Increased Resource Use Associated With Catheter-Related Bloodstream Infection in the Surgical Intensive Care Unit FREE

INFECTION OF intravascular catheters remains an important source of morbidity and mortality for critically ill patients.^1- 5 New technology for the diagnosis and prevention of catheter-related bloodstream infection (CRBSI) is rapidly evolving.⁶ Antimicrobial-coated catheters, ionic silver cuffs, antibiotic-impregnated hubs, and intraluminal antibiotic locks have all been shown^6- 10 to reduce the incidence of CRBSI in critically ill patients. However, these prevention strategies are expensive and might increase hospital costs. In the current health care economic climate, the clinical and economic benefit of this new technology must outweigh the increased costs. Population-specific attributable cost estimates for CRBSI will aid in the formal cost-effective analysis of new technology as it becomes available.

Previous studies have estimated the increased length of stay (LOS) and cost associated with CRBSI in a variety of settings. Most studies^3- 5 focus on nonsurgical patients, lack appropriate severity of disease adjustment, or include primary and secondary nosocomial bacteremias. Critically ill surgical patients with prolonged intensive care unit (ICU) stays are at high risk for infection. Therefore, they are the most likely to benefit from improved diagnosis and prevention.^11- 12 This study was conducted to estimate the increased resource use associated with CRBSI specifically for critically ill surgical patients after adjusting for severity of illness.

Patients were enrolled from January 7, 1998, to January 13, 1999, during a randomized clinical trial evaluating the efficacy of fluconazole for the prevention of fungal infections in critically ill surgical patients. Patients in the placebo (n = 130) and fluconazole (n = 130) arms are included in this analysis. The setting was a surgical ICU at a tertiary care medical center. Patients with an expected surgical ICU LOS greater than 3 days were included. Exclusion criteria included pregnancy, age younger than than 18 years, life expectancy less than 24 hours, and receipt of antifungal therapy within 7 days of ICU admission. Projected LOS was determined by a single ICU physician (P.A.L.) based on reason for admission, severity of illness, need for mechanical ventilation, comorbid disease, and hemodynamic instability. The protocol was approved by The Johns Hopkins Hospital Joint Committee on Clinical Investigation, Baltimore, Md. Signed informed consent was obtained for each patient or surrogate before enrollment. Once patients were enrolled, baseline data on relevant patient characteristics were recorded, including age, sex, race, ICU admission diagnosis, surgical procedure, medical comorbidities, and APACHE III (Apache Physiology and Chronic Health Evaluation III) scores. Clinical culture results were prospectively recorded and followed until definitive speciation and sensitivities were available or the specimen was deemed to have no growth.

Catheter colonization was defined as having greater than 15 colony-forming units of microorganisms on semiquantitative culture.¹³ The definition of CRBSI was the combination of catheter colonization and the same organism in a peripheral venous blood culture within 48 hours of the catheter culture. Diagnosis of CRBSI with coagulase-negative Staphylococcus required 2 positive blood cultures. Catheters were removed and cultured only for clinical suspicion of infection. The intradermal portion of the indwelling catheter was sent to the hospital microbiology laboratory for processing using the semiquantitative roll-plate technique.¹³

For purposes of billing, each patient hospital admission was assigned a unique identification number. Using this identifier, The Johns Hopkins Hospital discharge database (based on the uniform discharge data set) and the billing database were queried to determine total hospital charges and ICU-specific charges. Total hospital and ICU charges were converted to costs using the hospital-specific ratio of cost to charges during the study. Costs are reported in 1998 dollars. The hospital discharge and billing databases divide charges into several different categories, including total, routine (room and board), operating room, pharmacy, radiology, laboratory, supplies, therapy (physical, occupational, and speech), and other costs.

The Shapiro-Wilks test was applied to test the data for normality. Cost and LOS were not normally distributed and were log transformed for regression analysis. Univariate comparisons were performed with the Fisher exact test, χ² test, Wilcoxon rank sum test, simple linear regression, or simple logistic regression as appropriate. Severity of illness (APACHE III scores) and demographic data were included in the multivariate analysis if P<.05 in the univariate analysis (Table 1). Multiple logistic regression was used for in-hospital and ICU mortality. Multiple linear regression of log-transformed cost and LOS was performed and the regression coefficient for CRBSI was exponentiated to determine the associated percentage increase in the given variable. In all analyses, P = .05 was considered statistically significant. Analyses were performed using statistical software (Stata 6.0; Stata Corp, College Station, Tex).

