Local signs at insertion site and catheter-related bloodstream infections: an observational post hoc analysis using individual data of four RCTs

Background Little is known on the association between local signs and intravascular catheter infections. This study aimed to evaluate the association between local signs at removal and catheter-related bloodstream infections (CRBSI), and which clinical conditions may predict CRBSIs if inflammation at insertion site is present. Methods We used individual data from four multicenter randomized controlled trials in intensive care units (ICUs) that evaluated various prevention strategies for arterial and central venous catheters. We used multivariate logistic regressions in order to evaluate the association between ≥ 1 local sign, redness, pain, non-purulent discharge and purulent discharge, and CRBSI. Moreover, we assessed the probability for each local sign to observe CRBSI in subgroups of clinically relevant conditions. Results A total of 6976 patients and 14,590 catheters (101,182 catheter-days) and 114 CRBSI from 25 ICUs with described local signs were included. More than one local sign, redness, pain, non-purulent discharge, and purulent discharge at removal were observed in 1938 (13.3%), 1633 (11.2%), 59 (0.4%), 251 (1.7%), and 102 (0.7%) episodes, respectively. After adjusting on confounders, ≥ 1 local sign, redness, non-purulent discharge, and purulent discharge were associated with CRBSI. The presence of ≥ 1 local sign increased the probability to observe CRBSI in the first 7 days of catheter maintenance (OR 6.30 vs. 2.61 [> 7 catheter-days], pheterogeneity = 0.02). Conclusions Local signs were significantly associated with CRBSI in the ICU. In the first 7 days of catheter maintenance, local signs increased the probability to observe CRBSI.


Background
Infections due to central venous catheters and arterial catheters significantly increase hospitalization duration, hospital costs, patient morbidity and mortality in critically ill adult patients [1][2][3]. Moreover, intravascular catheter-related bloodstream infections are frequent events in the intensive care unit (ICU) setting [4]. Numerous risk factors associated with catheter-related bloodstream infections (CRBSIs) have been identified in several studies [5][6][7][8]. More specifically, in the last 20 years, only one old study assessed the exit-site signs as a predictor for intravascular catheter infections [9]. These data reflect practices from an arguably bygone era [10]. Indeed, the use of intravascular catheter bundles has Open Access *Correspondence: niccolo.buetti@gmail.com 1 University of Paris, INSERM, IAME, 75006 Paris, France Full list of author information is available at the end of the article recently changed the landscape of the risk factors [11] and, probably, of the clinical predictors for intravascular catheter infections. Moreover, recent studies suggested that the epidemiology of intravascular catheter-related infections is changing [12,13]. To the best of our knowledge, there are no recent data on the role of exit-site signs with regards to intravascular catheter infections. In short-term intravascular catheters, the extraluminal route of infection (i.e., originating from the dermal surface) predominates [14]: it is conceivable that local signs at catheter insertion may be linked to intravascular shortterm catheter infections. Our primary objective was to determine whether local symptoms and signs at insertion site were associated with CRBSI using data from four randomized controlled trials (RCTs). The secondary objective was to determine which clinical conditions increase the probability to observe CRBSI if inflammation at insertion site is present.

Data sources
This study included four longitudinal databases from four RCTs: DRESSING1 [15], DRESSING2 [16], ELVIS [17] and CLEAN [18]. Merging the data was facilitated by the fact that all these studies rely on the same definitions and similar inclusion criteria. These four studies had similar objectives: to assess the impact of specific prevention strategies on the rate of colonization and infection of intravascular catheters in ICU. Details on the studies are in the Additional file 1. The studies were approved by the national ethics committee. This post hoc analysis was not pre-planned. All RCTs complied with CONSORT guidelines and the current analysis complied with the STROBE guidelines for observational studies [19,20].

Study patients
The patients were recruited from 2006 to 2014 in various ICUs in France. The patients were included if they required a catheterization with a short-term central venous catheter, a short-term dialysis catheter, or peripheral arterial catheter (AC), with an expected duration of use of more than 48 h (see Additional file 1). The patients underwent follow-up until 48 h after ICU discharge.

