Long-Term Survival Following Sepsis

Abstract

Background

There have not yet been any prospective registry studies in Germany with active investigation of the long-term survival of patients with sepsis.

Methods

The Jena Sepsis Registry (JSR) included all patients with a diagnosis of sepsis in the four intensive care units of Jena University Hospital from January 2011 to December 2015. Long-term survival 6–48 months after diagnosis was documented by asking the treating general practitioners. The survival times were studied with Kaplan-Meier estimators. Cox regressions were calculated to show associations between possible predictors and survival time.

Results

1975 patients with sepsis or septic shock were included. The mean time of observation was 730 days. For 96.4% of the queries to the general practitioners, information on long-term survival was available. Mortality in the intensive care unit was 34% (95% confidence interval [32; 37]), and in-hospital mortality was 45% [42; 47]. The overall mortality six months after diagnosis was 59% [57; 62], the overall mortality 48 months after diagnosis was 74% [72; 78]. Predictors of shorter survival were age, nosocomial origin of sepsis, diabetes, cerebrovascular disease, duration of stay in the intensive care unit, and renal replacement therapy.

Conclusion

The nearly 75% mortality four years after diagnosis indicates that changes are needed both in the acute treatment of patients with sepsis and in their multi-sector long-term care. The applicability of these findings may be limited by their having been obtained in a single center.

Sepsis is defined as life-threatening organ dysfunction caused by a dysregulated host response to infection (1). The clinical diagnosis is made in the presence of infection-related dysfunction at least one of the following: lung, kidney, liver, cardiovascular system, blood count, or CNS (etable 1). Based on an analysis of national figures for ICD-10-GM-coded hospital discharge diagnoses, 136 542 people in Germany suffered from (severe) sepsis during 2015 (according to the “Sepsis-1 definition” [2], eTable 1). This corresponds to 158 cases of (severe) sepsis per 100 000 inhabitants, and 41.7 % of the patients involved died in hospital (3). Worldwide, the estimated figure for 2017 is 49 million patients affected—significantly more than previously estimated (4). About 11 million patients do not survive (5). However, due to major international differences in coding and documentation, the state of the data is uncertain. For the purposes of monitoring epidemiological indicators therefore, Rhee et al. (6) recommend the identification of clinical cases from patient records.

eTable 1. Criteria for the definitions of sepsis, from Weis 2017 (e3).

Sepsis-1 (1991)	Sepsis-2 (2001) (e4)	Sepsis-3 (2016)
>2 of 4 SIRS criteria ▪ Hypo-/hyperthermia ▪ Tachycardia ▪ Tachypnea ▪ Leukocytosis/leukopenia	“Some” of the following criteria: ▪ Hypo-/hyperthermia ▪ Tachycardia ▪ Hypotension ▪ Tachypnea ▪ SvO 2 < 70 % ▪ Cardiac index >3.5 L/min/m 2 ▪ Hyperglycemia ▪ Altered consciousness ▪ Marked edema/positive fluid balance ▪ Leukocytosis/leukopenia/ > 10% left shift ▪ CRP or PCT >2 SD above normal ▪ Increase in lactate/capillary refill time	At least one infection-related organ dysfunction, indicated by a change in SOFA Score by >2: ▪ Arterial hypoxemia ▪ Increased creatinine ▪ Hyperbilirubinemia ▪ Thrombocytopenia ▪ Hypotension/administration of vasopressors ▪ Obtundation (GCS) Quick SOFA score for risk assessment if: ▪ Altered consciousness ▪ Tachypnea ▪ Hypotension
Severe sepsis: Additionally, at least one infection-related organ dysfunction	Severe sepsis: Additionally, at least one infection-related organ dysfunction: ▪ Arterial hypoxemia ▪ Acute oliguria ▪ Increased creatinine ▪ Coagulopathy ▪ Ileus ▪ Thrombocytopenia ▪ Hyperbilirubinemia ▪ Serum lactate >1 mmol/L
Shock: Additionally: circulatory collapse that does not respond to fluid resuscitation	Shock: Additionally: Hypotension/circulatory collapse that does not respond to fluid resuscitation	Shock: Additionally: ▪ circulatory collapse, MAD >65 mmHg only with vasopressors ▪ Serum lactate >2 mmol/L
Presumed or proven infection

CRP, C-reactive protein; MAD, mean arterial pressure; PCT, procalcitonin; SD, standard deviation; SIRS, systemic inflammatory response syndrome; SOFA score, sepsis-related organ failure assessment score; SvO2, mixed venous oxygen saturation

Owing to a multiplicity of physical and psychological sequelae, the disease burden of sepsis survivors is not restricted to the period of acute care in hospital (7). Although the direct impact on long-term survival is currently a matter of debate (7, 8), the post-acute phase of sepsis is attracting increasing attention. To date, only a small amount of data on long-term survival after sepsis is available in Germany, based on small study populations (9, 10) or limited to 90-day mortality, as in a quality management program of the University Hospital Greifswald (11).

The Jena Sepsis Registry (JSR) at Jena University Hospital, a tertiary care center for about one million people in the state of Thuringia, was founded to meet the need for both clinical case identification and reliable capture of data on long-term survival. Since (like 90% of the population [12]) most patients who survive sepsis are treated by their family doctors, the latter were chosen as the source of information on long-term survival.

