Predictors of mortality in solid organ transplant recipients with bloodstream infections due to carbapenemase‐producing Enterobacterales: The impact of cytomegalovirus disease and lymphopenia

Treatment of carbapenemase‐producing Enterobacterales bloodstream infections in solid organ transplant recipients is challenging. The objective of this study was to develop a specific score to predict mortality in solid organ transplant recipients with carbapenemase‐producing Enterobacterales bloodstream infections. A multinational, retrospective (2004‐2016) cohort study (INCREMENT‐SOT, ClinicalTrials.gov NCT02852902) was performed. The main outcome variable was 30‐day all‐cause mortality. The INCREMENT‐SOT‐CPE score was developed using logistic regression. The global cohort included 216 patients. The final logistic regression model included the following variables: INCREMENT‐CPE mortality score ≥8 (8 points), no source control (3 points), inappropriate empirical therapy (2 points), cytomegalovirus disease (7 points), lymphopenia (4 points), and the interaction between INCREMENT‐CPE score ≥8 and CMV disease (minus 7 points). This score showed an area under the receiver operating characteristic curve of 0.82 (95% confidence interval [CI] 0.76‐0.88) and classified patients into 3 strata: 0‐7 (low mortality), 8‐11 (high mortality), and 12‐17 (very‐high mortality). We performed a stratified analysis of the effect of monotherapy vs combination therapy among 165 patients who received appropriate therapy. Monotherapy was associated with higher mortality only in the very‐high (adjusted hazard ratio [HR] 2.82, 95% CI 1.13‐7.06, P = .03) and high (HR 9.93, 95% CI 2.08‐47.40, P = .004) mortality risk strata. A score‐based algorithm is provided for therapy guidance.


K E Y W O R D S
antibiotic drug resistance, clinical research/practice, infection and infectious agentsbacterial, infectious disease, organ transplantation in general

| INTRODUC TI ON
Infections due to carbapenemase-producing Enterobacterales (CPE) are dramatically increasing worldwide. 1 Numerous transplant centers have been affected by outbreaks and many suffer a subsequent endemic situation. [2][3][4] The extreme difficulty of their treatment and the high mortality (30%-50%) associated with these infections explain their importance in the solid organ transplant (SOT) setting. 4,5 Their epidemiology has been extensively studied and specific recommendations for infection control and clinical management of these infections in SOT recipients have been published. [4][5][6][7][8] Nevertheless, current recommendations are based on observational studies conducted in the general population, [9][10][11][12][13] while the specific risk factors and clinical impact of infections due to CPE in SOT recipients remain to be elucidated. Large, multicenter studies, truly representative of the SOT patient population, are needed to develop risk-stratification tools to assist in guiding the management of these infections.
The objectives of this study were the following: (1) to validate the INCREMENT-CPE score to predict all-cause mortality of CPE bloodstream infections (CPE-BSI) in the SOT population; (2) to explore whether a new predictive score, INCREMENT-SOT CPE score, improves the predictive capacity, and (3) to check the utility of the new score to guide antibiotic therapy (monotherapy or combination) in different mortality risk groups.

| Variables and definitions
Clinically significant BSI was defined as the isolation of a carbapenemase-producing Enterobacterales in blood. 15  Microbiological variables included Enterobacterales species, carbapenemase type, and antimicrobial susceptibility data. Finally, we recorded INCREMENT-CPE mortality risk score 11,12 and the therapy administered (dates and doses of antibiotics). Empirical therapy was considered appropriate when an active drug was administered before the susceptibility profile. Targeted therapy was considered appropriate if it included an active drug and was administered within 5 days or earlier after the blood culture (day 0), and once the susceptibility profile was available. An active therapy was classified as monotherapy if it included 1 single active drug and as combined therapy if it included 2 or more active drugs. If the antibiotic regimen administered was changed, we considered that administered for ≥50% of the duration of therapy (for patients who died sooner than 48 hours after the start of therapy, 1 complete day of therapy was required). Meropenem and imipenem were considered active when MIC < 4 mg/L (monotherapy) or MIC 8-16 mg/L and administered in combination with ertapenem (monotherapy) or other active drugs (combination therapy). Tigecycline was not considered active for a urinary source. Variables were collected in a centralized electronic clinical research file. The database was curated and queries were sent to participating centers for missing or inconsistent data.

| Microbiological studies
The identification of microorganisms and susceptibility testing were performed at each participating center. The identification of microorganisms and susceptibility testing were performed at each participating center, using standard microbiological techniques.
Susceptibility was studied using automated systems or disk diffusion at each local laboratory and interpreted using the 2015 Clinical & Laboratory Standards Institute (CLSI) break points. 19 For isolates obtained before 2015, minimum inhibitory concentrations were reviewed and the susceptibility category was assigned accordingly; when the minimum inhibitory concentration (MIC) was not available or the available data had a MIC less than or equal to the older susceptibility break point, these were considered as susceptible if so reported by the local laboratory. Isolates were considered to be carbapenemase producers if a carbapenemase gene was detected by a molecular method.

