MS: 4753020981056172 – Quantifying the clinical virulence of Klebsiella pneumoniae
producing carbapenemase Klebsiella pneumoniae simultaneously though an in-vivo
insect model and translated patient outcomes
Editor's comment:
"While both reviewers found the topic important and the study interesting, they also have some
concerns on the sample size of patients and therefore the validity of the conclusion. It was
suggested that the manuscript will be much improved if the authors can either collect data on a
larger sample of patients to improve its statistical analysis power or focus their studies on the
animal model."
AU: We appreciate the guidance of the editor. We have revised our manuscript to focus on the
Galleria mellonella model. We have made changes to highlight that these are the most robust
results of our paper. The chronological order of our presentation now lists the insect model
results first. Second, we have now clearly denoted the “pilot” nature of human translational
studies and urged reader caution in interpretation. In the resubmitted manuscript, we have
informed readers in the title and 10 additional times that the study is indeed pilot (and thus our
results are focused on the insect model).
To address the first point, we are unfortunately unable to improve our study by amassing a
larger sample. A larger sample simply does not exist. Our data are highly unique because KPC
bacteremia is such a rare event. For instance, the CDC estimated that there are only 9,300
cases of carbapenem-resistant Enterobacteriaceae in the United States per year. Of these,
only a fraction are KPC. A smaller fraction yet constitutes bloodstream isolates. Thus our
numbers are quite large for a single center study. It should be noted that we are an 897 bed
hospital in downtown Chicago and are a tertiary care referral center for the city. To further
illustrate this point, our bloodstream infection numbers from 2012 follow: out of 386
Enterobacteriaceae blood stream infections, 12 infections were resistant to carbapenems and
only 6 were KPC. With accumulation of only 6 cases per year (before assessing any exclusion
criteria), it is unlikely that we can add additional cases in the near future. Furthermore, these
studies are expensive from a technician time standpoint and require funding. Our intention was
to present a pilot study such that we can attempt to secure additional grant funds. Larger
studies and multi-center studies will be required, but to this end, publication of “pilot” efforts is
needed. The current analysis is the only one of its kind to our knowledge and represents the
best available data from a very large tertiary care hospital. As such, we have clearly stated in
our resubmitted manuscript that this study is a “pilot” and should be interpreted as such.
***Please note. All line numbers refer to the “tracked changes” copy
Page 1 of 9
Reviewer: Jason J Pogue
Reviewer's report:
McLaughlin and colleagues present a very interesting analysis in an attempt to truly identify the
attributable mortality with KPC bloodstream infections. While, I commend them for their
translational approach, and do not necessarily disagree with their premise that it is the patient
population that acquire KPC that are associated with mortality rather than the virulence of the
organism, I think several limitations of these data limit the ability of the authors to make the
conclusions that they do. The major issues are discussed below
1. Probably the most important issue with these data are that there are only 15 KPC BSI
cases, and for the primary outcome of mortality there are only 12 events. This makes
multivariate modeling with three variables nearly impossible. While it appears in table 4 it
appears that the virulence score attenuated the independent risk of mortality with KPC, I do
not think this is an appropriate interpretation of these data. By forcing three variables into
the model the authors have made the model unstable. KPC falling out as a risk factor is
more an artifact of too many variables (as seen by the extremely large confidence interval
that on its upper end surprasses a 30-fold risk), rather than an effect of the virulence score.
In fact, the virulence score has an OR of 1.01, suggesting that it had zero impact on
mortality.
AU: We thank the reviewer for the comment. We have added this limitation to our manuscript
(lines2, 46, 96, 267, 335, 376-7, 489-503). We also have clarified the interpretation of the odds
ratio for the readership. The previous analysis was difficult for the readership to interpret as
odds ratios without converted probabilities are difficult to appreciate.
We have addressed the reviewer’s comment as follows. First, we have made the virulence
score more intuitive for the readership. It is now calculated as insect dead days so that a higher
score indicates higher virulence. As such, we have replaced Kaplan-Meier survival curves with
Kaplan-Meier failure curves.