The primary goal of the JSR was the systematic registration of sepsis patients in the intensive care units of a German university hospital, collecting data on clinical information, care parameters, in-hospital mortality, and long-term survival up to 48 months. In addition, the intention was to identify potential predictors of long-term survival on the basis of these data.

Method

In accordance with the definition of a clinical registry (13), all patients over 18 years of age who were treated in one of the four surgical and medical intensive care units of Jena University Hospital between 1 January 2011 and 15 December 2015 were included (in 2015, there were 72 beds with 3835 cases of treatment in these ICUs). The inclusion criterion was “severe sepsis” or septic shock as defined by the first Sepsis Consensus Conference of 1992 (“sepsis-1”) (2) (see eTable 1). Screening for signs of infection and organ dysfunction was done manually by four trained study assistants under physician supervision, based on a patient data management system (PDMS). There were no exclusion criteria. The web-based open-source software OpenClinica (14) was used for documentation. The documented baseline set of data at the time of diagnosis included infection site, APACHE II (15) and SOFA (16) scores, comorbidities, and subsequent microbiological pathogen identification (Table 1a, eTable 2). For patients discharged from hospital, mode of transfer, therapeutic and diagnostic procedures performed, and the number of treatment and ventilation days were also collected (table 1b).

Table 1a. Infectiological characteristics of the study population.

	Total (N = 1 975)
Site of infection
Pneumonia (n, %)	881 (44.6)
Respiratory tract (other) (n, %)	156 (7.9)
Primary bacteremia (n, %)	85 (4.3)
Cardiovascular (n, %)	123 (6.2)
Gastrointestinal/intra-abdominal (n, %)	546 (27.7)
Urogenital (n, %)	186 (9.4)
Surgical wound infection (n, %)	53 (2.7)
Bone/soft tissue (n, %)	126 (6.4)
Other (central nervous system, thoracic, catheter infections, etc.) (n, %)	160 (8.1)
Infection source
Community-acquired (n, %)	635 (32.2)
Hospital-acquired (n, %)	1340 (67.9)
Microbiologically identified (n, %)	1404 (67.1)
Clinical evidence (n, %)	571 (28.9)
Pathogen spectrum
Gram-positive (n, %)	957 (48.5)
Gram-negative (n, %)	779 (39.4)
Fungi (n, %)	447 (22.6)
Anaerobes (n, %)	56 (2.9)
Viruses (n, %)	36 (1.8)
Parasites (n, %)	1 (0.05)

eTable 2. Pathogens identified by microbiological study during the hospital stay of n = 1403 patients with available findings; multiple entries possible*.

Detected pathogens	Number (%)
Gram-positive	957 (48.46)
Staphylococcus aureus	224 (11.34)
Methicillin-resistant Staphylococcus aureus	49 (2.48)
Coagulase-negative staphylococci	191 (9.67)
Methicillin-resistant coagulase-negative staphylococci	200 (10.13)
Group D streptococci	375 (18.99)
Group A, B, C, or G streptococci	69 (3.49)
Streptococcus pneumoniae	38 (1.92)
Other streptococci	124 (6.28)
Other gram-positive cocci	56 (2.84)
Listeria monocytogenes	1 (0.05)
Other aerobic, gram-positive bacteria	51 (2.58)
Vancomycin-resistant enterococci	13 (0.66)
Gram-negative	779 (39.44)
Escherichia coli	390 (19.75)
Enterobacter spp.	99 (5.01)
Klebsiella spp.	165 (8.35)
Proteus spp.	83 (4.20)
Neisseria meningitides	0 (0.00)
Moraxella spp.	1 (0.05)
Neisseria gonorrhoeae	0 (0.00)
Salmonella spp.	3 (0.15)
Serratia spp.	50 (2.53)
Citrobacter spp.	29 (1.47)
Pseudomonas aeruginosa	132 (6.68)
Other pseudomonads	9 (0.46)
Stenotrophomonas maltophilia	39 (1.97)
Acinetobacter	29 (1.47)
Campylobacter spp.	1 (0.05)
Brucella spp.	1 (0.05)
Haemophilus spp.	29 (1.47)
Other enterobacteria	3 (0.15)
Other	58 (2.94)
Fungi	447 (22.63)
Candida albicans	282 (14.28)
Other Candida species	185 (9.37)
Aspergillus spp.	23 (1.16)
Other	51 (2.58)
Other	97 (4.91)
Anaerobes	56 (2.84)
Other micro-organisms	4 (0.20)
Viruses	36 (1.82)
Parasites	1 (0.05)
No findings available	572 (28.96)

*Percentages are based on the total sample of n = 1975 patients.

Table 1b. Clinical characteristics of the study population.

