Long-term survival in patients discharged alive from hospital following an intensive care unit admission with sepsis or septic shock in Australia and New Zealand: an observational cohort study

Summary

Background

Data on long-term survival after intensive care unit (ICU) admission for sepsis and septic shock are limited. This study aimed to evaluate survival over five years among critically ill sepsis or septic shock patients discharged alive from hospital.

Methods

This retrospective cohort study of adults who survived to hospital discharge after non-elective ICU admission (2018–2024) used Australian and New Zealand Intensive Care Society Adult Patient Database. Sepsis, septic shock, or non-sepsis patients were classified using diagnostic codes and physiological/laboratory criteria during first 24-h of ICU admission. Mixed-effects Cox models with time-varying covariate effects were used, adjusting for demographics, comorbidities, frailty, illness severity, and ICU interventions.

Findings

Of 557,538 hospital survivors, 7.3% had sepsis without shock, 11.5% septic shock, and 81.2% non-sepsis conditions. At five-years, unadjusted survival was lowest for septic shock (68.0%, 95% confidence interval/CI = 67.6–68.4%), sepsis without shock (74.2%, 73.7–74.6%), and non-sepsis (78.2%, 78.1–78.3%). After adjustment, hazards ratio/HR for sepsis without shock remained at or below the null relative to non-sepsis; 0–1 year: HR = 0.95, 0.92–0.98; 1–3 years: 0.96, 0.93–1.00; 3–5 years: 1.03, 0.97–1.08, while septic shock showed persistent independent excess mortality; 0–1 year: HR = 1.05,1.03–1.08; 1–3 years: 1.03, 1.00–1.06; and 3–5 years:1.09,1.05–1.14. Major predictors of mortality included age, frailty, comorbidities, and organ support.

Interpretation

Septic shock was associated with persistent independent excess mortality over five years after hospital discharge, whereas in sepsis without shock, this was largely explained by pre-existing comorbidity and frailty.

Funding

Medical Research Future Fund.

Research in context.

Evidence before this study

Sepsis and septic shock are leading causes of mortality and morbidity worldwide, often necessitating intensive care unit (ICU) admission. While short-term outcomes for sepsis have improved with advances in critical care, there is increasing recognition of substantial long-term sequelae, including persistent organ dysfunction, cognitive and functional decline, and increased health care utilisation among survivors. Most previous studies have concentrated on short-term mortality (ICU, in-hospital, or 90-day outcomes), with only a limited number reporting longer-term mortality following ICU or hospital discharge. Most long-term studies have included mixed patient populations or lacked granular diagnostic and comorbidity data. Furthermore, it remains unclear to what extent the excess long-term mortality observed among sepsis and septic shock survivors is attributable to sepsis itself, compared to underlying comorbidities, frailty, or other baseline patient factors. Comprehensive, contemporary, multicentre data with robust adjustment for confounding variables and extended follow-up are needed to clarify long-term outcomes for sepsis survivors compared to other ICU populations.

Added value of this study

This study is, to our knowledge, the largest bi-national analysis of long-term survival in hospital survivors following non-elective ICU admission with sepsis or septic shock, using high-quality, linked registry data from Australia and New Zealand. By analysing 557,538 adult patients discharged alive from hospital between 2018 and 2024, including 104,593 with sepsis or septic shock, and applying mixed-effects piecewise Cox models, adjusted for demographics, comorbidities, frailty, illness severity, and ICU interventions, this study provides risk estimates for mortality up to five years post-discharge. Adjusted HRs for sepsis without shock remained at or below the null, compared to non-sepsis conditions. In contrast, septic shock remained associated with a modest but persistent independent excess mortality over the five-year follow-up period. This study identified distinct long-term mortality patterns by sepsis groups, in that crude mortality differences were largely explained by patient characteristics in sepsis without shock, but not in septic shock.

Implications of all the available evidence

The findings of this study highlight that long-term mortality following sepsis was heterogeneous. For sepsis without shock, the absence of independent excess long-term mortality after adjustment suggests that post-discharge risk was primarily driven by pre-existing comorbidity and frailty burden. Accordingly, post-discharge care should focus on optimising management of underlying chronic conditions, frailty assessment, and rehabilitation, in addition to sepsis-specific interventions. For septic shock, the persistent excess mortality across five years indicates a sustained independent risk beyond baseline vulnerability, extending beyond the early post-discharge period. This highlights the need for extended post-discharge surveillance and targeted follow-up programs. Future studies should prioritise identifying modifiable risk factors and mechanisms, including persistent inflammation, immune dysfunction and organ impairment, evaluating structured post-discharge interventions, and exploring cause-specific mortality to inform strategies aimed at reducing long-term mortality in septic shock survivors.

Introduction

Sepsis is defined as life-threatening organ dysfunction caused by a dysregulated host response to infection.¹ It is a leading cause of death and accounted for 11 million (19.7%) deaths globally in 2017.^2,3 Septic shock, the most severe form, involves profound circulatory, cellular, and metabolic abnormalities.¹ Both conditions often require admission to the intensive care unit (ICU) as a result of severe clinical, laboratory, and pathophysiological abnormalities.¹ With earlier recognition and advances in critical care, ICU and in-hospital mortality of sepsis has declined, resulting in an increasing number of patients surviving to hospital discharge.^4,5 As such, there is growing recognition of long-term sequelae of sepsis, such as chronic inflammation, immunosuppression, organ dysfunction and muscle wasting,^1,6 impaired quality of life, cognitive and functional decline, and increased healthcare use.^1,7

Previous studies have reported an elevated risk of death for 1–5 years after ICU/hospital admission, particularly among patients with sepsis and septic shock.^6,8, 9, 10, 11, 12, 13, 14 However, most research has focused on short-term outcomes (ICU, in-hospital mortality, or 90-day survival).^14,15 Indeed, there are limited data describing long-term mortality among hospital survivors following ICU admissions with sepsis or septic shock, nor how this compares to the general ICU population.

