Mortality after Severe Sepsis and Septic Shock in Swedish Intensive Care Units 2008‐2016—A nationwide observational study

Recent studies have reported substantially decreased hospital mortality for sepsis, but data are scarcer on outcomes after hospital discharge. We studied mortality up to 1 year in Swedish intensive care unit (ICU) patients with and without sepsis.

in hospital mortality in critically ill patients which kept pace with a similar decrease in nonsepsis patients. 5 While decreasing hospital mortality suggest improved care, the need for data on longer term outcomes were immediately voiced. 10 There are numerous publications with time-fixed long-term outcomes after sepsis and septic shock but there is a scarcity of studies that analyse longitudinal trends of these outcomes. Timefixed mortality mitigates bias associated with hospital mortality, since outcome on discharge from hospital may be affected by mortality/morbidity trade-offs in ICU care and local discharge practices. 11 Verifying a mortality decrease sustained over time would increase understanding of how the epidemiology of severe sepsis has evolved.
Prior studies from the Swedish Intensive Care Registry (SIR) have demonstrated the feasibility to capture long-term outcomes with minimal loss to follow-up. 12,13 Hence, we set out to investigate whether mortality up to one year among critically ill patients with severe sepsis has changed from 2008 to 2016 and also to study the nonsepsis ICU population for comparison. Additionally, we wanted to investigate if patient demographics and comorbidity have changed over the same period, and whether these factors were associated with mortality.

| Ethics approval and consent
The study was approved by the local ethics committee, date of approval 20-01-2015, application no. 2015/519. Informed consent was waived as the data sources operate within the legal framework of Swedish Health Data Registries which allow use of anonymized data without informed consent, and personal identification numbers were blinded.

| Setting, participants and data
Data were retrieved from SIR on all adult general ICU patients (≥18

| Missing data and bias
We excluded 3.3% of patients from the mortality analysis because they were foreign citizens, had concealed personal identity numbers Finally, we performed sensitivity analyses by studying adjusted mortality odds ratios only including patients from units reporting to SIR from the start of the study period, and by repeating analyses without imputation of missing data (complete case analysis).

| RE SULTS
We identified 28 886 sepsis patients and 221 941 nonsepsis patients in the register. A flow chart describing the patient selection in the different analyses is presented in Figure 1.  Table 1. There was a small increase in mean age for both sepsis pa- Age, sex, SAPS3 score, CCI, hospital type (local, county, regional), invasive ventilation, CRRT and type of admission (medical, surgical), together with year of admission, were entered into a multivariable logistic regression model, calculating adjusted odds ratios for 30,90 and 365 day mortality, presented in Tables 2 and 3.

| D ISCUSS I ON
The principal finding in this nationwide observational study of ICU patients with sepsis was that we found no substantial and consist- shock. 19,20 Although several longitudinal studies report late mortality for large sepsis cohorts, most included patients were not ICU patients and mortality rates at time points corresponding to our study were much lower, suggesting that illness severity was lower than in our cohort. 21,22 Sepsis mortality comparable to our results were found in two studies on mostly elderly, non-ICU patients with presumably more co-morbidities in comparison to our septic study cohort. 23,24 This is in line with the finding in this and other studies that, apart from the acute illness requiring intensive care, co-morbidity is an important determinant of longer term mortality. 25 In this context, the decreasing mortality in the subgroup of sepsis patients without chronic comorbidity found in our data could suggest that any improvements in care may have a greater impact in this group. In a recently reported follow-up of a randomized controlled trial (RCT) 6-month mortality was 32% which is lower than the 90-day mortality in our unselected sepsis cohort, a result probably explained by differences in patient selection driven by the inclusion criteria of the RCT. 26 Interestingly, the mortality of the Scandinavian patients in this multicentre trial was higher than that of the patients from Australia and New Zealand.
In contrast with the trend reported in epidemiological studies, our data did not demonstrate a substantially reduced mortality between 2008 and 2016. 5,6,[27][28][29] The reasons for this are not clear.
Hypothetically, there might be a difference between the fixed time mortality assessed in our study, and hospital mortality. However, the hospital mortality in our study was similar to mortality at 30 days and did not decrease significantly during the study period. Early mortality contributed substantially to total mortality, with 10.1% mortality on the first day and 19.9% in the first week after ICU admission in the sepsis cohort.
In a recently published study comparing ICU populations identified by sepsis 2 and sepsis 3 criteria, a decreasing hospital mortality was seen in sepsis without shock, but not in septic shock identified by the sepsis 3 criteria. These findings suggest that the frequency and severity of circulatory dysfunction in the case mix could have a considerable impact on the trend in mortality. 8 Recent data suggest that increased recognition and coding of less severe cases might confound trends in mortality, creating the impression of a decrease. 30  Our reported ICU LoS is shorter than in some other epidemiologic studies, but this parameter is highly variable between different cohorts according to recent sepsis studies, and not necessarily associated with outcome. 5,6,32-36 Furthermore, short ICU stays seem to be a characteristic of Nordic countries, as shown by a joint analysis of intensive care in Sweden, Finland and Norway, and probably reflect a small number of ICU beds per capita. 33,37 Mortality in Swedish ICUs is low and the short ICU LoS will increase demand for a high level of post-ICU care as the outcome will be greatly influenced by this. 33 In some studies patients have been defined as septic or not at ICU admission or within the first 24 hours while in this study any sepsis occurring during the ICU episode was included. 5 This could potentially affect the case mix of the studies. However, regardless of the time of diagnosis, sepsis is a heterogeneous condition and the underlying comorbidity will vary. Also, the short average ICU LoS in this study implies that many patients are diagnosed within a short time frame.
In many epidemiological studies, the outcome measure is hospital mortality, which can be influenced by discharge practices. Our

TA B L E 3
Adjusted odds ratios for 30-90-and 365-day mortality in the nonsepsis population sepsis populations in different health care systems might provide clues to the reasons for the observed differences between recent studies, and facilitate more generalizable benchmarking.

| CON CLUS IONS
In the Swedish ICU population with severe sepsis and septic shock, as in the nonsepsis ICU population, we did not see a major decrease in mortality over recent years.

ACK N OWLED G EM ENTS
We thank Mr Henrik Renlund at Uppsala Clinical Research Centre for statistical support.