Prospective Evaluation of Risk Factors for Bloodstream Infection in Patients Receiving Home Infusion Therapy

Methods

Patients

Patients were enrolled at two study sites, one in Toronto, Ontario, Canada, and one in Cleveland, Ohio. The study protocol was approved by the institutional review boards at the Centers for Disease Control and Prevention and at all institutions involved at the two study sites. At the Toronto site, eligible patients included all patients referred to one of five home infusion agencies (which provide approximately 90% of infusion therapy in this region) from any Toronto-area hospital, patients in the Toronto Hospital total parenteral nutrition program, and patients in the Princess Margaret Hospital oncology program. At the Cleveland site, eligible patients included all patients at four hospitals, including The Cleveland Clinic Foundation Hospital, who received infusion therapy from the home infusion agency affiliated with The Cleveland Clinic Foundation. At each site, patients were first asked for verbal consent to participate. Patients who granted this consent were interviewed by a study nurse and were asked to provide written informed consent (in Toronto) or were mailed forms for written consent (in Cleveland).

Design

This prospective cohort study was performed from 1 April 1996 to 30 April 1997 in Toronto and from 1 May 1996 to 30 May 1997 in Cleveland. For all eligible patients, an anonymous pre-enrollment form without personal identifiers was completed; for patients who consented and were enrolled, additional forms were completed (Table 1). Data for completion of all forms were obtained from the hospital records and infusion agency records and by interviewing patients. Multiple sources were often used, especially to check information provided by patients. On-site study nurses completed all forms and entered information into computer databases. Patients were seen at frequencies ranging from two to three times per day to once per month, depending on diagnosis and indications for therapy.

An infection form was completed for episodes of suspected infection [Table 1]. Bloodstream infection was diagnosed if all of the following features were present: 1) one or more positive blood cultures, 2) antimicrobial therapy or catheter removal, and 3) no infection at another site that could have caused the bacteremia. For low-virulence organisms, such as coagulase-negative staphylococci, we required the additional criteria of one or more clinical features (fever, chills, or purulent exudate at the catheter insertion site) and at least two positive blood cultures. Episodes that occurred within 2 days of enrollment or hospital discharge were excluded because they would not represent home health care-related infections.

Statistical Analysis

Data from the catheter, infusion therapy, and infection-control forms were collected and analyzed as time-dependent variables. For example, in a single patient, catheter type A might have been used for days 1 to 14; that catheter might then have been removed and catheter type B inserted and used for days 15 to 45. For such variables, the unit of analysis is the patient, but the values of the variables are allowed to change at different times of follow-up.

Patients were included in the study only during periods in which they had a central or midline catheter at home; therefore, the number of days that a patient was followed in the study equals the number of catheter-days. Patients whose catheters had been placed before the study started were included, but bloodstream infections in these patients were included only if they occurred 30 or more days after enrollment, were associated with a different catheter, or were caused by an organism different from that which had caused bloodstream infection before enrollment. For patients who had a bloodstream infection during the study, data on patient-days and infections were excluded for 1 month after the infection and were included again thereafter; subsequent bloodstream infections were counted if the patient had a different catheter or if a different organism was isolated. Therefore, some patients had more than one bloodstream infection during the study.

By using SAS for Personal Computers (SAS Institute, Inc., Cary, North Carolina), the separate databases were merged into a common database for analysis. Some of the time-dependent risk factors may have changed between the day of microbial invasion and the day on which positive blood cultures were obtained; to account for this effect, bloodstream infections were included in the common database as of 2 days before the date on which they were reported. Incidence density rates (bloodstream infection rates per 1000 catheter-days) were calculated by dividing the number of bloodstream infections by the number of catheter-days and multiplying the result by 1000.

Univariable and multivariable analyses were performed by using Cox regression models for time-dependent covariates (8). Factors that were found to be statistically significant (P < 0.05) in univariable analysis were considered for inclusion in the multivariable model by using a forward stepwise algorithm. Independently statistically significant variables (P < 0.05) were included in the final multivariable model. Selected variables that were not independently significant were then individually inserted into the final regression model to determine hazard ratios and CIs after adjustment for the factors in the final model. We calculated 95% CIs for hazard ratios by subtracting from or adding to the variable estimate 1.96 times its SE and exponentiating the result. All reported P values are two-sided.

Role of the Funding Source

This study was funded by the U.S. federal government. No proprietary interest had a role in the collection, analysis, or interpretation of the data or in the decision to submit the paper for publication.

