Accuracy of bacterial DNA testing for central venous catheter-associated bloodstream infection in children with cancer

Article history

The research reported in this issue of the journal was commissioned by the HTA programme as project number 03/39/13. The contractual start date was in June 2005. The draft report began editorial review in January 2010 and was accepted for publication in September 2010. As the funder, by devising a commissioning brief, the HTA programme specified the research question and study design. The authors have been wholly responsible for all data collection, analysis and interpretation, and for writing up their work. The HTA editors and publisher have tried to ensure the accuracy of the authors’ report and would like to thank the referees for their constructive comments on the draft document. However, they do not accept liability for damages or losses arising from material published in this report.

Declared competing interests of authors

none

Copyright statement

© 2011 Queen’s Printer and Controller of HMSO. This journal is a member of and subscribes to the principles of the Committee on Publication Ethics (COPE) (http://www.publicationethics.org/). This journal may be freely reproduced for the purposes of private research and study and may be included in professional journals provided that suitable acknowledgement is made and the reproduction is not associated with any form of advertising. Applications for commercial reproduction should be addressed to: NETSCC, Health Technology Assessment, Alpha House, University of Southampton Science Park, Southampton SO16 7NS, UK.

2011 Queen’s Printer and Controller of HMSO

Children with cancer

The study took place under the auspices of the UK Children’s Cancer Study Group [UKCCSG, now the Children’s Cancer and Leukaemia Group (CCLG)]. Approximately 1500 children (up to the age of 15 years) are diagnosed with cancer in the UK every year, and leukaemia accounts for around 30% of these diagnoses. Approximately 90% of children with a cancer diagnosis in the UK are treated in a CCLG centre (www.cclg.org.uk).

The duration of treatment for cancer varies but is usually < 2 years. The majority of children are able to spend a large proportion of this time outside hospital in the community. Most children have a central venous catheter (CVC) inserted into a large vein, which remains in place for many months. This allows treatment to be given at home, or in hospital for more intensive treatment, while minimising interference with daily life. These devices are usually either tunnelled catheters (e.g. the Hickman catheter) or subcutaneous ports. After treatment, > 70% will eventually be cured of cancer (www.cclg.org.uk). However, infection is a major hazard for children undergoing treatment for cancer. Most will be admitted to hospital at least once for infection during their treatment for cancer. The dilemma facing clinicians is to distinguish between infections due to the CVC and other sources.

Diagnosis of central venous catheter-associated infection

Widespread use of CVCs has led to these devices becoming recognised as a major risk factor for hospital-acquired bloodstream infection in adults and children. 1–4 The rate of infection associated with CVCs varies from < 1 to 15 episodes per 1000 days of central line use, depending upon the patient population and a range of other factors. 5 The rate of CVC-associated infection in children undergoing treatment for cancer varies from 1.7 to > 5 per 1000 CVC days. 3,6,7 Complications include septic thrombophlebitis, endocarditis, septic shock and the dissemination of septic emboli. Studies in adults have reported an attributable mortality for CVC-associated infection of up to 25%, but rates for children have not been reported. 8 The cost of CVC-associated infection can be many thousands of pounds per episode, depending on the virulence of the infecting agent. 9

The CVC has been considered the source of nearly half of the episodes of bloodstream infection in some studies involving immunocompromised patients. 10,11 Discrimination between the CVC and other sources of bloodstream infection is important because treatment strategies differ. In addition to systemic antibiotics, CVC-associated infection requires either antibiotic treatment that is targeted at microbial colonisation of the CVC lumen by being left in the CVC lumen, or instilled slowly, or removal of the CVC. In children with cancer who have long-term surgically implanted CVCs, removal and reinsertion of a CVC carries operative and anaesthetic risks as well as costs, and risks using up venous access sites. It is this group of patients that particularly needs improved diagnostic methods. There is a variety of clinical and microbiological techniques for diagnosing CVC-associated infection.

CVC-associated infection is most apparent clinically when a patient with few other risk factors for infection develops signs and symptoms of infection associated with inflammation at the site of the device, or has fever, rigors, chills and/or hypotension associated with CVC manipulation (FRC), or develops septic shock. 12 A clinical diagnosis is more difficult in immunocompromised patients, in whom clinical presentation may be non-specific and there are other potential sources of infection. 13 Isolation of staphylococci or other skin bacteria from multiple blood cultures, Bacillus spp. or fungi raises the probability that the CVC is the source of infection.

In adults, a variety of culture methods are used to identify the CVC as the source of infection. These techniques include:

1. Comparison of blood cultures taken simultaneously from the CVC and a peripheral vein. Numerous studies have shown quantitative differences in the concentration of micro-organisms in blood collected through a CVC compared with blood collected from a peripheral vein when there is a CVC-associated infection. 14–16 A relatively cost-effective way of estimating the differences in microbial numbers between blood collected from a CVC and peripheral blood is to use the differential time to positivity. 17 When a blood culture bottle is continuously monitored using an automated microbial growth detection device (as is widely used in diagnostic laboratories), the time to detection of positivity is a function of microbial numbers in the inoculated blood. Assuming that the blood volumes are similar, detection of positivity in the blood drawn from the intravascular device > 2 hours before positivity in the blood drawn from the peripheral site is highly predictive of a CVC-associated infection. Other studies have shown a link between time to positivity (a marker of bacterial load) and outcome for both Staphylococcus aureus18 and Streptococcus pneumoniae19 bloodstream infections. An alternative method for quantifying organisms when there are large numbers of bacteria in blood drawn through a CVC is to use visualisation techniques such as acridine orange leucocyte cytospin staining, and this technique can provide a rapid diagnosis. 10,20 All these techniques for assessing the differential organism load are appropriate for CVCs that have been inserted for several weeks, in which CVC-associated infection is likely to be intraluminal, but less effective for detecting CVC-associated infection soon after insertion, when organisms may be colonising the outside of the catheter.

2. Comparison of blood culture samples from the CVC and CVC tip: semi-quantitative culture methods can be used to identify colonisation of a CVC once it has been removed [> 15 colony-forming units (CFUs)/ml from a 5-cm segment of the catheter tip]. 21,22 When indistinguishable isolates are cultured from blood cultures and from the device, that is strong evidence implicating the intravascular device in the aetiology of bacteraemia. 23–25

3. Other methods that have been used to diagnose intravascular device-associated infection include luminal brushing. 26,27

Rationale for the study

Many of the diagnostic techniques used in adults are not routinely feasible in children. Reliance on paired blood samples is problematic in children with cancer because of resistance by staff, patients and parents to the routine collection of peripheral blood samples. An additional problem is that children undergoing treatment for cancer frequently receive antibiotics both for prophylaxis and for treatment of infection, which reduces the reliability of diagnostic methods based on laboratory culture. CVC tip culture is not feasible because the CVC would not be removed early on in children with cancer unless the child was extremely ill. Finally, intraluminal brushing is not possible in children because of the narrow catheter gauge and the risk of dislodging thrombi. These problems have led to the development of a molecular method for the diagnosis of CVC-associated infection in children with cancer. 28

The principle underlying the molecular method is based on evidence that the concentration of bacteria and associated bacterial DNA (deoxyribonucleic acid) is high in blood drawn through a colonised CVC. The technique measures DNA that is common to all bacteria, from the 16S rDNA (ribosomal DNA) region. An advantage of the technique is that it can detect infection in patients in whom antibiotics have rendered bacteria non-viable and therefore undetectable by culture. The method has a relatively high detection level of around 10 genome copies per µl of blood (equivalent to 1000 CFUs/ml). The number of bacteria in the peripheral blood of a patient with bloodstream infection rarely exceeds 100 CFUs/ml. Previous studies have shown that a level of bacteria of 1000 CFUs/ml in blood drawn through the CVC discriminates between CVC-associated infection and infection associated with sources other than the CVC. 15 It also reduces the chances of a positive bacterial DNA test result arising as a consequence of sample contamination.