During the yearlong study, 1228 patients were admitted to the surgical ICU. Of these, 461 patients were eligible for the study and 260 (56%) were enrolled. The remaining patients were not enrolled because of lack of consent and LOS of 3 days (low risk); misidentification (projected LOS ≤3 days but actual LOS >3 days); or LOS greater than 3 days and lack of consent, age younger than 18 years, or entry into an alternate study. No patient who was projected to stay longer than 3 days died before the third day. Overall, 49% of patients were men, median patient age was 65 years (interquartile range [IQR], 51-74 years), and 73% were white (Table 1). Patients had a high level of critical illness, with a median APACHE III score of 64 (IQR, 51-79). Comorbid disease was also prevalent (Table 1).

The 260 patients in the study each had an average exposure of 18 catheter-days, for a total of 4712 catheter-days. Of enrolled patients, 136 (52%) had at least 1 catheter cultured for suspected infection. The median number of catheter cultures obtained was 2 per patient (IQR, 1-4; range, 1-26). A total of 89 colonized catheters in 54 patients (21%) yielded 120 microbiologic isolates. Colonization was monomicrobial in 73% of catheters and polymicrobial in 27% of catheters. The isolates were Gram-positive bacteria in 80%, Gram-negative bacteria in 15%, and yeast in 5% (Table 2). Eighty-six patients (33%) had at least a single positive blood culture during the study period. There were 17 episodes of CRBSI in 9 patients. The incidence of CRBSI was 3.6 episodes per 1000 catheter-days (95% confidence interval [CI], 2.1-5.8 episodes per 1000 catheter-days). In patients with CRBSI, there were 15 monomicrobial infections and 2 polymicrobial infections, yielding a total of 20 microbiologic isolates. Isolates were Gram-positive bacteria in 75%, Gram-negative bacteria in 20%, and yeast in 5% (Table 2).

Thirty-one patients (12%) died before ICU discharge and 58 (22%) died before hospital discharge. The in-hospital mortality rate for patients with CRBSI was 56% (5/9) compared with 21% (53/251) for those without CRBSI (P = .02) (Table 3). In this population, the difference between crude in-hospital mortality rates (attributable mortality) was 35%. Other univariate predictors of in-hospital mortality were APACHE III score (P<.001), age (P = .04), and immunosuppressive drug use on admission to the ICU (P = .03). In a multivariate analysis, only the APACHE III score remained a significant predictor of mortality (odds ratio [OR], 1.02; 95% CI, 1.04-1.06 per point increase in APACHE III score) (P<.001). However, there was a trend toward increased in-hospital mortality for patients with CRBSI (OR, 4.3; 95% CI, 0.9-19.9; P = .07).

Catheter-related bloodstream infection was associated with an increased risk of ICU death, with 44% mortality in patients with CRBSI vs 11% in patients without CRBSI (P = .01) (Table 3). Other univariate predictors of ICU death were APACHE III score (P<.001), age (P = .04), unscheduled admission (P = .007), gastrointestinal surgery (P = .003), and history of alcohol abuse (P = .04). In the multivariate analysis, CRBSI was associated with a 6-fold increased risk of ICU death (OR, 6.6; 95% CI, 1.8-24.4; P = .01). APACHE III score also remained a significant predictor of mortality in the multivariate analysis (OR, 1.04; 95% CI, 1.02-1.05 per point increase in APACHE III score) (P = .001).

The median total direct health care cost (excluding professional fees) was $40 722 (IQR, $22 755-$68 587) for patients enrolled in the study. For patients without CRBSI, the median total cost was $40 313 (IQR, $25 327-$67 899) vs $102 965 (IQR, $42 766-$238 698) for patients with CRBSI (P = .02) (Table 3). Comorbid diseases associated with increased cost include cirrhosis (P<.001) and dialysis-dependent renal disease (P = .03). Patients who underwent gastrointestinal surgery also had increased cost (P = .006). In a multivariate analysis, adjusting for severity of illness (APACHE III score) and demographic factors (age, sex, and univariate factors of signifance), CRBSI was associated with a 120% increase in total hospital costs (median, $56 167; 95% CI, $11 523-$165 735; P = .001). In this analysis, APACHE III score and increasing age were also associated with increased total hospital cost (P<.001 for both) (Table 4). Catheter-related bloodstream infection was associated with increased cost in all categories except for operating room cost and "other" cost (Figure 1). Routine cost (room and board) shows the largest increase between groups ($36 980 with CRBSI vs $13 836 without CRBSI; P<.001).

