Article Text


Long-term mortality after community-acquired sepsis: a longitudinal population-based cohort study
  1. Henry E Wang1,
  2. Jeff M Szychowski2,
  3. Russell Griffin3,
  4. Monika M Safford4,
  5. Nathan I Shapiro5,
  6. George Howard2
  1. 1Department of Emergency Medicine, University of Alabama School of Medicine, Birmingham, Alabama, USA
  2. 2Department of Biostatistics, School of Public Health, University of Alabama at Birmingham, Birmingham, Alabama, USA
  3. 3Department of Epidemiology, School of Public Health, University of Alabama at Birmingham, Birmingham, Alabama, USA
  4. 4Division of Preventive Medicine, Department of Medicine, University of Alabama School of Medicine, Birmingham, Alabama, USA
  5. 5Department of Emergency Medicine, Beth Israel Deaconess Medical Center, Boston, Massachusetts, USA
  1. Correspondence to Dr Henry E Wang; hwang{at}


Objective Prior studies have concentrated on the acute short-term outcomes of sepsis, with little focus on its long-term consequences. The objective of this study was to characterise long-term mortality following a sepsis event.

Design Population-based data from the 30 239 community-dwelling individuals in the REasons for Geographic and Racial Differences in Stroke (REGARDS) cohort.

Setting USA.

Participants Community-dwelling adults ≥45 years of age. Sepsis was defined as hospitalisation or emergency department treatment for a serious infection with the presence of ≥2 systemic inflammatory response syndrome criteria.

Outcomes 6-year all-cause mortality. The analysis utilised a time-varying Cox model adjusted for participant's age, demographic factors, health behaviours and chronic medical conditions.

Results The participants were observed for a median of 6.1 years (IQR 4.5–7.1). During this period, 975 individuals experienced a sepsis event. Sepsis hospital mortality was 8.9%. One-year, 2-year and 5-year all-cause mortality among individuals with sepsis were 23%, 28.8% and 43.8%, respectively, compared with death rates of 1%, 2.6% and 8.3% among those who never developed sepsis. On multivariable analysis, the association of sepsis with increased all-cause mortality persisted for up to 5 years, after adjustment for confounders; year 0.00–1.00, adjusted HR (aHR) 13.07 (95% CI 10.63 to 16.06); year 1.01–2.00 aHR 2.64 (1.85 to 3.77); year 2.01–3.00 aHR 2.18 (1.43 to 3.33); year 3.01–4.00 aHR 1.97 (1.19 to 3.25); year 4.01–5.00 aHR 2.08 (1.14 to 3.79); year 5.01+ aHR 1.41 (0.67 to 2.98).

Conclusions Individuals with sepsis exhibited increased rates of death for up to 5 years after the illness event, even after accounting for comorbidities. Sepsis is independently associated with increased risk of mortality well after hospital treatment.

Statistics from

Strengths and limitations of this study

  • Study uses data from the 30 329 participants in the REasons for Geographic And Racial Differences in Stroke REGARDS cohort.

  • Participants were followed for over 6 years.

  • Study characterised long-term survival after a sepsis episode.

  • Robust analytic approach accounted for confounding effect for a wide range of participant characteristics.


Sepsis, the syndrome of microbial infection complicated by systemic inflammation, is associated with an estimated 750 000 hospital admissions, 570 000 emergency department visits, 200 000 deaths and US$16.7 billion in medical expenditures annually in the USA.1–3 While prior studies describe the acute care and course of individuals developing sepsis, relatively limited data characterise the long-term consequences of a sepsis event.4 This gap in knowledge is important as the total public health impact of a disease encompasses not only the course of acute hospital care but also its downstream sequelae. Furthermore, compared with unaffected persons, an individual suffering from a disease—even after recovery from the acute illness—may also experience a higher risk of long-term death.

Prior studies describing mortality after sepsis have important limitations, including the lack of data describing health prior to the sepsis event, the use of hospital administrative data or data from single institutions or the focus on patients in intensive care units.4–8 Few studies have characterised the excess risk of long-term death attributable to a sepsis event or identified the independent predictors of early death after sepsis.

The REasons for Geographic and Racial Differences in Stroke (REGARDS) study is one of the largest population-based longitudinal cohorts of community-dwelling adults in the USA. In this study, we sought to characterise long-term mortality after the hospital treatment for sepsis in the REGARDS cohort.

Materials and methods

Study design

This study utilised data from REGARDS, a national population-based longitudinal cohort.