Furthermore, it remains uncertain whether the excess long-term mortality observed in survivors of sepsis and septic shock is explained primarily by underlying patient characteristics or whether sepsis and septic shock independently confer a sustained risk of mortality beyond hospital discharge.⁴ Comprehensive, multicentre data with adjustment for baseline characteristics and extended follow-up are needed to improve our understanding of long-term mortality in sepsis survivors, and inform post-discharge care, prognostication, and healthcare policy.

This study aimed to evaluate long-term survival over a five-year follow-up period among patients who were discharged from hospital alive, after an ICU admission with sepsis, septic shock, or a non-sepsis condition.

Methods

Study design and setting

This was a retrospective observational cohort study using data from the Australian and New Zealand Intensive Care Society (ANZICS) Adult Patient Database (APD).¹⁶ The ANZICS-APD was linked to the National Death Index (NDI, Australia) and National Health Index (NHI, New Zealand) to determine long-term mortality outcomes for all eligible patients.^17,18 The ANZICS-APD is a bi-national clinical quality registry that includes demographic, physiological and clinical data of ICU patients admitted to 226 participating ICUs (98% of ICUs in Australia, 68% of ICUs in New Zealand).¹⁹

Ethics approval

This study was approved by the Human Research Ethics Committee of the Alfred Hospital (Project-ID-697/23). Detailed data access, cleaning and linkage procedures are provided in Supplementary-material-1.

Study population

We included all adult patients (aged ≥ 16 years) who survived to hospital discharge and were admitted as non-elective (unplanned admissions) cases to any Australian or New Zealand ICU that contributed data to the ANZICS-APD from 2018 to 2024. For patients with multiple ICU admissions, only data from the index ICU admission were included.

Patients were excluded if they had an unknown diagnosis code,¹⁶ a combination of unknown hospital outcomes, missing death dates and no NDI/NHI linkage, were admitted solely for end-of-life care or potential organ donation, or incorrect/inaccurate survival times (NDI/NHI dates of death preceding hospital discharge dates or unknown hospital discharge dates).

Variables

The ANZICS-APD includes one diagnosis as the primary cause for each ICU admission, determined by treating clinicians or trained data collectors using the ANZICS modification of the Acute Physiology and Chronic Health Evaluation (APACHE-IV) diagnostic codes.¹⁶

Based on this ICU admission diagnostic category, patients were classified into one of three mutually exclusive groups: (1) sepsis; (2) septic shock; and (3) non-sepsis conditions (including all non-infective diagnoses and infective diagnoses not meeting sepsis criteria); see Supplementary Fig. S1. For patients without an explicit APACHE-IV sepsis/septic shock code sepsis/septic shock were defined using modified Sepsis-3 definitions²⁰ by combining an APACHE code for infection (Supplementary Table S1)^4,21 with physiological and laboratory parameters within the first 24 h of ICU admission; patients who developed sepsis or septic shock later during the ICU stay were not reclassified.

More specifically, sepsis was defined (Sepsis-3 definition) as either:

• An infective diagnosis plus a Sequential Organ Failure Assessment (SOFA) score of ≥2, or

• An APACHE primary diagnosis of sepsis (codes 501 or 502).²²

Septic shock was defined according to ANZICS-modified criteria, being:

• The presence of sepsis plus either:

  • a.
    A serum lactate >2 mmol/L, or
  • b.
    Vasopressor use with hypotension, where hypotension was defined as a minimum mean arterial pressure (MAP) < 65 mmHg, or, if MAP was unavailable, a systolic blood pressure < 90 mm Hg, or
  • a.
    Primary APACHE diagnosis of septic shock (codes 503 or 504).²⁰

All other cases were classified as “non-sepsis conditions”. Additional variables extracted for analysis included demographic characteristics (age, sex), year of ICU admission, ICU admission source, Clinical Frailty Scale (CFS) (frail ≥5 or not frail <5),²³ and comorbid conditions (defined by the ANZICS-APD data dictionary).¹⁶ Interventions during the ICU stay were captured, including invasive mechanical ventilation, total invasive ventilation hours, vasopressor or inotrope use, renal replacement therapy, extracorporeal membrane oxygenation (ECMO), and tracheostomy. Hospital outcomes, including length of stay, were also calculated. Illness severity was assessed using the Australian and New Zealand Intensive Care Risk of Death (ANZROD) prediction model.²⁴

Outcomes

The primary outcome was time-to-death up to five years after hospital discharge. Secondary outcomes were all-cause mortality across predefined follow-up periods (0–1, 1–3, and 3–5 years).

Statistical analysis

Baseline characteristics and outcomes were summarised and compared across the diagnostic categories (sepsis, septic shock, and non-sepsis) using appropriate statistical tests (chi-squared tests, ANOVA/Analysis of Variance, or Kruskal–Wallis tests). Summary statistics such as means or medians for continuous variables and proportions for categorical variables were presented.

Time-to-death was analysed using Kaplan–Meier survival curves, which estimated the probability of survival over time for each diagnostic group. For survival analyses using a dynamic cohort approach, patients were followed from the date of hospital discharge until death or censoring. For patients with NDI/NHI linkage, censoring occurred at the earliest of: (1) the end of the five-year follow-up period, or (2) the jurisdiction-specific censor date (the latest available death date in the NDI/NHI at the time of linkage). Patients who survived beyond five years were counted as alive at five years and were censored at that time point. For patients not linked to the NDI/NHI, censoring occurred at the date of the most recent hospital contact following hospital discharge or at five years if no further hospital contact was recorded. Overall, 9222 patients (1.7%) were censored using hospital-contact follow-up.