	Total (N = 1 975)
Age, years (mean, median, IQR)	66.4; 69 (58–76)
Male sex (n, %)	1287 (65.2)
APACHE II score (mean, SD)	23.6 (6.9)
Maximum SOFA score (mean, SD)	8.9 (3.5)
Septic shock, sepsis-1 definition (n, %)	1469 (74.4)
Septic shock, sepsis-3 definition (n, %)	959 (48.6)*
Pre-existing disease
Diabetes (n, %)	699 (35.4)
Heart failure – NYHA IV (n, %)	632 (32.0)
Cerebrovascular disease (n, %)	368 (18.6)
Renal failure (n, %)	559 (28.3)
COPD (n, %)	300 (15.2)
Liver cirrhosis (n, %)	99 (5.0)
Tumor (n, %)	145 (7.3)
Immune suppression (n, %)	219 (11.1)
Number of diagnoses (mean, median, IQR)	23.2; 22 (17– 28)
Intensive care
Length of stay in intensive care unit (days: mean, median, IQR)	15.9; 10 (5– 23)
Mechanical ventilation (n, %) (mean days, IQR) (mean hours, IQR)	1785 (90.4) 10 (4– 22) 250.8 (34.3–381.3)
Renal replacement therapy (n, %) (mean days, IQR) (mean hours, IQR)	828 (41.9) 11.2 (3– 16) 143.6 (27.2–189)
Overall length of hospital stay (days: mean, median, IQR)	33.2; 28 (15– 43)
Status post elective surgery (n, %)	236 (12)
Status post emergency surgery (n, %)	486 (24.6)
No surgery (n, %)	1253 (63.5)

* In 154 patients, arterial lactate values required for classification were not available.

APACHE, acute physiology and chronic health evaluation; COPD, chronic obstructive lung disease; IQR, interquartile range; NYHA, New York Heart Association; SD, standard deviation; SOFA, sepsis-related organ failure assessment

For patients with septic shock by the sepsis-1 definition, the maximum blood lactate value within the 24 hours after diagnosis was post-documented from the PDMS, enabling comparison for this subgroup with the current, sepsis-3 definition (1) (etable 1). It should be noted that the current sepsis-3 definition has dispensed with the term “severe sepsis,” replacing it with “sepsis.” In this article, we therefore use the form “(severe) sepsis.”

Information on long-term survival at 6, 12, 24, 36, and 48 months after diagnosis was collected by trained study assistants who contacted the treating primary care physicians by telephone. In addition to the date of death, the place of death was also recorded (see eMethod).

The first part of the analysis was descriptive (Table 1a, Table 1b, eTable 2). Survival times were presented using Kaplan-Meier estimators, as recommended by Pocock et al. (17) (figure). Cox regressions were calculated to show associations between possible predictors and survival time (for details see eMethod).

Figure.

Long-term survival of 1975 patients with (severe) sepsis and septic shock treated in intensive care. Kaplan–Meier estimators for the total sample (A) and stratified by sex (B), age (C), and APACHE II score (D) (tertiles in each case) are shown.

Results

Between January 2011 and December 2015, 1975 patients with (severe) sepsis or septic shock were identified and included (median annual number of cases: n = 388). Median age was 66 years (interquartile range [IQR] 58–76), and 65.2% of patients were male. In all, 90.4% were mechanically ventilated, with a median duration of ventilation of 250.8 hours (IQR 34.3–381.3). Further characteristics can be found in Table 1a and Table 1b, and information on microbiological findings in eTable 2.

For data collection on long-term survival, 2600 contacts with primary care physicians were documented, some of them after multiple attempts, at the relevant follow-up timepoints between 6 and 48 months after diagnosis. In 2507 of these contacts (96.4%), information on survival was obtained; 93 contacts (3.6%) brought no result. In these cases, no information was available to the primary care physician—usually because patients had either changed their doctor or moved away.

Mortality was 34.3% in the intensive care unit (ICU) (95% confidence interval: [32.2; 36.4]) and 44.7% in hospital [42.5; 46.9]. The criteria for septic shock by the sepsis-3 definition applied to a smaller proportion of patients (48.6%) than the criteria by the sepsis-1 definition (74.4%). The group fulfilling the sepsis-3 definition showed higher in-hospital mortality (etable 3).

eTable 3. Cumulative mortality in (severe) sepsis and septic shock for the sepsis-1 and sepsis-3 definitions of septic shock*.

Mortality	(Severe) sepsis	Septic shock (sepsis-1 def.)	Septic shock (sepsis-3 def.)	Total
JSR 2011–2015	n = 507	n = 1468	n = 959	N = 1975
ICU	122 (24.1%)	554 (37.7%)	399 (41.6%)	676 (34.3%)
Hospital	179 (35.3%)	703 (47.9%)	499 (52%)	882 (44.7%)
6 months	252 (49.7%)	922 (62.8%)	621 (64.7%)	1174 (59.4%)
12 months	280 (55.2%)	980 (66.8%)	660 (68.6%)	1259 (63.7%)
24 months	298 (58.8%)	1050 (71.5%)	703 (73.3%)	1347 (68.2%)
36 months	321 (63.3%)	1087 (74%)	727 (75.8%)	1407 (71.2%)
48 months	341 (67.3%)	1130 (77%)	747 (77.8%)	1470 (74.4%)
INSEP 2013 (18)	n = 218	n = 1 285	n = 848	N = 1 503
ICU	36/215 (16.7%)	473/1268 (37.3%)	371/838 (44.3%)	509/1483 (34.3%)
Hospital	49/209 (23.4%)	530/1224 (43.3%)	412/810 (50.9%)	579/1433 (40.4%)
DRG-ICD-10 2013 (22)	n = 81 606 (R65.1)	n = 33 815 (R57.2)	−	N = 115 421
Hospital	30 458 (37.3%)	19 891 (58.8%)	−	50 349 (43.6%)
Greifswald University2008–2013 (11)	n = 317	n = 852	−	N = 1169
ICU	63 (19.9%)	296 (34.7%)	−	359 (30.7%)
Hospital	90 (28.4%)	355 (41.6%)	−	445 (38.1%)
3 months	119 (37.5%)	407 (47.8%)	−	526 (45%)

*?IThe German DRG evaluation (22) also included patients who were not treated in intensive care. However, the overall mortality differs only slightly from the group of patients treated exclusively with intensive care (45.2% of N = 73 419). The results from Greifswald University Hospital refer to the period after implementation of a quality management program (11).

