Intensive Care Med (2017) 43:48–58
DOI 10.1007/s00134-016-4578-y
ORIGINAL
Increased incidence ofco-infection
incritically ill patients withinuenza
Ignacio Martin-Loeches
1,2*
, Marcus J Schultz
3
, Jean-Louis Vincent
4
, Francisco Alvarez-Lerma
5
, Lieuwe D. Bos
3
,
Jordi Solé-Violán
6
, Antoni Torres
7
and Alejandro Rodriguez
8,9
© 2016 Springer-Verlag Berlin Heidelberg and ESICM
Abstract
Background: Co-infection is frequently seen in critically ill patients with influenza, although the exact rate is
unknown. We determined the rate of co-infection, the risk factors and the outcomes associated with co-infection in
critically ill patients with influenza over a 7-year period in 148 Spanish intensive care units (ICUs).
Methods: This was a prospective, observational, multicentre study. Influenza was diagnosed using the polymerase chain
reaction. Co-infection had to be confirmed using standard bacteriological tests. The primary endpoint of this analysis was
the presence of community-acquired co-infection, with secondary endpoints including ICU, 28-day and hospital mortality.
Results: Of 2901 ICU patients diagnosed with influenza, 482 (16.6 %) had a co-infection. The proportion of cases of co-
infection increased from 11.4 % (110/968) in 2009 to 23.4 % (80/342) in 2015 (P < 0.001). Compared with patients with-
out co-infection, patients with co-infection were older [adjusted odds ratio (aOR) 1.1, 95 % confidence interval 1.1–1.2;
P < 0.001] and were more frequently immunosuppressed due to existing HIV infection (aOR 2.6 [1.5–4.5]; P < 0.001) or
preceding medication (aOR 1.4 [1.1–1.9]; P = 0.03). Co-infection was an independent risk factor for ICU mortality (aOR
1.4 [1.1–1.8]; P < 0.02), 28-day mortality (aOR 1.3 [1.1–1.7]; P = 0.04) and hospital mortality (aOR 1.9 [1.5–2.5]; P < 0.001).
Conclusions: Co-infection in critically ill patients with influenza has increased in recent years. In this Spanish cohort,
age and immunosuppression were risk factors for co-infection, and co-infection was an independent risk factor for
ICU, 28-day and hospital mortality.
Keywords: Influenza, Co-infection, Risk factors, Outcome, Intensive care
Introduction
Severe acute respiratory infection with H1N1 influenza
emerged in 2009 and was associated with high mortality
rates [
1]. e use of early antiviral therapy was one of the
cornerstones of treatment in severe respiratory infection
with influenza, and was associated with better outcomes.
Many patients were suspected of having a community-
acquired co-infection [2]. erefore, it was recommended
to consider antibacterial treatment on admission, until an
accompanying bacterial infection was excluded [
3].
Previous studies suggested temporal relationships
between influenza and co-infection [
4]. Indeed, retrospec
-
tive analysis of lung biopsies of patients who died from
influenza in the pandemic of 1918 suggested bacterial
super-infections of the lungs [
5]. is was also found for
the influenza pandemic in 1957 [
6]. Staphylococcus aureus,
Streptococcus pneumoniae, and Haemophilus influen
-
zae are the most-cited bacterial causes of co-infection.
*Correspondence: drmartinloeches@gmail.com
1
Multidisciplinary Intensive Care Research Organization (MICRO),
Wellcome Trust-HRB Clinical Research, Department of Clinical Medicine,
Trinity Centre for Health Sciences, St James’s University Hospital, Dublin,
Ireland
Full author information is available at the end of the article
Take-home message: Based on the data presented, co-infection is
a very frequent complication in critically ill patients with influenza.
Streptococcus pneumoniae is still the most frequent pathogen with
higher rates of potentially resistant pathogens. Immunosuppression is a
risk factor for co-infection.
H1N1 SEMICYUC Working Groupinvestigators are listed in Appendix
section .
