Long-term mortality following bloodstream infection

Authors

  • P. J. Lillie,

    Corresponding author
    1. Department of Infectious Diseases and Tropical Medicine, Hull and East Yorkshire Hospitals NHS Trust, Castle Hill Hospital, Cottingham, Hull, UK
    • Corresponding author: P. J. Lillie, Department of Acute Medicine, Kings Mill Hospital, Mansfield Road, Sutton in Ashfield, Nottinghamshire, NG17 4JL, UK

      E-mail: patricklillie@doctors.org.uk

    Search for more papers by this author
  • J. Allen,

    1. Department of Infectious Diseases and Tropical Medicine, Hull and East Yorkshire Hospitals NHS Trust, Castle Hill Hospital, Cottingham, Hull, UK
    Search for more papers by this author
  • C. Hall,

    1. Department of Infectious Diseases and Tropical Medicine, Hull and East Yorkshire Hospitals NHS Trust, Castle Hill Hospital, Cottingham, Hull, UK
    Search for more papers by this author
  • C. Walsh,

    1. Department of Infectious Diseases and Tropical Medicine, Hull and East Yorkshire Hospitals NHS Trust, Castle Hill Hospital, Cottingham, Hull, UK
    Search for more papers by this author
  • K. Adams,

    1. Department of Infectious Diseases and Tropical Medicine, Hull and East Yorkshire Hospitals NHS Trust, Castle Hill Hospital, Cottingham, Hull, UK
    Search for more papers by this author
  • H. Thaker,

    1. Department of Infectious Diseases and Tropical Medicine, Hull and East Yorkshire Hospitals NHS Trust, Castle Hill Hospital, Cottingham, Hull, UK
    Search for more papers by this author
  • P. Moss,

    1. Department of Infectious Diseases and Tropical Medicine, Hull and East Yorkshire Hospitals NHS Trust, Castle Hill Hospital, Cottingham, Hull, UK
    Search for more papers by this author
  • G. D. Barlow

    1. Department of Infectious Diseases and Tropical Medicine, Hull and East Yorkshire Hospitals NHS Trust, Castle Hill Hospital, Cottingham, Hull, UK
    Search for more papers by this author

Abstract

Bloodstream infection is associated with significant short-term mortality, but less is known about long-term outcome. We describe factors affecting mortality up to 3 years after bloodstream infection in a cohort of patients reviewed at the bedside by an infection specialist. Patients seen by the bacteraemia service of our infectious diseases department between June 2005 and November 2008 were included in analyses. Routine clinical data collected at the time of consultation, together with laboratory, demographic and outcome data were analysed to identify factors predicting death at 30 days and 3 years after bloodstream infection. Cox regression models for both time-points were constructed, together with Kaplan–Meier survival curves. In all, 322 bloodstream infections were recorded in 304 patients. The 30-day mortality was 15%, with a 3-year mortality of 49%. At 30 days after bacteraemia, in the Cox regression model, increasing age (p 0.003) and lower serum albumin (p 0.014) were predictive of death. At 3 years, age (p <0.0001) and albumin (p 0.004) remained significant predictors of death, with the presence of vascular disease (p 0.05) also significantly associated with mortality. If temperature was treated as a continuous variable then urea was significant (p 0.044); however, if temperature was categorized into hypothermia and non-hypothermia, then the presence of hypothermia (p 0.008) and chronic renal disease (p 0.034) became significant. There is an appreciable and gradual increase in mortality after an episode of bloodstream infection. Although many factors may not be amenable to intervention, patients at high risk of long-term mortality might require further follow up and assessment for potentially modifiable factors.

Introduction

Bacteraemia (bloodstream infection), whether associated with another site of infection or not, is a common cause of community and healthcare-associated infection and causes high early mortality in the range of 11–30% [1-8]. Most studies have concentrated on in-hospital or early mortality, few have looked at longer-term outcomes, particularly beyond 1 year. One study performed in Israel in the early 1990s showed a 1-year mortality of 48%, with a 4-year mortality of 63% [1]. A study from the same period, performed in Boston, USA, showed a 3-year mortality of 37% [2]. Both of these studies showed that patients who had had bacteraemia had a higher mortality than matched control people who had not. The reason for the apparent excess in longer-term mortality remains unclear, but possible explanations could be that bacteraemia is a marker of co-morbidity (e.g. in patients with renal impairment or malignancy) or it may trigger, for example, a long-term, low-grade inflammatory response, which subsequently leads to either destabilization of an existing co-morbidity or a new one, such as vascular disease. The latter, in particular, raises the possibility of early medical intervention following an episode of bacteraemia to reduce long-term mortality, although this approach is as yet unproven. Given the changes in patient demographics, co-morbidity and antibiotic resistance that have occurred over the last 20 years, little is known of the contemporary long-term mortality after bloodstream infection. We describe here the factors affecting mortality up to 3 years after bloodstream infection, in a UK teaching hospital bacteraemia service cohort.