DRG, diagnosis-related groups; ICD, International Classification of Disease; ICU, intensive care unit; JSR, Jena Sepsis Registry

Of the n = 1093 patients who survived their hospital stay, 38.9% were discharged home, 32.3% to another hospital, 23% to a rehabilitation facility, and 5.8% to a nursing home. For n = 327 of the patients who died after discharge, the place of death was known: 200 (61.2%) died in hospital, 49 (15.0%) in a nursing home, 47 (14.4%) at home, and 31 (9.4%) in a rehabilitation facility.

The Figure shows survival time curves over the 4 years from the time of diagnosis; eTable 3 shows cumulative mortality over the long term. Median follow-up was 730 days (= 24 months). After 70 days [55 days; 88 days], half of the patients had died (median survival time). One year after diagnosis the survival rate was 36% [34; 38], and after 4 years it was 25% [22; 28]. The Figure shows in addition how survival varies depending on key strata by sex, age tertile, and APACHE II score (as a measure of severity of disease at inclusion). A more detailed account of potential predictors of long-term survival identified in univariate and multiple Cox regression models is presented in Table 2. In multiple Cox regression, the following were found to be predictors of reduced long-term survival: age; hospital-acquired infection (infection source); diabetes and cerebrovascular disease (comorbidities); and length of stay in the ICU and renal replacement therapy (care parameters). In contrast, the classification “Reason for ICU admission: emergency surgery” was associated with improved long-term survival (all p ≤ 0.05). Associations with clinical variables at study inclusion for the entire patient population are shown in eTable 4. Among others, variables relating to sepsis severity stand out as predictors of acute death.

eTable 4. Associations between hazard ratio (HR; instantaneous mortality rate) and clinical variables at the time of study inclusion in the univariate and multiple Cox regression models*.

Variable	Univariate Cox model		Multiple Cox model
	HR [95% CI]	P	HR [95% CI]	P
Demographic variables
Age (per year)	1.02 [1.01; 1.02]	<0.001	1.02 [1.01; 1.02]	<0.001
Sex (male) (female, reference)	0.97 [0.86; 1.08] 1.00	0.58
Sepsis severity
APACHE II score (per point)	1.05 [1.04; 1.06]	<0.001	1.03 [1.02; 1.04]	<0.001
SOFA score (per point)	1.09 [1.08; 1.11]	<0.001	1.06 [1.04; 1.08]	<0.001
Septic shock (sepsis-1) (y/n)	1.43 [1.26; 1.63]	<0.001
Septic shock (sepsis-3) (y/n)	1.44 [1.29; 1.61]	<0.001	1.23 [1.10; 1.39]	<0.001
Comorbidities
Diabetes (y/n)	1.05 [0.94; 1.18]	0.38
Heart failure (y/n)	0.87 [0.62; 1.21]	0.40
Cerebrovascular disease (y/n)	0.95 [0.83; 1.1]	0.51
Renal failure (y/n)	1.25 [0.99; 1.57]	0.06	1.01 [0.79; 1.28]	0.97
Chronic obstructive lung disease (y/n)	1.07 [0.92; 1.24]	0.36
Liver cirrhosis (y/n)	1.6 [1.33; 1.93]	<0.001	1.61 [1.31; 1.98]	<0.001
Tumor (y/n)	1.46 [1.21; 1.76]	<0.001	1.33 [1.08; 1.64]	<0.01
Immune suppression (y/n)	1.05 [0.86; 1.28]	0.63
Site of infection
Cardiovascular infection (y/n)	1.18 [0.95; 1.47]	0.14
Pneumonia (y/n)	1.13 [1.01; 1.26]	0.03	1.04 [0.93; 1.17]	0.47
Other respiratory tract infection f(y/n)	1.08 [0.89; 1.3]	0.46
Intra-abdominal/gastrointestinal infection (y/n)	1.05 [0.93; 1.18]	0.43
Bone/soft tissue infection (y/n)	0.84 [0.67; 1.06]	0.14
Surgical wound infection (y/n)	1.06 [0.77; 1.46]	0.72
Primary bacteremia (y/n)	0.88 [0.67; 1.15]	0.35
Urogenital infection (y/n)	0.87 [0.72; 1.06]	0.16
Other infection (y/n)	0.96 [0.79; 1.16]	0.65
Infection source
Infection source (hospital-acquired) (community-acquired, reference)	1.34 [1.19; 1.51] 1.00	<0.001	1.30 [1.15; 1.48]	<0.001
Evidence of infection (microbiologically confirmed) (clinical evidence, reference)	1.09 [0.97; 1.23] 1.00	0.16
Pathogen spectrum (gram-negative) (gram-positive) (both gram-negative and gram-positive) (unknown or other, reference)	1.08 [0.92; 1.27] 1.10 [0.92; 1.27] 1.03 [0.88; 1.20] 1.00	0.33 0.19 0.74
Reason for ICU admission (elective procedure) (emergency surgery) (nonsurgical emergency, reference)	1.02 [0.86; 1.21] 0.94 [0.82; 1.07] 1.00	0.82 0.36

* Hazard ratios > 1 express a positive association with mortality rate, values < 1 a negative association. Reference group is all N = 1975 patients included in the study. For further details see eMethods.