Results

Enrollment

Of 1528 eligible patients, 827 (54%) were enrolled: 62% (354 of 569) of eligible patients at the Toronto site and 49% (473 of 959) of those at the Cleveland site. At the Toronto site, the primary reason given for nonenrollment was that patients were overwhelmed by their illness and were not able to cope with anything else. At the Cleveland site, the lower enrollment rate was attributed to the fact that patients were asked to provide informed consent by mail rather than at a face-to-face meeting. All diagnoses included on the pre-enrollment form were listed more often for enrolled than nonenrolled patients; the most common diagnoses of enrolled patients were infections other than HIV (67%) and cancer (24%) (Table 2). Compared with nonenrolled patients, enrolled patients were more likely to receive total parenteral nutrition and less likely to receive chemotherapy.

The 827 enrolled patients had 988 catheters. Of these, 433 (44%) were centrally inserted venous catheters, including 215 Hickman catheters and 125 Cook catheters; 324 (33%) were peripherally inserted central catheters, including 218 Per-Q-Cath catheters (Bard Access Systems, Murray Hill, New York); 155 (16%) were midline catheters, including 96 Landmark catheters (Menlo Care, Inc., Menlo Park, California); and 76 (8%) were implanted ports, including 70 Port-A-Cath catheters (SIMS Deltec, Inc., St. Paul, Minnesota). The median follow-up time was 67 days for patients with centrally inserted venous catheters, 27 days for those with peripherally inserted central catheters, 17 days for those with midline catheters, and 155 for those with implanted ports. Among the centrally inserted venous catheters, all were tunneled and cuffed, and none were impregnated with antiseptic agents. We recorded dates of insertion and removal of catheters but not whether catheters were removed because of occlusion. For the 988 catheters, needle access was used in 108 (11%), needleless access was used in 877 (89%), and protected needle access was used in 3 (<1%). Of the needleless accesses, 792 (89%) were Interlink systems (Baxter International, Deerfield, Illinois).

Bloodstream Infections

A total of 69 bloodstream infections occurred among 59 patients; 50 patients had 1 bloodstream infection, 8 had 2 bloodstream infections, and 1 had 3 bloodstream infections. Among patients with 2 or more bloodstream infections, the median duration between infections was 119 days (range, 40 to 187 days). The blood specimen showing bacteremia was taken from a peripheral vein in 34 infections (49%), the catheter in 21 infections (30%), and an unknown site in 14 infections (20%). Symptoms included fever in 58 infections (84%), chills in 43 infections (62%), redness at the catheter insertion site in 10 infections (14%), and purulence at the catheter insertion site in 5 infections (7%). At the time of infection, the leukocyte count was 100 to 1000 cells × 10⁹/L in 10.1% of patients and more than 10 000 cells × 10⁹/L in 23.2% (median leukocyte count, 5800 cells × 10⁹/L [range, 100 to 24 700 cells × 10⁹/L]). Ten episodes of concomitant infection (7 urine, 2 lung, and 1 other) occurred. The catheter was removed in 36 (52%) episodes of bloodstream infection, and intravenous antimicrobial therapy was begun in 68 (99%) episodes.

The 69 bloodstream infections included 14 (20%) that were polymicrobial (2 microorganisms were isolated for all episodes). The 83 isolates consisted of 48 (58%) gram-positive cocci, including 23 coagulase-negative staphylococci, 8 Staphylococcus aureus, 6 enterococci, 5 corynebacteria, and 6 others; 27 (32%) gram-negative rods, including 7 Klebsiella species, 5 Escherichia coli, 4 Acinetobacter species, 3 Enterobacter species, and 8 others; and 8 (10%) fungi, including 7 Candida species.

Risk Factors for Bloodstream Infections

The 827 patients were followed for a total of 69 532 catheter-days (median, 44 catheter-days per patient [range, 1 to 395 catheter-days per patient]), during which 69 bloodstream infections occurred (0.99 infections per 1000 catheter-days).

Eighteen variables were found to be statistically significant in univariable analyses (Table 1) and were considered for inclusion in the multivariate model. Of these, 5 were found to be independently significant and were retained in the final model (Table 3). When all 69 bloodstream infections were considered, bone marrow transplantation was associated with a significant increase in infection events (hazard ratio, 5.8 [95% CI, 3.0 to 11.3]); this estimate was slightly higher when only first infections were included (hazard ratio, 6.6 [CI, 3.3 to 13.3]). The risk associated with receipt of total parenteral nutrition (hazard ratio, 4.1 [CI, 2.3 to 7.2]) was similar regardless of whether lipids were included. Other independent risk factors were receipt of infusion therapy outside the home (hazard ratio, 3.6 [CI, 2.2 to 5.9]), such as in an outpatient clinic or physician's office; therapy through a multilumen (≥ 2 lumens) catheter (hazard ratio, 2.8 [CI, 1.7 to 4.7]); and previous bloodstream infection, either before or during the study (hazard ratio, 2.5 [CI, 1.5 to 4.2]). Interaction terms between time and each of these variables were nonsignificant, indicating that the proportional hazards assumption was appropriate. These five variables also remained statistically significant in models that excluded bloodstream infection caused by low-virulence organisms (analysis based on 48 events) and in models that included only bloodstream infection diagnosed from a peripheral blood culture (analysis based on 34 events) (data not shown).

