This study found decreased virulence for KPC (+) isolates compared to KPC(−) isolates in a G. mellonella model. The study analyzing patient outcomes from the patients that contributed the isolates found the opposite association at the bivariate level (i.e. KPC (+) patients experienced greater mortality). Specifically, this clinical study identified a 20% absolute increase in hospital mortality for KPC(+) patients. Minimally, the discordances between these two models should give researchers pause. Our pilot study imputing a virulence score, which should be interpreted with caution due to low power, demonstrated that attributable virulence in patients was decreased when considering the insect model virulence results. In a separate analysis designed to limit the number of predictor variables entered into the multivariate logistic model, increasing organism virulence score from the G. mellonella model was associated with downward trending mortality rates at the bivariate level. The length of stay post-infection for survivors was 11.5 days longer in the KPC(+) group when considering the unadjusted model. Similarly here, the magnitude of difference and association was attenuated with the addition of the virulence score. The G. mellonella model indicated that KPC(−) K. pneumoniae isolates were more virulent than KPC(+) K. pneumoniae isolates.
G. mellonella larvae have been shown as a validated invertebrate host model for deciphering virulence in K. pneumoniae[11, 12] as well as a number of other bacterial pathogens including: Listeria spp. , Staphylococcus aureus, Acinetobacter baumannii, Pseudomonas aeruginosa, among others. Several studies have also used the G. mellonella model to demonstrate the virulence between different strains or genetic mutations of organisms [11, 15, 16]. Results from this host-pathogen interaction model have also been shown to produce similar results to mouse models with K. pneumoniae. Importantly known K. pneumoniae virulence factors such as capsule polysaccharide and biochemical manipulations to Lipid A when studied as knock-out mutants and compared to isogenic parent isolates, demonstrate that G. mellonella models are highly capable of discerning virulence . Concordance has been seen with other organisms using virulence models with G. mellonella and higher order animal studies. Wand et al. studied the virulence of several species of Burkholderia and found that the results found in the G. mellonella model reflected those found in a similar murine infection model . Jander et al. reported a positive correlation (p < 0.001) of virulence patterns between Pseudomonas aeruginosa isolates tested in G. mellonella and a burned mouse model .
Prior work, as described below, suggests that patients with KPC(+) K. pneumoniae blood stream infections have a higher mortality than patients with KPC(−) isolates. Based upon the results of this study, the higher mortality may be due to patient factors and not virulence of the resistant organism. In the G. mellonella model, KPC(+) K. pneumoniae isolates were less virulent than KPC(−) isolates; there may be a fitness cost to carriage of the resistant plasmid, or less fit organisms might be more likely to acquire the resistance. While additional study is certainly needed, the results of this pilot study suggest that it might not be the virulence of KPC(+) K. pneumoniae that leads to poor patient outcomes.
Previous studies have evaluated outcomes of patients with KPC(+) K. pneumoniae blood stream infections; however, these studies used study designs different from ours. Borer et al. conducted a matched, retrospective, cohort study and found a 50% attributable mortality rate for KPC(+) K. pneumoniae blood stream infections when compared to hospitalized patients that were infected but without bacteremia . In a case–control study evaluating patient outcomes, Mouloudi et al. reported in-hospital mortality of 68% in the KPC(+) K. pneumoniae blood stream infection group compared to 41% in the carbapenem susceptible K. pneumoniae group and 44% in the metallo-beta-lactamase producing K. pneumoniae blood stream infection group . Similarly, Ben-David et al. conducted a retrospective cohort study of patients with KPC(+) K. pneumoniae blood stream infections compared to extended spectrum beta-lactamase (ESBL) producing K. pneumoniae blood stream infections and carbapenem susceptible K. pneumoniae blood stream infections . The authors found that infection-related mortality was significantly higher among patients with KPC(+) infections compared to ESBL producing or carbapenem susceptible infections (48%, 22%, and 17%, respectively). All three of these studies have found a higher mortality rate in carbapenem resistant blood stream infections, but like all clinical outcomes studies, they may suffer from an inability to completely correct for all comorbidity variables. Our use of the G. mellonella model allowed for KPC status to be isolated as the only changing variable.