Study catheters
For the current analysis, short-term dialysis catheters were considered as central venous catheters (CVCs). All catheters in a given patient were managed in the same way, and all study centers complied with the French recommendations for catheter insertion and care, which are similar to CDC recommendations [21] (see Additional file 1). Catheters were removed if no longer needed, in the case of dysfunction or thrombosis or if an infection was suspected. Catheter tips were cultured using quantitative culture techniques.

Definitions and evaluation criteria
Each study patient was evaluated daily by a team of research nurses. The patient was asked about discomfort at the insertion site, and the site was visually inspected for inflammation. These data were routinely collected at catheter removal. Local symptom (i.e., "pain") or signs as redness (i.e., redness ≥ 0.5 cm), pain, non-purulent discharge and purulent discharge were noted as either absent or present. We mostly focused on a composite variable including "≥ 1 local signs or symptom. " For sake of simplicity, we used the term "local signs" instead of "local signs or symptom" throughout the text. Of note, the center investigators reported whether a catheter infection was subjectively suspected ("suspicion of infection"). The information on pathological body temperature (body temperature ≥ 38·5 °C or hypothermia with body temperature ≤ 36·5 °C) 24 h before catheter removal was collected and analyzed.
According to French [22] and American guidelines [23], the following definitions were used. A CRBSI (or "infected catheter") was a combination of (outcome): (1) one or more positive peripheral blood cultures sampled 48 h before or after catheter removal; (2) isolation of the same organism (same species and same susceptibility pattern) from the colonized catheter or from the catheter insertion site, or a blood culture differential time to positivity of 2 h or more [24]; and (3) no apparent source of bacteremia other than the catheter. If a patient had a positive blood culture for coagulase-negative staphylococci (CoNS), a CoNS CRBSI was diagnosed solely if the pulsotype was the same among the strains recovered from the catheter and blood culture. Alternatively, at least two positive cultures with CoNS from separate blood samples were required. Catheter colonization was defined as a quantitative catheter tip culture yielding ≥ 1000 colonyforming units/mL [25]. For patients without catheter cultures, a masked adjudication committee determined whether bloodstream infections were classified as catheter related. The skin colonization was evaluated using semi-quantitative insertion-site cultures: the insertion site was sampled immediately before catheter removal. Because the size of the counting surface was different across studies, we created a semi-quantitative variable with sterile, low-grade colonization, and high-grade colonization according to the median of quantitative cultures obtained in each study.

Statistical analysis
Characteristics of patients and catheters were described as count (percent) or median (interquartile range) for qualitative and quantitative variables, respectively.
The statistical plan had two steps: (1) The primary objective was to evaluate the association between local sign and CRBSI after adjusting for the other CRBSI confounders; (2) the secondary objective was to evaluate the role of the different local signs for observing CRBSI in subgroup of clinically relevant populations. Moreover, we added an explanatory analysis including skin colonization at removal.
We used univariate logistic regressions in order to identify variable associated with CRBSI, and we calculated odds ratios (ORs) for ≥ 1 local sign, redness, pain, non-purulent discharge, and purulent discharge. We then performed multivariate logistic regression forcing the "local sign" (i.e., ≥ 1 local sign, redness, pain, non-purulent discharge, and purulent discharge) variables and adjusting for the other variables associated CRBSI (i.e., outcome) in order to determine the association between local signs and CRBSI. Of note, the final choice of adjustment variables was based on the following clinical well-known risk factors for CRBSI [26]: sex, SOFA score, duration of catheter maintenance, experience of the operator, catheter type, insertion site, skin antisepsis, and antibiotic at insertion [5-7, 18, 27]. Logistic models were stratified for the different centers included in the analysis. Moreover, to evaluate the possible clustering effect of multiple catheters per patient, we performed a sensitivity analysis for the first catheter inserted in an individual patient.
The risk to observe CRBSI for the different local signs in subgroup of clinically relevant populations (i.e., suspicion of catheter infection vs. any suspicion of infection, pathological temperature within the 24 h before removal vs. physiological temperature, duration of catheter maintenance ≤ 7 days vs. > 7 days, catheter type CVC vs. AC, presence of immunosuppression or not, or SOFA score ≤ 11 points vs. > 11 points) was calculated with univariate logistic regression. The heterogeneity between each subgroup of clinically relevant populations was assessed using the Cochran's Q test.
For the explanatory analyses including skin colonization, we compared groups using Chi-square or Fisher tests, as appropriate.