Selection of participants

Prior studies have described twofold increased stroke mortality in the Southeastern US (the ‘stroke belt’) and threefold increased stroke mortality along the coastal plains of North Carolina, South Carolina and Georgia (the ‘stroke buckle’).9 In addition, other studies highlight the increased stroke mortality among African-Americans. Being one of the largest ongoing national cohorts of community-dwelling individuals in the USA, REGARDS was designed to identify the reasons for the geographical and racial disparities.9

REGARDS includes 30 239 community-dwelling adults ≥45 years from all regions of the continental US. Participant representation oversampled the Southeastern US, with 21% of the cohort originating from the coastal plains of North Carolina, South Carolina and Georgia (the ‘stroke buckle’), and 35% originating from the remainder of North Carolina, South Carolina and Georgia plus Tennessee, Mississippi, Alabama, Louisiana and Arkansas (the ‘stroke belt’). The cohort is 42% African-American with 45% men and 69% of individuals are >60 years. The cohort does not include Hispanics where stroke mortality disparities are small to non-existent. The REGARDS cohort encompasses healthy community-dwelling adults—not just individuals with a history of stroke.

REGARDS enrolled participants during 2003–2007. REGARDS obtained baseline data for each participant using the phone interview and in-person evaluations, including medical history, functional status, health behaviours, physical characteristics (height and weight), physiological measures (blood pressure, pulse and ECG) and an inventory of medications. Each participant provided blood and urine specimens. Self-administered questionnaires evaluated diet, family history of diseases, psychosocial factors and prior residences. The study contacted participants at 6-month intervals by telephone, identifying the date, location and attributed reason for all emergency department visits and hospitalisations during the follow-up period. The study then retrieved medical records for specific health events. If the participant died, the study team reviewed death certificates and medical records and interviewed proxies to ascertain the circumstances of the participant's death.

Identification of sepsis events

Using the taxonomy of serious infections by Angus et al,1 we identified all hospitalisations (emergency department visits and/or hospital admission) attributed by participants to a serious infection. Two trained abstractors independently reviewed all relevant medical records to identify clinical and laboratory information, confirm the presence of a serious infection on initial hospital presentation and to verify the relevance of the serious infection as a major reason for hospitalisation. Initial review of 1349 hospital records indicated excellent inter-rater agreement for the presence of a serious infection (κ=0.92) and the presence of sepsis (κ=0.90) on hospital presentation.

Sepsis consisted of presentation to the hospital with an infection plus two or more systemic inflammatory response syndrome (SIRS) criteria, including (1) heart rate >90 bpm, (2) fever (temperature >38.3°C or <36°C), (3) tachypnoea (>20 breaths/min) or partial pressure of carbon dioxide <32 mm Hg and (4) leukocytosis (white cell count >12 000 or <4000 cells/mm3 or >10% band forms).10 We defined SIRS using any asynchronous combination of the worst vital signs and laboratory test results for the initial 28 h of hospitalisation. We chose a 28 h time frame to account for emergency department and up to one full day of inpatient treatment. Because our study focused on ‘community-acquired’ sepsis rather than ‘hospital-acquired’ sepsis, we did not utilise vital signs or laboratory findings from later points during hospitalisation. The study follow-up period was from 5 February 2003 to 30 July 2012.


The primary outcome was all-cause mortality. The REGARDS study ascertained all-cause mortality through active follow-up with participants or proxies, plus searches of the Social Security Administration's Master Death File and the National Death Index. Observations were censored at the date of death or last follow-up. Available for a portion (74.5%) of deaths, cause of death was determined by dual-physician review and adjudication of death records, next-of-kin or proxy reports and hospitalisation records.


Demographic characteristics included age, sex, race, geographical region and self-reported annual household income and education (years of school). As conducted for the parent REGARDS cohort, geographical region consisted of participant's residence in the stroke ‘buckle,’ stroke ‘belt’ and elsewhere.9 Health behaviours included smoking status and alcohol use. Alcohol use categories included none, moderate (1 drink per day for women or 2 drinks per day for men) and heavy (>1 drink per day for women and >2 drinks per day for men).11 To characterise baseline health status, we used quintiles of the physical composite score and mental composite score of the Short Form-12 (SF-12) health survey.

