Causes of late mortality among ICU‐treated patients with sepsis

Patients with sepsis may have an increased risk of late mortality, but the causes of late death are unclear. This retrospective matched cohort study aimed to determine the causes of late death (≥1 year) among patients with sepsis compared to patients without sepsis.


| BACKG ROU N D
Severe sepsis involves an infection that causes the host's immune response to damage its own tissues and organs, which can lead to organ dysfunction and death in some cases. [1][2][3] Despite improved awareness in recent years, the acute mortality rates remain high for sepsis (15% based on the Sepsis-3 definition) 3 and septic shock (50%). 4,5 There is on-going debate regarding whether the sepsis episode itself contributes to long-term mortality, or if late mortality is primarily influenced by comorbidities that existed before the septic event. 6 The actual causes of late death in patients with a previous sepsis episode are to a large extent unknown.
In a systematic review from 2010, Winters et al concluded that patients with sepsis have increased mortality, even years after their admission for sepsis, 7 although the 1-year mortality rate after discharge varied substantially (7%-43%). In addition, Winters et al noted that many of the studies regarding late mortality after sepsis were of poor quality, with small cohorts and occasionally no control group. A more recent meta-analysis by Shankar-Hari et al revealed that the post-acute mortality rate (difference between cumulative 1-year mortality and acute mortality) was 16.1%. 6 In studies with non-septic control groups, sepsis was not consistently associated with a higher risk of post-acute mortality, and the risk of late mortality among patients with sepsis was greatest when they were compared to the general population. Shankar-Hari et al also noted that many studies regarding long-term mortality after sepsis were of insufficient quality, which raises questions regarding the relationship between sepsis and additional post-acute mortality. Prescott et al attempted to address whether late mortality (31 days to 2 years) after sepsis was driven by pre-existing disease or was the result of sepsis itself. 8 They used matched control groups (non-hospitalized adults, patients admitted with non-septic infections and patients admitted with sterile inflammatory conditions), and found that sepsis was associated with a 22% absolute increase in late mortality relative to non-hospitalized adults. This increase in late mortality was also observed, albeit at lower magnitudes, when the other control groups were used. In contrast, a recent study focused on ICU-treated critically ill could not confirm an increase in late deaths after sepsis. 9 Many of the studies regarding late mortality in sepsis are register-based. This implies some general limitations, for example potential lack of validation for registration of diagnoses, and lack of data.
On the other hand, register-based studies may often present results that is necessary for the design of prospective studies.

More information about actual causes of late death in patients
with previous episodes of sepsis may in the long-run result in better post-ICU care for this group of patients. It may also generate new hypotheses regarding longstanding physiological or immunological alterations in patients with sepsis.
In summary, the results of studies regarding late mortality in sepsis are disparate. With that in mind, information about the actual causes of death after the initial sepsis event could provide additional insight in this field. Therefore, the present study was undertaken to determine the causes of late deaths in sepsis patients treated in the ICU compared to a matched cohort of non-sepsis ICU patients with similar disease severity.

| ME THODS
The present cohort study identified patients in the Swedish Intensive care Registry (SIR) database who were treated in an ICU All Swedish residents have a personal identification number, 10,11 and this number was used to link the SIR data to the Swedish Cause of Death Register, which is administered by the Swedish National Board of Health and Welfare. 12 The data linking process was per- The primary outcomes of interest were the primary and secondary causes of death at ≥365 days after the initial ICU admission (day 0). The secondary outcomes were mortality rates (in the ICU and at 30 days, 90 days, 180 days and 365 days), as well as causes of death at ≤365 days after the initial ICU admission. The study's protocol was approved by the Regional Ethics Review Board in Linköping, Sweden (2014/31-31 and 2017/238-32), and the requirement for informed consent was waived.

Editorial Comment
In this nationwide cohort study, causes of death of ICU patients with and without sepsis were characterized. The most common causes of death in general were heart disease and cancer, however, in patients with sepsis a predominant cause of late death was infectious disease.

| Statistical analysis
Demographic data were reported as mean (standard deviation) or median (95% confidence intervals). All statistical analyses were performed using STATA SE software (version 14; StataCorp).

