Background. Central venous catheters are universally used during the treatment of critically ill patients. Their use, however, is associated with a substantial infection risk, potentially leading to increased mortality and costs. We evaluate clinical and economic outcomes associated with nosocomial central venous catheter-related bloodstream infection (CR-BSI) in intensive care unit (ICU) patients.
Methods. A retrospective (1992–2002), pairwise-matched (ratio of case patients to control subjects, 1:2 or 1:1), risk-adjusted cohort study was performed at a 54-bed general ICU at a university hospital. ICU patients with microbiologically documented CR-BSI (n = 176) were matched with control subjects (n = 315) on the basis of disease severity, diagnostic category, and length of ICU stay (equivalent or longer) before the onset of CR-BSI in the index case patient. Clinical outcome was principally evaluated by in-hospital mortality. Economic outcome was evaluated on the basis of duration of mechanical ventilation, length of ICU and hospital stays, and total hospital costs, as derived from the patient's hospital invoices.
Results. The attributable mortality rate for CR-BSI was estimated to be 1.8% (95% confidence interval, -6.4% to 10.0%); in-hospital mortality rates for patients with CR-BSI and matched control subjects were 27.8% and 26.0%, respectively. CR-BSI was associated with significant excesses in duration of mechanical ventilation, duration of ICU and hospital stays, and a significant increase in total hospital cost. Linear regression analysis with adjustment for duration of hospitalization and clinical covariates, revealed that CR-BSI is independently associated with higher costs.
Conclusions. In ICU patients, CR-BSI does not result in increased mortality. It is, however, associated with a significant economic burden, emphasizing the importance of continuous efforts in prevention.
Central venous catheters are indispensable in the treatment of intensive care unit (ICU) patients, but use of these catheters is associated with a risk of infectious complications. Central venous catheters are the most common source of nosocomial bloodstream infection, and it has been estimated that>250,000 episodes occur annually in the United States . The attributable mortality rate for nosocomial bloodstream infection has been estimated to be 25% . The outcome for catheter-related bloodstream infection (CR-BSI) may be better than that for secondary bloodstream infections, because the source of the former is more easily eradicated and because its etiologic agents are often coagulase-negative staphylococci, which are less pathogenic than Staphylococcus aureus, gram-negative bacteria, and Candida species [3-8]. In general, bloodstream infections also impose an important financial burden associated with prolongation of hospitalization and the substantial added cost of health care, but the economic consequences of CR-BSI are rarely investigated [9, 10]. The objective of this study is to evaluate the clinical and economic effects of nosocomial central venous CR-BSI in adult ICU patients.
Setting and design. This study was conducted at the 1060-bed Ghent University Hospital. The ICU has 54 beds and includes medical, surgical, and cardiosurgical ICUs and a burn unit. Approximately 3300 patients are admitted to the ICU each year. No changes in mean age, duration of ICU stay, or APACHE II scores  were observed during the study period.
A retrospective, pairwise-matched (ratio of case patients to control subjects, 1:2 or 1:1), risk-adjusted cohort study was performed with adult patients admitted to the ICU during an 11-year period (January 1992 through December 2002). Every ICU patient with CR-BSI was matched with 2 ICU patients who did not have evidence of bloodstream infection at any site during their ICU stay.
Case finding. A laboratory-based surveillance of positive blood culture results is performed by the infection control team. Clinical significance, determination of whether the infection was nosocomial or community acquired, and the origin of the infection are registered. This prospective, case-based surveillance program was used for the retrospective search for all ICU patients with nosocomial CR-BSI. For patients with multiple episodes of CR-BSI, only the first episode was taken into account.