Chronic medical conditions included atrial fibrillation, cancer history, chronic lung disease, chronic kidney disease, coronary artery disease, deep vein thrombosis, diabetes, dyslipidaemia, hypertension, myocardial infarction, obesity, peripheral artery disease and stroke. We identified atrial fibrillation based on participants’ self-report or baseline ECG evidence. Chronic kidney disease included individuals with an estimated glomerular filtration rate <60 mL/min/1.73 m2 calculated using the Chronic Kidney Disease Epidemiology Collaboration CKD-EPIequation.12 Coronary artery disease included a history of myocardial infarction or coronary intervention. Diabetes included a fasting glucose ≥126 mg/L (or a glucose ≥200 mg/L for those not fasting) or the use of insulin or oral hypoglycaemic agents. Dyslipidaemia included individuals with self-reported high cholesterol or the use of lipid-lowering medications. Hypertension consisted of systolic blood pressure ≥140 mm Hg, diastolic blood pressure ≥90 mm Hg or the self-reported use of antihypertensive agents. Myocardial infarction included individuals with a self-reported history of myocardial infarction or baseline ECG evidence of myocardial infarction.

Obesity included those with a waist circumference >102 cm for men or >88 cm for women, or body mass index ≥30 mg/cm2.13 Participants self-reported the history of stroke (including transient ischaemic attacks) or deep vein thrombosis. Peripheral artery disease included a self-reported history of lower extremity arterial bypass or leg amputation. Because REGARDS did not collect information on pulmonary conditions such as asthma and chronic obstructive pulmonary disease, we defined participants’ use of pulmonary medications as a surrogate for chronic lung disease, including β agonists, leukotriene inhibitors, inhaled corticosteroids, combination inhalers, ipratropium, cromolyn, aminophylline and theophylline.

For sepsis hospitalisations, we also identified organ dysfunctions using the Sequential Organ Failure Assessment (SOFA) based on the worst laboratory and physiological findings during the first 28 h of hospitalisation for respiratory, renal, hepatic, cardiovascular, haematological and neurological systems.14

Data analysis

We sought to compare the adjusted risk of all-cause mortality between sepsis and non-sepsis individuals as a function of follow-up time. Follow-up time consisted of (1) years after first sepsis event for individuals with sepsis, and (2) years after study enrolment for non-sepsis individuals. We assumed that non-sepsis participants were representative of the general population, exhibited relative constant risk and entered the study relative to no particular health event. Therefore, the start of the follow-up period for non-sepsis individuals could be defined relative to any arbitrary event.

Because of the non-proportional nature of the relative hazards of death, we fit a time-varying Cox regression model in a piecewise manner, calculating the hazards of death in 1-year intervals.15 To account for differing observation start times, we adjusted the model for age at the start of the follow-up period (age decile at sepsis event for individuals with sepsis, and age decile at REGARDS enrolment for non-sepsis participants). We adjusted the hazard estimates for participants’ demographic characteristics, baseline function, health behaviours and chronic medical conditions.

To account for potential clustering of deaths within each stroke belt region (stroke belt, stroke buckle and non-belt), we repeated the analysis (1) using a robust variance estimator, and (2) modelling stroke belt region as a shared frailty, which are two common approaches to analysing clustered data in time-to-event analysis. Owing to the time lag in observations and medical record retrieval, we could not review medical records for a portion of participants with reported hospitalisations for a serious infection over the observation period. We therefore repeated the analysis excluding these individuals from the non-sepsis group. We also repeated the analysis stratifying by uncomplicated sepsis (infection+SIRS criteria only) versus severe sepsis (sepsis+organ dysfunction).

To identify participant characteristics independently associated with early death after sepsis, among individuals with sepsis, only we fit a Cox model with age at sepsis, sex, race, tobacco and alcohol use, chronic medical conditions, infection type, admission to intensive care unit and SOFA score on hospital presentation in the regression model.

We conducted all analyses using Stata V.12.1 (Stata, Inc, College Station, Texas, USA).


Valid follow-up data were available for 29 664 REGARDS participants, including 970 sepsis and 28 694 non-sepsis individuals. Median follow-up time was 6.1 years (IQR 4.5–7.1). Sepsis incidence was 5.8/1000 person-years (95% CI 5.4 to 6.2). Median time to the first sepsis event was 1.9 years (IQR 3.5–5.0). The most common infection types associated with incident sepsis were pneumonia, kidney and urinary tract infections and abdominal infections (table 1). Compared with non-sepsis participants, individuals with sepsis were older, more likely to be male, had lower income and education, were more likely to use alcohol or tobacco and were more likely to have chronic medical conditions (table 2).

Table 1

Infection types associated with first hospitalisations for sepsis

Table 2

Baseline patient characteristics

There were 324 deaths among 970 individuals with sepsis (33.4%) and 3155 deaths among 28 694 non-sepsis individuals (11%). The incidence of death was 141/1000 person-years for sepsis and 19.2/1000 person-years for non-sepsis individuals (table 3). Adjudicated cause of death was available for 84.6% of sepsis and 73.5% of non-sepsis deaths; the most common adjudicated causes of death among individuals with sepsis were infection, lung disease and cancer (table 4).

