Number of ICD-10 diagnosis fields required to capture sepsis in administrative data and truncation bias: A nationwide prospective registry study.

BACKGROUND

In observational studies that use administrative data, it is essential to report technical details such as the number of International Classification of Disease (ICD) coding fields extracted. This information is crucial for ensuring comparability between studies and for avoiding truncation bias in estimates, particularly for complex conditions like sepsis. Specific sepsis codes (explicit sepsis) are suggested to be identified by extracting 15 diagnosis fields, while for implicit sepsis, which comprises an infection code combined with acute organ failure, the number of diagnosis field remains unknown.

OBJECTIVE

The objective was to explore the necessary number of diagnosis fields to capture explicit and implicit sepsis.

MATERIALS AND METHODS

We conducted a study utilizing The Norwegian Patient Register (NPR), which encompasses all medical ICD-10 codes from specialized health services in Norway. Data were extracted for all adult patients with hospital discharges registered with explicit and implicit sepsis codes from all Norwegian hospitals between 2008 through 2021.

RESULTS

Out of 317,705 sepsis admissions, we identified 105,499 ICD-10 codes for explicit sepsis, while implicit sepsis was identified through 270,346 codes for infection in combination with 240,789 codes for acute organ failure. Through our analysis, we found that 55%, 37%, and 10% of the explicit, infection, and acute organ failure codes, respectively, were documented as the main diagnosis. The proportion of explicit and infection codes peaked in the primary diagnosis field, while for acute organ failure codes, this was true in the third secondary diagnosis field. Notably, the cumulative proportion reached 99% in diagnosis field 10 for explicit codes and in diagnosis field 13 for implicit codes.

CONCLUSION

Expanding the utilization of multiple diagnosis fields can enhance the comparability of data in epidemiological studies, both internationally and within countries. To make truncation bias visible, reporting guidelines should specify the number of diagnosis fields when extracting ICD-10 codes.