Matching procedure. Control patients were preferably selected from the year of the matched case patient's hospital admission. Matching was based on the APACHE II classification system (i.e., an equal APACHE II score [±4 points] and an equal principal diagnosis leading to ICU admission) . Control patients were also required to have had an ICU stay at least as long as the matched case patient's stay before onset of the CR-BSI, as well as short-term use of a central venous catheter throughout this period [12, 13]. APACHE II is a standard for the comparison of disease severity in ICU patients. Because the expected mortality can be derived from the APACHE II system (APACHE II score and diagnostic category), this matching procedure results in an equal a priori expected mortality and allows one to assess the impact of a subsequent complication on outcome. Therefore, this matching procedure has been repeatedly used in matched cohort studies in ICU settings [14-17]. Control patients were selected without knowledge of outcome. If there were>2 potential control subjects, matching was based on the admission date nearest to that for the patient with CR-BSI.
Definitions. CR-BSI was defined as the presence of a positive culture result for at least 1 peripheral blood sample, a catheter tip culture positive for an identical microorganism, clinical signs of sepsis, and the absence of any other source of sepsis . Microorganisms isolated from catheter tip and blood cultures were considered to be identical when there was a match for bacterial species and antibiogram. In the presence of a positive catheter tip culture result, 1 positive blood culture result was considered to be sufficient for the diagnosis of coagulase-negative staphylococcal CR-BSI. No confirmation by DNA fingerprinting analysis was performed. Patients with a blood culture that yielded>1 type of microorganism were defined as having polymicrobial infection. Polymicrobial episodes in which one of the microorganisms was linked with a possible source other than the catheter were excluded.
Blood samples were routinely obtained for culture when the patient's temperature increased to>38.4°C or when bacteremia was suspected because of hemodynamic instability, chills, or new organ failure. They were processed using BacT/Alert (Organon Teknika). Central venous catheters were removed and tips were cultured when clinical infection occurs and when catheters have been in place for>5 days. Catheter tips were processed in accordance with the method of Maki et al. . Catheters were considered to be colonized when a semiquantitative culture revealed ⩾15 cfu/catheter segment.
The date on which the sample for the first positive blood culture result was obtained is considered the date of onset of CR-BSI. CR-BSI was considered to nosocomial when it was detected>48 h after hospital admission.
Therapy is considered to be appropriate if the regimen contains an antimicrobial agent with in vitro and clinical activity against the causative pathogen(s) and if it was initiated ⩽48 h after the onset of the CR-BSI. Susceptibility testing was performed in accordance with the latest guidelines recommended by the NCCLS at any time during the study period.
Main outcome measures. Clinical outcome evaluation is based on the rate of major organ derangements and in-hospital mortality. Major organ derangements considered include need for mechanical ventilation, need for renal replacement therapy, and need for vasopressors or inotropics. Attributable mortality is defined as the excess mortality caused by the CR-BSI and is calculated by subtracting the mortality rate of the control subjects from the mortality rate of the case patients with CR-BSI .
Economic outcome was evaluated on basis of duration of mechanical ventilation, length of ICU and hospital stays, and hospital costs. Length of hospitalization was calculated from the moment of ICU admission. Hospital costs were obtained from each patient's hospital invoice. On the basis of data from these invoices, we performed a breakdown of the total hospital cost into 3 major types of expenditures: per diem costs (also called “hotel costs” and including nursing costs), pharmacy costs, and medical expenses (i.e., laboratory expenses and the costs of imaging, consultations, and procedures). The monetary amounts from hospital invoices from years before the last year of the study period were actualized (on the basis of the Belgian health index) to 2002 costs.
Statistical analyses. Continuous variables are described as median values (interquartile range). The Mann-Whitney U test and χ2 test were used, as appropriate. Because all variables for economic evaluation were not normally distributed (P < .001, by Kolmogorov-Smirnov test), differences in economic indicators were calculated by subtracting the median value of the matched control subjects from that of the patients with CR-BSI. Survival curves were prepared by means of the Kaplan-Meier method. Univariate survival distributions were compared with the log-rank test. To assess the relationship between CR-BSI and in-hospital mortality, a Cox proportional-hazards regression analysis was performed on the matched cohort study population. This model was calculated starting from the time of onset of CR-BSI or from the corresponding time for control patients. The enter method was used, and hazard ratios with 95% CIs are reported. Covariates with a plausible relationship with mortality or a P value of <.1 in univariate analysis were included in the model. Relationships between total cost and CR-BSI were assessed using linear regression analysis. Covariates included in this model are length of hospitalization, age, APACHE II score, and surgical or medical admission diagnosis, as well as need for mechanical ventilation, renal replacement therapy, vasopressors, or inotropic treatments. Backward, stepwise selection was used, and covariates remained in this model if the P value was <.1. All tests were 2-tailed. Statistical significance is defined as P < .05. Power and sample-size calculations were performed using an online calculator (available at http://members.aol.com/johnp71/proppowr.html), which is part of the Interactive Statistical Pages Web site (http://www.StatPages.net). All other statistical analyses were performed using SPSS software, version 11.0 (SPSS).