Table 3

Incidence of death for sepsis and non-sepsis individuals

Table 4

Adjudicated causes of death.

Among all first-sepsis events, 86 (8.9%) patients died in the hospital, with a median time to death of 7 days (IQR 3–15). One-year, 2-year and 5-year all-cause mortality among individuals with sepsis were 23%, 28.8% and 43.8%, respectively (figure 1). One-year, 2-year and 5-year all-cause mortality among non-sepsis individuals were 1%, 2.6% and 8.3%, respectively. Compared with non-sepsis persons, the rate of death was high among individuals with sepsis in the first year after the event (adjusted HR 13.07; 95% CI 10.63 to 16.06; figure 2 and see online supplementary appendix 1). Among those who survived for at least 1 year, individuals with sepsis exhibited twofold increased rates of death for up to 5 years after the sepsis event.

Figure 1

Kaplan-Meier curves and 95% CIs for long-term survival after sepsis. Sepsis versus non-sepsis. Includes 970 sepsis and 28 694 non-sepsis individuals. Analysis censored at 6 years.

Figure 2

Adjusted HRs for association of sepsis with long-term survival. Time-dependent effects described using a Cox regression model with all sepsis defined as a time-varying covariate. Analysis censored at 6 years. Each bar reflects the HR for death among individuals alive at the beginning of that time segment. Graph depicts observation time ≥6 months only. Adjusted for age, sex, race, region, health status, smoking, alcohol use, obesity and chronic medical conditions, as reported at the beginning of the REasons for Geographic And Racial Differences in Stroke REGARDS study.

To account for potential clustering of deaths within stroke belt region, we repeated the analysis using a robust variance estimator; we observed similar associations between sepsis and rates of long-term mortality; however, the higher rate of death among individuals with sepsis was statistically significant in the sixth year of follow-up. When repeating the analysis modelling stroke belt region as a shared frailty, we observed no major changes in the associations between sepsis and rates of all-cause mortality. When repeating the analysis excluding the 1310 non-sepsis individuals who had unadjudicated possible sepsis events, we observed similar associations between sepsis and long-term mortality.

We stratified the analysis by uncomplicated sepsis (n=257) versus severe sepsis (n=713). Compared with non-sepsis individuals, the adjusted rate of death was twofold higher for 1 year among those experiencing uncomplicated sepsis (see online supplementary appendix 2). Among those who survived at least 1 year, the adjusted rates of death were twofold higher for up to 5 years among those experiencing severe sepsis.

In the sepsis subset, participant characteristics independently associated with increased rates of death after the sepsis event included male sex, health status (physical component score), cancer history, chronic kidney disease, deep vein thrombosis, diabetes, dyslipidaemia, hypertension and obesity (table 5 and see online supplementary appendix 3). Among hospital course characteristics, death rates were higher among participants with increased sequential organ failure scores or admission to the intensive care unit. Rates of long-term death were lower for sepsis due to abdominal infections.

Table 5

Participant characteristics associated with rates of mortality after first sepsis events


For almost a decade, the Surviving Sepsis Campaign guidelines have advocated best practices for sepsis care with the goal of reducing the mortality associated with this condition.16 Drawing on REGARDS, one of the nation's largest population-based cohorts of community-dwelling adults, our study highlights that excess long-term all-cause mortality persists among those suffering from sepsis. Compared with individuals who did not develop sepsis, rates of death among individuals experiencing sepsis were twofold higher for up to 5 years after the sepsis event. Our analysis adjusted for a range of potential confounders and observed relatively small attenuation of risk, suggesting that comorbidities are not playing the major role in this extended risk.

Our findings add to the body of knowledge highlighting the downstream consequences of sepsis. As highlighted in Winters et al's4 systematic review, most sepsis mortality studies have focused on shorter endpoints, inadequately controlled for baseline comorbidities, and relied on population norms as matched controls. Other studies focused on intensive care unit patients with sepsis, used data from single or smaller groups of centres or utilised discharge diagnoses to identify sepsis events.6–8 Our contrasting study utilised a large (>30 000 persons) and diverse population-based sample of community-dwelling individuals in the USA and entailed follow-up of up to 6 years. Rather than relying on hospital discharge diagnoses, a process that may underdetect sepsis cases, we identified sepsis through review of hospital records.1 ,17 We were able to account and adjust for baseline comorbidities that may affect the outcomes of sepsis and non-sepsis individuals.