95% CI, 95% confidence interval; APACHE, acute physiology and chronic health evaluation; ICU, intensive care unit; SOFA, sepsis-related organ failure assessment

Discussion

This paper describes the systematic and prospective documentation of almost 2000 patients treated in intensive care units for (severe) sepsis or septic shock, with up to 4 years’ follow-up. High acute and long-term mortality were revealed: 44.6% of sepsis patients died during their stay in hospital and another 19.1% within the first year after diagnosis. Median survival after sepsis diagnosis was 70 days.

The age and sex distribution of the patients included in the JSR are within the range of large population-based studies of sepsis patients in Germany (3, 18) and the USA (6). Against the background of the increased mortality rates when the new sepsis-3 definition of septic shock is applied, this new definition proved to be highly selective, confirming findings in the recent literature (18– 20). As to the effect of the sepsis definition employed on (severe) sepsis, no conclusion can be drawn on the basis of our data. However, based on the results of a large British comparative study (21), we assume that application of the sepsis-3 definition would not have led to significant change, because the patient overlap is very large.

In-hospital and ICU mortality rates are comparable with results from national diagnosis-related group (DRG) analysis (22) and a German multicenter point prevalence study (19) (etable 3). However, the German estimates are significantly higher than those reported in large retrospective cohort studies of >10 000 patients in Australia/New Zealand and the United Kingdom, which give ICU and in-hospital mortality rates of 16% and 24% (23) and 24% and 32% (for 2012) (24), respectively.

The JSR also showed a lower long-term survival rate than the few international prospective long-term studies of sepsis patients treated in intensive care: The multicenter Finnsepsis study reported a 2-year mortality of 44.9% (25), and another multicenter study, from Scotland, reported a mortality rate of 61% for the 5-year period after diagnosis (26). In a single-center Canadian study, by the end of 10 years 67% of patients had died (27) (etable 5).

eTable 5. Nonsystematic review of international studies on long-term mortality after sepsis. Studies with N = 50 patients or more and >1 year of follow-up are included. Due to widely differing methodologies, documentation quality, and populations studied, results are not comparable between studies.

Study, database	Sample size	Population/definition	Country	Design	Mortality after…
					1 year	2/3 years	4+ years
Jena Sepsis Registry mixed ICU 2010–2015	N = 1976	Severe sepsis/ sept. shock (sepsis-1)	D	Single-center, prospective	63.7%	68.2% (2 yr) 71.2% (3 yr)	74.4% (4 yr)
Linder 2014 (27) mixed ICU 2000–2004	N = 1030	Severe sepsis/ sept. shock (sepsis-2)	CAN		36.5%		50.5% (5 yr) 67% (10 yr)
Francisco 2018 (e5) mixed ICU 2008/09	N = 383	“Severe infection”	P				37% (5 yr)
Davis 2014 (38) tertiary hospital 2007–2008	N = 228	Severe sepsis (sepsis-1)	AUS				33.7% (5 yr)
Storgaard 2013 (e6) tertiary hospital 2004–2007	N = 211	Severe sepsis/sept. shock	DK		41.7%		57.4% (4 yr)
Haraldsen 2003 (e7) surgical ICU 1983–1995	N = 210	“Abdominal sepsis”	S				49.5% (6 yr)
Koch 2013 (9) medical ICU(Aachen)	N = 164	Severe sepsis/sept. shock (sepsis-1)	D			53.7% (3 yr)
Korosec 2006 (e8) surgical ICU 2003	N = 66		SLO			67% (2 yr)
Karlsson 2009 (25) 24 ICUs 2004–2005	N = 470	Severe sepsis/sept. shock (sepsis-1)	FIN	Multicenter, prospective	40.9%	44.9% (2 yr)
Cuthbertson 2013 (26) 26 ICUs in Scotland 2003	N = 439	Severe sepsis (sepsis-1)	UK			58% (3.5 yr)	61% (5 yr)
Shankar-Hari 2019 (29) 192 ICUs in England	N = 94 748*4	Sepsis/sept. shock (sepsis-3)	UK	Multicenter, retrospective, clinical documentation	14.6%*1	20%*1(2 yr)/ 23.3%*1(3 yr)	25%*1 (4 yr) 25.7%*1 (5 yr)
Angus 2004 (e9) PROWESS 1998–2000	N = 1220	Severe sepsis (sepsis-1)	*3			49.5% (2.5 yr)
Henriksen 2016 (e10) AMAU 2010–2011	N = 1092	Severe sepsis/sept. shock (sepsis-1)	DK	Single-center, retrospective, clin. document.	38.7%	46.9% (2 yr) 51.1% (3 yr)
Iwashyna 2012 (e11) Medicare 2003–2008	N = 657 766	ICD-9-CM >65 years	USA	Retrospective, health insurance claims data		71.3% (3 yr)	81.6% (5 yr)
Rahmel 2020 (e12) “Knappschaf”t health insurance 2009–2016	N = 55 525*4	ICD-10-GM	D				61.8%*4 (5 yr)
Ou 2016 (32) NHIRD 2000–2002	N = 67 926	ICD-9-CM	TWN		23%		43.5% (5 yr)
Weycker 2003 (e13) Protocare 1991–2000	N = 16 019		USA		51.4%		74.2% (5 yr)
Shen 2016 (e14) NHIRD 2000–2008	N = 10 818*2		TWN				26.5% (8 yr)
Yende 2014 (e15) Medicare 2002–2006	N = 4179		USA		40.8%	51.2% (2 yr) 58.9% (3 yr)
Lemay 2014 (e16) Veterans 2002–2007	N = 2727	ICD-9-CM > 65 years	USA		31%	43% (2 yr)
Quartin 1997 (e17) Veterans 1983–1986	N = 1505	SIRS + bacteremia/suspected infection	USA		62.3%		82% (8 yr)
Wang 2014 (e1) REGARDS 2003–2007	N = 975	Infection + > 2 SIRS criteria	USA		23%	28.8% (2 yr)	43.8% (5 yr)
Prescott 2016 (37) Medicare 1998–2008	N = 960	ICD-9-CM	USA		48.5%	56.5% (2 yr)