The method described in this study avoids the need for paired blood cultures from the CVC and a peripheral vein, and uses a small volume (< 2 ml) of blood that is normally discarded when the CVC is accessed. 28 The method can be automated and results can be generated within 2 hours, rather than the 48 hours required for blood culture. DNA testing therefore has the potential to lead to earlier initiation of appropriate treatment than is currently possible with reliance on blood cultures.

Overview of the study

The overall aim of our study was to improve the detection and treatment of CVC-associated bloodstream infection in children with cancer who are admitted with fever. In Chapter 2 we report the first step in this process: determination of the accuracy of bacterial DNA testing for detecting CVC-associated infection. Knowing the accuracy of the test allows us to estimate a child’s risk of CVC-associated bloodstream infection. However, to be useful, the test needs to help clinicians decide which children are most likely to benefit from different treatment options. The original plan for the study was to conduct a randomised controlled trial (RCT) comparing DNA testing with standard testing followed by treatment conditional on the test results. However, the accuracy study, and other studies, revealed no consensus about what treatment should be given. 29 We found wide variation in the types of CVC-targeted treatment offered and which children were treated. For example, the duration of ‘CVC-targeted’ treatment (e.g. antibiotic lock treatment or slow infusion) varied from 5 days in one centre to 2 weeks in another. Moreover, several centres did not offer CVC-targeted treatment at all, and none routinely removed CVCs for infection. Partly the reason for this lack of consensus relates to clinicians’ uncertainty about the evidence of what works for CVC-associated infection and whether the evidence applies to children with cancer. Information is also lacking on the prognosis, given standard care, of serious adverse events such as eventual CVC removal for infection, recurrent infection or complications of infection. In summary, it was not possible to proceed immediately to a trial. It was agreed that an evidence synthesis was required to determine how tests on admission predict adverse prognosis for children admitted with fever, what interventions are effective and which groups of patients stand to benefit most from improved detection and treatment.

The three components of the evidence synthesis are:

1. An analysis of the prognosis of serious adverse events, given standard practice (i.e. no targeted treatment for CVC-associated infection), for children admitted with suspected CVC-associated bloodstream infection. This section uses follow-up data for children included in the accuracy study (see Chapter 2) to determine the prognosis for CVC removal or recurrent infection. Our premise was that clinicians would use information from DNA results, in combination with information from the clinical history and examination and the blood culture taken on admission, to decide on whether bloodstream infection is sufficiently likely to warrant immediate treatment, and what treatment should be given.

2. An overview of the effectiveness of different treatment options for CVC-associated infection in children with cancer. This section reports systematic reviews of three intervention options and the findings of a survey of practice regarding use of antimicrobial lock solutions for preventing or treating CVC-associated infection.

3. An analysis of the clinical effectiveness of different test–treatment strategies. In this section, we compile a balance sheet of outcomes to illustrate the consequences of different test–treatment strategies.

The detailed objectives, methods and results of each of these analyses are reported in the ensuing chapters. The final chapter includes a discussion of the implications of our findings for practice and the priorities for further research.

Introduction

In this section, we report findings from a prospective study to determine the accuracy of bacterial DNA for discriminating between children with and without a CVC-associated bloodstream infection. CVC-associated infection was measured by a composite reference standard based on blood culture results, clinical findings and clinicians’ judgement. The results of this evaluation were published in 2008. 30

Methods

The accuracy study involved eight UK paediatric oncology centres [Belfast, Bristol, Great Ormond Street (London), Liverpool, Newcastle upon Tyne, Nottingham, Royal Marsden (London) and University College Hospital (London)] and was co-ordinated by the Supportive Care Group of the CCLG. The protocol for the study was agreed by the CCLG (following a national meeting) and received ethical approval through the Trent Multicentre Research Ethics Committee (reference number 05/MRE04/23). A summary of the protocol for the study is in Appendix 1. A copy of the full protocol and ethics approval is available from the CCLG website (www.cclg.org.uk) or from the principal investigator, Mike Millar.

Results

Children admitted to hospital with fever were recruited into the study between 7 November 2005 and 6 November 2006. Samples and clinical data sheets were collected from 301 episodes of fever in 207 children. The numbers recruited by each centre were Belfast 15, Bristol 51, Great Ormond Street 2, Liverpool 63, Newcastle upon Tyne 63, Nottingham 19, Royal Marsden 19 and University College Hospital 31. We were unable to accurately estimate the number of eligible patients who were not recruited.

Exclusions from the analyses

Forty-one episodes were excluded from analysis.

The reasons for exclusion in 10 episodes were no written consent form, inappropriate sample storage or loss of sample, or antibiotics given intravenously during a 14- to 3-day period before the onset of fever. A further 26 episodes were excluded because of failure to collect samples from all lumens.

Five episodes were excluded because CVC-associated infection was considered to have been acquired post admission to hospital (diagnosed 5–23 days after initial presentation). Four of these five episodes were associated with positive blood cultures, and one episode was a tunnel infection. In one of these episodes, a sample was collected for 16S rDNA analysis at the time of fever recurrence (5 days after initial presentation), and this sample gave a bacterial DNA concentration of 0.34 pg/µl blood, while blood cultures taken at the same time grew Stenotrophomonas maltophilia.

The proportion of eligible episodes excluded from analysis ranged from 0% to 33.3% for each centre. CVC tips were sent for culture from 16 (84%) of 19 episodes in which the CVC was removed. The numbers of episodes overall, the number with different microbial DNA results and the numbers within each reference group are shown in Figure 1.

FIGURE 1.

Eligible episodes, exclusions and numbers with different microbial DNA results and in each reference group. Pr, probable; Po, possible; Ul, unlikely.

Discussion

The 16S rDNA test yielded sensitivity for episodes defined as probable CVC-associated infection, specificity and positive predictive values similar to those reported for paired quantitative blood cultures. 32 Unlike many reported evaluations, this study was performed by laboratory staff working at a distant site unaware of the clinical details of individual patients, and the results were achieved despite the frequent exposure of patients to oral antibiotics in the 2-week period preceding fever presentation.