For all study patients, median ICU cost was $14 859 (IQR, $7416-$35 003). Patients without CRBSI had a median ICU cost of $14 240 (IQR, $7273-$32 734) vs $91 812 (IQR, $21 616-$192 756) for patients with CRBSI (P = .002) (Table 3). Other univariate predictors of increased ICU cost include gastrointestinal surgery (P = .003) and a history of cirrhosis (P = .05). In the multivariate analysis, adjusting for severity of illness and demographic factors, CRBSI was associated with a 300% increase in ICU cost (median, $71 443; 95% CI, $11 960-$195 628; P<.001). In the adjusted analysis, APACHE III score was also associated with increased ICU cost (P<.001) (Table 4). The difference between ICU cost for patients with and without CRBSI divided into cost categories is similar to the findings in total hospital cost. Each cost category demonstrates a statistically significant difference in cost except for other cost, which is a small contributor to overall total ICU cost. Once again, routine cost (room and board) demonstrates the largest difference between the groups ($28 567 with CRBSI vs $5106 without CRBSI; P<.001).

For all patients enrolled, the median hospital LOS was 18 days (IQR, 11-29 days). Patients without CRBSI had a median hospital LOS of 18 days (IQR, 16-20 days) vs 40 days (IQR, 15-165 days) in patients with CRBSI (P<.001) (Table 3). Other univariate predictors included abdominal surgery within the last month (P = .01) and APACHE III score (P = .005). In a multivariate analysis correcting for severity of disease and demographics, CRBSI was associated with a 124% increase in LOS (22 days; CI, 7-70 days; P = .002). In this analysis, increasing APACHE III score was also associated with increased hospital LOS (P<.001), and there was a trend toward an association of increasing age and hospital LOS (P = .08).

The median length of ICU stay was 7 days (IQR, 3-14 days) for all patients enrolled. For patients without CRBSI, median ICU stay was 7 days (IQR, 6-7 days) compared with 22 days (IQR, 4-111 days) for patients with CRBSI (P<.001) (Table 3). The only other variable with a univariate association was APACHE III score (P = .004). An adjusted multivariate analysis showed a 200% increase in ICU days (20 days; 95% CI, 0.1-58 days; P = .002) associated with CRBSI. Higher APACHE III scores were also associated with increased ICU days in this analysis (P = .02).

Catheter-related bloodstream infection remains an important cause of morbidity and mortality in critically ill surgical patients. This prospective cohort study estimates the associated increase in cost and LOS associated with CRBSI in a well-defined critically ill population of surgical patients. Patients with CRBSI have increases in total hospital cost of $56 167 and increases in ICU cost of $71 443. Hospital LOS was increased by 22 days and ICU LOS was increased by 20 days. The increase in resource use occurs predominantly in the ICU. Because of the magnitude of resources consumed in this population, efforts at prevention, early diagnosis, and improved therapy are extremely important. Further research concerning the cost-effectiveness of new technology for diagnosis and prevention of central line infections in surgical patients should use this population-specific attributable cost estimate.

The cost and LOS associated with CRBSI in our study are similar to those from previous studies of ICU patients. Pittet and colleagues⁵ conducted a matched case-control study in a surgical ICU at a large tertiary care center. They reported excess cost of $40 000 (in 1994 dollars) and increased hospital and ICU LOSs of 24 and 8 days, respectively. These values are similar to ours, especially taking into account inflation of medical care costs from 1994 to 1998. However, patients in their study had primary and secondary nosocomial bloodstream infections. Only 20% of patients had bloodstream infection related to an indwelling intravascular catheter. In contrast, the estimate in our study is specific for primary bloodstream infections (CRBSIs) in critically ill surgical patients.