*¹ Mortality after discharge from the ICU; to obtain comparable values with the other studies, a hospital mortality of 34.6% can be added as an approximation; however, this does not take into account n = 919 patients without data

*² Patients who survived the first 3 months after diagnosis

*³ Multicenter study in accordance with the German Drug Law in 11 countries

*⁴ Patients who survived after discharge from ICU

ICU, intensive care unit; NHIRD, National Health Insurance Research Database (Taiwan); SIRS, systemic inflammatory response syndrome

Even if one looks at just studies with similar design (prospective inclusion of sepsis patients treated in intensive care), comparability both with the JSR and with each other is limited by several factors:

Case identification: The retrospective case identification used in large studies based on routine data is only comparable in a limited way to prospective clinical patient enrollment because of the limited validity of sepsis coding: Studies in the United States by Rhee et al. and Gaieski et al. (6, 28) found up to 3.5-fold differences in incidence and mortality when they compared health insurance claims data with clinical record data. Even for (automated) clinical case identification from electronic patient records, which is to be preferred, Rhee et al. (6) reported a positive predictive value of only 70.4% in comparison with manual clinical validation as the gold standard.

Patient population: As measured by the APACHE II score (25), the mean initial severity of sepsis patients in the JSR was higher than that in the above-mentioned English (24) and Australian cohorts (score >25 only in about 18%) (23). This is problably related to a higher age and a higher rate of comorbidities: The mean age in these studies was 2 to 3 years lower than the mean age of patients in the JSR. Patients in the Canadian and Finnish papers cited above with similar APACHE II scores were also notably younger, with mean ages of 58 and 60.4 years, respectively (25, 27).

Care: International differences are also noticeable with regard to care structures. The median length of stay in an ICU/hospital is reported as 4/21 days for the United Kingdom (29), 3.2/13.5 days for Australia (23), and 6.9/19.2 days for Canada (27). This compares with 15.9/33.2 days in the JSR. In addition, an international comparison showed Germany to have the highest number of intensive care beds in relation to population size. In the Anglophone countries, capacities are significantly lower (30). In 2008, a comparative cross-sectional study showed that the intensive care admission rate was ten times lower in the United Kingdom than in Germany (216/100 000 versus 2353/100 000 inhabitants) (21). Moreover, in the United States an average of 6.2% of patients discharged from hospital are transferred directly to a hospice (6). In Germany, this is true for only 0.3% of patients (3). Thus, more sepsis patients appear to die in the hospital in Germany than in the US, which inevitably increases hospital mortality rates.

In addition to recording mortality rates, the JSR investigated a number of clinical variables for a possible association with survival (Table 2, eTable 4). All the associations noted have been previously described in the literature. These include, first, clinical variables that cannot be influenced by optimization of acute care, such as age and chronic pre-existing conditions (31). Secondly, indicators of sepsis severity at diagnosis, such as APACHE II and SOFA scores (29), and the presence of septic shock were shown to be predictors of acute survival. Length of stay in the ICU and the need for renal replacement therapy, as additional care parameters assessed (32), were associated with decreased long-term survival. They too should be understood as an expression of the severity of sepsis.

Table 2. Associations with instantaneous mortality rate.

