Clinical Infectious Diseases

MAJOR ARTICLE

Association Between Infectious Diseases Consultation and

Mortality in Hospitalized Patients With Gram-negative
Bloodstream Infection: A Retrospective Population-wide
Cohort Study
Sean W. X. Ong,1,2,3,4, Jin Luo,4 Daniel J. Fridman,4 Samantha M. Lee,4 Jennie Johnstone,5,6,7 Kevin L. Schwartz,4,5,8,9 Christina Diong,4 Samir N. Patel,7,8
Derek R. MacFadden,10 Bradley J. Langford,5,8 Steven Y. C. Tong,11,12 Kevin A. Brown,4,5,8 and Nick Daneman1,3,4,8
1
Institute of Health Policy, Management and Evaluation, University of Toronto, Toronto, Ontario, Canada; 2Faculty of Medicine, Dentistry and Health Sciences, University of Melbourne, Melbourne,

Downloaded from https://academic.oup.com/cid/article/79/4/855/7676103 by guest on 30 October 2024

Victoria, Australia; 3Division of Infectious Diseases, Sunnybrook Health Sciences Centre, Toronto, Ontario, Canada; 4ICES, Toronto, Ontario, Canada; 5Dalla Lana School of Public Health, University of
Toronto, Toronto, Ontario, Canada; 6Division of Infectious Diseases, Sinai Health, Toronto, Ontario, Canada; 7Department of Laboratory Medicine and Pathobiology, University of Toronto, Toronto,
Ontario, Canada; 8Public Health Ontario, Toronto, Ontario, Canada; 9Li Ka Shing Knowledge Institute, Unity Health Toronto, Toronto, Ontario, Canada; 10Division of Infectious Diseases, Ottawa
Hospital Research Institute, Ottawa, Ontario, Canada; 11Department of Infectious Diseases, University of Melbourne, Peter Doherty Institute for Infection and Immunity, Melbourne, Victoria,
Australia; and 12Victorian Infectious Diseases Service, Royal Melbourne Hospital, Peter Doherty Institute for Infection and Immunity, Melbourne, Victoria, Australia

Objectives. Data supporting routine infectious diseases (ID) consultation in gram-negative bloodstream infection (GN-BSI) are
limited. We evaluated the association between ID consultation and mortality in patients with GN-BSI in a retrospective population-
wide cohort study in Ontario using linked health administrative databases.
Methods. Hospitalized adult patients with GN-BSI between April 2017 and December 2021 were included. The primary outcome
was time to all-cause mortality censored at 30 days, analyzed using a mixed effects Cox proportional hazards model with hospital as a
random effect. ID consultation 1–10 days after the first positive blood culture was treated as a time-varying exposure.
Results. Of 30 159 patients with GN-BSI across 53 hospitals, 11 013 (36.5%) received ID consultation. Median prevalence of ID
consultation for patients with GN-BSI across hospitals was 35.0% with wide variability (range 2.7%–76.1%, interquartile range 19.6%–
41.1%). In total, 1041 (9.5%) patients who received ID consultation died within 30 days, compared to 1797 (9.4%) patients without ID
consultation. In the fully adjusted multivariable model, ID consultation was associated with mortality benefit (adjusted hazard ratio [HR]
0.82, 95% confidence interval [CI] .77–.88, P < .0001; translating to absolute risk reduction of −3.8% or number needed to treat [NNT] of
27). Exploratory subgroup analyses of the primary outcome showed that ID consultation could have greater benefit in patients with high-
risk features (nosocomial infection, polymicrobial or non-Enterobacterales infection, antimicrobial resistance, or non-urinary tract source).
Conclusions. Early ID consultation was associated with reduced mortality in patients with GN-BSI. If resources permit, routine ID
consultation for this patient population should be considered to improve patient outcomes.
Keywords. bloodstream infections; gram-negative bacteria; health services research; infectious diseases; health administrative data.

BACKGROUND from 12% to 25% at 30 days [1–3]. An international surveil­

lance program showed that the proportion of BSIs caused by
Gram-negative bloodstream infections (GN-BSI) are common
and are associated with high mortality, with estimates ranging gram-negative bacilli (GNB), in particular multi-drug resistant
(MDR) GNB, has increased over the 20-year period from 1997
to 2016 [4]. The burden of GN-BSI will likely continue to in­

Received 27 March 2024; editorial decision 09 May 2024; published online 17 May 2024
crease over time with increasing antimicrobial resistance, cou­
Correspondence: S. W. X. Ong, Institute of Health Policy, Management and Evaluation, pled with an increasingly susceptible aging population [5, 6]. It
University of Toronto, 155 College St 4th Floor, Toronto, ON M5T 3M6, Canada (s.ong@
mail.utoronto.ca).
is thus essential that we find ways to optimize management and
Clinical Infectious Diseases ®
2024;79(4):855–63 improve clinical outcomes.
© The Author(s) 2024. Published by Oxford University Press on behalf of Infectious Diseases
The ”bundle-of-care” approach, comprising a checklist of in­
Society of America.
This is an Open Access article distributed under the terms of the Creative Commons Attribution- terventions to improve quality-of-care, has been used exten­
NonCommercial-NoDerivs licence (https://creativecommons.org/licenses/by-nc-nd/4.0/), which
sively in the management of Staphylococcus aureus and
permits non-commercial reproduction and distribution of the work, in any medium, provided
the original work is not altered or transformed in any way, and that the work is properly cited. Candida BSI, and has been associated with improved clinical
For commercial re-use, please contact [email protected] for reprints and translation rights for
reprints. All other permissions can be obtained through our RightsLink service via the outcomes including mortality [7–10]. A component common
Permissions link on the article page on our site—for further information please contact journal­ to these care bundles is routine infectious diseases (ID) consul­
[email protected].
https://doi.org/10.1093/cid/ciae282 tation, which has also been independently associated with