During the study period, 36,836 patients were admitted to the ICU. CR-BSI was diagnosed in 192 ICU patients (5.2 cases per 1000 admissions, or ∼1 case per 1000 catheter-days). Six cases of CR-BSI were accompanied by a bloodstream infection originating from another identifiable source and were excluded. Thus, 186 patients with CR-BSI were eligible for investigation. Thirty-one episodes were polymicrobial; in 15 of these episodes, both types of microorganisms were cultured from the catheter. Table 1 summarizes findings regarding the involved microorganisms.
Clinical outcomes. Of 186 patients with CR-BSI, 10 could not be matched, because no control subjects were found who fulfilled the duration of ICU stay requirement. For 37 patients, only 1 suitable control subject was found. Therefore, the matching procedure resulted in a matched cohort study with 176 patients who had CR-BSI and 315 control subjects. Matching for year of admission (±2 years) was successful for 253 control patients (80.3%).
For 154 patients with CR-BSI (87.5%), the catheter was removed ⩽24 h after the onset of the infection, and 127 (77.0%) of 165 patients received appropriate antimicrobial therapy.
Table 2 summarizes the characteristics of patients with CR-BSI and of the matched control subjects. After the onset of CR-BSI, additional renal replacement therapy was needed, compared with the amount needed by control subjects. The in-hospital mortality rates for patients with CR-BSI and matched control subjects were 27.8% and 26.0%, respectively (P = .672). Thus, the attributable mortality rate was 1.8% (95% CI, -6.4% to 10.0%). Figure 1 illustrates survival curves for both groups. No differences in mortality between patients with CR-BSI and matched control subjects were observed after subgroup analyses for causative pathogen, monomicrobial or polymicrobial CR-BSI, timing of catheter removal, and appropriateness of antimicrobial therapy (table 3). Results of the multivariate survival analysis are in table 4.
Economic outcomes. Data concerning economic outcomes are summarized in table 2. Patients with CR-BSI underwent mechanical ventilation for a longer period and had longer durations of ICU stay and hospitalization. Total costs on hospital invoices, as well as the components related to the per diem cost, pharmacy expenditures, and medical expenditures, were also higher for patients with CR-BSI. Figure 2 reveals the differences in economic determinants. After adjustment for duration of hospitalization and clinical covariates, linear regression analysis identified CR-BSI as being independently associated with increased costs (table 5).
In this study CR-BSI did not contribute to mortality, but it did cause additional morbidity, as expressed in an increase in the number of ICU-days and ventilator-days, as well as a need for renal replacement therapy. This added morbidity resulted in an important excess of economic costs.
Although there exists a broad consensus that bloodstream infections in ICUs cause excess morbidity, data about mortality are less consistent, with attributable mortality rates ranging from 0% to>30% [6, 9, 10, 15, 17, 21-27]. These divergence between studies may have resulted from differences in the case mix, causative pathogens, source of infection, quality of care (e.g., the rate of appropriate therapy), and methodological issues, such as different matching procedures. In the present study, CR-BSI was associated with a nonsignificant attributable mortality rate of 1.8% (95% CI, -6.4% to 10.0%). This low figure cannot be solely attributed to the high proportion of CR-BSIs caused by coagulase-negative staphylococci (72 [40.9%] of 176 episodes were monomicrobial), because exclusion of these cases caused only a marginal increase in the attributable mortality rate to 4.3% (95% CI, -6.6% to 14.8%). In accordance with previous studies, a lack of attributable mortality was also observed in the subgroup of patients with CR-BSI due to gram-negative organisms [21-24, 26]. It has to be remarked that, in the subgroup of patients with catheter-related candidemia, the attributable mortality rate was 22.8%; however, this failed to reach statistical significance. Also, for CR-BSI due to S. aureus, another feared pathogen, the number of case patients was too small to allow useful conclusions (table 3).