Differences in survival between admission diagnostic groups were assessed using the log-rank test. Cox proportional hazards regression models were used to estimate hazard ratios (HRs) and 95% confidence intervals (CI) for associations between admission diagnostic groups and time to event (death). The proportional hazards assumption was assessed using Schoenfeld residuals and diagnostic plots. To account for clustering of patients within hospitals, multivariate mixed-effects Cox proportional hazards models were fitted, including hospital site as a random effect.²⁵ Models were adjusted for covariates based on differences across diagnostic groups.

Assessment of the Cox proportional hazards assumption indicated violations for multiple covariates (demographics, ICU admission year and source, comorbidities, frailty, illness severity, diagnostic groups and ICU interventions), suggesting the effect of these covariates on mortality was not constant over time. All covariates were baseline variables measured during the index ICU admission and remained fixed during follow-up, but their associations with mortality were allowed to vary over time.

Thus, we evaluated two continuous time-varying approaches, log (time) covariate interactions and natural cubic spline interactions (three degrees of freedom, knots at the 33rd and 67th percentiles of log event times) before adopting the piecewise specification. Both continuous specifications produced biologically implausible covariate trajectories, consistent with over-extrapolation of parametric functional forms in data-sparse later follow-up periods. Therefore, we specified a mixed-effects piecewise Cox regression model as the primary analysis, allowing covariate effects to vary across follow-up intervals. Follow-up time was partitioned into three intervals: 0–1 year, 1–3 years, and 3–5 years after hospital discharge. ICU admission year was included as a stratification variable to account for incomplete follow-up in more recent years of the dynamic cohort approach and secular trends in outcomes (temporal changes in case mix or sepsis management). Interval-specific hazard ratios (HR) for all covariates were estimated by combining main effect and time-band interaction coefficients, with 95% confidence intervals (CI). Covariates were included primarily for confounding adjustment, and their coefficients should not be interpreted as causal effects. For categorical variables, missing data were assigned as a separate category to retain all available cases in multivariate analyses. For continuous variables (Age, ANZROD), cases with missing data were excluded in multivariate analysis (<0.1%).

Sensitivity analyses using 1-year intervals, across finer follow-up periods within the first year after hospital discharge (0–3, 3–6, 6–9, and 9–12 months), 1–3 and 3–5 years, excluding age, excluding ICU interventions, stratifying sepsis by source of infection (respiratory-source, non-respiratory-source sepsis compared with non-sepsis) and excluding prognostically extreme groups, were also conducted. In addition, all analyses were repeated using the Sepsis-3 definition for sepsis and septic shock. Sepsis was defined as described above. Septic shock was defined as sepsis with both a vasopressor requirement and a serum lactate >2 mmol/L or a primary APACHE diagnosis of septic shock (codes 503 or 504). Data analyses were conducted using R.

Role of funding source

Medical Research Future Fund (no role in study design, data collection, analysis, interpretation, writing).

Results

Details of study cohort

Between 2018 and 2024, 1,370,519 ICU admissions were reported to the ANZICS-APD (Fig. 1). After restricting to the index ICU admission and applying exclusion criteria (majority of excluded admissions were readmissions and elective admissions), 557,538 patients who survived to hospital discharge were included in the analysis. Supplementary Table S2 describes the distribution of admission diagnostic categories and APACHE diagnostic codes by sepsis status and infection site/system. Most non-sepsis patients had non-infectious diagnoses. Within the sepsis and septic shock groups, the most common primary sites were unspecified/other and the respiratory tract (bacterial and viral pneumonia), followed by genitourinary, gastrointestinal, and skin infections. Most common non-sepsis diagnoses included drug overdose, cardiovascular diseases, multiple trauma, and post-surgical conditions.

Fig. 1.

Flow chart of selection of admission records for analysis.

Differences by diagnostic groups

Among 557,538 patients, 7.3% had sepsis without shock, 11.5% septic shock, and 81.2% non-sepsis conditions (Table 1). Males comprised approximately 55% in each diagnostic group. The proportion of sepsis admissions, particularly septic shock, increased over time. Septic shock patients were the oldest, had the highest comorbidity burden (immunosuppression, lymphoma, leukaemia, metastatic cancer, chronic renal, liver, and cardiovascular disease), the greatest frailty prevalence, and the highest illness severity by ANZROD, SOFA, and lactate. In terms of admission source, septic shock patients most commonly presented via the emergency department, while patients without sepsis or septic shock more often came from the operating theatre or recovery. Invasive mechanical ventilation was used more frequently in non-sepsis conditions, although septic shock patients required substantially longer ventilatory support (Table 2). Advanced ICU interventions, vasopressors/inotropes, renal replacement therapy, ECMO and tracheostomy were more frequently used in septic shock. Patients with sepsis without shock and septic shock were less likely to be discharged home compared to those with non-sepsis conditions, and more likely to require transfer to rehabilitation, chronic care or another acute hospital. Median hospital and ICU lengths of stay were longest in septic shock patients. ICU readmission rates, both within the same admission and during a different hospital stay, were highest in the sepsis groups (22.3% in septic shock). Median duration from hospital discharge to death among those who died was shortest with septic shock. Crude one-year, three-year, and five-year mortality by groups is provided in Table 2.

Table 1.

Demographic, admission characteristics and prognostic scores of ICU patients according to admission diagnostic categories.