Second, as the reviewer suggests that only one variable can be added for a 12 mortality event
analysis, we have conducted a separate analysis to specifically address this concern. Below, is
the result of multivariate analysis that the reviewer questioned. Note, the new virulence score
returns the inverse odds ratio and 95% CI (i.e. it is isometric to the original model without the
inversion).
Page 2 of 9
OLD MODEL
Logistic regression
Number of obs
=
75
=
75
---------------------------------------------------------------died | Odds Ratio
[95% Conf. Interval]
P>|z|
-------------------+-------------------------------------------kpc |
2.505322
.2015842
31.13656
0.475
wormday2 |
1.008754
.9469172
1.074629
0.787
modapacheiiscore~0 |
1.127818
.9653754
1.317595
0.130
_cons |
.0300346
.0036317
.2483922
----------------------------------------------------------------
NEW MODEL
Logistic regression
Number of obs
--------------------------------------------------------------died | Odds Ratio
[95% Conf. Interval]
P>|z|
-------------------+------------------------------------------kpc |
2.505322
.2015842
31.13656
0.475
virulence_new
|
.9913221
.9305541
1.056059
0.787
modapacheiiscore~0 |
1.127818
.9653754
1.317595
0.130
_cons |
.0506681
.0010676
2.404766
---------------------------------------------------------------
This minimal contribution of the virulence score as an interval variable can be appreciated
below. After controlling for the covariates of KPC and the modified APACHE score, there is no
effect on mortality (i.e. the mean predicted value is just over 20% mortality at a virulence score
of 0 and ~13% at a score of 60). This result is not statistically significant.
-.2
0
.2
.4
.6
Predictive Margins with 95% CIs
0
20
40
60
Virulence Score (new)
Page 3 of 9
Analysis by information criterion analyses follows. This is more fully discussed below.
----------------------------------------------------------------------------Model |
Obs
ll(null)
ll(model)
df
AIC
BIC
-------------+--------------------------------------------------------------A |
75
-32.97524
-29.79754
4
67.59508
76.86503
-----------------------------------------------------------------------------
To address the reviewer’s concern that only a single variable should be entered, we have now
done that and have only inserted the virulence score.
Logistic regression
Number of obs
=
75
--------------------------------------------------------------died | Odds Ratio
[95% Conf. Interval]
P>|z|
-----------------+--------------------------------------------virulence_inverted |
.9696771
..9379116
1.002518
0.070
_cons
|
.6397821
.1619283
2.527792
---------------------------------------------------------------
Here, we see trends toward an effect. As worm virulence increases, mortality decreases.
Again, this can be more easily appreciated by plotting the probability of mortality by incremental
virulence score (with attendant 95% CIs).
0
.2
.4
.6
.8
Adjusted Predictions with 95% CIs
0
20
40
60
Virulence Score (new)
Thus, we believe that our pilot results in the translational model suggest that further study is
necessary. An opposite relationship could be expected on the basis of previously published
clinical studies alone. We have provided these additional analyses (lines 226-7, 341-44) and
de-emphasized their value (as instructed by the reviewer and described above).
Page 4 of 9
----------------------------------------------------------------------------Model |
Obs
ll(null)
ll(model)
df
AIC
BIC
-------------+--------------------------------------------------------------B |
75
-32.97524
-31.41979
2
66.83959
71.47456
-----------------------------------------------------------------------------
Finally, the reviewer has described a “rule of thumb” for multivariate model building.
That is, “a model should only have one predictor per 10 outcomes”. The actual model
built; however, can be tested to account for the benefits of additional predictors and the
associated degrees of freedom that they require. One well described way of doing this
is by the Information Criterion (either Bayesian or Akaike) (see Hosmer DW and
Lemeshow S. Applied Regression. 2nd Ed. for an in-depth discussion on appropriate
mathematical model building).
Calculations were performed as follows: Akaike’s information criterion was defined as:
AIC = -2 lnL + 2k where lnL is the maximized log-likelihood of the model and k is the
number of parameters estimated. Schwarz’s Bayesian information criterion was defined
as: BIC = -2 lnL + k lnN where N is the sample size. (STATABASE REFERENCE
MANUAL. RELEASE 13)
.