This study analyzed the virulence of K. pneumoniae based on KPC status as a clinical variable and has shown that KPC(+) status was not associated with mortality when the virulence score was included in the model. Virulence was directly calculated from a translational in-vivo model in which all variables were held constant except for KPC status. Prospective group assignment in the insect model allowed unknown confounders to be balanced between the groups and was able to provide an unbiased estimate of the effect of KPC status.
Studies conducted by Zarkotou et al.  and Tumbarello et al.  both found that increasing APACHE scores were risk factors for mortality in patients with KPC(+) K. pneumoniae blood stream infections. This would indicate that patient factors other than resistance may drive poor patient outcomes. Similarly, Ben-David et al. and Mouloudi et al. reported that increasing Pitt bacteremia score, Charlson score, and Sequential Organ Failure Assessment (SOFA) score were independent risk factors for mortality, respectively [2, 4]. Both of these studies also found that carbapenem resistance was an independent risk factor for mortality. These studies suggest that KPC(+) K. pneumoniae are being isolated from critically ill patients, and it may be impossible with conventional clinical study methodologies to statistically separate out the attributable virulence from KPC status.
One of our statistical findings may require additional explanation. The bivariate logistic regression model, utilizing the Chi-square test, identified a significant effect of KPC on mortality (p = 0.050). The same data analyzed with the more conservative Fisher’s Exact test resulted in findings with p = 0.055. While bivariate logistic regression is appropriate for multivariate model building, the Fisher’s Exact test is mathematically most appropriate. Despite interpretive differences, the findings demonstrate a trend towards statistical significance that is consistent with previous literature reporting that KPC(+) status portends increased mortality. Had our sample size been larger, the increased study power would have likely resulted in statistical significance (i.e. p < 0.05) regardless of the test used.
While we believe that this novel methodology can improve effect estimates in statistically complex clinical data models, this pilot study has limitations that must be considered. First, the clinical study was retrospective and had a limited number of patients available. Still, the G. mellonella studies demonstrated decreased virulence for KPC(+) strains), and minimally the results demonstrate discordance between the models. Our imputation model attempted to combine these results, but care should be taken in interpreting these results until a larger study can confirm results. The limited patient enrollment concern was partially addressed by matching KPC(+) patients on a 1:4 basis with KPC(−) patients in order to increase the sample size, yet numbers were constrained. Unfortunately, the model cannot easily be improved by simply adding patients. Our data are highly unique because KPC blood infections remain a rare event. For instance, the CDC estimated that there are only 9,300 cases  of carbapenem-resistant Enterobacteriaceae infections in the United States per year. Only a fraction of these are KPC and even a smaller fraction is bloodstream isolates. Thus our numbers are quite large for a single center study. To further illustrate this point, 386 Enterobacteriaceae blood stream infections occurred at our 897-bed, tertiary-care, academic center in 2012. Of these, only 6 isolates were KPC. Multicenter trials may be necessary for future analyses.
Second, there may have been inter-experimental variation in the robustness of the G. mellonella; however, we accounted for this by using an inter-experimental control isolate to in order to adjust for any differences in insects. Third, the potential exists for differences in regional strains of K. pneumoniae; a larger multi-center study could address organism epidemiology. Fourth, it is possible that the G. mellonella results may not correlate with human outcomes, but this possibility is less likely given the translatability of the G. mellonella model with other bacterial virulence studies [12–17] and animal studies [10, 11]. Fifth, optimal therapy for these infections has not been fully defined, and the effect of antibiotic therapy on survival is difficult to assess in this study due to limited power. Finally, due to 1:1 matching of isolates for the insect model, median virulence scores were extrapolated for 45 patients. However, the findings obtained when we analyzed a subset of 1:1 matched patients (i.e. patients for which actual virulence scores were obtained and not calculated) were concordant to those of the entire cohort (data not shown).