	Sepsis without shock	Septic shock	Non-sepsis conditionsa	Total
N	40,599 (7.3%)	63,994 (11.5%)	452,945 (81.2%)	557,538 (100.0%)
Age — year	64 (49–75)	68 (55–77)	62 (45–75)	557,514
Sex
Male	22,531 (55.5%)	35,161 (54.9%)	247,256 (54.6%)	304,948 (54.7%)
Female	18,030 (44.4%)	28,795 (45.0%)	205,275 (45.3%)	252,100 (45.2%)
Unknown	38 (0.1%)	38 (0.1%)	414 (0.1%)	490 (0.1%)
Year of ICU admission
2018	7361 (18.1%)	8409 (13.1%)	68,404 (15.1%)	84,174 (15.1%)
2019	6517 (16.1%)	9005 (14.1%)	68,669 (15.2%)	84,191 (15.1%)
2020	5206 (12.8%)	8450 (13.2%)	65,599 (14.5%)	79,255 (14.2%)
2021	5860 (14.4%)	8425 (13.2%)	63,194 (14.0%)	77,479 (13.9%)
2022	5335 (13.1%)	9089 (14.2%)	60,128 (13.3%)	74,552 (13.4%)
2023	5190 (12.8%)	10,314 (16.1%)	62,705 (13.8%)	78,209 (14.0%)
2024	5130 (12.6%)	10,302 (16.1%)	64,246 (14.2%)	79,678 (14.3%)
ICU admission source
Operating theatre/Recovery	6084 (15.0%)	5067 (7.9%)	150,109 (33.1%)	161,260 (28.9%)
Emergency department	19,369 (47.7%)	36,726 (57.4%)	190,817 (42.1%)	246,912 (44.3%)
Ward	11,884 (29.3%)	16,082 (25.1%)	79,292 (17.5%)	107,258 (19.2%)
Other hospital/Other hospital ICU	3157 (7.8%)	6013 (9.4%)	30,281 (6.7%)	39,451 (7.1%)
Other/Unknown	105 (0.3%)	106 (0.2%)	2446 (0.5%)	2657 (0.5%)
Comorbiditiesb
Chronic respiratory	3599 (8.9%)	4970 (7.8%)	31,751 (7.0%)	40,320 (7.2%)
Chronic liver (cirrhosis)	436 (1.1%)	1370 (2.1%)	7191 (1.6%)	8997 (1.6%)
Chronic cardiovascular	3163 (7.8%)	5144 (8.0%)	33,554 (7.4%)	41,861 (7.5%)
Chronic renal (dialysis dependent)	1484 (3.7%)	2762 (4.3%)	13,237 (2.9%)	17,483 (3.1%)
Immunosuppression	3604 (8.9%)	7335 (11.5%)	16,511 (3.6%)	27,450 (4.9%)
Lymphoma	645 (1.6%)	1563 (2.4%)	3294 (0.7%)	5502 (1.0%)
Metastatic cancer	1410 (3.5%)	3227 (5.0%)	11,938 (2.6%)	16,575 (3.0%)
Leukaemia	1004 (2.5%)	2133 (3.3%)	3550 (0.8%)	6687 (1.2%)
Unknown	5 (0.0%)	21 (0.0%)	66 (0.0%)	92 (0.0%)
Clinical Frailty Scale score
Not frail (<5)	22,952 (56.5%)	37,341 (58.4%)	285,107 (62.9%)	345,400 (62.0%)
Frail (≥5)	7125 (17.5%)	14,725 (23.0%)	63,718 (14.1%)	85,568 (15.3%)
Unknown	10,522 (25.9%)	11,928 (18.6%)	104,120 (23.0%)	126,570 (22.7%)
ANZROD risk of death (%)	3.5 (1.4–8.7)	6.5 (2.6–16.2)	2.1 (0.7–7.0)	557,063
ANZROD risk of death (%) Mean (SD)	7.3 (10.2)	12.5 (15.3)	7.1 (12.7)	557,063
Total SOFA score	3.0 (2.0–5.0)	5.0 (3.0–7.0)	3.0 (1.0–5.0)	557,538
Arterial lactate — mmol.L-1	1.2 (1.0–1.6)	2.1 (1.3–3.2)	1.7 (1.0–2.6)	425,156

^aAll other emergency admissions.

^bCategories are not mutually exclusive; patients have more than 1 chronic diseases, ANZROD; Australian and New Zealand Risk of Death, SOFA; Sequential Organ Failure Assessment, ICU; intensive care unit, Continuous variables are presented as median (IQR/Interquartile range), SD; Standard deviation, All comparisons across admission diagnostic categories were statistically significant at p < 0.001.

Table 2.

Interventions and outcomes of ICU patients according to admission diagnostic categories.

	Sepsis without shock	Septic shock	Non-sepsis conditionsa	Total
N	40,599 (7.3%)	63,994 (11.5%)	452,945 (81.2%)	557,538 (100.0%)
Invasively ventilatedd	6622/35,549 (18.6%)	14,656/60,130 (24.4%)	136,021/411,864 (33.0%)	157,299/507,543 (31.0%)
Total invasive ventilation hours during ICU stayb	44 (14–119)	68 (22–158)	21 (9–62)	138,089
Non-invasive ventilationd	7988/35,301 (22.6%)	9597/59,732 (16.1%)	45,873/406,382 (11.3%)	63,458/501,415 (12.7%)
Vasopressor/inotrope used	6964/34,989 (19.9%)	49,392/60,526 (81.6%)	137,331/406,326 (33.8%)	193,687/501,841 (38.6%)
Renal replacement therapyd	853/34,503 (2.5%)	4478/58,996 (7.6%)	13,936/401,069 (3.5%)	19,267/494,568 (3.9%)
ECMOd	54/34,370 (0.2%)	166/58,953 (0.3%)	1011/400,196 (0.3%)	1231/493,519 (0.2%)
Tracheostomyd	374/34,160 (1.1%)	836/58,779 (1.4%)	5364/398,566 (1.3%)	6574/491,505 (1.3%)
Hospital outcome
Home/Hospital in the home	31,222 (76.9%)	45,958 (71.8%)	336,041 (74.2%)	413,221 (74.1%)
Transferred to Nursing home/Chronic care/Palliative care	1053 (2.6%)	2115 (3.3%)	11,366 (2.5%)	14,534 (2.6%)
Transferred to other hospital—ICU	1246 (3.1%)	2397 (3.7%)	11,017 (2.4%)	14,660 (2.6%)
Transferred to other acute hospital	4900 (12.1%)	8455 (13.2%)	55,781 (12.3%)	69,136 (12.4%)
Transferred to Rehabilitation	1919 (4.7%)	4621 (7.2%)	31,381 (6.9%)	37,921 (6.8%)
Transferred to Mental health/Other	259 (0.6%)	448 (0.7%)	7359 (1.6%)	8066 (1.4%)
Hospital length of stay, days	9.0 (5.0–16.0)	10.0 (6.0–20.0)	8.0 (4.0–15.0)	557,528
ICU length of stay, days	2.3 (1.2–4.2)	2.9 (1.8–5.2)	1.9 (1.0–3.6)	556,743
Readmitted to ICU within the same hospital stayc	2310 (5.7%)	4024 (6.3%)	24,804 (5.5%)	31,138 (5.6%)
Readmitted to ICU during a different hospital stayc	8531 (21.0%)	14,294 (22.3%)	83,805 (18.5%)	106,630 (19.1%)
Duration from hospital discharge to death (days) (among those who died during 5-year follow-up)	381 (94–882)	282 (65–754)	360 (88–849)	111,566
Duration from hospital discharge to censoring (days) (among those survived during 5-year follow-up)	1741 (998–1826)	1611 (833–1826)	1721 (949–1826)	445,972
1-year mortalitye	4339/36,929 (11.7%)	9530/56,578 (16.8%)	41,782/407,945 (10.2%)	55,651/501,452 (11.1%)
3-year mortalitye	5915/26,617 (22.2%)	11,097/36,717 (30.2%)	55,346/282,748 (19.6%)	72,358/346,082 (20.9%)
5-year mortalitye	4942/15,572 (31.7%)	8028/20,074 (40.0%)	41,630/155,921 (26.7%)	54,600/191,567 (28.5%)