Variable	Univariate Cox model		Multiple Cox model
	HR [95% CI]	P	HR [95% CI]	P
Demographic variables
Age (per year)	1.03 [1.02; 1.04]	<0.001	1.03 [1.02; 1.04]	<0.001
Sex (male) (female, reference)	1.01 [0.83; 1.24] 1.00	0.89
Sepsis severity
APACHE II score (per point)	1.04 [1.03; 1.06]	<0.001	1.01 [0.99; 1.02]	0.47
SOFA score (per point)	1.07 [1.04; 1.1]	<0.001	1.03 [0.99; 1.06]	0.17
Septic shock (sepsis-1) (y/n)	1.24 [1.00; 1.53]	0.05	1.05 [0.84; 1.30]	0.66
Septic shock (sepsis-3) (y/n)	1.01 [0.83; 1.23]	0.89
Comorbidities
Diabetes (y/n)	1.45 [1.20; 1.75]	<0.001	1.29 [1.06; 1.57]	0.01
Heart failure (y/n)	1.06 [0.64; 1.78]	0.81
Cerebrovascular disease (y/n)	1.41 [1.14; 1.75]	<0.01	1.29 [1.03; 1.61]	0.03
Renal failure (y/n)	1.47 [0.98; 2.20]	0.06	1.15 [0.74; 1.76]	0.54
Chronic obstructive lung disease (y/n)	1.27 [1.00; 1.63]	0.05	1.1 [0.85; 1.42]	0.45
Liver cirrhosis (y/n)	0.91 [0.57; 1.44]	0.69
Tumor (y/n)	1.39 [0.97; 1.99]	0.08	1.29 [0.89; 1.87]	0.18
Immune suppression (y/n)	1.04 [0.72; 1.50]	0.84
Site of infection
Cardiovascular infection (y/n)	0.99 [0.65; 1.52]	0.97
Pneumonia (y/n)	1.25 [1.03; 1.51]	0.02	1.08 [0.88; 1.33]	0.44
Other respiratory tract infection f(y/n)	1.27 [0.93; 1.75]	0.13
Intra-abdominal/gastrointestinal infection (y/n)	0.84 [0.68; 1.04]	0.117
Bone/soft tissue infection (y/n)	0.73 [0.48; 1.10]	0.13
Surgical wound infection (y/n)	1.19 [0.70; 2.03]	0.52
Primary bacteremia (y/n)	1.09 [0.72; 1.66]	0.68
Urogenital infection (y/n)	1.12 [0.84; 1.51]	0.44
Other infection (y/n)	1.08 [0.78; 1.49]	0.65
Infection source
Infection source (hospital-acquired) (community-acquired, reference)	1.45 [1.18; 1.78] 1.00	<0.001	1.44 [1.16; 1.79]	<0.01
Evidence of infection (microbiologically confirmed) (clinical evidence, reference)	1.15 [0.93; 1.42] 1.00	0.20
Pathogen spectrum (gram-negative) (gram-positive) (both gram-negative and gram-positive) (unknown or other, reference)	1.24 [0.86; 1.52] 1.04 [0.81; 1.35] 1.24 [0.96; 1.59] 1.00	0.35 0.74 0.09
Reason for ICU admission (elective procedure) (emergency surgery) (nonsurgical emergency, reference)	1.09 [0.81; 1.44] 0.73 [0.57; 0.93] 1.00	0.57 0.01	0.76 [0.56; 1.03] 0.77 [0.60; 0.99]	0.08 0.05
Care parameters
Length of stay on ICU (per day)	1.01 [1.00; 1.01]	<0.001	1.01 [1.00; 1.01]	0.01
Mechanical ventilation (y/n)	1.21 [0.90; 1.64]	0.21
Renal replacement therapy (y/n)	1.52 [1.25; 1.84]	<0.001	1.36 [1.06; 1.73]	0.01

Associations between hazard ratio (HR; instantaneous mortality rate) and clinical variables at the time of study inclusion, and between HR and parameters of hospital care, in the univariate and multiple Cox regression models; HR > 1 expresses a positive association with mortality rate, values <1 a negative association. Reference group is the N = 1093 patients who survived their hospital stay. For further details see eMethods.95% CI, 95% confidence interval; APACHE, acute physiology and chronic health evaluation; ICU, intensive care unit; SOFA, sepsis-related organ failure assessment

One of the initiatives to improve acute care in Germany has been started at Greifswald University Hospital. There, a quality management program reduced 90-day mortality from 64.2% to 45% (11) (etable 3). The key point is staff training in early detection and implementation of evidence-based treatment protocols including (among other things) immediate antibiotic administration and collection of blood for culture. Since hospital-acquired infections were also negatively associated with long-term survival in the JSR (33), the introduction of hospital-wide access to an infectious disease consultatant seems to be another possible measure to reduce severe cases of hospital-acquired sepsis (34).

To better understand and treat the long-term consequences of sepsis (35) will require trans-sectoral initiatives and structured outpatient care. In the JSR, long-term survival was assessed on the basis of primary care physician records with only minimal data loss. In contrast, contacting patients directly by telephone resulted in a failure rate of almost 30% after 5 years in one Scottish study (26). The Ohio State University Sepsis Registry was even closed in 2011 due to inadequate follow-up rates (36).

The findings derived from the JSR are limited by the facts that it is a single center dataset and that data collection is confined to mortality over the long term, precluding any conclusions about long-term morbidity. Although the JSR findings show parallels to national or multicenter data (e.g., regarding demographic characteristics or hospital mortality), the extent to which the results are comparable with other locations remains unclear. The lack of an appropriate control group means that no conclusions can be drawn about whether sepsis survival has increased long-term mortality rates in our population. However, indications from the recent literature suggest that it has: The US observational studies by Prescott et al. (37) and Wang et al. (e1) respectively showed increased mortality up to 2 years after sepsis among patients over age 65, and up to 5 years after sepsis among those over age 45, compared with the control group. Most comparable with the present work is a single-center prospective cohort study from Australia (38), which found, in comparison with national mortality tables, a (probably) sepsis-induced increase in mortality rates for up to 2 years after sepsis (etable 5).