The large attributable cost in our patients is significant across all cost categories except operating room cost. This scenario is plausible because most operating room costs are acquired before ICU stay and the onset of CRBSI. Cost of room and board (routine cost) is the single largest contributor to increased cost in these patients. This finding is consistent with other studies^{3- 5,14- 15} that implicate increased LOS as the dominant contributor to increased cost. In the present study, however, laboratory, supply, and pharmacy costs are also important contributors to overall increased cost. The increase in room cost makes up only approximately half of the estimated attributable cost. Furthermore, we estimated a 22-day increased hospital LOS and a 20-day increased ICU LOS; therefore, most extra days in the hospital were in the ICU setting. This suggests that in a critically ill population with prolonged ICU stay, CRBSI further complicates their underlying disease condition and prevents progression and recovery. All of these consequences will result in prolonged ICU stay and increased cost. The adverse effects of CRBSI will be more pronounced in ICU patients compared with non-ICU patients.

Infection of indwelling vascular catheters occurs from 2 routes. First, endogenous skin flora at the insertion site migrates along the external surface of the catheter and colonizes the intravascular tip. Second, pathogens from hub contamination move along the internal surface of the catheter and colonize the lumen.⁶ New technology aimed at preventing CRBSI is directed at both of these routes. Appropriate topical antiseptic use coupled with maximal barrier precautions at the time of catheter insertion decreases the rate of catheter infection.^16- 17 Antimicrobial-impregnated catheters coated on the inside and outside are beneficial at reducing catheter colonization and CRBSI.^7- 10 A recent cost-effective analysis concluded that antiseptic-impregnated catheters were efficacious and less expensive compared with standard catheters. The attributable cost estimate used in this cost-effectiveness analysis was determined by multiplying the excess LOS by the per diem costs of hospitalization, yielding a value of approximately $10 000.¹⁵ Because in our study population the estimate of increased cost is much higher, these catheters would be cost-effective.

Current diagnostic modalities for catheter-related infection require removal of the catheter, subjecting patients to the risks associated with reinsertion. New methods for diagnosis of catheter-related infections that obviate the need for catheter removal are being developed. For example, the differential time to positivity of paired quantitative blood cultures drawn simultaneously from the catheter hub and peripheral sites has been shown to have 91% specificity and 94% sensitivity.¹⁸ In addition, the Gram stain and acridine-orange leukocyte cytopsin test, another rapid-diagnostic technique that does not require catheter removal, has sensitivity of 96% and specificity of 92%.¹⁹ However, the cost-effectiveness of these new diagnostic strategies has not been studied.

Estimating the attributable cost of an infectious complication has several limitations. There are 2 study designs commonly used to estimate the cost associated with an infectious complication.²⁰ The first method, used in our study, is a multivariate analysis of a population of patients adjusting for potential confounding variables. The second method is to match patients with the study condition to those without the condition based on age, sex, race, severity of disease, and other variables associated with outcomes. Both of these methods allow for estimation of the increased cost or LOS associated with a certain disease condition. However, the precise contribution to the outcome of interest (cost, LOS, or mortality) cannot directly be determined. Hence, an associative (not causative) relation can be postulated. In a large observational cohort study, Asensio and Torres²¹ used both methods (matched pairs and multivariate analysis) to determine the excess cost and LOS associated with deep surgical site infections after open heart surgery. They concluded that the multivariate method yields a more accurate estimate of excess resource use attributable to nosocomial infections.²¹

Another limitation of this study is the small number of infections. There were only 17 episodes of CRBSI in this population during the study. Because of the low number, we have limited power to detect and adjust for all patient characteristics associated with outcomes. However, we were able to adjust for demographics and severity of illness (APACHE III score) in our analysis, which both had a strong association with outcomes. The other univariate predictors had much weaker associations with the outcome variables. Therefore, because we accounted for the most important variables, the additional contribution of other confounders is not likely to contribute much validity to our model.

In conclusion, this study demonstrates a profound increase in health care cost and LOS associated with CRBSI in critically ill surgical patients. The increased cost was incurred in the ICU setting and was mostly due to increased LOS. New technology for prevention, diagnosis, and treatment of catheter-related infections is under continual development and evaluation. Formal cost-effectiveness analyses of strategies to diagnose and prevent CRBSI for surgical patients should use population-specific attributable cost estimates.

Presented at the 20th Annual Meeting of the Surgical Infection Society, Providence, RI, April 27, 2000.

Corresponding author and reprints: Pamela A. Lipsett, MD, Department of Surgery, The Johns Hopkins Hospital, 600 N Wolfe St, Blalock 685/683, Baltimore, MD 21287-4685 (e-mail: plipsett@jhmi.edu).