Previous studies have also failed to demonstrate attributable mortality for ICU-acquired CR-BSI. In a study that compared 38 patients with 76 matched control subjects, Soufir et al.  found a mortality rate of 53% for ICU patients with CR-BSI, compared with 27% for matched control subjects (relative risk, 2.1; 95% CI, 1.2–3.7). After adjustment for prognostic covariates, however, the estimated risk of death was no longer significant (relative risk, 1.3; 95% CI, 0.7–2.5). Using a similar matched cohort design, Renaud et al.  reported a nonsignificant attributable mortality rate of 12% (95% CI, -15% to 38%). In this study, the mortality rates for patients with and for those without CR-BSI were 39% and 27%, respectively. Also, Rello et al.  could not demonstrate excess mortality, because the death rates for patients with CR-BSI and for matched control subjects were 22% and 35%, respectively. The results of these 3 studies might be influenced by under powering, particularly the latter 2, containing only 26 and 49 patients with CR-BSI, respectively. To estimate the required study power, we balanced between matched cohort studies of nosocomial bloodstream infection describing very high fatality rates [10, 16, 27, 30–33] and those reporting no significant excess mortality [6, 15-17, 21-24, 26, 28, 29]. For a 15% difference in mortality to be statistically significant (40% vs. 25%), a sample size of 125 patients with CR-BSI and 250 control subjects is needed (α = 0.05; power, 80%), a requirement that our study fulfills. However, to reach significance with an absolute difference in mortality of 5%, which can still be considered clinically relevant, the sample size is too small.
Prompt catheter removal probably contributed to the favorable clinical outcome, because this is associated with shorter duration of bloodstream infection, fewer recurrent infections, and improved outcome . In our study population, 87.5% of infected catheters were removed within 24 h, and 97.2% were removed within 72 h. Yet, the attributable mortality in patients without immediate catheter removal was not significant (table 3). It is probable that patients whose catheters were removed after 24 h represented a subgroup with less pronounced clinical signs of infection, in which case catheters were removed only after blood culture results had become positive. A more recent study demonstrated that, in hemodynamically stable patients, delayed catheter removal can be acceptable .
Together with accurate catheter management, the high rate of early initiated appropriate antimicrobial therapy might have contributed to the favorable outcomes observed [15, 35]. Subgroup analysis, however, could not confirm that CR-BSI without appropriate therapy resulted in a higher fatality rate (table 3). Again, it can be presumed that patients who received inappropriate therapy represented a group with a less severe clinical course, so that clinicians did not initiate empirical broad-spectrum antimicrobial therapy. In our ICU, empirical use of carbapenems and glycopeptides is restricted for patients with overt septic shock, in whom infection caused by antibiotic-resistant bacteria is likely to occur, as determined on the basis of colonization status or medical history . Insufficient study power in this particular subgroup can also explain the absence of differences in mortality.
In the present study, CR-BSI was associated with a 12-day increase in the total duration of hospitalization, with 8 of these days being in the ICU. This is particularly remarkable, because the median duration of ICU stay prior to CR-BSI was 6 days shorter than the duration of ICU stay for the control group (exposure time is shown in table 2). Patients with CR-BSI underwent mechanical ventilation for a more extended time. Increased time of mechanical ventilation translates into an important economical burden, because this represents an increased time of severe illness with a substantial amount of associated procedures. The cost attributable to CR-BSI was €13,585 per patient. Linear regression analysis revealed that CR-BSI was independently associated with higher costs. Renal replacement therapy was also associated with increased hospital costs, and receipt of this treatment occurred significantly more often after the onset of CR-BSI.