Bloodstream infection rates per 1000 catheter-days were 0.16 (4 infections during 25 019 catheter-days) for patients with none of these five risk factors, 0.46 (13 infections during 28 039 catheter-days) for those with one risk factor, 2.22 (29 infections during 13 075 catheter-days) for those with two risk factors, and 6.77 (23 infections during 3399 catheter-days) for patients with three or more risk factors.

Because receipt of infusion therapy outside the home was a strong risk factor for bloodstream infection, we explored diagnoses in 264 patients receiving such therapy. Of these patients, 132 (50%) had infections other than HIV, 117 (44%) had cancer, 19 (7%) had HIV infection, and 14 (5%) had undergone bone marrow transplantation. The total is more than 100% because some patients had more than one diagnosis.

Bloodstream infection rates and adjusted hazard ratios for selected factors that were not independently significant are shown in Table 4. The rate of bloodstream infection was similar at the Cleveland and Toronto sites. Compared with implanted ports, centrally inserted venous catheters carried a higher but nonsignificant risk (hazard ratio, 2.4 [CI, 0.8 to 7.4]). Time since catheter insertion did not influence the risk for bloodstream infection; compared with the reference category (hazard ratio, 1.0) of 1 to 33 days, catheters in place for 244 days or longer carried a hazard ratio of 0.6 [CI, 0.3 to 1.5]. The hazard ratio was lower during periods of antibacterial treatment (hazard ratio, 0.4 [CI, 0.1 to 1.1]) and when chlorhexidine was used at dressing changes (hazard ratio, 0.6 [CI, 0.2 to 1.3]) and was higher when transparent dressings were used (hazard ratio, 1.3 [0.8 to 2.1]), but these effects were not statistically significant.

Needleless devices were used during 76% of total catheter-days; however, these devices did not increase risk (hazard ratio, 0.9 [CI, 0.5 to 1.6]) (Table 4). The most common interval for changing of needleless devices was four times per month; risk was elevated when the devices were changed less frequently (hazard ratio, 2.2 [CI, 0.9 to 5.7]), but this was not statistically significant.

Discussion

In a prospective study of 827 patients receiving home infusion therapy through central or midline catheters, we found five probable risk factors for bloodstream infection. Patient-associated risk factors were recent bone marrow transplantation and previous bloodstream infection. Patient care-related features were receipt of total parenteral nutrition, administration of intravenous therapy in an outpatient setting other than the home (such as a hospital clinic or physician's office), and use of a multilumen catheter. The rate of bloodstream infection increased as the number of risk factors present increased, ranging from 0.16 infections per 1000 catheter-days in patients with no risk factors to 6.77 infections per 1000 catheter-days in patients with three or more risk factors.

Bone marrow transplantation increases susceptibility to bacterial infections, primarily during the neutropenic period in the first 1 to 2 months after transplantation (9, 10). Our data do not explain how bone marrow transplantation increases risk, but a previous study also identified neutropenia as a risk factor for bloodstream infection (11). We recorded leukocyte counts of patients at the time of bloodstream infection but not in patients without bloodstream infection; thus, we could not evaluate leukopenia or neutropenia as a risk factor for bloodstream infection. Neutrophil counts should be recorded in future studies of this topic.

Infusion therapy outside the home, which was received by patients with diverse diagnoses, has not been reported previously as a risk factor for bloodstream infection. It may be a marker for increased catheter manipulation and line entries, or it may indicate poor infection-control technique in outpatient clinics or physician's offices. Receipt of total parenteral nutrition (3, 12), receipt of therapy through a multilumen catheter (13), and previous bloodstream infection (4) have been associated previously with bloodstream infection. In our study, HIV infection may have provided an apparently protective but nonsignificant effect; however, patients with HIV may have received oral antibiotic prophylaxis against opportunistic infections, may have been better educated, and may have had greater experience with infusion techniques.