Infectious Diseases Consultation in Gram-Negative Bloodstream Infection • CID 2024:79 (15 October) • 855
reduced mortality in these 2 infections in large cohort and ID consultations in the Ontario Health Insurance Plan, which
quasi-experimental pre-post studies [11–14]. can only be charged by accredited ID specialists. To mitigate
In contrast, the data supporting routine ID consultation in immortal time bias, we treated ID consultation as a time-
GN-BSI are less robust. A prior systematic review examined dependent exposure, such that patients were considered in
this question and identified six studies, however five of these fo­ the non-exposed group (and contributed unexposed person-
cused only on Pseudomonas or MDR-GNB infections rather time in the Cox model) until the day of ID consultation (as de­
than GN-BSI in general [15]. Two cohort studies found a signif­ termined by the date of billing code), after which they were
icant mortality benefit associated with ID consultation in switched over to the exposed group [21].
GN-BSI with hazard ratios (HR) of 0.60 and 0.33 [16, 17]; how­
ever, these studies did not account for immortal time bias and
Outcomes
may have thus resulted in over-inflated effect estimates [18].
The collection date of the first positive blood culture that diag­
To re-examine this question, we conducted a retrospective
nosed the GN-BSI was considered the index date and start of
population-wide cohort study of patients with GN-BSI in
the follow-up period. The primary outcome was time to all-
Ontario, Canada. Our primary objective was to determine if

Downloaded from https://academic.oup.com/cid/article/79/4/855/7676103 by guest on 30 October 2024

cause mortality censored at 30 days, counted from the index
ID consultation was associated with a reduction in 30-day all-
date. Secondary outcomes were 90-day all-cause mortality,
cause mortality in adult patients hospitalized with GN-BSI.
and—for the sub-group of patients ≥65 years old—the duration
of oral antibiotics prescribed on discharge from index hospital­
METHODS ization. We could only explore antibiotic prescriptions on dis­
Study Setting and Data Sources charge in this sub-group as population-level data are not
The study cohort has been previously described in a separate available in Ontario for inpatient medications or outpatient
paper that evaluated the impact of follow-up blood culture col­ medications in those <65 years old.
lection on mortality in patients with GN-BSI [19]. This
population-wide cohort comprised all hospitalized adult pa­ Statistical Analyses
tients >18 years old with GN-BSI between April 2017 and To describe variation in ID consultation practices across hospi­
December 2021 in Ontario, Canada. Data were obtained from tals, we calculated the proportion of patients receiving ID con­
multiple linked health administrative databases at ICES, sultation; we tested if this was associated with size of hospital
Ontario. ICES is an independent, non-profit research institute (measured by total number of patients with GN-BSI) using
whose legal status under Ontario’s health information privacy Pearson’s correlation test. To account for clustering by site,
law allows it to collect and analyse healthcare and demographic the primary outcome was analyzed using a frailty model, an ex­
data, without consent, for health system evaluation and im­ tension of the Cox proportional hazards model with hospital as
provement. Blood culture data were obtained from the a random effect (ie, each hospital could have a varying baseline
Ontario Laboratories Information System (OLIS), which com­ hazard, also known as a frailty term) [22]. We identified a priori
bines microbiology laboratory data from >100 laboratories confounders based on clinical reasoning for inclusion in the
across Ontario in a single repository [2, 20]. The full list of model. These variables were included as time-fixed covariates:
data sources is reported in Supplementary Methods. patient age, sex, material resources index, Deyo-Charlson
comorbidity index [23], immunosuppression, nosocomial in­
Site and Patient Selection fection, infection source, pathogen sub-group, and antimicro­
To minimize confounding by hospital, we only included hospi­ bial resistance. Disease severity, as measured by intensive
tals with ID services available, defined as having at least 5 ID care unit (ICU) admission, was treated as a time-varying covar­
consultations during the 5-year study period. We also excluded iate. Detailed covariate definitions are specified in the
patients from hospitals with fewer than 25 patients with Supplementary Methods. To quantify potential residual con­
GN-BSI in total as these were likely facilities that do not founding, we calculated E-values to determine the minimum
typically manage acute medical problems. Patients who died strength of association for an unmeasured confounder to sig­
within 48 hours of the index blood culture collection were ex­ nificantly change our conclusions [24]. To identify subgroups
cluded. Other patient-level exclusion criteria are reported in in whom ID consultation may have a greater benefit, we also
Supplementary Methods. conducted a range of subgroup analyses using the primary
model. Subgroup analyses were not corrected for multiple test­
Primary Exposure: ID Consultation ing and findings should be considered exploratory. The second­
The primary exposure was ID consultation, defined as any ID ary outcome of duration of oral antibiotics prescribed on
physician review between 1 and 10 days following the index discharged was analyzed using a negative binomial mixed
date. This was identified using a list of unique fee codes for model adjusting for the same confounders as the primary model.