It is unclear whether increased sepsis mortality reflects the increased susceptibility of those with heightened comorbid burden or whether sepsis triggers an independent pathophysiological process leading to early death.18 Where adjudicated cause of death was available, 70% of deaths following a sepsis event were attributed to a range of other conditions, including cardiovascular and pulmonary diseases. We also identified that select chronic medical conditions such as diabetes, chronic kidney disease and chronic lung disease were independent predictors of early death among individuals with sepsis. Other described sequelae of sepsis include acute kidney injury, atrial fibrillation, cognitive impairment, functional disability and impaired quality of life.19–22 Additional study must confirm the mechanisms by which sepsis creates or complicates the management of these other conditions. For example, does sepsis increase long-term mortality by causing acute kidney injury and subsequent chronic kidney disease? Another important question is whether prevention or optimal management of these parallel conditions might reduce the long-term rates of sepsis death. For example, could optimal management of sepsis-triggered chronic kidney alter an individual's risk of death?

Some limitations of this analysis should be noted. The REGARDS cohort contains individuals over 45 years only, and thus we could not characterise sepsis in younger individuals, who would likely exhibit lower rates of long-term mortality after a sepsis event. We could not identify sepsis events not reported by participants. We did not examine the influence of repeat sepsis events, which could conceivably increase the risk of mortality. We identified individuals who presented to the hospital with sepsis, but did not include those who acquired sepsis during their hospitalisation; we would expect inclusion of the latter individuals to amplify the observed mortality differences. We did not evaluate cognitive impairment, functional disability or impacts on quality of life.

REGARDS was designed to study stroke, not sepsis. However, our novel study takes advantage of important features of REGARDS, including the large participant base, extensive baseline information and extended observation period. There have been no population-based cohorts designed specifically to study sepsis, and few studies on sepsis have had access to comprehensive ‘baseline’ information as is in our study. It would be logistically and financially difficult to replicate a cohort of this scale.

By design, the REGARDS cohort includes only African-Americans and Caucasians, and thus these results may not generalise to other ethnic groups. Select medical conditions not identified by REGARDS may have exhibited associations with risk of sepsis as well as mortality; for example, a history of immunosuppression. As with all observational studies, residual confounding is always a concern. However, our risk adjustment strategy accounted for a comprehensive range of variables that were systematically identified for each participant at the beginning of the REGARDS study. While we accounted for the presence of baseline participant characteristics and chronic medical conditions, we could not account for changes in these patterns over time.


In the REGARDS cohort, individuals with sepsis exhibited increased rates of long-term death, even after accounting for comorbidities. Sepsis is independently associated with increased mortality risk well after hospital treatment.


The authors thank the other investigators, the staff, and the participants of the REGARDS (REasons for Geographic and Racial Differences in Stroke) study for their valuable contributions. A full list of participating REGARDS investigators and institutions can be found at and


View Abstract
  • Supplementary Data

    This web only file has been produced by the BMJ Publishing Group from an electronic file supplied by the author(s) and has not been edited for content.

    Files in this Data Supplement:


  • Contributors HEW, JMS and GH conceived the study. HEW and MMS organised and oversaw data collection. HEW, NIS, MMS and GH obtained funding for the study. HEW, JMS and RG conducted the analysis, and all authors contributed to review of results. HEW drafted the manuscript, and all authors contributed to its editorial review and revision. HEW assumes responsibility for the work as a whole.

  • Funding This study was supported by award R01-NR012726 from the National Institute for Nursing Research, UL1-RR025777 from the National Center for Research Resources, as well as by grants from the Center for Clinical and Translational Science and the Lister Hill Center for Health Policy of the University of Alabama at Birmingham. The parent REGARDS study was supported by cooperative agreement U01-NS041588 from the National Institute of Neurological Disorders and Stroke, National Institutes of Health, Department of Health and Human Service.

  • Competing interests MMS reports the following potential conflicts of interest: Amgen—salary support to study patterns of statin use in Medicare and other large databases; diaDexus—salary support for a research grant on lipids and coronary heart disease outcomes; diaDexus—consulting to help with Food and Drug Administration FDA application; National Institutes of Health, Agency for Healthcare Research and Quality—salary support for research grants. Dr Wang, Dr Szychowski, Dr Griffin, Dr Shapiro and Dr Howard do not report any related conflicts of interest.

  • Ethics approval The study was approved by the Institutional Review Board of the University of Alabama at Birmingham (Approval #X090531004).

  • Provenance and peer review Not commissioned; externally peer reviewed.

  • Data sharing statement No additional data are available.

Request permissions

If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.