Risk of death after first-time blood stream infection in incident dialysis patients with specific consideration on vascular access and comorbidity

This study was prepared as a retrospective, epidemiologic cohort study concerning all patients more than 17 years of age, who initiated dialysis (PD or HD) between 1.1.2010 and 1.1.2014, using the Danish Nephrology Registry (DNR), MiBa, and the Danish nationwide administrative registries as data source.

All citizens in Denmark have a unique personal identification number, which allows for individual-level linkage of information from the applied registries. DNR was established in 1990 and contains comprehensive data concerning all patients receiving RRT [20]. MiBa contains nationwide data on microbiological diagnoses, including all positive blood cultures identified in Denmark since 2010 [21]. Since 1978, information on all admissions to Danish hospitals, according to the international classification code (ICD) and the Nordic classification of surgical procedures (NCSP), respectively, have been registered in the National Patients Registry [22, 23]. Information on age, sex, renal diagnosis, date of referral to a nephrology department, date of dialysis initiation, access type at first access, and date of RRT modality changes (HD, PD or transplantation (RT)) was obtained from the DNR. This registry also supplied information on initial and subsequent dialysis modality and access type, i.e., PD, AVF, NTC and TC. AVGs are not regularly used in Denmark. Date of all surgical TC and AVF procedures, changes from one vascular access type to another, and information on comorbidity at the day of dialysis initiation were obtained from the National Patient Registry. The primary cause of death was obtained from the DNR. We applied a modified Charlson comorbidity index (MCCI), in which each of the following comorbidities was assigned one index point: diabetes (DM), heart failure, ischemic heart disease, peripheral vascular disease, ischemic/hemorrhagic/neurologic CNS disease, chronic obstructive pulmonary disease (COPD), collagenoses, gastric ulcer, liver disease, and cancer (all types, including leukemia). All patients were automatically allocated one index point for CKD. The index was further divided in three levels, i.e. MCCI = 1, MCCI = 2–3 and MCCI > 3 for use in the main statistical model.

The date and result of all positive cultures for the included patients were obtained from MiBa. In specific situations with unreproducible positive cultivations of e.g. opportunistic skin-associated low virulence microorganisms and lack of clinically signs of BSI, positive cultures were registered as ‘possible contamination’ by the corresponding microbiologic departments. Positive cultures identified before the date of first dialysis were excluded. Multiple positive cultures on the same day were classified as polymicrobial. Multiple positive cultures within a 7-day period were regarded as one BSI episode, with the first positive culture being registered as the causative agent. Bacterial diagnoses were classified into 13 major groups for analysis: S. aureus, Coagulase Negative Staphylococci (CNS), Unspecified Staphylococci (where only the genus was reported, and not the species), Streptococci, Enterococci, Other gram-positive bacteria, Escherichia coli (E. coli), Enterobacter, Klebsiella pneumoniae (klebsiella), Pseudomonas aeruginosa (pseudomonas), Other gram-negative bacteria, Fungi, and Other (where a positive blood cultivation was reported without genus or species). In the current study, CNS was defined as the ability of the staphylococcus species to coagulate plasma, which basically included all other Staphylococci than S. aureus. For a full presentation of the microorganisms included in each group, please see Additional file 1.

Crude incidence rates (IRs) for each study outcome per 100 person-years were calculated for patients receiving PD and HD, as well as each HD vascular access type. Differences in baseline characteristics were analyzed with Chi-square test for categorical covariates and Student’s t-test or the non-parametric Kruskal-Wallis and Mann-Whitney tests for continuous variables, respectively. Time dependent Cox proportional hazard models were used to model survival and other endpoints, and outcomes were expressed as hazard ratios (HRs). Patients entered the model at the day of dialysis initiation, and the cohort was followed until occurrence of a study outcome or until 1.1.2015. The following outcomes were assessed in different models: BSI, all-cause mortality, subsequent BSIs (after first-time BSI) and post BSI mortality. In analyses of all-cause mortality and first-time BSI, each observation was split at any change in dialysis modality or vascular access, so that these variables could change time-dependently. In analyses on the effect of first-time BSI in risk of all-cause mortality as compared to patients without BSI, each observation was also split at the day the causal microorganism was isolated, to avoid immortal time bias. In the absence of registered TC or AVF creation, all NTCs were censored 90 days after initiation. In addition, all analyses were adjusted for age, sex, calendar year, MCCI, and Gram stain. The latter was only included in consideration of patients who have had a first-time BSI. Results were tested in sensitivity analyses where all covariates were included at baseline. Furthermore, we modeled the risk of all-cause mortality and first-time BSI in baseline analyses where a 60-day qualifying period, i.e., the date of dialysis initiation (index date) plus 60 days, was added to avoid an anticipated high risk of complications in close temporal proximity to RRT initiation. As we did not have information on the specific infectious focus preceding a BSI, we performed a baseline subgroup analysis of patients with first-time BSI caused by S. aureus, which are most likely to be caused by the vascular/peritoneal access. Also, we investigated the effect of vascular access change subsequent to first-time BSI on risk of post BSI mortality. Time-dependent analyses were censored at the day of RT, and change of modality. Baseline analyses were additionally censored at the day of change in vascular access. Finally, each variable confined by the MCCI was evaluated in a stepwise selection model with alternating forward selection and backward elimination, i.e., only variables with a univariate P value < 0.25 were allowed to be tested further in the stepwise model, and only variables that contributed with a P value of < 0.15 were contained in the fitted model. The proportional hazard assumption, linearity of continuous variables and absence of interaction between variables were fulfilled if not otherwise stated. A two-sided P value of 0.05 or less was considered significant. P values > 0.10 are denoted in the text as ‘not significant’ (NS) unless stated otherwise. Analyses and data management were performed in SAS version 9.4 (SAS Institute Inc. Cary, North Carolina) and R version 3.3.3 (R Foundation for Statistical Computing, Vienna, Austria).