| Statistical procedures
Continuous variables were compared using the Kruskal-Wallis test.
Categorical variables were compared using the χ 2 test or Fisher exact tests. Survival distributions were compared using the log-rank test and were graphically displayed using Kaplan-Meier curves.
Validation of the INCREMENT-CPE score 12 was performed by calculating the area under the receiver operating characteristic curve (AUROC) for observed data, the sensitivity (Se) and specificity.
Multivariable logistic regression was used to develop a new score. The original INCREMENT-CPE score (modified by excluding the variable "inappropriate empirical and early targeted therapy," since we aimed to investigate different aspects of treatment for the new score) was dichotomized into 2 previously validated categories of risk (<8, low risk vs ≥8, high risk). 11 To control for the site effect, we classified centers into low mortality-risk and high mortality-risk using TreeNet (Salford Predictive Modeller software) and considering all other variables ( Figure S1). The study period (to control for changes in clinical management over time), the source of BSI and lymphocyte count were dichotomized by CART (Classification and Regression Tree, Salford Predictive Modeller Software; Table S2 and Figure S2). The variance inflation factor (VIF) value for every variable was calculated to control the influence of multicollinearity. We assumed lack of multicollinearity if all variables had a VIF value <2. The variable "high-mortality risk center" was included in the analysis to obtain a predictive model for which this effect was controlled but was not considered for the score. Potential interactions between variables were explored using TreeNet and those selected were included in the models. Variables with a P ≤ .20 in the final models were selected for the assignment of a score, provided their inclusion significantly improved the predictive capacity of the model. A weighted score for each variable was calculated dividing each regression coefficient by one-half of the smallest coefficient and rounding to the nearest integer. The prediction ability of a model was examined by calculating its AUROC with a 95% CI; Se, specificity, positive predictive value (PPV), negative predictive value (NVP), and accuracy (Ac) were calculated for different breakpoints.
Sensitivity analysis for the INCREMENT-SOT-CPE score was performed using Salford Predictive Modeller Software to check the robustness of its predictive ability. Fifteen subgroups of the cohort with a 20% sample size were randomly extracted (43 cases per subgroup), and the AUROC of the score to predict 30-day all-cause mortality was calculated for each subgroup. A minimum, maximum, and median value of AUROC was obtained. The process was repeated another 7 times, extracting 15 subgroups with sample sizes ranging 30% to 90% (10% intervals, thus obtaining 8 average AUROCs, maximum, and minimum values).
For the analysis of the association of monotherapy vs combination therapy with mortality, a propensity score for receiving combination therapy was calculated using a nonparsimonious logistic regression model. The impact of combination therapy was studied by Cox-Regression, adjusting by propensity score and other potential confounders, after checking for collinearity.
The analyses were carried out using R software (version 3.0.1), SPSS 25.0 (SPSS Inc), and Salford Predictive Modeller software 8.2 (includes CART and TreeNet).

| Cohort features and validation of the INCREMENT-CPE mortality score (objective I)
Among 228 patients included, 216 fulfilled inclusion criteria ( Figure 1).
The predictive value of the INCREMENT-CPE mortality score 12 was studied. We found that this score was associated with 30-day all-cause mortality (odds ratio, 1.40 per unit; 95% CI, 1.  Table S3. For an INCREMENT-CPE score value ≥8, previously validated as a cut-off value predictive for low vs high mortality in non-SOT patients, 11,12 the calculated NPV and PPV in the SOT cohort were 84.7% and 50.4%, respectively.

| Development of the new INCREMENT-SOT-CPE mortality score (objective II)
We explored SOT-related variables that could improve the predictive capacity of the INCREMENT-CPE score in our population.

PÉREZ-NADALES Et AL.
Variables associated with 30-day mortality in the final model were: INCREMENT-CPE score ≥8 (excluding the variable about therapy from this score), CMV disease in the previous 30 days, lymphocytes ≤600 units per mm 3 , and lack of source control ( Table 2); the interaction between CMV disease and INCREMENT-CPE score ≥8 was negative and with a similar (but negative) β coefficient as CMV disease, indicating that CMV disease does not further increase the risk of death if the INCREMENT-CPE-score is ≥8, but do so only if the score is <8 ( Figure S3). The variable inappropriate empirical therapy was kept in the final model since its inclusion improved the predictive capacity. None of the final variables included in the multivariate model showed multicollinearity (VIF ≤ 1.06, Table S4). The AUROC of the resulting logistic regression model was 0.84 (95% CI, 0.78-0.89). The score assigned to each variable according to its beta regression coefficient is shown in Table 3 Table S5. The NPV and PPV for a score value ≥8 were 89.4% and 53.4%, respectively; and for a score ≥12, NPV and PPV were 78.8% and 72.3%, respectively. A classification into low (score 0-7), high (score [8][9][10][11], and very-high (score 12-17) mortality was developed, with mortality rates of 11.4% (10/87), 35.3% (23/65), and 71.8% (46/64), respectively (Table S6). The sensitivity analysis (see Methods for details) confirmed the robustness of the model; the minimum value of the AUROCs for all subcohorts was always >0.70 and the average AUROC value was >0.80 ( Figure S4).