49
However, Aspergillus spp. have also been identified as
important pathogens [
7]. e exact rate of co-infection and
its risk factors, however, remained largely unknown. ere
is also a lack of understanding of the potential impact of
co-infection on the outcome of patients with influenza [
8].
We hypothesized that community-acquired co-infec
-
tion is common and independently associated with mor-
tality in intensive care unit (ICU) patients with influenza.
erefore, we reanalysed the data of a prospective obser
-
vational study on influenza in critically ill patients in
Spain from 2009 to 2015, covering four influenza seasons.
In addition, we determined risk factors for co-infection.
Patients andmethods
Study design
is was a prospective, observational study conducted from
2009 to 2015 in a large cohort of ICUs in Spain. ere were
four seasons of influenza, based on epidemic threshold rates
developed by the Spanish Ministry of Health [
9]: one in
2009 during the influenza H1N1 pandemic, one in the win
-
ter of 2010 to 2011, one in the winter of 2014, and one in the
winter of 2015. During these four seasons (2009, 2010, 2014
and 2015), all patients admitted to the ICU with influenza-
like symptoms were systematically tested to confirm res
-
piratory infection with influenza A or bacterial pathogens.
Local investigators registered data of consecutive influenza
patients in a national registry created by the Spanish Society
of Intensive Care Medicine. e institutional review board of
Joan XXIII University Hospital approved the original study
(IRBref#11,809) and waived the requirement for patients to
give individual informed consent due to the observational
nature of the study. e participation of 148 ICUs meant that
we could monitor and prospectively follow approximately
80% of the patients admitted to Spanish ICUs with influenza.
Inclusion andexclusion criteria
is reanalysis did not use inclusion or exclusion cri-
teria other than those employed in the original study.
However, patients under the age of 16years and patients
admitted from nursing homes or other healthcare facili
-
ties were excluded.
Standard care andcollection ofsamples fordiagnostic
purposes
e Ministry of Health and competent authorities in
Spain intensively monitored and audited management
of influenza in the national ICUs. Standardized guide
-
lines were used in all centres [
10]. Oseltamivir therapy
was considered early treatment (ET) if administered
within 2 days of the onset of influenza symptoms [
2],
and empirical antibiotics were started after obtain
-
ing a nasopharyngeal swab, endotracheal aspirates and
blood. Nasopharyngeal swabs were used for viral testing,
respiratory secretions for quantitative cultures, and blood
samples were cultured and used for serological tests.
Bronchoalveolar lavage fluids were not obtained because
of the high risk of generating aerosols. If present, pleural
effusions were punctured for microbiological culture.
Denitions
Co-infection was suspected if a patient had an acute onset
of signs and symptoms suggesting lower respiratory tract
infection, with radiographic evidence of a pulmonary
infiltrate that had no other known cause [
11]. Co-infec
-
tion had to be laboratory confirmed using the Centers for
Disease Control and Prevention criteria. If the co-infec
-
tion was diagnosed within 2days of hospital admission,
it was considered a community-acquired co-infection.
e diagnosis was considered definitive if respiratory
pathogens were isolated from blood or pleural fluid and if
serological tests confirmed a fourfold increase of atypical
pathogens, including Chlamydia spp., Coxiella burnetii
and Moraxella pneumoniae. Respiratory aspergillosis was
considered a ‘definite’ diagnosis only if Aspergillus spp.
were identified on histopathology. e diagnosis was con
-
sidered ‘probable’ if respiratory pathogens were isolated in
endotracheal aspirates. Respiratory aspergillosis was con
-
sidered a ‘probable’ diagnosis in the presence of halo or
air-crescent signs on computed tomography of the lungs
with positive determination of serum galactomannan,
and ‘possible’ if Aspergillus spp. were found in endotra
-
cheal aspirates [
7]. Appropriateness of antibiotic therapy
was defined as administration of at least one antimicrobial
agent effective against the isolated pathogen.
