Discussion
In this observational study of sepsis spanning from 1 January 2008 to 31 December 2021, we found a significant variation in coding practices among the four regional health authorities, with a 12.9 % absolute difference in the proportion of sepsis codes with a known focus between Western Norway Regional Health Authority and Central Norway Regional Health Authority. The use of codes for known focus was also shown to have increased, while the use of codes with an unknown focus and double registration decreased during the study period.
To the best of our knowledge, the wide variation in the use of medical codes for sepsis between health authorities has not previously been shown in any study, indicating the need for a review of coding practices. Examples of an undesirable variation have, however, been observed previously: Kaspersen et al. found in 2018 that hospitals adhere to both national and local sepsis protocols, and only six hospitals used SOFA in their definition of sepsis (8). Coding practices between clinicians, departments and hospitals varied considerably, and access to clinical coders can vary across health regions (9). Previous studies have shown that clinical coders have been able to provide correct diagnostic codes in 90–95 % of cases, while clinicians have been shown to code correctly in 65–75 % of cases (9). To achieve the most consistent coding between hospitals and health regions, it is important to provide training in diagnostic coding for clinicians and to use clinical coders.
In 2018, Fleischmann-Struzek et al. showed that the use of sepsis codes for infection plus organ failure and unknown focus in Germany increased in the period 2007–13 (10). The opposite trend observed in our study in the use of codes with an unknown focus can be attributed to the more stable incidence of sepsis in Norway during the period 2008–21 compared to that reported in Germany (11, 12). In addition, the coding of acute organ failure is omitted from the coding guidelines' flowchart, leaving room for discretion in coding practices (2). We believe that including coding for acute organ failure in the coding guidelines' flowchart could make it easier for clinicians to adhere to the desired coding practice and help improve the quality of administrative data, thus enhancing the utility value of subsequent use of such data.
In Norwegian coding practices, coding is based on known and unknown focus with corresponding 'R' codes for sepsis, whereas in international literature, known focus is more broadly defined by including the agent of infection, in addition to focus plus organ failure (2, 4). The variation in coding practices complicates comparisons of epidemiological data between countries. It is hoped, however, that implementation of ICD-11, which is intended to facilitate a new classification of sepsis, will pave the way for more homogeneous coding across borders (13).
Admitting and treating patients with suspected sepsis and an unknown focus is a familiar problem for experienced clinicians, and the goal in such cases is to localise the infection. Codes for unknown focus, such as A40 (streptococcal sepsis), mean that the localisation of the focus is not known (2). Best practice should involve localising the infection and coding known focus where possible. Increasing the use of sepsis codes with a known focus will generate more detailed information about the sepsis diagnosis itself, which can be important for further follow-up and rehabilitation.
In this study, we found double registration of known and unknown focus in almost 10 % of admissions, with an annual reduction of 0.9 % between 2008 and 2021. The reduction in double registration and use of codes for unknown focus from 2012 can be explained by the new coding guidelines for sepsis, with the new codes for 'Systemic inflammatory response syndrome' and 'Septic shock', which were published in 2011. In 2021, over 80 % of all sepsis cases were coded with 'known focus', and this trend is contributing to more uniform coding and thus a higher quality of administrative data.
Our findings reveal major variations between the health regions in the coding of sepsis. These are believed to be multifactorial and may be due to unequal access to rapid microbiological diagnostics, geographical variations in primary care services, distance to hospitals, and variations in patient populations in the different health regions. Major variations in coding practices between health regions can lead to over- or under-funding, which indirectly creates geographical disparities in healthcare provision. Given the extent of the variation in our study, we recommend a new national review of sepsis coding.
A weakness of the study is that we used the Sepsis-3 definition as an extraction criterion from the Norwegian Patient Registry, despite this definition not coming into effect until 2016 (1). Sepsis is a heterogeneous syndrome, and in research and quality work in this field, the absence of a uniform method for identifying patients with sepsis is a challenge.
The definition of sepsis has changed over the years. From 1991 to 2015, sepsis was defined as infection plus deviation from at least two normal values measured by systemic inflammatory response syndrome (14, 15). The Sepsis-3 definition did not generate any new ICD-10 codes. However, the code for systemic inflammatory response syndrome after 2016 could be used when a patient's baseline in SOFA increased by at least two points. This change in coding practice may have compromised the comparability of data before and after the Sepsis-3 definition. There is also some uncertainty about the number of acute organ failures, since not all codes for organ failure need to be triggered by an infection, which may have led to an overestimation of sepsis codes with a known focus. The increased emphasis on sepsis in recent years through the Surviving Sepsis Campaign and the 2016–18 report by the Norwegian Board of Health Supervision may also have led to more cases of less severe infection being coded as sepsis, which in turn may have resulted in an overestimation (16, 17) . However, this mechanism, known as the Will Rogers phenomenon (18), is of less significance for this study since the incidence of all sepsis admissions in Norway was stable in the period 2008–21 (12). Furthermore, our dataset only includes information about the four regional health authorities, not the individual hospital trusts or hospitals, a decision made in consultation with the Regional Committees for Medical Research Ethics and the Norwegian Patient Registry for data protection purposes. It is also likely that disparities between health authorities are due to a combination of chance and the variation of patient populations, given the different role of university hospitals compared to local hospitals.
This study has numerous strengths, including our use of data from all public hospitals in Norway over a period of 14 years. Another strength lies in our adherence to the recommendation that a search for codes in a minimum of 15 diagnosis fields is needed to capture all relevant ICD-10 codes (6). Additionally, the validation of our medical codes for identifying sepsis and their use by other researchers further enhances the strength of our study (4, 5, 19). The fact that we adapted the codes to the Norwegian context in collaboration with a clinical research group (Mid-Norway Centre for Sepsis Research) and involved infectious disease specialists, microbiologists, anaesthesiologists and experts in medical coding ensured that the codes reflect actual clinical practice.