| Proposed algorithm for clinical practice
In order to apply these results to the clinical management of SOT patients with CPE-BSI, we propose an algorithm that requires calculation of INCREMENT-CPE score 11,12

| D ISCUSS I ON
Our results indicate that being a recipient of SOT does not seem to worsen the prognosis of CPE-BSI. Thirty-day all-cause mortality in our INCREMENT-SOT cohort was 36.6%, higher that in the pre-CPE era 20 and similar to that previously reported in other series in SOT,4 and in the general population. 12 Some studies and a meta-analysis have reported a higher mortality (>40%) when CPE-BSI is caused by K pneumoniae. In our study, the type of Enterobacterales was not associated with mortality in the analysis after adjusting by other exposures, as previously observed. 15,21,22 The development of the new INCREMENT-SOT-CPE score was based on the INCREMENT-CPE score, which had been previously validated in the general population in different studies. 9,11,12,23,24 We used this strategy because there are no specific studies in SOT and many transplant groups use this predictive model, which takes into account variables important in any type of patient with BSI, including SOT. Besides, the inclusion of this general model in our new score reinforces the utility of the new model in posttransplant periods, such as the postoperative period, when the full impact of immunosuppression-derived from prolonged exposure to suppressive therapies-is still absent. 25 Finally, an alternative model including individual variables-transplant and nontransplant-instead of the INCREMENT-CPE score, showed a lower predictive capacity and was thus not considered.
We additionally studied the impact of specific transplant variables that complemented the INCREMENT-CPE score on the prognosis of this type of infections in SOT patients. Thus, our results indicate that the predictive capacity of the INCREMENT-CPE score can be improved when it is combined with other mortality predictors such as source control, appropriate empirical therapy, and variables related to immunosuppression, ie, lymphopenia and CMV disease.
The application of these additional predictors is very important in patients with INCREMENT-CPE score <8, when the score can be applied to indicate monotherapy or combined therapy ( Figure 2).
It is obvious that an adequate control of the source and an appropriate empirical treatment can improve the prognosis of the bacterial infection. Lymphopenia can be a surrogate marker of over-immunosuppression. Nevertheless, some experts believe that a reduction in immunosuppression may lead to higher mortality by increasing the capacity of the immune system to induce a systemic inflammatory response. 26 CMV is an immunomodulatory virus that cans favor bacterial infections. 27 Theoretically, CMV prevention could reduce this increased risk, 28 although recent consensus does not recommend CMV prophylaxis in the scenario of solid organ transplantation. 27,29 This is further complicated by the fact that sepsis may increase CMV reactivation. 29,30 Our results suggest that CMV disease increases mortality in SOT recipients with CPE-BSI, although CMV disease may also be a mere marker of the net-state of over-immunosuppression, which would be ultimately associated with all-cause mortality.
Interestingly, the data from our study suggest that CMV disease does not increase the risk of death further in SOT recipients with a high underlying risk of death, as measured by the INCREMENT-CPE score, but only in patients with a lower underlying risk. Unfortunately, data on CMV prophylaxis was not collected in this study, but our results open the door to further research about whether prevention of CMV may be beneficial in SOT recipients colonized by CPE in order to improve their outcomes in case of an invasive infection due to these bacteria.
Our study has the limitations of retrospective studies, despite applying a rigorous definitions and statistical analyses to control biases. A second limitation is that we have analyzed patients not treated with the newly available drugs (ie, ceftazidime-avibactam or meropenem-vaborbactam). The impact of the new drugs on the TA B L E 3 INCREMENT-SOT-CPE score: assignment of scores based on the regression coefficients obtained for the selected variables using multivariable logistic regression observational studies in the SOT population treated with the "classic" drugs will still be relevant in many areas. Another limitation is that the sample size of our cohort precluded the selection of derivation and validation subcohorts (see reference 12). The sensitivity analysis confirmed the internal robustness of our model; nevertheless, an external validation in a prospective cohort would be desirable. Finally, KPC carbapenemase may be overrepresented in our cohort, as compared to other carbapenemases.
To conclude, in transplant centers with outbreaks or endemia by CPE, identification of colonized patients is important so that empirical treatment with CPE coverage can be readily administered in case of BSI development. In this study, we have identified transplant-related variables specifically associated with the risk of mortality in SOT recipients with CPE-BSI. We expect this will help to identify patients at high risk of death and allow a more personalized clinical management (ie, prevention of cytomegalovirus disease and the judicious use of immunosuppression in order to avoid lymphopenia).

ACK N OWLED G M ENTS
and grants from Astellas and Merck, outside the submitted work; PAG reports grants from MSD and personal fees from MSD, Biotest, Angelini, Paratek, Gilead, Becton Dickinson, and Nordic Pharma, outside the submitted work. The other authors have no conflicts of interest to disclose.

DATA AVA I L A B I L I T Y S TAT E M E N T
The data that support the findings of this study are available from FIBICO (Fundación para la Investigación Biomédica de Córdoba).
Restrictions apply to the availability of these data, which were used under license for this study. Data are available from the authors with the permission of FIBICO.