Description of patients and catheters
A total of 6976 patients and 14,590 catheters (101,182 catheter-days) from 25 ICUs with described local signs were included in this study (see Additional file 1: Figure  E1): 2033 (29.1%) from the CLEAN study, 1460 (20.9%) from ELVIS, 1614 (23.1%) from DRESSING1 and 1869 (26.8%) from DRESSING2. The characteristics of the patients and catheters are illustrated in Tables 1, 2, and 3.

At least one local sign
At least one local sign was associated with CRBSI in the univariate (OR 4.72, 95% confidence interval [CI] 3.18-7.00, p < 0.01, Fig. 1 and Additional file 1: Table E1) and after adjusting on confounders (OR 4.42, 95% CI 2.93-6.69, p < 0.01, Fig. 1 and Additional file 1: Table E2). For ≥ 1 local sign, similar results were observed only when the first catheter in an individual patient was considered (OR 4.63, 95% CI 2.35-9.12, p < 0.01). This association remained statistically significant from 2007 to 2014 (data not shown).
At least one sign was present in 40.4% of the infected catheters (vs. 13.1% of the non-infected catheters, p < 0.01, Fig. 2), and its probability to observe CRBSI was higher in the first 7 days of catheter maintenance (OR 6.30 vs. OR 2.61 for > 7 days, p for heterogeneity = 0.02).

Sensitivity, specificity, positive predictive value (PPV), negative predictive value (NPV), positive and negative likelihood ratio
The sensitivity, specificity, and PPV, NPV of each local sign for predicting CRBSI are illustrated in Table 4.

Skin colonization at catheter removal
The skin colonization at catheter removal was significantly more often colonized in case of ≥ 1 local sign (p < 0.01), redness (p < 0.01), non-purulent (p = 0.01), and purulent discharge (p < 0.01, Additional file 1: Table E8). The skin colonization did not significantly differ according to pain at removal (p = 0.74).