856 • CID 2024:79 (15 October) • Ong et al

 We conducted 3 secondary analyses to (a) evaluate ID consul­ The mortality benefit with ID consultation was still observed
tation as both a fixed and random effect, (b) evaluate the impact when the follow-up period was prolonged from 30 days to 90
of day of ID consultation among the subset of patients who re­ days (aHR 0.87, 95% CI .83–.91). It was also robust to changing
ceived ID consultation, and (c) estimate from the primary model the definition of ID consultation to 1–5 days instead of 1–10
the absolute risk reduction in 30-day mortality if the entire cohort days (aHR 0.80, 95% CI .75–.86), or including patients already
were to receive ID consultation on day 2 of illness. We also con­ on active ID follow-up from before the GN-BSI episode (aHR
ducted multiple sensitivity analyses, which (a) varied hospital in­ 0.82, 95% CI .76–.86). The effect estimate was consistent across
clusion criteria, (b) varied definition of time window for ID multiple sensitivity analyses where we modified hospital exclu­
consultation, and (c) excluded outlier sites identified in the sec­ sion criteria by changing the minimum number of total pa­
ondary random effect model. Methodologic details of these addi­ tients or total consults required for inclusion in the analysis
tional analyses are provided in Supplementary Methods. (Supplementary Table 1).
P values < .05 were considered statistically significant, and The effect of ID consultation was similar in a mixed model
all hypothesis tests were 2-sided. All analyses were conducted where ID consultation was included as both a fixed and random
in R version 3.6.1 (R Foundation for Statistical Computing, effect (fixed effect aHR 0.85, 95% CI .78–.93). Supplementary

Downloaded from https://academic.oup.com/cid/article/79/4/855/7676103 by guest on 30 October 2024

Vienna, Austria) using the coxme, lme4, survival, and forester Figure 1 shows the distribution of site-specific HR. There were
packages. 2 outliers identified in this model, where >75% of patients
with GN-BSI received ID consultation and the observed effect
was much larger (HR < 0.6). The mortality benefit of ID consul­
RESULTS
tation was consistent in a sensitivity analysis which excluded
Among 58 335 patients with GN-BSI during the study period, these 2 potential influential hospitals (aHR 0.86, 95% CI .80–.92).
30 159 patients from 53 hospitals were eligible for inclusion Subgroup analyses of the primary outcome showed that ID
(study flowchart in Figure 1). The mean age was 71.0 years, consultation had a greater benefit in patients with several high-
and 15 882 (52.7%) were male. 11 013 (36.5%) patients received risk features (nosocomial infection, non-Enterobacterales or­
an ID consultation between 1 and 10 days after the index blood ganisms, antimicrobial resistance, non-urinary tract source,
culture. Patients with an ID consultation were younger but had or polymicrobial infection; Figure 3). The direction of effect es­
a greater comorbidity burden, with a higher proportion of timate was consistent in the direction of benefit across all sub­
patients with Deyo-Charlson score ≥2 and who were immuno­ group analyses.
suppressed (Table 1). Patients with an ID consultation also Unadjusted 30-day mortality appeared to be lower for pa­
had more high-risk features, such as infection by non- tients who received ID consultation earlier in the course of ill­
Enterobacterales species, infection by MDR organisms, noso­ ness compared to those who received ID consultation later
comial infection, and polymicrobial infection. (Supplementary Figure 2). After adjustment for the same
The proportion of patients receiving ID consultation varied covariates in the primary model, those who received earlier
widely by hospital (median 35.0%, range 2.7%–76.1%, interquar­ consultation had lower odds of mortality (adjusted odds
tile range [IQR] 19.6%–41.1%, Figure 2A). The size of the hospi­ ratio [aOR] 0.93, 95% CI .89–.96 per each earlier day of
tal, as indicated by the total number of patients with GN-BSI, consultation).
and the proportion of patients with ID consultations were weak­ To quantify the potential bias introduced through residual
ly correlated (Pearson’s correlation coefficient 0.277, P = .044). confounding, we calculated E-values required to shift the point
Most patients received ID consultation early in the course of ill­ estimate and upper bound of the confidence interval to an HR
ness, with the number of ID consultations falling with increasing of 1.0 (ie, null effect) [24]. To shift the point estimate or upper
time from index blood culture (Figure 2B). CI to 1.0, the E-values were 1.74 or 1.56, respectively. This can
1041 (9.5%) of patients who received ID consultation died be interpreted as the minimum strength of association that a
within 30 days, compared to 1797 (9.4%) of patients without confounder (or set of confounders) must have on the same
ID consultation. In the fully adjusted multivariable model in­ HR scale, to “explain away” the exposure-outcome associa­
cluding potential confounders and with ID consultation as a tion observed. Put another way, patients in the no ID consul­
time-varying exposure, ID consultation was associated with a tation group must be significantly different from the ID
mortality benefit (adjusted hazard ratio [aHR] 0.82, 95% confi­ consultation group by a set of confounders that increases
dence interval [CI] .77–.88, P < .0001; Table 2). Obtaining pre­ the hazard for mortality in the unexposed group by 1.74- or
dictions using this adjusted model on the entire study cohort 1.56-fold for our findings to be invalid.
resulted in a predicted 30-day mortality of 10.5% if no patient We had access to outpatient antibiotic prescription data in
received ID consultation versus 6.7% if all patients received ID the subset of patients ≥65 years old (21 277 patients or 70.5%
consultation on day 2 of illness, translating to an estimated ab­ of the cohort). Also, 699 patients who were on long-term anti­
solute risk reduction of 3.8% or number needed to treat of 27. biotics (defined as having received antibiotics >28 days) were

Infectious Diseases Consultation in Gram-Negative Bloodstream Infection • CID 2024:79 (15 October) • 857
 Downloaded from https://academic.oup.com/cid/article/79/4/855/7676103 by guest on 30 October 2024
Figure 1. Study flowchart. Abbreviations: GN-BSI, gram-negative bloodstream infection; ID, infectious diseases.