The method reported here has a relatively high minimum detection level of c. 10 genome copies/µl of blood. This relatively high minimum detection level probably explains the episodes with positive blood culture and undetectable bacterial DNA (although the possibility of blood culture contamination cannot be excluded). This high detection level also reduces the chances of a positive bacterial DNA test result arising as a consequence of sample contamination. A limitation of the methodology used in this study was the use of the discard sample. The implicit assumption was that this sample would represent microbial colonisation throughout the CVC lumen. This assumption may not be correct. We would recommend the collection of a sufficient sample volume to ensure that the whole volume of the CVC lumen is sampled. Extraction of microbial DNA from a larger volume would also potentially increase test sensitivity.

The manual DNA extraction method described in this study is time-consuming, but subsequent evaluations have obtained comparable results using automated DNA extraction systems, with considerable savings in technicians’ time (results not shown). The quantitative bacterial DNA method used in the present study does not generate a product that is sufficiently informative to allow bacterial identification. When the bacterial DNA concentration was > 0.5 pg/µl, it was possible to identify bacteria by amplification of a discriminatory 16S rDNA region, followed by sequencing of the amplified product. The majority of identifications according to molecular and conventional laboratory methods were consistent. Discrepant identifications probably reflect the limitations of routine laboratory standard operating procedures.

Previous reports have suggested a link between time to positivity (a marker of bacterial load) and outcome for both S. aureus18 and S. pneumoniae19 bloodstream infections. In the present study, increasing bacterial DNA load in blood samples drawn through the CVC was associated with an increasing risk of CVC removal for suspected infection. Information was collected for only 4 weeks after fever presentation. Prolonging the period of data collection might allow a better assessment of the implications for outcomes in patients with high bacterial load CVC-associated infection (see Chapter 3). Bacterial load is an important determinant of the efficiency of sterilisation and disinfection processes, so it is perhaps not surprising to find a relationship between the effectiveness of antimicrobial treatment of CVC-associated infection and bacterial load.

Whether measurement of high or low levels of DNA is most useful depends on how the test will be used in practice. In the original proposal for this study we envisaged the main benefit of using a molecular test for CVC-associated infection to be a reduction in unnecessary CVC removal – based on estimates that 60% of CVCs removed for suspected infection were removed unnecessarily (see background to study in trial protocol, Appendix 1, and Farr33). However, the results showed that removal of a CVC for suspected infection without FRC or a DNA result of < 0.125 pg/µl was uncommon, so the number of ‘unnecessary’ CVC removals defined by clinical criteria or DNA level was small. Hence, the potential benefit of the DNA test in reducing unnecessary CVC removal is also small. This finding concurs with the stated current practice in paediatric oncology in the UK and much of Europe,34 which is to retain CVCs if removal can be avoided. On the other hand, the bacterial DNA test does identify children with episodes of probable CVC-associated infection in whom improved treatment strategies for CVC-associated infection could be targeted. The test was not sufficiently sensitive to rule out patients at low risk of CVC-associated infection.

The main limitation of the accuracy study was the lack of an adequate reference standard. We used criteria for a CVC-associated bloodstream infection that combined blood culture results with clinical signs of CVC-associated infection and response to treatment, based on clinicians’ judgement. These judgements may have been strongly influenced by the blood culture result, which could have biased results in favour of underestimating the accuracy of bacterial DNA testing. This means that we were unable to determine whether DNA testing is more accurate than blood culture, and whether DNA testing would improve the prediction of outcomes over and above information currently available from clinical signs and blood culture. In Chapter 3, we aim to address this question by comparing the prediction of DNA and other tests for prognostic outcomes, including CVC removal and recurrence of bloodstream infection.

Background

To decide on the introduction of a new test, clinicians need to know the added value of bacterial DNA testing over and above information that would be available from other tests that would usually be performed. Clinicians can opt to add a new test to an existing set of tests, use the new test instead of an existing test, or not use the new test at all. In Chapter 2, we compared bacterial DNA testing with a clinical reference standard for CVC-associated bloodstream infection. However, this reference standard is imperfect and, because it includes one of the existing tests used on admission (blood culture), the accuracy study does not provide information on whether DNA testing is more effective than blood culture. To address this question, we compared DNA testing and blood culture as predictors of the consequences of CVC-associated infection. This makes sense clinically as the usual intervention for children admitted with fever is at least 5 days of i.v. antibiotics. Many children with a bloodstream infection from any source (CVC or other sites) will be adequately treated by this regimen. Clinicians particularly want to identify children who are unlikely to respond to such treatment and need additional interventions targeted at CVC-associated infection, or even need their CVC removed. Clinicians also need to be able to identify children who do not need the 5 days of antibiotics at all, and could be discharged early on. This last question cannot be addressed by this study as no centre routinely discharged children early.

To predict the children likely to need additional targeted interventions, we analysed outcomes for a cohort of children derived from the accuracy study reported in Chapter 2. We determined the prognosis for outcomes at 28 days and 6 months after admission, according to tests and markers assessed on all children at admission. The primary outcomes were days of i.v. antibiotic treatment and CVC removal.

In addition to these secondary analyses of existing data, we searched the literature for studies on prognostic outcomes in children with cancer who had suspected CVC-associated infection (see Chapter 5, Figure 5 and search strategy in Appendix 5). As we found no relevant studies, we explored using other data sets for secondary analysis.

We obtained a data set from a longitudinal study of children with cancer conducted in the 1990s by Tweddle et al. 35 The study, UKCCSG SC 9403, was instituted jointly by the UKCCSG (now CCLG) and the Paediatric Oncology Nurses’ Forum (PONF) of the Royal College of Nursing. The design was a prospective observational study examining both mechanical and infective complications of CVCs in children being treated for cancer.

Eligible participants were all patients requiring central venous access for cancer therapy administered by a UKCCSG centre over a 20-month period from 1994 to 1996. Infection episodes requiring i.v. antibiotic treatment were recorded in the data set and we assumed that children were admitted to hospital for these episodes. To reproduce the cohort derived from the accuracy study reported in Chapter 2, we randomly selected one i.v. treatment period for each child, provided treatment started > 2 weeks after a previous treatment period. The cohort comprised 1069 patients, of whom 339 had at least one admission meeting our criteria. During these analyses we were notified of concerns about data errors by the Clinical Trials Unit (CTU) in Leicester, where the data were held. As the CTU was in the process of closing, no further data checking was possible. In view of the lack of confidence in the data expressed by the custodians, we have not used these analyses. Characteristics of the cohort are summarised in Appendix 8.