Study endpoints
e primary endpoint of this analysis was the presence of
community-acquired co-infection. Secondary endpoints
included ICU, 28-day and hospital mortality, the num
-
ber of ventilator-free days and patient’s survival at day
28. Ventilator-free days were defined as days of successful
and complete weaning from mechanical ventilation up
to day 28. For subjects who died during this period, the
ventilator-free days were counted as 0 [
12].
Analysis plan
Firstly, the proportion of cases and rate of co-infection
were determined. is rate was calculated per season and
comparisons made among seasons. e first season acted
as reference season, and calculations were carried out
using logistic regression and odds ratios with confidence
intervals. is was repeated for each pathogen.
Associations between co-infection and the clini
-
cal outcome measures were studied by logistic regres-
sion and corrected for potential confounders, which
included gender, age, disease severity (APACHE II
50
score), comorbidities (asthma, chronic obstructive pul-
monary disease, chronic heart failure, chronic kidney
disease, haematological disease, diabetes mellitus, HIV
and immunodeficiency), pregnancy, obesity, oseltamivir
treatment, appropriateness of initial antibiotic therapy,
acute kidney injury, need for renal replacement therapy,
need for invasive mechanical ventilation and presence of
septic shock. Potential chronic comorbidities and states
that were risk factors for the occurrence of co-infection
included asthma, chronic obstructive pulmonary disease,
pregnancy, obesity, diabetes mellitus, HIV and immuno
-
deficiency and were also identified by logistic regression.
Both analyses started with all potential confounders and
backward selection based on P value was performed.
Statistical analysis
Discrete variables are expressed as counts with percentage
and continuous variables, as means and standard devia
-
tion (SD) or as medians with the 25th to 75th interquartile
range (IQR). Parametric or nonparametric tests were used
for continuous variables as appropriate after the normal
-
ity of the distribution had been tested. A P value <0.05
was considered significant. Differences in patients’ demo
-
graphic and clinical characteristics were assessed using
the Chi squared test or Fisher’s exact test for categorical
variables and Student’s t test or the Mann–Whitney U test
for continuous and ordinal variables, where appropriate.
Trends in the rate and proportion of cases of co-infec
-
tion and causative pathogens were assessed by logistic
regression, with 2009 selected as the year of reference. A
stepwise backward-selection logistic regression analysis
was performed to study the association with outcome.
Variable selection was done based on P values (<0.10).
For all models that had ICU mortality as the dependent
variable, the APACHE II score was included as covariate,
irrespective of the associated P value. Potential explana
-
tory variables were checked for co-linearity prior to
inclusion in the regression models using tolerance and
variance inflation factor.
All statistical analysis was performed using SPSS v.20.0
for Mac (IBM Corp., Armonk, NY, USA).
Results
Patients
A total of 2901 ICU patients with polymerase chain
reaction (PCR)-confirmed influenza were included
and analysed (Table
1; Fig. 1); 1581 patients were male
(59.1 %) and the mean age was 51.6 ± 15.9 years. All
patients were severely ill, with a mean APACHE II score
of 16.1±7.6. e mean ICU and hospital length of stay
were 13.5±14.6 and 21.4±18.8days, respectively. ICU
mortality, 28-day mortality and hospital mortality were
22.1, 19.7 and 26.2 %, respectively. S. pneumoniae was
the bacterium most often identified, followed by Pseu
-
domonas aeruginosa and methicillin-sensitive S. aureus
(MSSA) (Table
2).
Relative rate ofco-infection
Overall, co-infection was diagnosed in 16.6% of patients.
An increasing trend was observed over the years of the
study: 11.4 % in 2009, 17.3 % in 2010, 18.8 % in 2014,
and as high as 23.4% in 2015. e odds ratios (OR) for
co-infection were 1.6 [1.2–2.2], 1.8 [1.4–2.4] and 2.4
[1.7–3.3] in 2010, 2014 and 2015 respectively (Fig.