^aAll other emergency admissions, ICU; intensive care unit, ECMO; extracorporeal membrane oxygenation, Continuous variables are presented as median (IQR/Interquartile range).

^bReported among patients who received invasive mechanical ventilation.

^cRe-admission data were obtained by linking the re-admission data to the analysed cohort of unique patients with index ICU admissions, All comparisons across admission diagnostic categories were statistically significant at p < 0.001.

^dPercentages for interventions calculated using the denominator of patients with known data for each variable (excludes patients with unknown data).

^ePercentages for mortality calculated using the denominator of patients with potential for complete follow-up at each time point (Crude mortality proportions were calculated only among patients with potential for complete follow-up at each time point and are therefore not directly comparable with Kaplan–Meier survival estimates derived from the full dynamic cohort using censoring methods).

Long term mortality

Up to five years after hospital discharge, the unadjusted probability of survival (Kaplan–Meier estimates) was lowest for septic shock patients with 84.5% (95% CI = 84.2–84.8%) at one year, 74.4% (95% CI = 74.0–74.7%) at three years, and 68.0% (95% CI = 67.6–68.4%) at five years (Fig. 2). For sepsis without shock, survival was 89.0% (95% CI = 88.7–89.3%) at one year, 80.3% (95% CI = 79.9–80.7%) at three years, and 74.2% (95% CI = 73.7–74.6%) at five years. For non-sepsis conditions, survival was 90.5% (95% CI = 90.4–90.6%) at one year, 83.0% (95% CI = 82.9–83.1%) at three years, and 78.2% (95% CI = 78.1–78.3%) at five years. Compared with non-sepsis conditions, this represented an absolute difference of 4.0% lower 5-year survival in sepsis without shock and 10.2% lower survival in septic shock (log-rank p < 0.0001).

Fig. 2.

Kaplan–Meier survival curves by admission diagnostic category following hospital discharge (5- year follow-up).

In univariate piecewise Cox regression, unadjusted mortality was higher for both sepsis groups relative to non-sepsis conditions across all follow-up intervals. For sepsis without shock, unadjusted HR were 1.14 (95% CI 1.11–1.18) at 0–1 year, 1.18 (1.14–1.22) at 1–3 years, and 1.29 (1.22–1.36) at 3–5 years. For septic shock, unadjusted HR were 1.68 (1.64–1.72) at 0–1 year, 1.49 (1.44–1.53) at 1–3 years, and 1.49 (1.43–1.56) at 3–5 years.

Table 3 presents the adjusted HR from the primary mixed-effects piecewise Cox regression model. After full adjustment for demographics, comorbidities, frailty, illness severity, and ICU interventions, the HR for sepsis without shock remained at or below the null relative to non-sepsis conditions across all follow-up intervals (0–1 year: HR = 0.95 [0.92–0.98]; 1–3 years: 0.96 [0.93–1.00]; 3–5 years: 1.03 [0.97–1.08]). Septic shock showed persistent independent excess mortality throughout the five-year follow-up period after full adjustment, with HR of 1.05 (1.03–1.08) at 0–1 year, 1.03 (1.00–1.06) at 1–3 years, and 1.09 (1.05–1.14) at 3–5 years (Fig. 3). Increasing age, frailty, chronic comorbidities (chronic respiratory, liver, renal diseases, immunosuppression, lymphoma, metastatic cancer, leukaemia), ANZROD risk of death and renal replacement therapy were associated with increased mortality throughout the entire follow-up. Being female and receipt of invasive mechanical ventilation were associated with lower mortality across all intervals.

Table 3.

Adjusted hazards ratio for mortality across three follow-up intervals after hospital discharge from mixed-effects piecewise Cox regression model with time-varying covariate effects (0–1 year, 1–3 years, and 3–5 years).