Below, I have compared the models:
Likelihood-ratio test
(Assumption: B nested in A)
LR chi2(2) =
Prob > chi2 =
3.24
0.1975
Akaike's information criterion and Bayesian information criterion
------------------------------------------------------------------------------------------------Model
|
Obs
ll(null)
ll(model)
df
AIC
BIC
-------------+----------------------------------------------------------------------------------Model with virulence score only |
75
-32.97524
-31.41979
2
66.83959
71.47456
Model with KPC+Virulence+APACHE |
75
-32.97524
-29.79754
4
67.59508
76.86503
-------------------------------------------------------------------------------------------------
Here, we see that the model using only the virulence score is considered the better model (as
per lower AIC and BIC) with less complexity, yet equal descriptiveness. Thus, per the
reviewer’s request as well as directed by the mathematical calculation, we will additionally
present the model including only virulence score.
2. I would strongly consider using patients with ESBL infections as the source population for
the controls. These patients are likely more similar to KPC patients and would serve as a
better reference than pan-susceptible isolates. While a case-case control study is ideal for
risk factor studies, it would not be appropriate as uninfected controls could not make it into
the animal model.
AU: We thank the reviewer for the comment. Certainly, the reviewer is correct that one can ask
the question: “are KPC isolate blood stream infections more virulent than ESBL blood stream
Page 5 of 9
infections?” This question, while important, cannot be answered with insect or animal models.
As the G.mellonella model has NOT been validated for comparisons across organism species
(nor genus), the application here would be questionable. Thus, a large body of work would be
required to justify the comparisons. However, the question that we have asked, “is KPC K.
pneumoniae more virulent than non-KPC K.pneumoniae?” is assessable with these
methodologies. Two new studies (not available when our article was originally submitted) also
support the comparison in the G. mellonella model [1, 2].
When we complete future clinical analyses, we will keep the reviewer’s query in mind as it is
very relevant to describe virulence between pathogens in clinical studies.
3. What was the source of bacteremia in patients, and was it the same between the groups?
AU: We thank the reviewer for the comment. We have previously classified the source of
bacteremia in several previous studies [3-5]. However in our studies, we have found that the
source of bacteremia is often highly subjective/speculative as a function of the fact that the
patients are complex, have numerous lines, concomitant infections, etc. Therefore, we did not
collect this data in the present analysis. We have added this limitation to our manuscript.
4. I would consider refocusing this study on the animal model. I find those data to be very
intriguing and of interest. If the authors diagree with this then they will need to collect a
larger sample of KPC patients to validate their model
AU: We thank the reviewer for the comment. As above (see question 1), we have deemphasized the human conclusions and have clearly labeled our human studies as “pilot”. Also
per the reviewer’s comment, we hope that our pilot data is interesting and hope to continue to
contribute more isolates for study and provide better estimates in the future. Until then, we
believe that the current pilot results can be instructive to clinicians.
Page 6 of 9
Reviewer: Susan M Logan
Reviewer's report:
It is clearly recognised in the literature that there is a need to more reliably determine the
relative virulence of KPC isolates and the current study is focused towards this goal. It was
attempted in this study to evaluate quantitatively KPC virulence by
a) determination of clinical virulence using standard clinical data modeling techniques
b) assessing relative virulence of KPC isolates by examining virulence in a Galleria model of
infection
c) using the virulence score from insect model to recalculate KPC virulence in clinical model.
Clinical data modeling techniques are clearly described and data is presented in comprehensive
fashion. I am not qualified to aseess the rigour of the statistical analysis.
1. While the initial analysis found an increased virulence for KPC+ patients, this virulence was
decreased when measured organism virulence (from insect model) was incorporated into
the multivariate models. The insect model data revealed that KPC+ isolates were less
virulent than KPCisolates suggesting that poor patient outcomes of KPC+ patients may not
be due to the virulence of KPC+ strains. The insect model data as presented are very
restricted in scope and could be expanded to better explore the virulence phenotypes
observed (See recent paper -Insua et al Infect. Immun 81:3552-3565)
AU: We thank the reviewer for the comment. Certainly, this is one direction of study, and
genetic knock-out studies (such as performed in the cited article) certainly would be interesting.