additionally excluded, leaving 20 578 or 68.2% of the cohort in It is likely that these patients are the patients for whom ID con­
this analysis). Patients with ID consultation were less likely to sultation would make the greatest impact.
be prescribed antibiotics on discharge (2050 [30.0%] vs 5882 Although we could not determine the mechanism of ID con­
[42.8%]) and had significantly shorter antibiotic durations on sultation that led to this downstream reduction in mortality in
discharge (adjusted RR 0.82, 95% CI .76–.87, P < .0001). our cohort, we hypothesize that ID consultation could lead to
more appropriate selection of antibiotics, identification of pa­
tients with complicated disease, improved diagnostic workup
DISCUSSION
identifying infectious foci or complications, and early and
In a large population-wide cohort of patients with GN-BSI, we aggressive source control interventions. These associations
found that ID consultation was associated with significantly have been demonstrated in other BSIs caused by S. aureus,
lower all-cause mortality at 30 and 90 days. We believe that Candida, P. aeruginosa, and Enterococcus, where patients re­
this finding is genuine for several reasons. First, the mortality ceiving ID consultation were more likely to have appropriate
benefit was consistent across multiple sensitivity analyses antimicrobial selection and adequate source control [11, 13,
where we varied hospital exclusion criteria, changed the con­ 26, 27]. Inadequate antibiotic therapy, or a longer time to ap­
sultation definition time window, and excluded potential outli­ propriate antibiotic selection, has been associated with in­
er sites. Second, there was suggestion that earlier consultation creased mortality in GN-BSI in multiple studies [28–30].
could be associated with greater benefit compared to later con­ A shorter time to optimal management could explain why great­
sultation, consistent with a gradient effect. Third, in subgroup er benefit was seen among patients with earlier consultation.
analyses, the observed benefit appeared to be greater in patients Separate from this impact on mortality, we also show that
with specific high-risk features, which is biologically plausible. ID consultation has additional benefits in reducing oral

858 • CID 2024:79 (15 October) • Ong et al

Table 1. Baseline Characteristics of Patients With Gram Negative Bloodstream Infection (GN-BSI) Overall and According to Receipt of Infectious
Diseases (ID) Consultation

All Patients (N = 30 159) ID Consult (n = 11 013) No ID Consult (n = 19 146) SMDa

Male sex 15882 (52.7) 6152 (55.9) 9730 (50.8) 0.101
Age, years 71.0 (15.9) 68.5 (16.4) 72.4 (15.4) 0.245
Material resources indicator quintileb
1st 5276 (17.7) 1881 (17.2) 3395 (18.0) 0.020
2nd 5600 (18.8) 1972 (18.1) 3628 (19.2) 0.030
3rd 5728 (19.2) 2164 (19.8) 3564 (18.9) 0.024
4th 5979 (20.1) 2182 (20.0) 3797 (20.1) 0.003
5th 7216 (24.2) 2717 (24.9) 4499 (23.8) 0.025
Rural residence
Yes 1426 (4.7) 417 (3.8) 1009 (5.3) 0.071
No 28585 (94.8) 10547 (95.8) 18038 (94.2) 0.071
Missing 148 (0.5) 49 (0.4) 99 (0.5) 0.010

Downloaded from https://academic.oup.com/cid/article/79/4/855/7676103 by guest on 30 October 2024

Deyo-Charlson comorbidity index
No previous hospitalizationc 9836 (32.6) 3070 (27.9) 6766 (35.3) 0.161
0 5636 (18.7) 2039 (18.5) 3597 (18.8) 0.007
1 3121 (10.3) 1071 (9.7) 2050 (10.7) 0.032
2+ 11566 (38.4) 4833 (43.9) 6733 (35.2) 0.179
Immunosuppressedd 2541 (8.4) 1303 (11.8) 1238 (6.5) 0.187
Gram-negative organism in index blood culturee
E. coli 16150 (53.5) 4931 (44.8) 11219 (58.6) 0.279
Klebsiella spp. 5530 (18.3) 2065 (18.8) 3465 (18.1) 0.017
Pseudomonas spp. 1988 (6.6) 1007 (9.1) 981 (5.1) 0.157
Enterobacter spp. 1684 (5.6) 800 (7.3) 884 (4.6) 0.112
Proteus spp. 1456 (4.8) 479 (4.3) 977 (5.1) 0.036
Bacteroides spp. 1091 (3.6) 526 (4.8) 565 (3.0) 0.095
Serratia spp. 590 (2.0) 326 (3.0) 264 (1.4) 0.109
Citrobacter spp. 464 (1.5) 221 (2.0) 243 (1.3) 0.058
Acinetobacter spp. 393 (1.3) 213 (1.9) 180 (0.9) 0.084
Haemophilus spp. 342 (1.1) 143 (1.3) 199 (1.0) 0.024
Other gram-negativesf 1882 (6.2) 955 (8.7) 927 (4.8) 0.153
Organism subgroup
Enterobacterales 23819 (79.0) 7825 (71.1) 15994 (83.5) 0.301
Anaerobes 1268 (4.2) 613 (5.6) 655 (3.4) 0.104
Non-fermenters 2021 (6.7) 1021 (9.3) 1000 (5.2) 0.157
Others 724 (2.4) 379 (3.4) 345 (1.8) 0.103
Multiple 2327 (7.7) 1175 (10.7) 1152 (6.0) 0.169
Polymicrobial index blood culture 3168 (10.5) 1515 (13.8) 1653 (8.6) 0.163
Multi-drug resistant organismg 2800 (9.3) 1399 (12.7) 1401 (7.3) 0.180
Nosocomial infectionh 5565 (18.5) 2461 (22.3) 3104 (16.2) 0.156
Infection source
Urinary tract 15594 (51.7) 4724 (42.9) 10870 (56.8) 0.280
Respiratory 2109 (7.0) 945 (8.6) 1164 (6.1) 0.096
Intra-abdominal 4655 (15.4) 1747 (15.9) 2908 (15.2) 0.019
Skin and soft tissue 1395 (4.6) 770 (7.0) 625 (3.3) 0.170
Other/multiple 6406 (21.2) 2827 (25.7) 3579 (18.7) 0.169
ICU admission 3418 (11.3) 1459 (13.2) 1959 (10.2) 0.094
Numbers indicate mean (standard deviation) for continuous variables and number (percentage) for categorical/binary variables.
Abbreviations: ICU, intensive care unit; SMD, standardized mean difference.
a
Standardized mean difference >0.1 suggests a significant difference between the two groups.
b
The material resources index is a component of the Ontario marginalization index that combines multiple socioeconomic indicators and determinants of health at the neighborhood level.
Individuals in the 1st quintile experience the least socioeconomic marginalization, while those in the 5th quintile experience the most socioeconomic marginalization.
c
Therefore unable to calculate Deyo-Charlson comorbidity index.
d
Defined as active hematologic malignancy in the preceding 12 m, hematopoetic stem cell transplant in the preceding 12 m, or any previous solid organ transplant.
e
Total proportions do not add up to 100% as some index blood cultures had >1 organism.
f
In order of decreasing frequency: Morganella spp., Fusobacterium spp., Salmonella spp., Providencia spp., Stenotrophomonas spp., Neisseria spp., Moraxella spp., Prevotella spp., Veillonella
spp., Pantoea spp., Campylobacter spp., Raoultella spp., Aeromonas spp., Pasteurella spp., Achromobacter spp., Sphingomonas spp., Yersinia spp., Shigella spp.
g
Definition adapted from Dutch antimicrobial resistance surveillance guidelines; refer to Supplementary Methods for full definition.
h
Defined as index blood culture positive >48 h from admission.