Our findings may differ from those of previous studies because of variation in study samples, definitions of bloodstream infection and other study methods, infection-control practices, and types of catheters and other devices used. In contrast to some recent investigations (3-7), we found no association between needleless devices and bloodstream infection. Although the number of patients in our study in whom needle access devices were used was small, the hazard ratio CI of 0.5 to 1.6 for this variable is consistent with at most a modest risk.

We found no significant difference in risk among types of catheter used, but centrally inserted venous catheters had the highest hazard ratio (2.4 [CI, 0.8 to 7.4]) and implanted ports had the lowest hazard ratio (1.0 [reference category]). The hazard ratio for peripherally inserted central catheters was intermediate (1.5 [CI, 0.3 to 6.8]). Peripherally inserted central catheters are a sensible option for ambulatory delivery of total parenteral nutrition, antibiotics, and chemotherapy because of their relative ease of insertion and removal, as well as their relatively low rates of associated infection (12).

Our findings generally agree with a recent guideline for preventing catheter-related bloodstream infections (14). The guideline recommends using single-lumen catheters if possible (we found a higher risk associated with multilumen catheters), does not recommend antibacterial ointment under dressings (we found no significant effect for ointment), does not specify a frequency of dressing changes (we found no significant effect for this factor), and has no recommendation for or against the use of transparent dressings (we found no significant risk associated with this factor, in contrast to some studies suggesting that transparent dressings increase risk [15]).

We found that days since catheter insertion did not influence risk for bloodstream infection; “old” catheters have a risk per day similar to that of “young” catheters. This is in agreement with results of another recent study (16) and supports the recommendation against routine catheter changes (14).

Previous studies have reported a range of bloodstream infection rates. Among patients with centrally inserted venous catheters, rates were 4.5 to 14.6 infections per 1000 catheter-days among hospital inpatients (17) and 0.49 to 3.9 infections per 1000 catheter-days among outpatients (11, 18, 19). Among patients with peripherally inserted central catheters, rates were 0 to 1.6 infections per 1000 days (12, 20-22).

Compared with data from hospital studies (23), we report a generally similar distribution of species causing bloodstream infection but found a higher proportion of corynebacteria (6% compared with <1%). This finding may indicate increasing importance of these low-virulence organisms or that some of the episodes that we classified as bloodstream infections represented catheter colonization or contamination of the blood culture specimen. We reported a higher proportion of polymicrobial bloodstream infections (20%) than that in recent studies in hospital (14%) (24) and outpatient (11%) (5) settings.

The external validity of our study—the ability to generalize results to other populations—may be limited because enrolled patients were more likely than nonenrolled patients to have received transplants or total parenteral nutrition and therefore may have been at higher risk for bloodstream infection. The internal validity was limited by the fact that this was an observational study without randomization; thus, potential confounding factors may not have been balanced among various comparison groups. In addition, the sample size limited the power to detect small effects. Some bacteremias that were diagnosed from blood cultures drawn through the catheter may have been due to catheter colonization rather than to bloodstream infection, but 5 major risk factors remained significant when these events were excluded. Although many potential risk factors were examined, it is unlikely that the reported associations were spurious, as evidenced by the high hazard ratios and the highly significant P values. Some bloodstream infections may have been missed if catheters were removed, possibly because of occlusion, and blood cultures were not performed. Finally, our study contained a small number of infections relative to the number of candidate risk factors. Although our final model included only 5 risk factors, we started with 18 candidate variables. Future studies with more data will need to replicate our finding that these 5 factors are significant predictors in multivariable analyses.

Strengths of our study are its large sample size and prospective cohort design. In addition, data on several potential risk factors were collected in a time-dependent manner, allowing a detailed multivariable analysis that adjusted for potential confounding variables and accounted for time at risk.

These data help to define the frequency of and risk factors for bloodstream infection during home infusion therapy. Of the five independently significant risk factors identified, two—use of multilumen catheters and receipt of infusion therapy outside the home—were associated with most bloodstream infections and may be modifiable. Interventions aimed at these two factors may have the greatest impact in reducing bloodstream infections. Multilumen catheters (associated with 42 of 69 bloodstream infections) are required to administer some infusion regimens, but the risk associated with their use should be carefully weighed so that they are not used unnecessarily. Receipt of infusion therapy outside the home (associated with 36 bloodstream infections) is a newly reported risk factor; further studies are needed to confirm this association and to determine the specific mechanism by which it may increase bloodstream infections. In the interim, personnel in outpatient clinics and physician offices should review infection-control procedures and ensure that proper technique is used when catheters are accessed.