Methods

Study population

The bacteraemia service at Hull and East Yorkshire Hospitals (a 1500-bed, two-site hospital trust, serving a population of approximately 1.2 million) has been described previously[6]. Briefly, between June 2005 and November 2008, patients with a presumed infecting pathogen isolated from blood cultures were reviewed at the bedside by an infectious diseases physician, either a consultant or a Specialist Trainee (Fellow), without previous solicitation by the patient's clinical team. Patients with haematological malignancy and those with solid organ tumours who were undergoing chemotherapy, patients already on the infectious diseases unit and children aged <16 years were not routinely reviewed because other existing infection services cover these areas. Initial antibiotic advice was provided by either a consultant or a Specialist Trainee (Fellow) in microbiology by telephone, with a structured and more detailed report provided following a bedside consultation by a consultant or Specialist Trainee (Fellow) in infectious diseases (a median of 3 days after blood cultures were taken).

Data collection

Data were recorded or dictated at the time of consultation, using a structured and pre-piloted, paper-based proforma, which also formed the report sent to the patient's clinical team as part of routine clinical practice. These included demographic data (gender and age), clinical observations (pulse, temperature, blood pressure and respiratory rate), relevant examination data (source of infection), microbiological and antibiotic data. These data were entered onto an Excel spreadsheet to which, via the hospital computer system, were added standard laboratory data from the time the blood culture was taken (or within 24 h), data regarding socio-economic deprivation (postal code of patient's address was used to derive the Index of Multiple Deprivation Score, using UK Office of National Statistics census data), co-morbidities and date of death. Data collection and analysis were approved for ongoing quality improvement, service evaluation and audit by the Hull and East Yorkshire Hospitals NHS Trust. No funding was required for this work.

Data analysis

Factors associated with mortality at both 30 days and 3 years were ascertained. Two separate analyses were performed; one using patients with a complete data set and subsequent analyses including patients with missing values, with the missing values imputed using the expectation-maximization method. Initial analyses were univariate, using Mann–Whitney and Fisher's exact tests as appropriate. Factors significant at a level of p ≤0.1 in univariate analyses were entered into a Cox regression model, with backwards removal of variables until the final model was achieved, with co-morbidities being entered into the models to adjust for them. Kaplan–Meier survival curves for significant predictors of 3-year mortality were constructed and analysed using the log-rank test. Analyses were performed using GraphPad Prism 5 and SPSS (version 19.0).

Results

Patient cohort

Three hundred and twenty-two bacteraemic episodes were reviewed in 304 separate patients between June 2005 and November 2008. The median age of patients was 70 years (interquartile range 57–79) and 60.6% were women. The organisms isolated are shown in Table 1, with methicillin-sensitive Staphylococcus aureus, methicillin-resistant S. aureus and Escherichia coli being the commonest single isolates. Of the 322 patients, 263 (81.6%) were receiving effective antibiotic therapy, based on antibiotic sensitivity, before being reviewed by an infectious diseases physician at the bedside, increasing to 292 (90.7%) after review (p <0.0001, McNemar's test). However, receipt of ineffective therapy was not associated with either 30-day (p 0.85) or 3-year (p 0.77) mortality. The overall mortality is shown in Figure 1, with a 30-day mortality of 15% and a 3-year mortality of 48.75%.

Table 1. Common organisms isolated
OrganismNumber isolated
Methicillin-sensitive Staphylococccus aureus (MSSA)94
Methicillin-resistant Staphylococcus aureus (MRSA)51
Coagulase-negative staphylococci47
Streptococci44
Escherichia coli 36
Non-specified coliform29
Enterococci24
Candida spp.9
Pseudomonas sp.7
Klebsiella sp6
Polymicrobial cases40
Figure 1.

Overall survival.

Thirty-day mortality

Table 2 shows the univariate and Cox regression predictors of mortality by 30 days. Both lower serum albumin and increasing age remained significant predictors of mortality in secondary analyses when missing values were imputed using an expectation-maximization method (data not shown).

Table 2. Predictors of 30-day mortality at the time of bacteraemia
VariableDied by 30 days (n = 61)Survived to 30 days (n = 261)Univariate p valueCox regression p valueOR (95% CI)
  1. a

    Values are medians or absolute values (%).

  2. IMDS, Index of Multiple Deprivation Score; MRSA, methicillin-resistant Staphylococcus aureus; BP, blood pressure; CRP, C-reactive protein.