Infectious Diseases Consultation in Gram-Negative Bloodstream Infection • CID 2024:79 (15 October) • 859
 Downloaded from https://academic.oup.com/cid/article/79/4/855/7676103 by guest on 30 October 2024
Figure 2. Variation in ID consult proportion across sites and breakdown of day of ID consult among patients with ID consult. Panel (A) depicts the variation in proportion of
GN-BSI patients receiving ID consultation across all hospitals included. Each point represents one hospital, with the size of the point proportional to the number of ID consults
at that hospital. The x-axis shows the total number of patients with GN-BSI, whereas the y-axis shows the proportion of GN-BSI patients who received an ID consultation.
Panel (B) depicts the breakdown of timing of ID consultation among all patients who received ID consultation in the entire study cohort. Abbreviations: GN-BSI, gram-negative
bloodstream infection; ID, infectious diseases.

antibiotic use on discharge. Although we could not explore methodology to address immortal time bias by treating the ex­
inpatient antibiotic durations in our cohort, it is possible posure of interest (ID consultation) as a time-varying exposure.
that this effect applied to inpatient intravenous antibiotic This increases confidence in our conclusions and may also ex­
durations as well. Multiple previous studies have demon­ plain why our estimate of benefit may be lower in magnitude
strated that ID consultation is associated with more appro­ than prior studies. Our population-wide cohort allowed us to
priate antibiotic prescribing [31–34]. ID consultation can evaluate the heterogeneity in uptake and effect of ID consulta­
work synergistically with antimicrobial stewardship pro­ tion across different hospitals.
grams to increase real-world implementation of shorter Our study has several limitations. First, given its retrospec­
treatment durations. tive observational design, we cannot exclude the possibility of
Our findings are consistent with those of previous studies residual confounding. We did not have access to inpatient pa­
evaluating the impact of ID consultation on outcomes in pa­ rameters to calculate quantitative severity indexes such as
tients with GN-BSI [16, 17]. However, our study has several ad­ SOFA scores and hence could only adjust for ICU admission
vantages over previous work. We used appropriate as a surrogate indicator for disease severity. However, it is likely

860 • CID 2024:79 (15 October) • Ong et al

Table 2. Results of Primary Analysis for Primary Outcome of Time to increased mortality. Second, we could not directly identify
All-cause Mortality Censored at 30 Days the mechanism through which ID consultation impacted mor­
tality, as we had no data available on inpatient antibiotic ther­
Variable Adjusted HR 95% CI P Value
apy or source control interventions. Third, the billing code we
ID consultation 0.82 .77–.88 <.0001
Male sex 1.03 .97–1.09 .32
used to define ID consultation only captures consultations by a
Age (per year increase in age) 1.02 1.02–1.03 <.0001 registered ID specialist. Other interventions that may review
Organism subgroup microbiologic data or appropriateness of antibiotic therapy
Enterobacterales Ref Ref Ref would not be captured (eg, antimicrobial stewardship or phar­
Anaerobes 1.62 1.44–1.81 <.0001 macy review), and our findings cannot be extrapolated to
Non-fermenters 1.33 1.21–1.46 <.0001
those interventions. Fourth, although we had data available
Others 1.00 .83–1.21 .99
Multiple 1.09 .92–1.29 .32
for all patients with GN-BSI in the province, we excluded a
Material resources indexa large number who were admitted to hospitals where no ID
1st quintile Ref Ref Ref consultation service was routinely available. This may com­
2nd quintile 1.03 .93–1.13 .56 promise the applicability of our findings to similarly less

