The implementation of an automated coding system for cause-of-death statistics

ABSTRACT

Background: The production of cause-of-death statistics requires the coding and selection of an underlying cause of death from death certificates. Nowadays, this is done manually in many countries around the world. However, automated coding systems have been available since the 1970s and more and more countries are switching from manual to automated coding. The introduction of an automated coding system is expected to change the coding process and its outcomes in a fundamental way. Therefore, we studied the implementation of such a system, called IRIS, in the Netherlands.

Methods: We adapted the system to our situation step-by-step and measured the number of death certificates processed without any manual intervention. Medical coders analyzed and qualified death certificates that could not be processed by the system. We also performed a bridge (double) coding study on a set of death certificates so that we could compare the underlying cause of death assigned by IRIS with the underlying cause of death assigned by a medical coder to the same death certificate.

Results: IRIS could handle 68.5% of the death certificates without any manual intervention. Of the (31.5%) rejected death certificates, the main reasons for rejection were the following: editing and/or coding problems (76%), not able to select an underlying cause of death (12%), ambiguous causal connection between codes (5%), a combination of the reasons mentioned (7%). In the bridge coding study, 78% of the death certificates coded by IRIS without any manual intervention showed exactly the same underlying cause of death (ICD-10, four digits) as death certificates coded manually.

Conclusions: An automated coding system for causes of death reduces the workload for medical coders considerably. When editing or coding problems can be solved the system is expected to handle up to 85% of the death certificates without manual intervention. The performance of the system is strongly dependent on the quality of death certificates. A change from manual to automated coding brings about changes in the frequency of occurrence of major causes of death. Users of death statistics should be aware of these changes when studying trends in time or regional variations of causes of death.

KEYWORDS:

• Cause-of-death statistics
• cause of death
• automated coding
• IRIS
• implementation study
• bridge coding study

Summary Table

• The change from manual to automated coding will change the coding process and its outcomes in a fundamental way.

• The number of entries in the dictionary and the power of regular expressions to process medical diagnostic expressions determine the performance of an automated coding system.

• A bridge coding study should be performed to identify changes in frequency of occurrence of causes of death.

• Some of these changes should be accepted with regard to international comparability of data and some should be corrected in order to avoid artifacts of automated coding.

• Users of cause-of-death statistics studying time trends or regional variations should be informed about shifts in statistics by publication of bridge coding studies.

Acknowledgments

The author thanks the four coders of Statistics Netherlands for their participation in this study. The author also thanks the staff members of Health Statistics for their critical comments on earlier drafts of this paper and the IRIS working group for their help in using and implementing IRIS.

Declaration of interest

There was no conflict of interest. IRIS is open source software for coding causes of death. It operates under Windows XP®. The software is available by (free) membership of the IRIS user group, a platform for further development and quality assurance of IRIS (see www.Iris -institute.org).

Notes

¹ For example, cachexia, dehydration, or delirium is often reported as direct cause of death. When these expressions do not fit in a causal chain of morbid events, IRIS selects them as underlying cause of death. Reporting a direct cause of death differs by certifier, region or country. Such bias of selection should be identified and corrected by the user of IRIS.