Matched cohort studies reporting on economic consequences of bloodstream infection in ICU patients are available, but such studies focusing specifically on CR-BSI are lacking. Pittet et al.  found that bloodstream infections contributed to an increase in the length of ICU stay of 8 days and an excess duration of hospitalization of 24 days (P < .1). The mean extra cost was $40,890 per survivor (P < .01). DiGiovine et al.  found excess lengths of ICU stay and hospitalization of 5 and 7 days, respectively (P < .01), for patients with primary bloodstream infection. The estimated attributable cost was $15,965 (P < .001). On the basis of an unmatched cohort analysis, Dimick et al.  reported a median excess cost of $56,167 and a 22-day increase in the duration of ICU stay for patients with CR-BSI. Their cohort, however, included only 17 patients with CR-BSI. Arnow et al.  found that CR-BSI was associated with an excess hospital cost of only $6000, but their study was confounded by case mix (ICU and non-ICU patients) and the inclusion of central venous catheters as well as peripheral catheters as sources of CR-BSI. Differences in national hospital billing systems hamper detailed comparisons between studies from diverse countries. However, in all studies, the excess costs are substantial and economically relevant. Moreover, the Belgian health care setting is relevant within a broader international context, because its health care-related costs and therapeutic/medical approach are in accordance with the averages for Europe .
Our study has limitations inherent to the single-center, matched-cohort design. The single-center design may act as a strength, however, because this has led to a homogenous standard of care applicable to both patients with CR-BSI and to control subjects, who were matched to the same period. Matched-cohort studies are prone to selection bias if many potential patients are excluded. We failed to match 10 patients with CR-BSI (5.4%) with control subjects. Two of these patients died (20%); thus, inclusion of these subjects into the cohort would not have altered the study findings.
Crucial in the interpretation of matched cohort studies is the reliability of the control group. Matching should be based on determinants of outcome [12, 13]. In the present study, matching was based on disease severity and diagnostic category, because these are the utmost important prognostic indicators for mortality and costs. Matching by disease severity was based on APACHE II scores at the time of ICU admission and not on the scores for the day prior to onset of CR-BSI. The latter method might be preferred on the basis of the assumption that patients who develop nosocomial infection may have a worse clinical course between ICU admission and the onset of infection. Yet, we have previously reported that disease severity remains stable between ICU admission and onset of nosocomial sepsis at our ICU . Moreover, because there was no difference in mortality rates in the present study, this limitation is of minor importance.
The reliability of the control group might be questioned, because the mortality rate among control subjects was slightly lower than would be predicted by the APACHE II score (table 2). Again, this poses no problem to the overall conclusions, because although a lower-than-expected mortality rate often leads to an overestimation of the attributable mortality rate, in fact, no difference in outcome was observed.
Some aspects limit the generalization of the study results. The favorable clinical outcome must be interpreted in the context of a high rate of early initiation of appropriate therapy and accurate catheter management. Also, the overall conclusion might not be valid for CR-BSI caused by Candida species and S. aureus, because there was a limited number of patients affected by these pathogens.
In conclusion, in the presence of prompt catheter removal and initiation of antimicrobial therapy, no significant attributable mortality could be documented in critically ill patients. However, increases in the durations of ICU and hospital stay contribute to an important economic burden. These significant increases in cost underscore the need to vigorous application of evidence-based, cost-effective preventive measures.
We thank Mr. J. De Schuijmer (Infection Control Team, Ghent University Hospital) for his continuous efforts in collecting and organizing data concerning nosocomial bloodstream infection. We also want to thank Ms. K. Kint (Department of Hospital Pharmacy, Ghent University Hospital), for providing data on antimicrobial therapy, and Mr. K. Duthoy (Department of Informatics, Ghent University Hospital), for providing data on hospital bills.
Potential conflicts of interest. All authors: no conflicts.
- critical illness
- hospital costs
- hospital mortality
- hospitals, university
- intensive care unit
- length of stay
- treatment outcome
- mechanical ventilation
- linear regression
- central venous catheter
- symptom onset
- severity of illness
- infection risk
- catheter related blood stream infection