Methods for prognostic analyses of the accuracy study cohort

The aim of the secondary analyses of the accuracy cohort (described in Chapter 2) was to determine how clinical signs or test characteristics recorded at admission discriminate between children with and without sequelae of CVC-associated bloodstream infection. We assumed that this cohort represents the baseline prognosis in patients treated with standard care rather than targeted treatment for CVC-associated infection. This assumption is based on practice reported by collaborating centres and the fact that CVC-targeted treatment was recorded for only 24/260 infection episodes in the accuracy study (see Table 2).

The study was co-ordinated through the Supportive Care Group of the CCLG and involved eight UK centres. They were Belfast, Bristol, Great Ormond Street (London), Liverpool, Newcastle upon Tyne, Nottingham, Royal Marsden (London) and University College Hospital (London).

Statistical analysis

Survival analysis was used for time-to-event outcomes associated with each prognostic variable, and hazard ratios (HRs), CIs and p-values were calculated. Survival curves were plotted using Kaplan–Meier estimates for time-to-event outcomes for each of the three tests (DNA, blood culture and clinical signs of FRC). Poisson regression was used to calculate rate ratios for recurrent i.v. treatment periods, taking into account multiple recurrences in some patients. Linear regression analysis was used to determine the effect of prognostic markers on the total duration of i.v. treatment.

The multivariable analyses were confined to two primary outcomes: time to CVC removal and the total duration of i.v. treatment during follow-up. We did not undertake multivariable analyses for recurrent treatment episodes owing to lack of power. To determine the added predictive value of DNA status, we analysed multivariable models that included blood culture and clinical signs, with and without DNA status. We compared the goodness of fit of these models using the Akaike’s information criterion (AIC) statistic. We included additional variables that were associated with time to CVC removal or the total duration of i.v. treatment, provided that they were not strongly correlated with other variables in the model. Statistical analysis was performed using r v.2.9.2 (R Foundation for Statistical Computing, Vienna, Austria). 37 We carried out a sensitivity analysis restricted to prognostic markers available on the day of admission and excluding blood culture results.

Results

The distribution of prognostic markers and outcome variables are shown in Table 7 for 179 children with sufficient data for inclusion in the 28-day follow-up and for 99 children included in the 6-month follow-up. The age distribution between the two data sets was similar, with 20% of children aged < 3 years and 65% with haematological cancer. The median duration of CVC insertion before the index admission was 4 months. Few children (7%) had FRC. However, one-fifth had a positive blood culture, although few of these contained pathogens, as defined in Table 2.

The relationship between clinical signs, DNA test and blood culture results is shown in Table 8. All test results are negative for 73% of children (131/179). The univariate analyses for follow-up to 28 days show relatively few associations at a 5% level of significance. Implanted CVCs and those with a single lumen were associated with earlier stopping of initial i.v. antibiotic treatment than external ports or CVCs with multiple lumens (i.e. HR for stopping treatment was > 1.0; Table 9). Children with FRC or with a positive blood culture were less likely to stop initial treatment early (HR < 1.0) and had a longer overall duration of i.v. treatment. They were also more likely to have their CVC removed. These findings are not surprising as duration of treatment and the decision to remove the CVC will be partly determined by the blood culture result and by the presence of clinical signs of CVC-associated infection. The effect of high levels of bacterial DNA varied according to whether the CVC was removed or not. Subgroup analyses in the lower part of Table 9 and Appendix 3 show that a high level of DNA was associated with increased days of i.v. treatment in patients in whom the CVC was not removed, but this relationship was not observed for patients with the CVC removed, partly because so few patients were studied.

Similar patterns were observed for follow-up at 6 months, although associations were weaker and fewer were significant at the 5% level (see Appendix 3). In the 6-month follow-up, use of oral antibiotics in the 2 weeks prior to index admission appeared to be protective for recurrent i.v. treatment and CVC removal, and children younger than 3 years appeared to have a reduced risk of recurrent infection. These results should be regarded with caution because the cohort represents just over 50% of those eligible for inclusion. In addition, the large number of comparisons also increases the chance of associations being statistically significant by chance.

Figures 2 and 3 show survival plots for the three tests of clinical signs (FRC), bacterial DNA and blood culture for two outcomes: recurrent i.v. treatment episode and CVC removal. Consistent with the effects shown in Table 9 for the 28-day outcomes, these tests do not discriminate between children with recent infection, but show a clear effect for CVC removal. Similar patterns are seen for the cohort followed up for 6 months (see Appendix 3).

FIGURE 2.

Kaplan–Meier plots of time to recurrent infection requiring i.v. treatment in cohort followed up for 28 days according to three test results: (a) FRC; (b) bacterial DNA concentration; (c) blood culture (BC) result.

FIGURE 3.

Kaplan–Meier plots of time to CVC removal in cohort followed up for 28 days according to three test results: (a) FRC; (b) bacterial DNA; (c) blood culture (BC).

Multivariable analyses

The strong associations between clinical signs of FRC and pathogens isolated on blood culture and an increased risk of CVC removal persisted in the multivariable analyses. The addition of bacterial DNA to the model attenuated this relationship slightly. Single-lumen CVCs were associated with a reduced overall duration of i.v. treatment: this effect was not altered by inclusion of DNA level in the model. DNA level was not significantly predictive of any outcome and did not significantly improve the fit of the model, as measured by the AIC. Similar results were found for the cohort followed up for 6 months (see Appendix 3, Table 22).

The sensitivity analyses showed that if only markers available on the day of admission were considered, bacterial DNA did contribute significantly to the prediction of CVC removal and duration of i.v. treatment (Table 10 and Figure 4). If high DNA (> 0.5 pg/µl) or clinical signs of FRC were considered as a combined marker (vs any other result for FRC or DNA), a positive result was highly predictive of CVC removal and i.v. treatment duration.

FIGURE 4.

Sensitivity analysis showing survival analysis for time to CVC removal at 28 days given FRC-positive or DNA level > 0.5 pg/µl (vs FRC-negative or DNA ≤ 0.5 pg/µl).

TABLE 10a - Multivariable analyses of predictors of outcomes related to CVC-associated infection

DNA, bacterial DNA result.

Significant values at the 0.05 level are shown in bold.

Explanatory variables	Time to CVC removal				Total duration of i.v. treatment [adjusted for indicator variable: CVC removal with follow-up period (yes/no)]
	Model without DNA		Model with DNA		Model without DNA		Model with DNA
	HR (95% CI)	p-value	HR (95% CI)	p-value	Coefficient (95% CI)	p-value	Coefficient (95% CI)	p-value
Single lumen	0.17 (0.02 to 1.37)	0.096	0.17 (0.02 to 1.37)	0.096	–2.94 (–4.50 to –1.37)	0.000	–2.79 (–4.42 to –1.17)	0.001
With FRC	13.50 (3.87 to 47.08)	< 0.0005	6.93 (1.63 to 29.33)	0.009	2.65 (–0.57 to 5.87)	0.109	2.33 (–0.98 to 5.64)	0.170
DNA (> 0.5 pg/µl)			4.51 (1.06 to 19.20)	0.042			1.44 (–1.88 to 4.77)	0.395
DNA (0.125–0.5 pg/µl)			0.00 (0 to ∝)	0.998			0.73 (–2.15 to 3.61)	0.620
AIC	87.55		84.60		596.21		599.28

TABLE 10b - Sensitivity analyses restricted to variables available on day of admission

DNA, bacterial DNA result.