2). A
significant increase in the rates of S. pneumoniae, P. aer
-
uginosa, MSSA and H. influenzae co-infection over the
years was found (Fig.
2). e relative frequency of Asper
-
gillus spp. did not increase over the years of the study
(Fig.
2).
Risk factors forco-infection
Comorbidities in patients with and without co-infection
are shown in Table 3. e likelihood of co-infection
increased with age (adjusted OR 1.01 [1.01–1.02]), pre
-
ceding HIV infection (adjusted OR 2.6 [1.5–4.5]) and
immunosuppressive medication (adjusted OR 1.4 [1.02–
1.9]). e numbers of days from onset of clinical symp
-
toms to hospital admission, from hospital admission to
start of antiviral therapy, and from onset of clinical symp
-
toms to start of antiviral therapy did not differ between
patients with and without co-infection (Supplementary
Table1) (Fig.
3).
Clinical outcomes
ICU mortality was not significantly different among influ-
enza types (A-H1N1: 21.9%; A-H3N2: 24.2%; B: 18.9%;
Table 1 Characteristics of patients included in the study
(N=2901)
Variable n=2901
Sex (male) (n, %) 1706, 59.1 %
Age (mean ± SD) 51.6 ± 15.9
APACHE II score (mean ± SD) 16.1 ± 7.6
Asthma (n, %) 291, 10.1 %
Chronic obstructive pulmonary disease (n, %) 608, 21.2 %
Chronic heart failure (n, %) 331, 11.5 %
Chronic kidney disease (n, %) 246, 8.6 %
Haematological diseases (n, %) 197, 6.9 %
Pregnancy (n, %) 109, 3.8 %
Obesity (n, %) 962, 33.5 %
Obesity (BMI > 40) (n, %) 406, 14.1 %
Diabetes mellitus (n, %) 477, 16.6 %
HIV (n, %) 70, 2.4 %
Immunodeficiency (n, %) 311, 10.8 %
51
C: 18.8%; P=0.7) for patients with or without co-infec-
tion. Patients with co-infection more often received early
oseltamivir treatment than those without co-infection
(1428/2419, 59% vs. 314/482, 65.1%; P=0.01). However,
early oseltamivir treatment was not associated with a sig
-
nificantly lower ICU mortality in patients with (171/259;
66.6 vs. 122/192; 63.5%; P=0.6) or without co-infection
(1187/1982; 59.9 vs. 419/702; 59.7%; P=0.9),. Continu
-
ous renal replacement therapy, invasive mechanical ven-
tilation and immunosuppression were independently
associated with ICU mortality; the adjusted OR (aOR)
values are summarized in Table
4. Co-infection was also
independently associated with increased ICU mortal
-
ity (aOR 1.4, 95% CI 1.1–1.8; P<0.02; Table
4), 28-day
mortality (aOR 1.3, 95% CI 1.1–1.7; P=0.04) and hos
-
pital mortality (aOR 1.9 95% CI 1.5–2.5; P<0.001). e
mean number of ventilator-free days and survival at day
28 were lower in patients with co-infection (12.9, IQR
10.6–14.2 vs. 10.3, IQR 9.6–12.1; P<0.001). A subgroup
analysis showed that only positive cultures for P. aerugi
-
nosa (aOR 2.6, 95% CI 1.3–5.1; P=0.004) or Aspergillus
spp. (aOR 4.1, 95 % CI 1.9–9.6; P = 0.001) were inde-
pendent risk factors for ICU mortality when corrected
for APACHE II score.
Discussion
We have reported data from the largest prospective study
to date evaluating patients with severe influenza admitted
to the ICU. e most significant finding was the high rate
of co-infection, complicating the clinical course in one
out of six critically ill patients with influenza. Moreover,
the rate of co-infection steadily increased over the study
period and was independently associated with increased
mortality.
Previous studies have provided conflicting results
regarding the impact of co-infection on patient outcome.