	Adjusted hazard ratio (95% CI) at 0–1 yeara	Adjusted hazard ratio (95% CI) at 1–3 yearsa	Adjusted hazard ratio (95% CI) at 3–5 yearsa
	556,874	480,780	328,600
Sepsis status
Sepsis without shock	0.95 (0.92–0.98)	0.96 (0.93–1.00)	1.03 (0.97–1.08)
Septic shock	1.05 (1.03–1.08)	1.03 (1.00–1.06)	1.09 (1.05–1.14)
Non-sepsis conditionsb	Reference	Reference	Reference
Age — year	1.03 (1.03–1.03)	1.04 (1.04–1.04)	1.05 (1.05–1.05)
Sex
Male	Reference	Reference	Reference
Female	0.85 (0.84–0.87)	0.82 (0.80–0.83)	0.82 (0.79–0.84)
Unknown	0.91 (0.66–1.25)	0.76 (0.50–1.16)	0.54 (0.27–1.08)
ICU admission source
Emergency department	Reference	Reference	Reference
Ward	1.15 (1.12–1.17)	1.01 (0.98–1.04)	1.00 (0.96–1.04)
Operating theatre/Recovery	0.87 (0.85–0.89)	0.85 (0.83–0.87)	0.85 (0.82–0.88)
Other hospital/Other hospital ICU	1.08 (1.04–1.11)	0.97 (0.93–1.01)	0.95 (0.89–1.01)
Other/Unknown	1.15 (1.01–1.30)	1.05 (0.91–1.22)	0.94 (0.75–1.19)
Comorbidities
Chronic respiratory	1.25 (1.22–1.29)	1.57 (1.53–1.62)	1.56 (1.49–1.64)
Chronic liver	2.21 (2.12–2.31)	2.42 (2.28–2.57)	2.06 (1.86–2.28)
Chronic cardiovascular	1.06 (1.03–1.08)	1.05 (1.02–1.09)	1.09 (1.04–1.15)
Chronic renal	1.48 (1.43–1.53)	1.79 (1.72–1.87)	1.83 (1.71–1.95)
Immunosuppression	1.37 (1.33–1.40)	1.42 (1.37–1.48)	1.30 (1.21–1.39)
Lymphoma	1.70 (1.61–1.79)	1.16 (1.06–1.27)	1.07 (0.92–1.23)
Metastatic Cancer	3.86 (3.76–3.97)	3.11 (2.98–3.26)	1.75 (1.60–1.92)
Leukaemia	1.80 (1.71–1.88)	1.81 (1.69–1.94)	1.37 (1.20–1.55)
Clinical Frailty Scale
Not frail (<5)	Reference	Reference	Reference
Frail (≥5)	1.87 (1.83–1.91)	1.93 (1.88–1.99)	1.95 (1.87–2.03)
Unknown	1.24 (1.21–1.28)	1.28 (1.24–1.31)	1.18 (1.13–1.23)
ANZROD risk of death (%)	1.02 (1.02–1.02)	1.01 (1.01–1.01)	1.01 (1.00–1.01)
Invasively ventilated
No	Reference	Reference	Reference
Yes	0.82 (0.80–0.84)	0.82 (0.80–0.84)	0.85 (0.81–0.88)
Unknown	0.81 (0.75–0.86)	0.82 (0.76–0.89)	0.81 (0.74–0.89)
Renal replacement therapy
No	Reference	Reference	Reference
Yes	1.10 (1.06–1.15)	1.23 (1.16–1.29)	1.31 (1.22–1.42)
Unknown	0.94 (0.86–1.03)	1.06 (0.96–1.17)	0.97 (0.87–1.09)
ECMO
No	Reference	Reference	Reference
Yes	1.57 (1.32–1.86)	0.56 (0.38–0.82)	0.70 (0.42–1.16)
Unknown	0.92 (0.84–1.00)	0.85 (0.77–0.94)	0.91 (0.81–1.03)
Tracheostomy
No	Reference	Reference	Reference
Yes	1.03 (0.95–1.13)	0.99 (0.89–1.11)	0.94 (0.79–1.11)
Unknown	1.25 (1.14–1.38)	1.18 (1.07–1.31)	1.14 (1.01–1.30)

^aAdjusted for demographics, ICU admission year & source, comorbidities, frailty, illness severity, ICU interventions, (ICU admission year was included as a stratification variable (strata) to account for secular trends in outcomes over time), 95% CI; 95% confidence interval.

^bAll other emergency admissions, ANZROD; Australian and New Zealand Risk of Death, ECMO; extracorporeal membrane oxygenation, ICU; intensive care unit, n = 556,874 included in modelling with n = 664 excluded due to missing covariate data out of included cohort of 557,538.

Fig. 3.

Adjusted hazard ratios of mortality for sepsis groups versus non-sepsis conditions (reference) from mixed-effects piecewise Cox regression model with time-varying covariate effects (0–1 year, 1–3 years, and 3–5 years) (adjusted for demographics, ICU admission year and source, comorbidities, frailty, illness severity, ICU interventions) Shaded regions represent 95% confidence intervals.

All sensitivity analyses were consistent with the primary findings. The five-band annual-interval model confirmed the absence of within-band heterogeneity (Supplementary Table S3). Further analysis examining three-month intervals within the first year after discharge showed similar patterns. Sepsis without shock showed no independent association with mortality; the marginally significant 0–1-year estimate in the primary model was driven by the 3–6-month interval, with non-significant estimates at other times. Septic shock was independently associated with modestly elevated mortality, being highest at 3–5 years and in the first 6 months (Supplementary Table S4). Excluding age substantially increased the HR of sepsis without shock at 3–5 years and septic shock HRs throughout five-year follow-up, confirming independent confounding (Supplementary Table S5). Excluding ICU interventions produced marginally higher septic shock estimates with virtually unchanged sepsis without shock HRs, indicating ICU interventions did not materially mediate the septic shock mortality risk (Supplementary Table S5). When stratified by infection source, respiratory sepsis was associated with lower early mortality relative to non-sepsis, but higher mortality at 3–5 years, whereas non-respiratory sepsis showed modestly higher mortality over five years (Supplementary Table S5). Restricting the comparator to prognostically intermediate non-sepsis admissions yielded results consistent with the primary analysis for HRs of sepsis without shock across all intervals, while estimates for septic shock were slightly attenuated but remained statistically significant, including at 3–5 years (Supplementary Table S5). Results using strict Sepsis-3 criteria were consistent with the primary analysis (Supplementary Fig. S3, Supplementary Table S6).