Significant values at the 0.05 level are shown in bold.

Explanatory variables	Time to CVC removal				Total duration of i.v. treatment (Adjusted for indicator variable: CVC removal with follow up period (Yes/No))
	Model without DNA		Model with DNA		Model without DNA		Model with DNA
	HR (95% CI)	p-value	HR (95% CI)	p-value	Coefficient (95% CI)	p-value	Coefficient (95% CI)	p-value
DNA as independent variable
Single lumen	0.17 (0.02 to 1.37)	0.096	0.17 (0.02 to 1.37)	0.096	–2.94 (–4.50 to –1.37)	0.000	–2.79 (–4.42 to –1.17)	0.001
With FRC	13.50 (3.87 to 47.08)	<0.0005	6.93 (1.63 to 29.33)	0.009	2.65 (–0.57 to 5.87)	0.109	2.33 (–0.98 to 5.64)	0.170
DNA (> 0.5 pg/µl)			4.51 (1.06 to 19.20)	0.042			1.44 (–1.88 to 4.77)	0.395
DNA (0.125–0.5 pg/µl)			0.00 (0 to ∝)	0.998			0.73 (–2.15 to 3.61)	0.620
AIC	87.55		84.60		596.21		599.28
			Model DNA + FRC				Model DNA + FRC
Single lumen			0.26 (0.03 to 2.07)	0.202			–2.82 (–4.38 to –1.26)	0.001
FRC (+) or DNA > 0.5 pg/µl			29.63 (6.17 to 142.23)	0.000			2.97 (0.20 to 5.75)	0.037
AIC			87.55				596.21

Discussion

These findings provide no evidence that bacterial DNA, used as a single test on admission in this patient population, improved the prediction of outcomes at 28 days related to CVC-associated bloodstream infection. Analyses based on follow-up to 6 months did not change these conclusions but were underpowered to detect potentially important effects. Limitations of the study are discussed in Chapter 6.

Overview of the systematic reviews

Scope of the reviews: rationale

We conducted systematic reviews for three treatment comparisons. These were selected based on clinical opinion (Mike Millar in discussion with the CCLG) and the available research literature. 38 The clinically important outcomes were resolution of infection, removal of the CVC owing to infection and recurrence of infection. The intervention options reviewed are outlined in brief below.

1. Early CVC removal compared with retaining the CVC and treatment in situ Central venous catheter access is particularly important in children not only for the administration of cytotoxic drugs but also to avoid the trauma of repeated venepuncture. In paediatric oncology practice, removal of the CVC for suspected CVC-associated infection is seen as a last resort reserved for children with complicated or unresolving CVC-associated infection. This is because, compared with adults, CVC reinsertion carries a greater anaesthetic and operative risk, involves greater technical difficulties and is more likely to ‘use up’ venous access sites that might be needed in future.

2. Treat the suspected CVC-associated infection with an antimicrobial lock solution compared with a standard heparin lock Antibiotic or antiseptic lock solution (together referred to as antimicrobial) can be used to treat suspected CVC-associated infection, and in children might be used to ‘salvage’ the line – avoiding removal.

3. Antimicrobial locks to prevent CVC-associated infection We conducted a review of the effectiveness of locks to prevent CVC-associated infection, as this option might reduce the overall incidence of admission of children with suspected CVC-associated infection.

The section on locks includes a survey of paediatric oncology units in the UK: on the use of antimicrobial locks in practice, and on the formulation of locks used.

The inclusion criteria for these reviews distinguished between interventions affecting infection within the CVC and interventions involving the whole patient. We reasoned that antimicrobial locks are interventions that act on infection within the CVC and should therefore have a similar effect across patient groups. We included any studies in adults or children in these reviews (2 and 3). On the other hand, the criteria for CVC removal are related to problems with venous access, as well as the underlying clinical problem, and are likely to vary between children and adults. We therefore restricted the review of CVC removal to studies in children and adolescents with cancer.

One further strategy, used by some clinicians for children with suspected CVC-associated bloodstream infection, is to administer systemic antibiotics by slow infusion instead of bolus i.v. injection. The rationale, to increase the duration of exposure of organisms colonising the CVC to antibiotics, is based on evidence that bactericidal action is time dependent. 39 We did not conduct a systematic review of this option for two reasons. First, we found no relevant studies for slow infusion compared with bolus administration of antibiotics (see Figure 5). Second, our searches highlighted difficulties in the definition of slow and bolus. For example, we found one systematic review of 17 RCTs that compared continuous infusion with intermittent administration of antibiotics. 40 Clinical failure was lower, albeit with equivocal statistical significance, in patients randomised to continuous infusion. We also found one study protocol (see Appendix 4) that randomised patients either to bolus injection with teicoplanin or to prolonged teicoplanin exposure, which could involve slow infusion (over 1–2 hours) or a teicoplanin lock (either given at the clinician’s discretion). There is a need for further research using clearly defined criteria for the duration and dosage of antibiotic administration.

FIGURE 5.

Flow diagram of searches.

No protocol for any of these reviews has been published elsewhere and there is no registration number for these reviews.

Search strategy, selection of studies and data extraction

The results of the searches are summarised in Figure 5. Search terms are given in Appendix 5. First, we conducted a broad search of the Cochrane Central Register of Controlled Trials (CENTRAL) to identify any RCTs that included terms relating to CVC and infection. Second, we conducted a sensitive search for prognostic studies, using terms related to follow-up or prognosis combined with terms for CVC and infection or removal, and restricted to children or adolescents. Third, we devised searches for comparisons involving locks and removal of the CVC. Fourth, we carried out a search for infusion versus bolus treatment, prior to aborting this review (reasons given above). Because of changes in the quality of supportive care for patients with CVCs during the last 15 years, we restricted studies to those published after 1994. We did not impose any language restrictions.

Mike Millar and Ruth Gilbert scanned all the abstracts from the two broad-based searches to identify potentially eligible studies for any of the reviews. Mike Millar scanned all abstracts from all of the searches and Ruth Gilbert scanned all the searches labelled as ‘prognosis’ and those from CENTRAL.

Full copies of potentially eligible studies were reviewed by one author (RG, MM or WZ) and included and excluded studies were decided by discussion within the group. Data extraction was initially carried out by Weiwei Zhou and checked by Ruth Gilbert. No additional data were sought directly from investigators.