For example, a study performed in Europe, identifying
S. pneumoniae as the most frequent pathogen isolated
in co-infection, demonstrated no significant association
between co-infection and ICU mortality after adjustment
Fig. 1 Inclusion diagram and rate of bacterial co-infection per epi-
demic period. Patients from four influenza epidemics were included.
The total number of patients with a positive PCR for influenza was
2901. Of these, 482 had a bacterial co-infection. The lower panel gives
the rate of co-infection in each period. The error bars indicate the
95 % confidence interval
Table 2 Numbers and proportions of the pathogens iso-
lated in critically ill patients with bacterial co-infection
(N=482)
MRSA methicillin-resistant Staphylococcus aureus, MSSA methicillin-sensitive
Staphylococcus aureus
*Histopathological conrmation
**CT ndings compatible with invasive aspergillosis
+
Percentage of all microorganisms
Pathogen N %
+
Denitive Probable Possible
S. pneumoniae 246 51.04 17 229 0
P. aeruginosa 55 11.4 2 53 0
MSSA 42 8.7 2 40 0
Aspergillus spp. 35 7.2 2* 25** 8
H. influenza 17 3.5 0 17 0
A. baumannii 14 2.9 0 14 0
MRSA 12 2.4 3 9 0
K. pneumoniae 12 2.4 1 11 0
E. coli 11 2.2 1 10 0
L. pneumophila 5 1.1 1 4 0
S. marcescens 4 0.8 1 3 0
S. hominis 4 0.8 4 0 0
E. cloacae 4 0.8 2 2 0
P. jirovecii 4 0.8 0 4 0
M. pneumoniae 4 0.8 1 3 0
C. pneumoniae 3 0.6 1 2 0
M. tuberculosis 3 0.6 0 3 0
S. maltophila 2 0.4 0 2 0
K. oxytoca 2 0.4 0 2 0
M. morganii 1 0.2 0 1 0
Shewanella spp. 1 0.2 0 1 0
B. fragilis 1 0.2 0 1 0
Nocardia spp. 1 0.2 0 1 0
52
for confounding factors [13]. In contrast, a retrospective
study analysing 683 critically ill patients in the USA with
confirmed or probable 2009 influenza A, found that bac
-
terial co-infection, especially with S. aureus, was associ-
ated with significantly higher mortality [
14]. e main
differences between these studies were that in the USA
study only 62.1% of the patients had confirmed co-infec
-
tion and there was a higher rate of S. aureus.
All the studies published to date in critically ill patients
have focused on only one influenza season, the vast
majority of them on the 2009–2010 pandemic season
[
14–19]. Some studies also attempted to analyse the
occurrence and impact of bacterial organisms compli
-
cating critical care illness during the previous 12months
[
20]. In the current study we present the clinical charac
-
teristics and trend of co-infection over the past 7 years
(2009–2015), providing useful information for the man
-
agement of patients with severe influenza.
Studies analysing the frequency of influenza and bac
-
terial co-infection have reported high heterogeneity. A
recent systematic review and meta-analysis of 27 studies
analysed the frequency of bacterial co-infection in influ
-
enza patients. e results from these studies were highly
variable, ranging from 2 to 65%, although the majority
of studies ranged between 11 and 35% [
21]. Our results
show a significant increase in occurrence from 11.4 %
in 2009 to 23.4% in 2015. e most frequent pathogen
identified in the seven-year period was S. pneumoniae
followed by P. aeruginosa and MSSA. In the last few years
the rate of isolation of S. pneumoniae has been declin
-
ing and the rates of P. aeruginosa and H. influenzae have
increased. It is worth mentioning the reappearance of
Fig. 2 Odds ratios for co-infection, stratified by pathogen. Odds ratio and 95 % confidence intervals are shown per epidemic period for all co-
infection (upper left) and per pathogen. The dotted line indicates an odds ratio of 1. If the error bars cross this line, the rate is not significantly different
from the rate in 2009, the reference year