Accuracy of coded cause of death data: a study based on primary liver cancer

Introduction: Coded cause of death data from death certificates is important for estimating mortality related to cancers and their treatments. However, coding may be inaccurate or complicated by variation in mode of diagnosis (for example, radiology versus pathology, or in methodology between government agencies). This may have important reporting and funding implications.

Aims: (1) To investigate the level of agreement for cause of death data between the Australian Bureau of Statistics (ABS) and Tasmanian Cancer Registry (TCR), and to compare these with independent, blinded medical practitioner opinions on cause of death. (2) To estimate the impact of different coding practices and resulting cause of death data on cause-specific survival.

Methods: Causes of death were compared between the ABS and TCR, and discrepancies were independently reviewed by specialist medical practitioners. Cohen’s Kappa statistics were applied to evaluate the degree of concordance between the ABS, TCR and medical practitioner opinions regarding cause of death. The cumulative incidence function was used to estimate cause-specific survival time in the presence of a competing risk framework according to sex, place of residence, country of birth, and type of PLC.

Results:A minimal level of agreement (Kappa=0.35) was observed when comparing the TCR and ABS cause of death data. Agreement between the TCR and medical practitioners was weak (Kappa=0.51), moderate between the ABS and medical practitioners (Kappa=0.61), strong (Kappa=0.87) between the medical practitioners. These results reflect a greater level of agreement between medical practitioners than between coding agencies. Overall, cause-specific survival time was similar across the TCR, ABS and medical practitioners by sex, place of residence and country of birth, however, regarding type of PLC, only a small difference was observed.

Conclusions: Reporting of mortality data can vary substantially between agencies/institutions. Utilisation of specialist clinician oversight might improve data cohesion and fidelity. Overall, cause-specific survival time was similar across the TCR, ABS and medical practitioners, but a small difference was observed for the type of PLC. As PLC is a low survival cancer, these results may not apply to cancers with better survival such as breast cancer.