Systematic review of early central venous catheter removal compared with retention and treatment in situ

Background to reviews of antimicrobial locks for treatment or prevention

Injection of lock solution to fill the CVC lumens is standard practice to maintain CVC patency when not in use for infusion of fluids or administration of drugs. The standard lock solution is heparinised saline. The amount injected is typically 1–2 ml in children (0.4 ml in neonates). 72 Antimicrobial and antiseptic agents have been added to heparinised saline and other solutions to prevent adherence and multiplication of bacteria in the lumen and to eradicate bacteria that adhere to the CVC tubing. Antibiotic lock solutions can achieve much higher levels of antibiotic within the CVC lumen than could be safely achieved within the bloodstream. 34,73–75 The lock solution is left in the CVC until the next time the lumen needs to be used for fluid or drug administration to the patient, usually for at least 2 hours, though dwell times from 20–60 minutes up to several days have been reported. 76 Recommended practice is to withdraw and discard the lock solution before using the CVC again for infusion or administration of drugs in order to avoid adverse effects due to excessive blood levels of antibiotics or antiseptic solution.

Advantages of antibiotic or antiseptic locks include their low cost and the simple substitution of a different type of lock solution. Disadvantages include selection pressure, either for antibiotic-resistant organisms or for particular pathogens. 5 A further disadvantage is the need to allow a minimum dwell time, when the CVC lumen cannot be accessed, which can be associated with adverse effects, such as hypoglycaemia in neonates. 77 When used for treatment, antibiotic or antiseptic locks are almost always given in addition to systemic i.v. therapy. The type of antibiotic depends on the infecting organism and the dwell time is usually hours (e.g. 8–12 hours per day36) rather than days. The primary aim is to treat the current episode of infection and to reduce the risk of infection recurrence and complications. When used for prevention, antibiotic or antiseptic locks are given without other systemic treatment and the dwell time can be several days (e.g. between treatments or dialysis episodes). The primary aim is to reduce the risk of a CVC-associated infection developing. In view of these differences, we have reviewed the use of locks for treatment and prevention separately.

Systematic review of antibiotic locks for treating central venous catheter-associated infection

Results

We found four systematic reviews,73,75,80,81 one other review,34 and four comparative studies (see Table 12). 36,78,79,82 One of these was excluded as the authors compared different types of antimicrobial lock solutions. 82 Of the remaining three studies, only one was an RCT. 36 Rijnders et al. 36 compared vancomycin or ceftazidime and heparin lock with heparin alone, in children and adults with proven or suspected CVC-related bloodstream infection (see Table 12). This was a good-quality trial in which treatment and placebo were randomised by the hospital pharmacist and allocation was concealed from clinicians entering patients into the trial. In practice, the study population was selected to favour patients with coagulase-negative staphylococcal infection (29/46) or unproven CVC-related infection. A large number of potentially eligible patients were excluded from the trial because their physician requested removal of the CVC (e.g. all CVC-associated bloodstream infections due to S. aureus, fungal or Gram-negative infections). These recruitment problems, in addition to failing to recruit the planned sample size, led to the study being stopped.

The remaining two studies78,79 compared cohorts of adult patients in whom antibiotic locks were used with a historical cohort of patients in the same centre before antibiotics locks were introduced. The main weakness of these before–after comparisons is that other practices may have changed, apart from the use of antibiotic locks. For example, the US study by Poole et al. 79 excluded patients with enterococcal bloodstream infection in the phase when antibiotic locks were used, but did not specify this exclusion for the historical cohort.

Patients, interventions and outcomes differed between the three studies. In the study by Rijnders et al. 36 patients included a mix of adults and children who required a CVC, Poole et al. 79 studied adults undergoing haemodialysis, and Fortun et al. 78 studied adults receiving chemotherapy or parenteral nutrition. Most patients in both intervention and control groups had CVC-associated bloodstream infection due to coagulase-negative staphylococci. However, all three studies had an imbalance between the comparison groups, with fewer patients with Gram-positive bloodstream infection in the control group than in the group treated with antibiotic locks.

Interventions also differed. In the study by Poole et al.,79 patients did not receive concomitant systemic antibiotic therapy, but patients in the control group underwent CVC exchange. Finally, the criteria for the outcome – treatment failure – differed, ranging from follow-up of 1 month78 to 24 weeks. 36 The baseline risk of treatment failure ranged from 34% to 57% (see Table 13). As none of the studies reported sufficient information to allow a pooled analysis of the time to treatment failure, we calculated a relative risk and pooled relative risk for treatment failure. In view of the differences between the studies, we used a random effects model.

The relative risk of treatment failure reported in the RCT by Rijnders et al. 36 was 0.59 with a 95% CI that included 1.0 (0.29 to 1.19; Table 13). The pooled relative risk, based on all three included studies, did not provide evidence of a significant benefit of antibiotic lock solution at the 5% level (pooled relative risk 0.70, 95% CI 0.47 to 1.05; Figure 6). The results were moderately heterogenous (I²-value 0, 95% CI 0 to 90).

FIGURE 6.

Forest plot and pooled relative risk (random effects model).

Systematic review of antimicrobial locks for prevention

Results

We found 17 reviews. 5,73,74,77,106–116 Two of these reviews77,106 contained 21 of 24 RCTs included in our review. Scrutiny of published guidelines did not yield additional studies. 13,38,117,118 One further systematic review119 was published after the searches were complete, and contained four further studies. 99,120–122 Only one of these, involving gentamicin (total 140 patients),99 was an RCT and contained data that could be used in the meta-analysis. In total, we found eight systematic reviews. 5,77,106,108–111,115,119 We excluded a further 17 studies because they were not trials, did not compare antimicrobial locks or compared a single perioperative administration of lock solution (see Table 23, Appendix 6). We found one trial in progress using TauroLock in children with cancer (NCTs 00735813, 00545831, 0074916).

Of the 245 included trials, 18 compared various antibiotic lock solutions with heparinised saline (see Table 14). 72,83–99 The remaining six studies included citrate or taurolidine. 100–105,123 Most studies were conducted in patients undergoing haemodialysis (n = 17), but there were four studies in children with cancer. 85–87,100 No studies compared locks in adults or children receiving parenteral nutrition.

Under half of the studies (n = 10) reported adequate allocation concealment, and nine of these were placebo controlled (see Table 14). At least five studies randomised more CVCs than there were patients, so that the same patient was included in the trial more than once. 87,89,96,98 In other studies, poor reporting made it hard to assess repeated inclusion of the same patient. Similarly, loss to follow-up was hard to assess because of poor reporting but was ≤ 10% where numbers randomised and assessed for outcome were reported. 84,92,96,105 The duration of follow-up ranged from 20 to 368 days (see Table 14). There was substantial variation in the baseline rate of bloodstream infection in the control group. The rate was lowest (< 1/1000 CVC days) in adults undergoing haemodialysis and highest (up to 21/1000 CVC days) in children in intensive care (Figure 7).

FIGURE 7.

Baseline rate of bloodstream infection by age and condition.