Downloaded from https://academic.oup.com/cid/article/79/4/855/7676103 by guest on 30 October 2024

3rd quintile 1.05 .95–1.15 .34
well resourced, community, or rural settings without ID cov­
4th quintile 1.06 .97–1.17 .2
5th quintile 1.12 1.02–1.22 .02
erage. A previous survey of the distribution of ID specialists in
Deyo-Charlson comorbidity index the United States showed that 80% of counties are not served
No previous hospitalizationb Ref Ref Ref by even a single ID physician [35]. We did not explore access
0 0.78 .7–.86 <.0001 to concomitant services such as surgery or interventional ra­
1 1.06 .96–1.19 .25 diology for source control. It is plausible that ID consultation
2+ 1.48 1.37–1.6 <.0001
may be more common in centres where these services are
Immunosuppressedc 1.35 1.24–1.48 <.0001
ICU admission 6.33 5.93–6.76 <.0001
more easily available, resulting in potential confounding.
Infection source Nonetheless, the mortality benefit observed was consistent
Urinary tract Ref Ref Ref in our sensitivity analyses restricted to only hospitals with
Respiratory 1.86 1.70–2.04 <.0001 high volume of GN-BSI patients and a high number of ID con­
Intra-abdominal 1.18 1.07–1.29 .0006
sultations. These are likely to be academic or tertiary centres
Skin and soft tissue 1.80 1.60–2.02 <.0001
Other/multiple 1.89 1.76–2.03 <.0001
with all these services available.
Multi-drug resistant organismd 1.34 1.22–1.46 <.0001 Implementation of routine ID consultation for all patients
Nosocomial infectione 1.11 1.04–1.19 .0032 with GN-BSI may be difficult given resource limitations, even
Polymicrobial infection 1.24 1.07–1.44 .0048 in high-resource settings. ID physicians are in short supply
Abbreviations: CI, confidence interval; HR, hazard ratio; ICU, intensive care unit; ID, and training positions for ID physicians are regularly under-
infectious diseases; Ref, reference group.
a
The material resources index is a component of the Ontario marginalization index that
subscribed [35–37]. One major reason is the lack of competitive
combines multiple socioeconomic indicators and determinants of health at the remuneration for ID physician services [38]. Similar studies
neighborhood level. Individuals in the 1st quintile experience the least socioeconomic
marginalization, whereas those in the 5th quintile experience the most socioeconomic like ours demonstrating the value of ID consultation can sup­
marginalization.
b
port advocacy for improved remuneration and increased ca­
Therefore unable to calculate Deyo-Charlson comorbidity index.
c
Defined as active hematologic malignancy in the preceding 12 m, hematopoietic stem cell
pacity for ID physician services in all healthcare settings.
transplant in the preceding 12 m, or any previous solid organ transplant. Even if routine ID consultation is not logistically feasible
d
Definition adapted from Dutch antimicrobial resistance surveillance guidelines [25]; refer to
eMethods for full definition.
for all patients with GN-BSI, we should minimally be prioritiz­
e
Defined as index blood culture positive >48 h from admission. ing routine ID consultation for the highest risk patients
where it is likely to make the greatest impact. We identified
several high-risk features (eg, antimicrobial resistance, non-
that residual confounding should shift the effect estimate in the Enterobacterales infection, non-UTI source) in our subgroup
opposite direction—patients who receive ID consultation are analyses that were associated with greater benefit. Further
more likely to be those with more severe disease or with a great­ research should confirm our findings in a prospective
er comorbidity burden. Nonetheless, we quantified that any po­ quasi-experimental pre-post implementation study or cluster-
tential unmeasured confounding must be in the opposite randomized trial. These should also identify the specific
direction by a relatively large margin (1.74-fold) to explain “value-add” components of ID consultation such as the impact
away the observed association. Residual confounding may on antibiotic choice, added investigations, or source control in­
also affect the finding that earlier consultation is superior to lat­ terventions. Given limited resource availabilities, it will also be
er consultation—for example, patients may be referred later in important to examine the cost-effectiveness of different imple­
the illness course if they develop complications or have persis­ mentation policies for routine ID consultation in GN-BSI pa­
tent bacteremia, which may independently be associated with tient populations.

Infectious Diseases Consultation in Gram-Negative Bloodstream Infection • CID 2024:79 (15 October) • 861
 Downloaded from https://academic.oup.com/cid/article/79/4/855/7676103 by guest on 30 October 2024
Figure 3. Forest plot of association of ID consult and 30-day mortality among cohort subgroups. Abbreviations: CI, confidence interval; GN-BSI, gram-negative bloodstream
infection; HR, hazard ratio; ID, infectious diseases; UTI, urinary tract infection.