The strength of the JSR is its follow-up period, which for a prospective observational study is quite long even by international standards, and its relatively high number of cases and data completeness, required by many authors as an indicator of quality (8). It seems possible that this approach to data collection may be transferable to other locations. This is currently under way in the Mid-German Sepsis Cohort (39): This multicenter patient cohort study also addresses, on the basis of extensive interviews, the issue of long-term morbidity in sepsis patients.

Conclusion

For the specific setting of a German tertiary care center, results regarding the long-term survival of almost 2000 patients who received a diagnosis and intensive treatment for sepsis are now available. Prospective case identification produced data of high quality. Primary care physician records proved to be a practicable data source for long-term registry projects.

As indicated by the WHO Sepsis Resolution (40), the reduced survival rates after sepsis, even in the long term, underscore the need for additional efforts, both hospital-based and on a wider intersectoral basis, to further reduce the number of preventable sepsis-associated deaths and to improve the long-term care of sepsis survivors.

Supplementary Material

eMethod

Documentation sources

The documentation of the Jena Sepsis Registry (JSR) comprises three sources:

Follow-up telephone interviews to collect data on mortality

After each of the preset follow-up timepoints (6, 12, 24, 36, and 48 months after sepsis onset), we waited between 1 and 3 months before contacting the primary care physician, because patients do not go to the doctor every week. If no information was available on the first call because no visit to the primary care physician had yet been made, another follow-up call was carried out. Thus, when we refer to “contacts” in our report, this may mean several calls; the number of calls made in each case was not recorded.

In parallel with this, we investigated whether patients had been readmitted as inpatients after the relevant cut-off date. If they had, no further inquiries were made to the primary care physician for this cut-off date.

Only patients who remained under observation were included in future telephone interviews. In addition to whether the patient had died, the place of death (home, acute hospital, rehabilitation facility, nursing home, other) was also recorded, if known.

Procedure for determining censoring and event timepoints

For patients who died during their first hospital stay (onset of sepsis), the date of death held in the hospital information system was used. For all patients who survived this first hospital stay (onset of sepsis), the information collected during the telephone interviews (see above) was used to determine censoring and event timepoints.

Because the time at which the telephone interviews were conducted was not recorded, the timepoints were calculated as 6, 12, 24, 36, and 48 months after sepsis onset. Then, for all patients who survived their initial hospital stay (onset of sepsis), it was determined whether they had died between discharge and telephone interview or, as applicable, between telephone interviews. If an exact date of death was available, this was used. If only information on the month or year of death was available, the 15th of that month (n = 76; 3.8%) or the 30th of June of that year (n = 13; 0.7%) was used as an approximation. If the only information available was that a patient had died, the time of the last telephone contact (see above) was used as an approximate date of death (n = 32; 1.6%).

For all patients for whom no information regarding death was found during follow-up, the record was set to “censored” at the time of last contact.

Analysis of possible predictors of survival

To explore associations between possible effect sizes and survival time, univariate and multiple Cox regression models were used and reported according to suggested formulations from the work of Zwiener et al. (e2). Two Cox regression analyses were performed. The models shown in eTable 4 include all patients registered in the study and all variables for which information was available at the time of inclusion in the study. In the models reported in Table 2, additional care parameters were added that could only be documented during the course of treatment. Therefore, this analysis only includes patients who survived and thus completed inpatient care.

All potential predictors with a univariate P-value ≤ 0.1 (based on the Wald test) were included in a multiple model. No indication of deviations from the Cox model assumptions was found. In the Tables for the Cox regression models, two-sided P-values were in each case calculated without correction for multiple comparisons, with the analysis focused on effect estimators (hazard ratios) and 95% confidence intervals (CI). Hazard ratios > 1 characterize poorer survival or higher instantaneous risk of death.

For the variables age, length of stay in intensive care, and SOFA and APACHE II scores, the effect estimators refer to the change in instantaneous risk of death when the variables increase by one unit (indicated in parentheses). For binary variables, effect estimators refer to the risk of death for “yes” compared with “no.” For sex, infection source, evidence of infection, and effect estimators refer in each case to the category indicated first compared with the reference category that follows it. The analyses were performed using the statistical software R 3.5.1 (The R Foundation).

Ethical and legal aspects

The JSR was carried out in accordance with the requirements of the current version of the Declaration of Helsinki. It was approved by the ethics committee of the Friedrich-Schiller University of Jena (approval no. 3218–08/11). All patient-related data were recorded electronically in pseudonymized form. Each patient was given a consecutive identification number (study subject ID) provided by an automatically generated web form (PID registration).

• eCRFs (electronic case report forms in OpenClinica) during the hospital stay (onset of sepsis).

• Information from hospital information systems (COPRA, SWISSLAB, and SAP i.s.h.med) during the hospital stay (onset of sepsis and hospital discharge).

• eCRFs (via OpenClinica) during follow-up telephone interviews (6, 12, 24, 36, and 48 months after sepsis onset).

Key messages.

• The Jena Sepsis Registry was able to prospectively enroll nearly 2000 patients who fulfilled clinical criteria of (severe) sepsis or septic shock and follow them for up to 4 years.

• Half the patients died within 70 days.

• A mortality rate of almost 75% at 4 years after diagnosis underlines the need for further hospital-based and wider intersectoral measures.

• Conducting registry studies drawing on primary care physician records appears feasible and may be a useful complement to other data sources.