Age (years)a7768<0.00010.0031.046 (1.016–1.078)
IMDSa21.830.9>0.1N/AN/A
MRSA (compared with all other isolates)20 (32.8)31 (11.9)0.0003>0.1N/A
Pulse (/min)a9094>0.1N/AN/A
Systolic BP (mmHg)a115120>0.1N/AN/A
Respiratory rate (/min)a20190.0942>0.1N/A
Temperature (°C)a37.938.0>0.1N/AN/A
Creatinine (μmol/L)a144119>0.1N/AN/A
Urea (mmol/L)a12.58.40.0011>0.1N/A
Haemoglobin (g/dL)a10.210.80.0452>0.1N/A
White cell count (×109)a11.912>0.1N/AN/A
Albumin (g/L)a22260.00010.0190.928 (0.872–0.988)
CRP (mg/L)a170.5165.5>0.1N/AN/A
Initial antibiotic active49 (80.3)214 (82)>0.1N/AN/A
Vascular co-morbidity18 (29.5)91 (34.9)>0.1N/AN/A
Renal co-morbidity16 (26.2)76 (29.1)>0.1N/AN/A
Respiratory co-morbidity4 (6.6)34 (13)>0.1N/AN/A
Diabetic10 (16.4)55 (21.1)>0.1N/AN/A

Three-year mortality and survival curves

The complete data set without missing values, was used to construct a Cox regression model, the results of which are shown in Table 3. When the data set was expanded to include missing variables that had been imputed as described previously, age and albumin remained significant predictors of mortality, with higher respiratory rate (p 0.021, OR 1.033/breaths/minute), lower body temperature (p 0.045, OR 0.868/°C increase) and chronic renal disease (p 0.047 OR 0.684 for absence of renal disease) but creatinine replaced urea as predictive of mortality (p 0.046, OR 1.001/μmol increase), with vascular disease becoming non-significant. In a second analysis, where temperature was treated as a binary variable (hypothermia/non-hypothermia), the presence of hypothermia was strongly associated with 3-year mortality (p 0.008, OR 2.738), replacing markers of renal dysfunction. Kaplan–Meier survival curves for quartiles of age, albumin, urea and temperature are shown in Figure 2.

Table 3. Predictors of 3-year mortality at the time of bacteraemia, temperature treated as a continuous variable
VariableDied by 3 years (n = 158)Survived to 3 years (n = 164)Univariate p valueCox regression p valueOR (95% CI)
  1. a

    Values are medians or absolute values (%).

  2. IMDS, Index of Multiple Deprivation Score; MRSA, methicillin-resistant Staphylococcus aureus; BP, blood pressure; CRP, C-reactive protein.

Age (years)a7563<0.0001<0.00011.034 (1.017–1.051)
IMDSa22.0224.63>0.1N/AN/A
MRSA (compared with all other isolates)38 (24.0)13 (7.9)0.0002>0.1N/A
Pulse (/min)a9591>0.1N/AN/A
Systolic BP (mmHg)a123120>0.1N/AN/A
Respiratory rate (/min)a20180.0063>0.1N/A
Temperature (°C)a38.038.10.0389>0.1N/A
Creatinine (μmol/L)a162105<0.0001>0.1N/A
Urea (mmol/L)a12.056.9<0.00010.0441.023 (1.001–1.045)
Haemoglobin (g/dL)a10.111.10.0009>0.1N/A
White cell count (×109)a11.8512.15>0.1N/AN/A
Albumin (g/L)a24270.00010.0040.949 (0.916–0.983)
CRP (mg/L)a165.5166>0.1N/AN/A
Initial antibiotic active130 (82.3)133 (81.1)>0.1N/AN/A
Vascular co-morbidity59 (37.3)50 (30.5)>0.10.050.635 (0.403–1.001)
Renal co-morbidity52 (32.9)40 (24.4)>0.1>0.1N/A
Respiratory co-morbidity19 (12.0)19 (11.6)>0.1>0.1N/A
Diabetic32 (20.2)33 (20.1)>0.1>0.1N/A
Figure 2.

Kaplan–Meier survival curves of age, urea, albumin and temperature.

Discussion

There are relatively few data regarding long-term outcome after bloodstream infection from the current era. Although some data from population-based studies are available, the other commonly cited papers are now nearly 20 years old and may not reflect the increasingly elderly population with high levels of co-morbidity. Conversely, with improved infection input and specialist review, it is possible that outcomes may be relatively improved compared with earlier reports. Of the factors not associated with mortality in our cohort, perhaps the most interesting and surprising is the lack of effect of inappropriate antibiotic therapy. Although this was shown to influence long-term mortality in the study from Israel [1], this may be less evident in our cohort given the large proportion of patients who received appropriate empirical antibacterial therapy; we were not able to accurately measure the time between onset of illness and receipt of empiric or appropriate antibiotic therapy. Another explanation might be that our study was not large enough to detect a small, but clinically relevant difference. The data regarding socio-economic deprivation and infection are limited, although our unit recently found that the incidence of invasive pneumococcal disease was higher in those with higher Index of Multiple Deprivation Scores (an index of global deprivation ranked 0–100, with higher scores indicating greater deprivation) [9]. The data from this cohort would however seem to suggest that this may not translate into a poorer outcome when infection occurs.