In the period January 2010–January 2014, 2646 incident men and women more than 17 years of age, receiving RRT were identified from DNR, of whom 1742 received HD, 764 PD and 140 underwent RT as their first registered RRT modality, leaving 2506 patient ready for inclusion. At baseline, 434, 479, and 782 HD patients were dialyzed by use of AVFs, TCs, and NTCs, respectively. Forty-seven patients did not have their initial HD access registered. Patients with missing data concerning initial access did not differ significantly from the other patients (data not shown). During follow up, 1027 patients died of which 366 occurred after BSI, 303 underwent RT, and 124 regained some degree of native renal function. The mean age at baseline in the total study population was 64.8 ± 15.2 years, with an overall overweight of male gender (65.4%). In general, patients who initially received a NTC were of the same age; however, they were burdened with excess comorbidity as compared to patients with AVF, TC and PD. Conversely, the group of patients who started on PD were slightly younger, and presented with a lower percentage of patients classified with high MCCI scores. Other baseline characteristics are shown in Table 1. Mean follow up for the total study population was 1.7 ± 1.4 (median 1.5, interquartile range (IQR) 0.3–2.7) years.

We identified 1069 episodes of bacteremia in 784 patients corresponding to an overall IR of 17.7 episodes per 100 person years. A total of 169, 33, and 14 bacteremia episodes occurred two, three and more than three times in the same person. In 71 cases, the putative causative microorganism was registered as ‘possible contamination’ and 30 cases were classified as polymicrobial. With S. aureus (32.1%) and E. coli (13.3%) being the primary individual isolates, the overall Gram-positive and Gram-negative bacteria accounted for 68.0 and 29.7% of the cultured microorganisms, respectively. E. coli was isolated in 19.5% of the bacteremia episodes among patients on PD, and in 11.8% among those on HD. Conversely, S. aureus appeared more frequently among HD patients as compared with PD (35.1% vs. 20.1%). As shown in Fig. 1, the composition of microorganisms in the first encountered bacteremia differed considerably from what was isolated after more than two bloodstream infections, with a slight reduction in Gram positive microorganisms. Where first-time bloodstream infections were dominated by S. aureus and E. coli; fungi, other Staphylococcus species, Enterococci, Klebsiella and Pseudomonas were significantly more prevalent among patients with more than two blood stream infections (Fig. 1). The overall distribution of the 13 pre-specified microbiologic groups for first-time BSI according to the vascular access modality as well as PD are shown in Table 1.

Crude IRs for first-time BSI and all-cause mortality are shown in Table 2. Censored IRs are listed in Figs. 2 and 3. In the fully adjusted time-dependent Cox proportional hazard analyses we found an increased risk of BSI for patients receiving HD by use of AVF [HR 1.35 (95% CI 1.10–1.67), P = 0.005], TC [HR 2.25 (95% CI 1.74–2.90), P < 0.001], and NTC [HR 6.87 (95% CI 5.57–8.48), P < 0.001] relative to patients receiving PD (Fig. 3). Also, there was an increased risk of all-cause mortality associated with TC and NTC with HRs of 1.42 (95% CI 1.14–1.77, P = 0.002) and 5.22 (95% CI 4.36–6.25, P < 0.001), respectively. However, there was no significant difference in risk of all-cause mortality for AVF [HR 0.86 (95% CI 0.72–1.03), NS] relative to PD (Fig. 3). Comparable results were found in baseline analyses where patients entered the model at the date of dialysis initiation, and when a 60-day qualifying period was applied (Additional file 2).

In time-dependent Cox analyses modeled to examine the risk of death after first-time BSI as compared to all patients without BSI, regardless of dialysis and vascular access type, we found a significantly increased risk of death associated with all modalities with HRs of 2.72 (95% CI 1.70–4.37, P < 0.001), 1.99 (95% CI 1.62–2.43, P < 0.001), 2.92 (95% CI 2.12–4.03, P < 0.001), and 6.44 (95% CI 5.15–8.08, P < 0.001) for PD, AVF, TC and NTC, respectively (Fig. 2). Confining the analysis to only include patients after their first-time BSI with PD as reference, there was no significant difference between AVF and TC relative to PD; whereas NTC remained significantly associated with a marked increase in risk of death (Fig. 3). Comparable results were found in analyses based on baseline variables, as well as in the subgroup analyses that only included patients with a first-time BSI caused by S. aureus (Additional file 2). Also, the risk of subsequent BSI episodes was significantly higher in HD patients who encountered their first BSI with a NTC [HR 2.29 (95% CI 1.09–4.82), p = 0.030], whereas no significant difference was found for patients with AVF [HR 0.94 (95% CI 0.46–1.93), NS] and TC [HR 1.68 (95% CI 0.77–3.70), NS]. In post hoc analyses, HRs for post BSI mortality and subsequent BSI with TCs, using AVF as reference, were 1.57 (95% CI 1.07–2.29, P < 0.021) and 1.78 (95% CI 1.13–2.83, P < 0.014), respectively. Finally, we found an overall decreased risk of all-cause mortality associated with patients who changed their vascular access in the period after first BSI episode [HR 0.61 (95% CI 0.47–0.79), P < 0.001]; however, this association was not evident in a subgroup analysis restricted to those who did not receive an AVF [HR 0.73 (95% CI 0.41–1.23), NS]. Estimates regarding the other included covariates in the main time-dependent Cox models, and from the two-step fitted model covering each of the covariates included in the modified comorbidity score, are shown in Tables 3 and 4, respectively.