| RE SULTS
During 2008-2013, the SIR database included 15 141 patients with severe sepsis or septic shock at their first admission. These patients were matched 1:1 to non-septic cases according to age, gender, length of ICU stay and SAPS3 probability for death. This produced 9,144 matched pairs of patients with and without sepsis, as well as 5997 patients with sepsis but no appropriate matching control patient in the SIR database. Compared to the matched patients with sepsis, the unmatched patients generally had higher SAPS3 probabilities for death (55% vs 45%) and longer ICU stays (6 days vs 1 day) ( Table 1). We excluded 384 matched pairs because of missing follow-up mortality data, and 8,760 matched pairs were ultimately included in the analyses ( Figure 1). Most included patients were male (58.3%) and the mean age was 68.8 years (14.4 years) ( Table 1). The   Analysing more recent deaths after the initial admission revealed several significant differences between the groups. When we considered mortality within 30 days, we found that cancer was significantly more common in the sepsis group, while heart diseases were more common in the control group (Table 3). In addition, the 30day and 180-day mortality rates were slightly higher in the control group (Table 4) (Table 5). In patients with infectious disesases as a cause of late death, pneumonia was the most common diagnose (Table 5).

| D ISCUSS I ON
The present study of critically ill ICU patients revealed that infectious disease as a cause of death were more common in initial survivors of sepsis than in a non-sepsis control group. To the best of our TA B L E 1 Factors used in the matching procedure  Note: Data are shown as percentage, mean (standard deviation) or number unless otherwise specified.
Abbreviations: LOS, length of stay; SAPS3, Simplified Acute Physiology Score version 3. knowledge, this is the first report to describe the causes of death among ICU patients who survived their initial septic event.

F I G U R E 1 Study flow chart
The most common causes of death in the general Swedish population involve circulatory diseases (including heart diseases) and cancer. 13 We observed similar results in this patient population, with heart diseases and cancer being the most common causes of death in both study groups. However, we also found that infectious diseases were more common causes of death in the sepsis group, compared to the control group. As expected, this difference was most apparent when analysing the most immediate deaths after the ICU event (0-30 days after their initial event). However, the difference between the groups persisted and remained when we examined deaths at ≥1 year after the patient's initial admission.
Further analysis of infectious disease-related deaths revealed that pneumonia was a common cause of death in both groups. In general, pneumonia is a common cause of sepsis, 14 but it is also a common complication among patients with COPD. 15,16 The prevalence of COPD was higher in our control group compared to the sepsis group, which may contribute to a high incidence of pneumonia in the control group. However, codes for sepsis as a cause of death were much more common in the sepsis group, even at  19 19 Intestinal infection (A04) 8 4 Chronic hepatitis (B18) 6 5 However, we were able to reduce the probability of inter-group differences in mortality by using SAPS3 probability for death as a matching variable, which is an effective approach because the SAPS3 score predicts mortality in ICU patients. 9, 19 We also believe that using severely ill ICU-treated patients as the control group is a strength of the present study, because this approach allowed us to examine differences in the causes of death that were specific to the sepsis itself.
Nevertheless, it is always difficult to design an adequate control group and we cannot exclude the possibility of unconsidered confounding of the matching process. Despite using SAPS3 score, which includes some comorbidities but not all, in the matching procedure, one confounding variable may be pre-existing diseases and comorbidities. Unfortunately, we were not able to include more detailed data on comorbidities, which is a limitation of this study.
Another obvious limitation in our study is that many of the sep- When discussing proper diagnosis recording, it is also important to remember that sepsis is a complex clinical condition that can mirror a broad range of other diagnoses. There can also be a subtle difference between severe sepsis and infections without sepsis. For example, a study of ICD coding accuracy using a Swedish register of national hospital admissions (the Swedish National Patient Register) revealed that ICD codes at discharge had a high positive predictive value but a low sensitivity, compared to patient chart reviews. 23 Furthermore, we have observed similar results, as only 55% of patients discharged after ICU treatment for sepsis were assigned ICD codes indicating sepsis. 24 Thus, this limited accuracy may also apply to the Cause of Death Register, which relies on retrospective ICD coding.
In summary, our results indicate that infectious disease were a more frequent cause of death in patients previously treated for sepsis compared to the matched control group. The study has some limitations, as detailed above. Many of the limitations are general to register-based research, and the results of our study may be further investigated in prospective studies. The explanation for the finding that infectious diseases are more common as a cause of death in patients with a previous ICU-treated sepsis episode compared to ICU-treated patients without sepsis is unclear.
Hypothetically, patients treated for sepsis may have persisting alterations in the immune response, warranting studies with longer follow-up periods as well as studies into potential underlying immunological mechanisms.

| CON CLUS IONS
The most common causes of late mortality among ICU-treated Swedish patients with and without sepsis were heart diseases and cancer. However, infectious diseases were a more frequent cause of death in the sepsis group, compared to the control group.