Our data confirm the effects of age on both short-term and long-term mortality after bloodstream infection [1-4, 8], together with the known association of low serum albumin with higher mortality at both time-points [1, 10]. Whether the latter can be modified by, for example, better management of existing co-morbidities that cause hypoalbuminaemia or better nutritional support is debatable, but our results suggest that physicians should review both of these in the recovery period after an episode of bacteraemia. Although mortality was considerably higher in older age quartiles, there was still notable mortality in the two younger quartiles with a steady increase in mortality over the 3-year follow-up period. Our results point to simple markers of renal function (urea or creatinine) as being important predictors of longer-term mortality. Whereas the effect of elevated creatinine has been noted previously [1], this may be of clinical utility as a target for intervention. As impaired renal function is an important marker for cardiovascular disease [11, 12], and a recent study from Canada showed that nearly 25% of deaths in the year following an episode of bacteraemia were due to vascular disease [3], it may be that the group of patients with impaired renal function are at higher risk of dying, possibly as the result of vascular events. This may also explain one of the reasons for the mortality benefit detected in patients with bacteraemia who were taking statins, particularly as the benefit may not be evident initially, but becomes apparent during longer follow up [13]. Bloodstream infection could therefore be a surrogate marker of risk of death from one or more other causes, or may precipitate or deteriorate certain disease states such as vascular disease or renal dysfunction, perhaps by the effect of infection-related inflammation on the progression of atherosclerosis [14, 15]. This may also be one of the reasons why patients with known vascular or renal disease in our study were at a greater risk of death over the 3-year period than those with other co-morbidities. The other potentially interesting finding from our cohort is the association between low body temperature at the time of bacteraemia and increased mortality at 3 years. A recent study in critical care patients with and without sepsis, found that whereas high fever was associated with poor short-term outcome, a moderate pyrexia was associated with reduced mortality, and anti-pyretic treatment was associated with a poorer outcome [16]. We have previously noted the effect of low temperature on early mortality in patients with pneumonia and in an early review of this bacteraemia cohort [17], and an older study found that patients who had experienced chills with their infection had a reduced mortality in long-term follow-up [18]. Why this should be associated with long-term, but not short-term, mortality is unclear, but potentially it could be a marker of an immune response, relating to an impaired ability to mount a fever to subsequent infections and as a result an increased risk of death.

Our study has several limitations. Although some data were collected at the time of initial presentation, retrospective supplementation was required. Additionally, as routinely collected data were used some data were inevitably not recorded and it is therefore possible that this may have affected our results. We did not include a control group and we are therefore unable to comment on mortality in a comparable population that had not suffered bacteraemia. Three earlier studies looking at long-term mortality after bacteraemia, however, used either a case–control design or compared bloodstream infection patients with those without infection [1, 2, 18] and found that patients with bacteraemia had increased mortality throughout follow up compared with the non-bacteraemic controls. We did not collect data on cause of death because this was not readily available, although we hope to collect these data in a subsequent study. These data would be necessary to fully understand the causes of death over time and the interventions that may be of use in modifying this. However, a recent population-based survey of 1-year mortality in patients who had community-acquired bloodstream infections, found that 39% died of malignancy and 24% from cardiovascular disease [3]. Because of the nature of our service, a low percentage of patients in this cohort had progressive haematological or other malignancy. Despite this, mortality in the follow-up period was still considerable suggesting that there may be an opportunity to address vascular or other causes of mortality and morbidity, perhaps by some form of risk-factor-based assessment of those patients who survive the initial period after a bacteraemic episode.

Our data, and that of other researchers, suggest that the time for a large and detailed prospective study aimed at understanding better the causes of mortality after bloodstream infection is long overdue and would be useful in providing evidence about who to follow up, the nature of the follow up required and the opportunity for cost-effective therapeutic intervention. In the interim, as specialist infection input has been shown to have a beneficial impact in several studies of bloodstream infection and may improve outcomes up to 1 year after bacteraemia [19-22], it would appear sensible to consider the appreciable and poorly recognized longer-term mortality after an episode of bloodstream infection by reviewing the stability of existing co-morbidities combined with an assessment of future cardiac risk.

Transparency Declaration

All authors declare that there are no conflicts of interest.

Ancillary