Despite the large number of studies and 3043 patients studied (see Table 14), this review lacked the power to detect CVC-associated bloodstream infection due to resistant organisms because of the low rate of bloodstream infection (see Table 15). Fourteen studies followed up patients for ≥ 3 months, three of these followed patients for > 10 months (see Table 14). 83,93,96 Very large studies with long-term follow-up would be required to provide clear evidence to support or refute widespread concerns about rare but serious infections with resistant organisms due to the use of antimicrobial locks. However, other studies have reported selection of antibiotic-resistant strains. Guerraoui et al. 124 reported resistant strains of S. epidermidis after prolonged use of gentamicin locks for prophylaxis in permanent haemodialysis catheters. Finally, in a before–after study comparing taurolidine with heparin saline locks in children with cancer, Simon et al. 100 reported a reduction in CVC-associated infection with staphylococci and an increase in CVC-associated infection with E. coli (10/27 in the taurolodine group vs 4/31 in the control group; odds ratio for E. coli occurrence 3.97, 95% CI 1.12 to 13.93).

The pooled incidence rate ratio for all 23 studies included in the meta-analysis (one excluded because of zero events) was 0.46 with narrow confidence limits (95% CI 0.39 to 0.53; Figure 8). There was moderate heterogeneity across studies (I²-value 34; see Table 15), but the chi-squared test for heterogeneity was not significant at the 5% level (p = 0.0579). All except one trial had a central estimate consistent with a beneficial effect of lock solution compared with heparinised saline (see Figure 8 and Table 15).

FIGURE 8.

Forest plot and pooled incidence rate ratio for bloodstream infection in patients randomised to antimicrobial lock prophylaxis or standard heparin lock. IRR, incidence rate ratio; M–H, weighting for Mantel–Haenszel model.

We explored sources of heterogeneity using subgroup analyses (Table 16). The pooled incidence rate ratio was similar in studies regardless of the quality of allocation concealment, but was lower in studies using antibiotic rather than antiseptic solution (see Table 16). There was moderate heterogeneity within subgroups according to condition and child or adult populations, and the pooled incidence rate ratio was closer to 1.0 in studies of children than in those of adults (see Table 16). Nevertheless, the effect of antimicrobial lock solution was substantial and significant at the 5% level for subgroups of children with cancer and for babies in intensive care (see Table 16 and Figure 8). There was evidence that rate ratios were lower for patients with a tunnelled CVC than for those with other forms, and where the lock solution had been explicitly administered as a flush (see Table 16).

TABLE 16 - Pooled incidence rate ratios (IRRs) for subgroups and measures of heterogeneity

BSI, bloodstream infection; CRBSI, catheter-related bloodstream infection.

Type of CVC not stated in Kim et al. ’s paper. 94

Explicit lock not stated in Zhang et al. ’s paper. 97

Outcome measured not stated in Zhang et al. 's paper. 97

stata (v.9.2) command metan was used for calculating Q-statistic, degrees of freedom and p-value of chi-squared test. stata command heterogi was used for obtaining I²-value by specifying Q-statistic and degrees of freedom from metan’s results.

	n	IRR (95% CI)	Q-statistic	p-value (chi-squared test)	I2-value (95% CI)
All studies	23	0.43 (0.36 to 0.51)	33.29	0.0579	34 (0 to 60)
Allocation concealment
Yes	10	0.43 (0.35 to 0.62)	8.48	0.4869	0 (0 to 62)
Not stated	13	0.43 (0.32 to 0.59)	24.7	0.0163	51 (8 to 74)
Type of lock
Antibiotic	18	0.40 (0.33 to 0.48)	24.51	0.1061	31 (0 to 61)
Non-antibiotic	5	0.56 (0.39 to 0.79)	7.01	0.1354	43 (0 to 79)
Age group
Adults	18	0.37 (0.30 to 0.46)	24.79	0.0994	31 (0 to 61)
Children	5	0.59 (0.44 to 0.79)	6.24	0.1821	36 (0 to 76)
Type of lock, age group, condition
Antibiotic_Adults_Cancer	2	0.29 (0.06 to 1.42)	0.85	0.3569
Antibiotic_Adults_Haemodialysis	12	0.36 (0.28 to 0.45)	18.16	0.0778	39 (0 to 69)
Antibiotic_Children_Cancer	3	0.56 (0.38 to 0.83)	4.46	0.1075	55 (0 to 87)
Antibiotic_Children_Intensive	1	0.38 (0.14 to 0.96)
Antiseptic_Adults_Haemodialysis	4	0.45 (0.28 to 0.71)	5.78	0.123	48 (0 to 83)
Antiseptic_Children_Cancer	1	0.76 (0.43 to 1.33)
Type of CVC
Tunnel-cuffeda	15	0.44 (0.36 to 0.54)	28.28	0.0131	50 (10 to 73)
Non-tunnelled	5	0.47 (0.35 to 0.64)	0.78	0.9405	0 (0 to 79)
Mixed	2	0.26 (0.12 to 0.54)	0.01	0.9342
Explicit lock
Yesb	15	0.43 (0.35 to 0.52)	22.31	0.0726	37 (0 to 66)
No	7	0.44 (0.32 to 0.61)	10.81	0.0945	44 (0 to 77)
Outcome measured
BSIc	7	0.47 (0.36 to 0.63)	13.67	0.0335	53 (0 to 79)
CRBSI	15	0.41 (0.33 to 0.51)	18.76	0.1742	25 (0 to 60)

The funnel plot showed weak evidence of asymmetry (Figure 9). The Egger test for asymmetry was of equivocal significance (Egger test p = 0.073). A L’Abbé plot did not indicate a strong relationship between the baseline event rate and the incidence rate ratio (plot not shown).

FIGURE 9.

Funnel plot for studies of antimicrobial lock prophylaxis compared with standard heparin solution for preventing CVC-associated infection. For five studies with at least one arm with zero events, we added 0.5 to each arm.

Survey of practice

Evidence for the effectiveness of antimicrobial lock solution for preventing CVC-associated bloodstream infection has now been summarised in nine systematic reviews (including our own). 5,77,106–111 These reviews consistently report a reduced rate of CVC-associated infection in patients given antimicrobial lock prophylaxis compared with heparin lock solution. According to our review (based on 24 RCTs), the rate of infection was halved in patients given antimicrobial lock prophylaxis. The evidence that antibiotic solutions are effective for treating CVC-associated infection is much weaker. There have been five systematic reviews including our own,73,75,80,112 but only one RCT36 and two before–after studies. 78,79 Our systematic review found no clear evidence for the effectiveness of antibiotic lock treatment for CVC-associated bloodstream infection.