However, the goal in this article was to define virulence of organisms that are clinically incident
in patient blood stream infections rather than identify exact virulence mechanisms. Specifically,
we wished to see if the classification of the KPC phenotype was more or less virulent. Many
additional works will be necessary to completely categorize the known phenotypes/genotypes
and search for those that have not yet been identified. We have plans to embark on some of
this work, but it is currently outside of the scope of this analysis.
2. Could the virulence properties measured in Galleria possibly not be those which are relevant
to virulence in humans ?
AU: We thank the reviewer for the comment. We have listed this as a possible study
confounder (lines 508-17) ; however, it is important to note that previous researchers have
found that there is good concordance between the Galleria model and animal models of higher
order of magnitude. In fact, the reference highlighted [6] by the reviewer concludes: “Taken
together, these results support the utility of G. mellonella as a surrogate host for assessing
infections with K. pneumoniae.”
Overall , while very preliminary in nature the results presented in this study which suggest that it
may not be the virulence of KPC+ isolates which lead to poor patient outcomes are somewhat
intriguing and deserve further study.
However in its current format , I feel the manuscript makes too small an advance to warrant
publication.
Page 7 of 9
AU: We respect the opinion of the scientific reviewer; however, we respectfully disagree. As
directly quoted in the manuscript that the reviewer has suggested [6] :
“the use of animal models to identify the virulence factors of human pathogens is indispensable.
Currently, identification and characterization of novel virulence factors rely largely on assessing
mutant bacteria for growth in the organs of infected mice. The dependence on mouse infection
models limits large-scale analysis of virulence due to the large number of animals needed to
obtain statistically significant results.”
Thus, the need is readily apparent to be able to translate the Galleria mellonella insect model to
humans. It is already in high use to discern virulence factors. The translation to humans is
paramount. This is the first study reviewing outcomes for K. pneumoniae in G.mellonella and
human bloodstream infections. While it is in PILOT form, as now clearly described, it is an
important foray into the human translation arena. Thus, the results, while not entirely what the
scientific community expects, require further investigation. Such investigations will only occur if
initial interest in the pilot studies is appreciated.
Page 8 of 9
Supporting references cited in this document
1.
2.
3.
4.
5.
6.
Wand ME, McCowen JW, Nugent PG, Sutton JM: Use of the Galleria mellonella
infection model to study the opportunistic pathogen Klebsiella pneumonia
supports the presence of a multifaceted interaction with the host immune
response. Journal of medical microbiology 2013.
Wand ME, Bock LJ, Turton JF, Nugent PG, Sutton JM: Acinetobacter baumannii
virulence is enhanced in Galleria mellonella following biofilm adaptation. Journal of
medical microbiology 2012, 61(Pt 4):470-477.
Esterly JS, Wagner J, McLaughlin MM, Postelnick MJ, Qi C, Scheetz MH: Evaluation of
clinical outcomes in patients with bloodstream infections due to Gram-negative
bacteria according to carbapenem MIC stratification. Antimicrob Agents Chemother
2012, 56(9):4885-4890.
Crank CW, Scheetz MH, Brielmaier B, Rose WE, Patel GP, Ritchie DJ, Segreti J:
Comparison of outcomes from daptomycin or linezolid treatment for vancomycinresistant enterococcal bloodstream infection: A retrospective, multicenter, cohort
study. Clin Ther 2010, 32(10):1713-1719.
Scheetz MH, Hoffman M, Bolon MK, Schulert G, Estrellado W, Baraboutis IG, Sriram P,
Dinh M, Owens LK, Hauser AR: Morbidity associated with Pseudomonas
aeruginosa bloodstream infections. Diagn Microbiol Infect Dis 2009, 64(3):311-319.
Insua JL, Llobet E, Moranta D, Perez-Gutierrez C, Tomas A, Garmendia J, Bengoechea
JA: Modeling Klebsiella pneumoniae Pathogenesis by Infection of the Wax Moth
Galleria mellonella. Infection and immunity 2013, 81(10):3552-3565.
Page 9 of 9