Discussion
Through high quality data from four multi-centric RCTs and after correction for other confounders, this post hoc analysis showed that local signs at the exit site were clearly associated with intravascular catheter infections in both CVCs and ACs. This finding was more pronounced for redness, non-purulent discharge, and purulent discharge. Three old studies examined the condition of the insertion site as a clinical predictor of CRBSI in adults. The first study included 169 patients, and the authors illustrated that infection was also associated with redness at the insertion site greater than 4 mm in diameter [30]. The second was a prospective, observational study that enrolled 1353 CVCs and showed that inflammation at the catheter site was absent in approximately 70% of CRBSIs [31]. To the best of our knowledge, only one observational study exhaustively explored this research question in the last 20 years [9]. In 2002, Safdar et al. showed that the insertion site appearance was not associated with catheter colonization or CVC-related bloodstream infections [9]. However, this study (1) reported old data collected from 1998 to 2000, probably reflecting the era prior to the routine implementation of infection prevention bundles; (2) included less than one-tenth of all catheters included in our post hoc analysis; (3) yielded a large proportion of CoNS among CRBSI; and (4) analyzed only CVC, without including AC. Similarly to the DRESSING1 and CLEAN trials, Safdar et al. also included patients from two RCTs [9], one evaluating chlorhexidine-gluconate sponge dressings [32] and the other evaluating chlorhexidine skin disinfection for prevention of CRBSI [33]. In contrast to Safdar et al., local signs in our analysis were significantly associated with an increased risk for CRBSI. To support our findings, we found that skin colonization at catheter removal occurred more frequently if local signs were present. Moreover, we observed that an important proportion of CRBSI were due to S. aureus and Gram-negative bacilli, organisms that elicit more inflammation compared to CoNS which are less virulent [34] and in our study represented only 13% of CRBSI. In contrast, in the Safdar's study, 87% of microorganisms identified were CoNS. In this context, and consistently with our findings, a change in the epidemiology of intravascular catheter infection toward lower prevalence of CoNS and increasing proportion of Gram-negative microorganisms has been documented in several recent studies [12,13,35]. In light of these considerations, the description of this cohort may probably better represent this issue.
Our findings have some important clinical Implications which, to date, have never been assessed. First, compared to the most recent literature available, local signs are associated with CRBSI, and their presence should elicit investigations for diagnosing potential intravascular catheter infections. Second, local signs may help clinicians in a specific clinical condition: if at least one local sign is present within the first 7 days of catheter maintenance, it further increases the probability to observe a catheter infection. Therefore, clinicians should deserve particular attention to the catheter insertion site in the first week after the insertion and a promptly catheter removal should be considered. Moreover, and not surprisingly, purulence at insertion site is a strong reason to remove a catheter. Interestingly, immunosuppression, pathological temperature at removal, catheter type, and severity of illness with the presence of local signs did not significantly help clinicians in predicting CRBSI. Moreover, the solely intuition of the physician who subjectively suspected an intravascular catheter infection in the presence of local signs did not increase the probability to observe a CRBSI. On the other hand, in catheters removed for suspected infection without any local signs, the probability to get infected remained low. An old RCT including a relatively low number of hemodynamically stable critically ill patients without proven bacteremia and local sign at insertion site illustrated that a "watchful waiting strategy" (versus immediate catheter removal) permitted a substantial decrease in the number of unnecessarily removed CVCs without increased morbidity [36]. In the case of suspicion of infection without any local signs, a catheter retention may be considered and would reduce unessentially removals. A post hoc analysis for this subpopulation in our cohort showed that CRBSI were not associated with increased mortality (OR 1.37, 95% CI 0.68-2.77, p = 0.38), thus suggesting, at least when catheters were routinely removed, a low mortality risk for this population.
Our study has important limitations. First, it was an observational study, and unmeasured factors may persist and cause residual confounding. However, we presented high quality exhaustive data that were prospectively collected by trained investigators and study monitors during all RCTs. Second, all RCTs were conducted in Universityaffiliated ICUs in France from 2006 to 2014 and included only selected short-term intravascular catheters, thus limiting the generalizability of our results. Third, we did not have any data reported for the variable "pain" in intubated or comatose patients at removal, thus probably leading to an underestimation of the proportion of this specific symptom. Fourth, redness was defined as ≥ 5 mm diameter: a smaller threshold compared to the IDSA guidelines which declared 2 cm as relevant for CRBSI [23]. We selected our cutoff in reason of the results of an old study which illustrated that infection was associated with redness > 4 mm in diameter [30]. Fifth, we described a large database designed to investigate the impact of certain prevention measures, and interactions may have occurred among the various study groups. Finally, PPVs of local signs were low, thus limiting the utilization of local signs to create algorithms for better decision-making in all patients with short-term intravascular catheters.

Conclusions
Using the largest dataset ever collected from four multicentric RCTs conducted with consistent catheter care, we showed that local signs were clearly associated with infections in short-term catheters in the ICU. In the first 7 days of catheter maintenance, local signs increased the probability to observe CRBSI.

Supplementary Information
The online version contains supplementary material available at https ://doi. org/10.1186/s1305 4-020-03425 -0.  Figure E1 flow-chart; Figure E2 Risk for developing CRBSI for non-purulent discharge and purulent discharge by different subgroups), supplementary Tables (Table E1: Univariate logistic model for catheter-related bloodstream infection, stratification by ICU; Table E2: Multivariate logistic models for CRBSI forcing the variable ≥ 1 local sign, stratification by ICU; Table E3: Multivariate logistic models for CRBSI forcing the variable redness, stratication by ICU;