplasm, 2,274 more deaths for heart diseases, and 2,144 fewer deaths for cerebrovascular diseases. In all, 1,165 deaths were recorded under the bronchitis, emphysema, and asthma category, which ranked 12th according to manual ICD-9. By contrast, 5,314 deaths were recorded under the chronic lower respiratory disease category, which ranked 7th according to automatic ICD-10. Three CODs had CR > 2: septicemia (2.751), viral hepatitis (2.385), and essential hypertension and hypertensive renal disease (2.102). By contrast, two CODs had CR < 0.7: Alzheimer's disease (0.250) and chronic and unspecified bronchitis (0.638). Conclusions: The results of this bridge coding study indicated large differences between the new and old methods in some COD categories. Two possible reasons are identified for this discrepancy. First, the level and range of codes included in the corresponding tabulation COD categories were different. Second, many medical certifiers recorded more specific COD diagnoses as the second and third diagnoses, which resulted in differences in judgment between coders and the automated system when assigning the UCOD. We recommend three measures when examining cause-specific mortality trends before and after 2008 in Taiwan. First, large differences in the level and range of codes included in corresponding COD categories should be evaluated. Second, the CR should be used for adjustment. Third, multiple COD data should be analyzed as a complement.