CONCLUSION Health Information (CIHI). This document used data adapted from the
Statistics Canada Postal CodeOM Conversion File, which is based on data li­
ID consultation could potentially reduce mortality in patients censed from Canada Post Corporation, and/or data adapted from the
with GN-BSI. If resources permit, routine consultation for Ontario Ministry of Health Postal Code Conversion File, which contains
data copied under license from ©Canada Post Corporation and Statistics
this patient population early in the course of illness should be
Canada. The authors thank the Toronto Community Health Profiles
considered to improve patient outcomes. Future studies should Partnership for providing access to the Ontario Marginalization Index. The
examine the feasibility and cost-effectiveness of widespread im­ authors thank IQVIA Solutions Canada Inc. for use of their Drug
plementation of ID consultation. Information File. The analyses, conclusions, opinions, and statements ex­
pressed herein are solely those of the authors and do not reflect those of the
funding or data sources; no endorsement is intended or should be inferred.
Supplementary Data Data sharing. The data set from this study is held securely in coded form
Supplementary materials are available at Clinical Infectious Diseases online. at ICES. Although data-sharing agreements prohibit ICES from making the
Consisting of data provided by the authors to benefit the reader, the posted data set publicly available, access may be granted to those who meet pre­
materials are not copyedited and are the sole responsibility of the authors, specified criteria for confidential access, available online at www.ices.on.
so questions or comments should be addressed to the corresponding ca/DAS. The full data set creation plan and underlying analytic code are
author. available from +the authors on request, with the understanding that the
programs may rely on coding templates that are unique to ICES.
Ethics approvals. ICES is a prescribed entity under Ontario’s Personal
Notes Health Information Protection Act (PHIPA). Section 45 of PHIPA autho­
Author contributions. Conceptualization: S. W. X. O., N. D.; Methodology: rizes ICES to collect personal health information, without consent, for the
S. W. X. O., N. D.; Formal analysis: S. W. X. O.; Resources: J. L., D. J. F., purpose of analysis or compiling statistical information with respect to the
S. N. P., C. D.; Data curation: J. L., D. J. F.; Writing—original draft: management of, evaluation, or monitoring of the allocation of resources to
S. W. X. O.; Writing—review and editing: J. J., K. L. S., D. M., B. L., S. Y. C. or planning for all or part of the health system. Projects that use data col­
T., N. D.; Visualization: S. W, X. O.; Supervision: K. A. B., N. D.; Project ad­ lected by ICES under section 45 of PHIPA, and use no other data, are ex­
ministration: S. M. L.; Funding acquisition: K. A. B., N. D. empt from REB review. The use of the data in this project is authorized
Acknowledgments. Parts of this material are based on data and informa­ under section 45 and approved by ICES” Privacy and Legal Office.
tion compiled and provided by the Ministry of Health, Ontario Financial support. This work was supported by a project grant from
Laboratories Information System (OLIS), and Canadian Institute for CIHR (grant number 159503 to K. A. B. and N. D.). This study was

862 • CID 2024:79 (15 October) • Ong et al

supported by ICES, which is funded by an annual grant from the Ontario 17. Tang G, Huang L, Zong Z. Impact of infectious disease consultation on clinical
Ministry of Health (MOH) and the Ministry of Long-Term Care (MLTC). management and outcome of patients with bloodstream infection: a retrospective
Potential conflicts of interest. The authors: No reported conflicts of in­ cohort study. Sci Rep 2017; 7:12898.
18. Yadav K, Lewis RJ. Immortal time bias in observational studies. JAMA 2021; 325:
terest. All authors have submitted the ICMJE Form for Disclosure of
686–7.
Potential Conflicts of Interest. Conflicts that the editors consider relevant
19. Ong SWX, Luo J, Fridman DJ, et al. Follow-up blood cultures do not reduce mor­
to the content of the manuscript have been disclosed. tality in hospitalized patients with gram-negative bloodstream infection: a retro­
spective population-wide cohort study. Clin Microbiol Infect 2024; 28:2200519.
20. Langford BJ, Daneman N, Diong C, et al. Antibiotic susceptibility reporting and
association with antibiotic prescribing: a cohort study. Clin Microbiol Infect 2021;
References 27:568–75.
1. Goto M, Al-Hasan MN. Overall burden of bloodstream infection and nosocomial 21. Jones M, Fowler R. Immortal time bias in observational studies of time-to-event
bloodstream infection in North America and Europe. Clin Microbiol Infect 2013; outcomes. J Crit Care 2016; 36:195–9.
19:501–9. 22. Balan TA, Putter H. A tutorial on frailty models. Stat Methods Med Res 2020; 29:
2. Verway M, Brown KA, Marchand-Austin A, et al. Prevalence and mortality asso­ 3424–54.
ciated with bloodstream organisms: a population-wide retrospective cohort study. 23. Deyo RA, Cherkin DC, Ciol MA. Adapting a clinical comorbidity index for use
J Clin Microbiol 2022; 60:e0242921. with ICD-9-CM administrative databases. J Clin Epidemiol 1992; 45:613–9.
3. Schechner V, Wulffhart L, Temkin E, et al. One-year mortality and years of po­ 24. VanderWeele TJ, Ding P. Sensitivity analysis in observational research: introduc­
tential life lost following bloodstream infection among adults: a nation-wide pop­ ing the E-value. Ann Intern Med 2017; 167:268–74.

Downloaded from https://academic.oup.com/cid/article/79/4/855/7676103 by guest on 30 October 2024