Discussion

Summary of study findings

In this large bi-national cohort of over 500,000 patients discharged alive from hospital in Australia and New Zealand, mixed-effects piecewise Cox regression showed distinct patterns of long-term mortality. Unadjusted analyses showed substantially decreased survival in both sepsis without shock and septic shock. After adjustment for demographics, comorbidities, frailty, illness severity, and ICU interventions, the HR for sepsis without shock remained at or below the null relative to non-sepsis conditions across all follow-up intervals. In contrast, septic shock showed modest but persistent independent excess mortality.

Comparison with previous studies

Prior studies have reported variable longer-term mortality for sepsis (ranging from 21.1% at 1-year to 63.3% at 3–5 years), and septic shock (36.5% at 1-year to 77.8% at 3–5 years).8, 9, 10, 11, 12^,26,27 In our study, the absolute mortality among sepsis and septic shock survivors was at the lower end of these ranges, likely reflecting advances in pre-hospital, in-hospital, and critical care, as well as differences in case mix, and variation in study design, population, and definitions.

Our cohort was restricted to non-elective ICU admissions surviving to hospital discharge, whereas many prior studies examined all hospitalised sepsis patients, including frailer community-acquired cohorts, with few studies focusing on ICU/hospital survivors. The non-sepsis group in our study included critically ill patients undergoing emergency surgery, trauma care, and other complex interventions, yielding a comparator with higher baseline acuity than other control populations used in prior studies. Relative to the ICD-based definitions used in earlier research, we applied Sepsis-3 definitions and APACHE coding to classify septic shock and sepsis separately, using mixed-effects piecewise Cox regression on a contemporary, large-scale ICU cohort with comprehensive follow-up, an important distinction not often explored in previous long-term outcomes studies. Although septic shock was defined in the primary analysis using ANZICS-modified criteria, sensitivity analyses applying strict Sepsis-3 definitions yielded consistent results.

While unadjusted analyses in this study found substantially decreased survival in the sepsis groups, this excess risk was attenuated after adjustment for baseline characteristics, illness severity, and ICU interventions. Some previous studies on sepsis survivors reported persistent long-term mortality risk and major cardiovascular events at one year and up to five years after discharge,^6,12,13 while others did not find any difference in long-term outcomes between sepsis and non-sepsis conditions.⁸ For sepsis without shock, full adjustment attenuated any statistically independent association with mortality across all intervals. Regarding infection source, higher mortality for respiratory sepsis at 3–5 years and non-respiratory sepsis throughout follow-up relative to non-sepsis highlights that the source of infection may influence the temporal trajectory of post-discharge mortality risk. Adjusted HRs for sepsis without shock below 1 at early intervals reflect the high long-term mortality burden of the critically-ill non-sepsis comparator, which included cardiac arrest, cardiogenic shock, advanced malignancy, and major trauma, rather than any protective effect of sepsis. They also reflect the attenuation of the crude excess mortality once higher baseline comorbidity and frailty in sepsis without shock patients were accounted for. For septic shock, a modest but persistent association remained after adjustment, consistent with previous research¹² suggesting that septic shock is an independent predictor of poor long-term outcomes. Higher illness severity, a greater burden of comorbidities and frailty, and more frequent ICU interventions in the septic shock group were reflected in prolonged ICU support (longer ventilation hours, vasopressor use, and renal replacement therapy) and greater acute physiological derangement.

Although long-term prognosis was also greatly determined by older age, frailty, chronic comorbidities, immunosuppression, and need for organ support, the independent excess mortality attributable to septic shock itself, over and above patient-level factors, was statistically significant and clinically important throughout follow-up. The finer interval analysis showed the highest excess mortality for septic shock in the 3–5 years and first six months periods, likely reflecting early incomplete recovery from acute organ dysfunction, and progressive comorbidity acceleration, recurrent infection vulnerability and chronic organ impairment later on. The negligible difference with the model excluding ICU interventions indicates that ICU interventions did not materially mediate the septic shock mortality risk and that the persistent excess mortality in this group was not explained by acute intervention requirements. The association between invasive mechanical ventilation and lower long-term mortality likely reflects residual differences in admission case-mix and selection bias among hospital survivors rather than a protective effect of ventilation itself. The analysis excluding prognostically extreme groups showed that the persistent septic shock mortality was not explained by prognostic heterogeneity of the non-sepsis comparator group. This aligns with other work suggesting that while much of the long-term mortality attributed to sepsis or septic shock was explained by underlying patient vulnerability and comorbidity,^8,13 septic shock itself also confers additional risk beyond baseline characteristics and ICU interventions.^12,28

The persistent higher mortality risk for septic shock over five years, alongside the absence of any independent risk for sepsis without shock, may reinforce the evolving recognition of sepsis as a disease with heterogeneous long-term consequences.¹ For sepsis without shock, the crude survival gap appears largely attributable to pre-existing vulnerability. The sustained excess risk following septic shock was consistent with the hypothesis that severe haemodynamic compromise, sustained organ dysfunction, and associated inflammatory and metabolic derangements may initiate or accelerate a trajectory of chronic health deterioration beyond what is captured by baseline characteristics.²⁹ Potential mechanisms underlying reduced survival include persistent inflammation, immune dysregulation, ongoing organ dysfunction, muscle wasting, neurocognitive decline, and acceleration of comorbidities via microvascular and endothelial injury.¹ However, it remains challenging to separate the effects of sepsis from differences in baseline characteristics and comorbidities, highlighting the need for future research in more homogeneous patient populations with and without sepsis. This could include the use of propensity-score matching and comparing outcomes with age-standardised mortality in the general population. Thus, longer-term outcome after sepsis reflects a complex interplay between patient characteristics, comorbidities, ICU treatments and sepsis itself.²⁹

