Evidence for a causal link between sepsis and long-term mortality: a systematic review of epidemiologic studies

Abstract

Background: In addition to acute hospital mortality, sepsis is associated with higher risk of death following hospital discharge. We assessed the strength of epidemiological evidence supporting a causal link between sepsis and mortality after hospital discharge by systematically evaluating the available literature for strength of association, bias, and techniques to address confounding.

Methods: We searched Medline and Embase using the following 'mp' terms, MESH headings and combinations thereof - sepsis, septic shock, septicemia, outcome. Studies published since 1992 where one-year post-acute mortality in adult survivors of acute sepsis could be calculated were included. Two authors independently selected studies and extracted data using predefined criteria and data extraction forms to assess risk of bias, confounding, and causality. The difference in proportion between cumulative one-year mortality and acute mortality was defined as post-acute mortality. Meta-analysis was done by sepsis definition categories with post-acute mortality as the primary outcome.

Results: The literature search identified 11,156 records, of which 59 studies met our inclusion criteria and 43 studies reported post-acute mortality. In patients who survived an index sepsis admission, the post-acute mortality was 16.1% (95% CI 14.1, 18.1%) with significant heterogeneity (p < 0.001), on random effects meta-analysis. In studies reporting non-sepsis control arm comparisons, sepsis was not consistently associated with a higher hazard ratio for post-acute mortality. The additional hazard associated with sepsis was greatest when compared to the general population. Older age, male sex, and presence of comorbidities were commonly reported independent predictors of post-acute mortality in sepsis survivors, challenging the causality relationship. Sensitivity analyses for post-acute mortality were consistent with primary analysis.

Conclusions: Epidemiologic criteria for a causal relationship between sepsis and post-acute mortality were not consistently observed. Additional epidemiologic studies with recent patient level data that address the pre-illness trajectory, confounding, and varying control groups are needed to estimate sepsis-attributable additional risk and modifiable risk factors to design interventional trials.

Keywords: Bias; Causality; Confounding factors (epidemiology); Mortality; Sepsis.

Fig. 1

Approach to the study question. Post-acute mortality was estimated as the difference between one-year mortality and acute mortality, to address the study questions as described (see “Methods”, “Approach”). We assessed whether the studies used cumulative mortality or post-acute mortality in regression models. This was done to identify risk factors that are associated with post-acute mortality, which will help future researchers delineate modifiable risk factors. This approach helps to generate a summary estimate of post-acute mortality and also explores the relationship between acute mortality and post-acute mortality at cohort level

Fig. 2

Flow diagram showing the literature search and results. ^#Non-duplicate articles were identified from the references list of two review articles [9, 10]. QOL quality of life. A/c refers to acute

Fig. 3

Random effect meta-analysis of post-acute mortality. First author, reference number and cohort recruitment year for each study are shown. X axis indicates mortality proportions. NR not reported, Recruitment Year refers to recruitment window reported in studies

Fig. 4

a One-sided contour-enhanced funnel (confunnel) plot. The figure either implies potential publication bias or consistency in results (implying causality) across the published studies identified by this systematic review. The confunnel plot adds contours of statistical significance to the standard funnel plot and assesses whether the areas where studies are potentially missing correspond to areas of low statistical significance, the assumption being that studies that do not attain statistical significance boundaries are less likely to be published. b Post-acute mortality versus acute mortality with linear fit superimposed and assessed between study dose-response effects. The equation, R-square statistics of the fit, the sample size and root mean square standard error (RMSE)(s) are also shown (referred to as aaplot and designed by Nicholas J. Cox)

Fig. 5

Additional post-acute hazard with sepsis. To illustrate the differences in additional hazard with sepsis, all four sub-graphs were generated with the same scale on the x axis. Dashed line at hazard ratio 1 is the reference line; shaded area shows the range between 0 and 1. If the same study reported risk-adjusted and unadjusted hazard ratios, these are presented together to highlight confounding. If only proportions are reported, they are presented as dots. Confidence intervals of hazard ratios are shown when reported in studies. a Additional hazard when compared to general population controls. b Additional hazard when compared to hospitalized controls. Hazard ratios associated with single episode of pneumonia are reported for Yende S et al. [40]. c Additional hazard when compared to critically ill controls. d Confounding from studies reporting multiple controls. In Linder et al, the additional risk of sepsis compared to critically ill controls (Model-1) and cardiovascular system (CVS) surgical controls (Model-2) between 1 and 5 years [37]. In Ghelani et al, the additional risk of sepsis compared to critically ill controls (Model-1) and hospital infected controls (Model-2) [30]. In all the graphs, the hazard ratios are either reported by the study or estimated by comparing the sepsis outcomes to reported control populations. HAI healthcare-associated infection, SMR standardized mortality ratio, HR @ 2 yrs refers to hazard ratio at 2 years, Model-1, Model-2 different risk-adjusted models reported in studies. Additional details are provided in Additional file 1: Table S6