ulation based study. Lancet Reg Health Eur 2022; 23:100511. 25. Kluytmans-Vandenbergh MF, Kluytmans JA, Voss A. Dutch guideline for pre­
4. Diekema DJ, Hsueh PR, Mendes RE, et al. The microbiology of bloodstream in­ venting nosocomial transmission of highly resistant microorganisms (HRMO).
fection: 20-year trends from the SENTRY antimicrobial surveillance program. Infection 2005; 33:309–13.
Antimicrob Agents Chemother 2019; 63:e00355-19. 26. Chiong F, Wasef MS, Liew KC, et al. The impact of infectious diseases consulta­
5. Antimicrobial Resistance Collaborators. Global burden of bacterial antimicrobial tion on the management and outcomes of Pseudomonas aeruginosa bacteraemia
resistance in 2019: a systematic analysis. Lancet 2022; 399:629–55. in adults: a retrospective cohort study. BMC Infect Dis 2021; 21:671.
6. Ljungquist O, Blomstergren A, Merkel A, Sunnerhagen T, Holm K, Torisson G. 27. Tholany J, Suzuki H, Livorsi DJ, Perencevich EN, Goto M. The association of in­
Incidence, aetiology and temporal trend of bloodstream infections in southern fectious diseases consultation and 30-day mortality rates among veterans with en­
Sweden from 2006 to 2019: a population-based study. Euro Surveill 2023; 28: terococcal bacteraemia: a propensity score-matched retrospective cohort study.
2200519. Clin Microbiol Infect 2023; 29:1039–44.
7. Lopez-Cortes LE, Del Toro MD, Galvez-Acebal J, et al. Impact of an evidence- 28. Xu S, Song Z, Han F, Zhang C. Effect of appropriate empirical antimicrobial ther­
based bundle intervention in the quality-of-care management and outcome of apy on mortality of patients with gram-negative bloodstream infections: a retro­
Staphylococcus aureus bacteremia. Clin Infect Dis 2013; 57:1225–33. spective cohort study. BMC Infect Dis 2023; 23:344.
8. Escrihuela-Vidal F, Kaasch AJ, Von Cube M, et al. Impact of adherence to indi­ 29. Baltas I, Stockdale T, Tausan M, et al. Impact of antibiotic timing on mortality
vidual quality-of-care indicators on the prognosis of bloodstream infection due to from gram-negative bacteraemia in an English district general hospital: the im­
Staphylococcus aureus: a prospective observational multicentre cohort. Clin portance of getting it right every time. J Antimicrob Chemother 2021; 76:813–9.
Microbiol Infect 2022; 29:498–505. 30. Ohnuma T, Chihara S, Costin B, et al. Association of appropriate empirical anti­
9. Green J, Howard J, Shankar A, et al. Assessing the impact of a “bundle of care” microbial therapy with in-hospital mortality in patients with bloodstream infec­
approach to Staphylococcus aureus bacteraemia in a tertiary hospital. Infect tions in the US. JAMA Netw Open 2023; 6:e2249353.
Prev Pract 2020; 2:100096. 31. Bork JT, Claeys KC, Heil EL, et al. A propensity score matched study of the pos­
10. Vena A, Bouza E, Corisco R, et al. Efficacy of a “checklist” intervention bundle on itive impact of infectious diseases consultation on antimicrobial appropriateness
the clinical outcome of patients with Candida bloodstream infections: a in hospitalized patients with antimicrobial stewardship oversight. Antimicrob
quasi-experimental pre-post study. Infect Dis Ther 2020; 9:119–35. Agents Chemother 2020; 64:e00307-20.
11. Bai AD, Showler A, Burry L, et al. Impact of infectious disease consultation on 32. Al-Tawfiq JA. The pattern and impact of infectious diseases consultation on an­
quality of care, mortality, and length of stay in Staphylococcus aureus bacteremia: timicrobial prescription. J Glob Infect Dis 2013; 5:45–8.
results from a large multicenter cohort study. Clin Infect Dis 2015; 60:1451–61. 33. Itoh N, Akazawa N, Kanawaku E, et al. Effects of infectious disease consultation
12. Vogel M, Schmitz RP, Hagel S, et al. Infectious disease consultation for and antimicrobial stewardship program at a Japanese cancer center: an interrupt­
Staphylococcus aureus bacteremia—a systematic review and meta-analysis. J ed time-series analysis. PLoS One 2022; 17:e0263095.
Infect 2016; 72:19–28. 34. Rinaldi M, Gatti M, Tonetti T, et al. Impact of a multidisciplinary management
13. Mejia-Chew C, O’Halloran JA, Olsen MA, et al. Effect of infectious disease con­ team on clinical outcome in ICU patients affected by gram-negative bloodstream
sultation on mortality and treatment of patients with Candida bloodstream infec­ infections: a pre-post quasi-experimental study. Ann Intensive Care 2024; 14:36.
tions: a retrospective, cohort study. Lancet Infect Dis 2019; 19:1336–44. 35. Walensky RP, McQuillen DP, Shahbazi S, Goodson JD. Where is the ID in
14. Kobayashi T, Marra AR, Schweizer ML, et al. Impact of infectious disease consul­ COVID-19? Ann Intern Med 2020; 173:587–9.
tation in patients with candidemia: a retrospective study, systematic literature re­ 36. Reece R, Beckwith CG. The infectious diseases specialist, at risk of extinction. J
view, and meta-analysis. Open Forum Infect Dis 2020; 7:ofaa270. Infect Dis 2023; 228:1649–51.
15. Hasegawa S, Kakiuchi S, Tholany J, et al. Impact of infectious diseases consulta­ 37. Lawrence S, Aggarwal D, Davies A, Partridge D, Ratnaraja N, Llewelyn MJ. The
tion among patients with infections caused by gram-negative rod bacteria: a sys­ state of hospital infection services in the UK: national workforce survey 2021.
tematic literature review and meta-analysis. Infect Dis (Lond) 2022; 54:618–21. Clin Infect Pract 2022; 15:100151.
16. Shulder S, Tamma PD, Fiawoo S, et al. Infectious diseases consultation associated 38. Swartz TH, Aberg JA. Preserving the future of infectious diseases: why we must
with reduced mortality in gram-negative bacteremia. Clin Infect Dis 2023; 77: address the decline in compensation for clinicians and researchers. Clin Infect
1234–7. Dis 2023; 77:1387–94.

Infectious Diseases Consultation in Gram-Negative Bloodstream Infection • CID 2024:79 (15 October) • 863