Implications

Our findings highlight distinct implications. For sepsis without shock, the absence of independent excess long-term mortality after adjustment suggests that post-discharge risk was primarily driven by pre-existing comorbidity and frailty burden. Accordingly, post-discharge care should focus on optimising management of underlying chronic conditions, frailty assessment, and rehabilitation, besides sepsis-specific interventions. For septic shock, the persistent independent mortality excess across the five-year follow-up period indicates that the excess risk was not confined to the early post-discharge phase and did not attenuate over time. This has important implications for post-discharge surveillance: structured follow-up programs should extend beyond the first year after discharge. Identifying the causes of this persistent excess mortality, whether infectious, cardiovascular, malignant, or due to ongoing organ dysfunction, should be a priority for future research, as this will inform which interventions may reduce preventable mortality.^7,12

Patients with sepsis without shock and septic shock were less likely to be discharged home and more likely to require transfer to rehabilitation, chronic care, or another acute hospital, suggesting greater post-acute care needs that may influence long-term outcomes. High-risk subgroups, particularly individuals with frailty, chronic renal disease, chronic liver disease, and metastatic malignancy, represent priority populations for targeted post-discharge care. Future research should evaluate whether structured follow-up, transitional care programs, medication optimisation, multidisciplinary rehabilitation, and monitoring for recurrent infection and infection-source-specific risks and cardiovascular complications can reduce the persistent mortality burden in septic shock survivors. Additionally, future research should examine hospital readmissions after sepsis, investigate modifiable risk factors, including persistent inflammation, immune dysfunction, and social determinants of health, and develop biomarkers to identify individuals at highest risk of late mortality following septic shock.

Strengths and limitations

Major strengths of this study include its large, contemporary, multicentre, bi-national cohorts and high-quality data from the majority of ICU admissions in Australia and New Zealand, which support the generalisability of our findings to other regions with similar resources. The analytic approach, incorporating mixed-effects piecewise Cox models, adjusting for confounders with minimal missing data, and ensuring comprehensive follow-up, enhances the robustness of our findings.

Several limitations should be considered. First, residual confounding might persist due to unmeasured variables including socio-economic factors, functional status and medication use at discharge. Second, while we relied on ICU admission diagnoses and available clinical parameters within the first 24 h of ICU admission to identify sepsis cases, some misclassification is likely to have occurred. Additionally, while our study captures all-cause mortality, we did not assess cause-specific mortality or non-fatal sequelae such as re-hospitalisations, quality of life, or functional outcomes. Our study was limited to non-elective ICU admissions, thereby excluding elective post-operative patients and those with sepsis managed outside the ICU, which may result in an underestimation of the true incidence and outcomes of sepsis.³⁰ Sepsis diagnoses were limited to cases identified within the first 24 h of ICU admission, so patients developing sepsis or septic shock later during their ICU stay would remain classified as non-sepsis conditions, potentially attenuating differences between groups. Moreover, 22.7% of patients had unknown CFS. Although retained as a separate category, the elevated mortality risk in this group suggests potential informative missingness and residual confounding by frailty. Also, the low frequency of ECMO in our cohort most likely reflects case selection rather than national regulation or reimbursement barriers. Patients without NDI/NHI linkage were censored at their last recorded hospital contact date, which may introduce informative censoring because patients with higher healthcare utilisation may have longer observable follow-up. Additionally, survivor bias is possible, as analysing hospital survivors might have preferentially included healthier or more resilient patients, potentially underestimating the true burden of long-term sequelae. Finally, combined shock states, such as septic shock with cardiogenic shock, were not captured separately and represent an important area for future research.

Conclusion

In this large, bi-national cohort of patients discharged alive from hospital after ICU admission, adjusted analyses using piecewise Cox regression identified distinct long-term mortality patterns by sepsis groups. Adjusted HRs for sepsis without shock remained at or below the null relative to non-sepsis conditions across all follow-up intervals, indicating that the crude survival disadvantage in this group was likely attributable to higher pre-existing comorbidity and frailty burden. Septic shock showed modest but persistent independent excess mortality over the five-year follow-up period, highlighting that this condition confers a persistent mortality burden beyond baseline patient vulnerability and illness severity. Future research should focus on identifying the causes of persistent post-discharge mortality in septic shock survivors to inform targeted post-discharge care strategies.

Contributors

Win Wah: Conceptualisation, Data Curation, Formal Analysis, Methodology, Software, Visualisation, Writing—original draft, Writing—review & editing. Alexis Paul Poole: Writing—review & editing. Lewis Campbell: Writing—review & editing. Jeremy Cohen: Writing—review & editing. Jamie Cooper: Writing—review & editing. Adam Deane: Writing—review & editing. Anthony Delaney: Writing—review & editing. Glenn Eastwood: Writing—review & editing. Simon Finfer: Writing—review & editing. Craig French: Writing—review & editing. Kimberley Haines: Writing—review & editing. Naomi Hammond: Writing—review & editing.Alisa Higgins: Writing—review & editing. Carol Hodgson: Writing—review & editing. Daryl Jones: Writing—review & editing. Edward Litton: Writing—review & editing. Zoe McQuilten: Writing—review & editing Ary Serpa Neto: Writing—review & editing.David Pilcher: Conceptualisation (Sepsis criteria), Writing—review & editing. Sandra Peake: Writing—review & editing. Manoj Saxena: Writing—review & editing Kiran Shekar: Writing—review & editing. Kelly Thompson: Writing—review & editing. Steven Webb: Writing—review & editing. Paul J Young: Writing—review & editing. Andrew A Udy: Conceptualisation, Writing—review & editing

Data access and verification: David Pilcher and Win Wah accessed and verified the data.

Data sharing statement

The data on which this study is based cannot be made available by the researchers, but data can be requested directly from the data custodians, the Australian and New Zealand Intensive Care Society (ANZICS) registry, through their data request processes. Conditions apply, including ethical approval.

Declaration of interests

The authors declare no competing interests. Research grant funding is declared in the Funding statement.