These findings from systematic reviews contrast with recommendations for practice in national and international guidelines. Most recently, the Infectious Disease Society of America recommended using lock treatment rather than CVC removal for CVC-associated bloodstream infection due to pathogens other than S. aureus, P. aeruginosa, Bacillus spp., Micrococcus spp., propionibacteria, fungi or mycobacteria in patients with long-term CVCs. 13 Despite the acknowledged weak evidence base, antibiotic lock treatment has been recommended for nearly a decade in patients with CVC-associated infection. 116 In contrast, national guidelines in the UK and USA, and international guidelines for specific disease groups, recommend against routine use of antimicrobial locks to prevent infection except for patients with recurrent CVC-associated bloodstream infection. 38,125–127 Reasons against using antimicrobial locks for preventing infection include the theoretical risk of antibiotic resistance and the potential for systemic toxicity from leakage of the lock solution into the bloodstream. 76 Some guidelines state that antimicrobial lock works only for preventing CVC-associated infection in neutropenic patients,125,126 a perception disproved by the consistent finding of benefit in patients receiving haemodialysis in our review. Views may be beginning to change, however. The British Society for Haematology guidelines do not explicitly recommend use of antimicrobial locks for prevention or treatment, but mention both applications,128 and the most recent guidelines from the USA advocate antimicrobial lock prophylaxis for patients with limited venous access. 13,76

The disparity in the literature between evidence for effectiveness of antimicrobial locks and recommended practice was mirrored in our discussions with paediatric oncologists. We therefore undertook a national survey of paediatric oncology centres to determine the extent to which antimicrobial locks were used for prevention or treatment of CVC-associated bloodstream infection, any experiences of adverse effects, and what were the perceived disadvantages that discouraged their use. In centres that reported using locks, we sent a questionnaire to the hospital pharmacist requesting details of the formulation of lock solutions used.

Survey of formulation of locks for treatment

An issue that became apparent following informal discussion with local pharmacists is the paucity of commercially available antibiotic lock solutions. In the UK, commercially available antimicrobial locks are limited to TauroLock (taurolidine citrate; Kimal, TauroPharm GmbH, Waldbuttelbrunn, Germany) and Citra-Lock (trisodium citrate; Dirinco, Eindhoven, Holland), which are used only for prevention. There are no commercially available antibiotic lock solutions for prevention or treatment so that, in practice, using antimicrobial locks requires either the use of a commercially available product or considerable support from local hospital pharmacies. For example, the provision of antimicrobial lock solution for prevention would require the manufacture of > 200 ampoules/pre-packed syringes of lock solution per child per year with a CVC or port. For a centre with 100 children undergoing active treatment for cancer this would be a significant undertaking (even for a large hospital pharmacy).

An additional problem is the limited data on stability of antibiotic solutions, particularly if stored in pre-filled plastic syringes. These uncertainties are further compounded when antibiotic solutions are prepared and stored in the presence of heparin, which interacts with some antibiotics. One option is not to use heparin and use saline instead, and some CVC manufacturers specify the use of heparin flushes or locks and some do not. However, a recent RCT with 203 children undergoing treatment for cancer randomised to either heparin or saline flush solutions showed a significantly higher rate of complications (CVC occlusion or bacteraemia) in the children in whom saline was used (as opposed to heparin),129 although there was no difference in the retention of CVCs between the two groups.

Concerns with respect to stability, the potential for interactions with heparin, plastics and combinations of antimicrobials, and the lack of standardisation or accepted best practice guidelines for the use of antimicrobial locks are reflected in recent literature. 130 In the small number of limited studies that have reported antimicrobial stability and activity in candidate lock solutions there is considerable variation in the concentrations, conditions and methodologies used. 131–136

In summary, uncertainties remain concerning:

• selection of antibiotic according to microbial aetiology

• interactions between antibiotics and other drugs, including heparin, and with plastics

• shelf life of pre-prepared solutions

• requirement for heparin

• antibiotic concentration

• optimum diluent

• frequency of use of locks, and dwell times.

In the light of discussions with local hospital pharmacists, we asked the 14 clinicians who had replied to the initial questionnaire and who reported the use of antimicrobial locks for prevention or treatment for details of the formulation of lock solutions used. To respond, clinicians usually referred to the hospital pharmacists for further details. We received replies from seven centres (Table 18).

The information from the seven centres that replied showed four different vancomycin preparations and four different gentamicin preparations. Teicoplanin was used instead of vancomycin in two centres (two different preparations) and amikacin was used as an alternative to gentamicin in two centres (two different preparations).

Discussion

Antimicrobial locks are used by two-thirds of the paediatric oncology units in the UK for treating suspected CVC-associated bloodstream infection, but by only one-fifth of units reported using locks for prevention. In practice, the description of prevention reflects prolonged use in patients who had been treated with antibiotic locks and does not represent prophylactic use in patients without an infection. These patterns of use contrast strongly with the available evidence. Despite this lack of evidence, clinicians are under enormous pressure to use any means that might save a ‘precious’ line and avoid the trauma of removal and CVC reinsertion. Yet interventions that delay removal of the line can increase the risk of serious complications or death owing to CVC-related infection. There is an urgent need for an RCT of lock treatment compared with standard therapy to guide practice. There is also a need to develop strategies to allow sufficient lock dwell time but minimise competition for access to CVC lumens for infusion of fluids and drug administration.

Clinicians’ reluctance to substitute antimicrobial locks for the current heparin locks in order to prevent CVC-associated infection, despite strong evidence of effectiveness, partly reflects the practical difficulties of accessing ready-mixed lock solution. Although seven units expressed concerns about costs, in practice these are small compared with the costs of admission for CVC-associated infection. Further short-term efficacy studies of antibiotic locks for prevention are not required, but there is a need to ensure ready access to lock solution for prevention and to minimise the potential for drug errors. There may still be a need for longer-term studies that focus on the potential for use of antimicrobial locks to select for antimicrobial-resistant microbes and the clinical consequences of selection of antimicrobial resistance. The evidence for efficacy of antimicrobial locks is strongest for antibiotic locks that are not commercially available in pre-filled syringes (e.g. gentamicin 10 mg/ml is available in a glass vial). The use of antimicrobial locks for prevention would require that antimicrobials for which there are efficacy data be made available in pre-filled syringes.

Rationale

The multivariable analyses indicated that the addition of bacterial DNA results did not improve the prediction of these outcomes, over and above the information provided by clinical signs and blood culture results (see Chapter 3). However, these analyses did not take into account the potential value of the earlier timing offered by DNA testing. Blood culture results are usually available 48 hours after sampling, whereas bacterial DNA results can be available within 2 hours, and, in combination with clinical signs on admission, could lead to early intervention. For example, early CVC removal in patients with a high DNA level and clinical signs of CVC infection, or early stopping of i.v. antibiotics in patients with negative results, could potentially avoid unnecessary days of admission for i.v. treatment, although the disadvantages could include unnecessary CVC removal or worsening of infection after stopping treatment.

The best way to assess the potential consequences of different test–treatment strategies is to undertake a cost-effectiveness analysis in which the various outcome states are measured in terms of quality-adjusted life-years (QALYs) and the entire health-care costs associated with each test–treatment strategy and its consequences are calculated. The cost/QALY gained from moving from one strategy to a more effective one can then be calculated and compared with costs that the health-care system is usually willing to pay (around £30,000/QALY).

We did not consider these complex analyses to be justified, given the quality of the available data. First, data on prognosis were limited in terms of sample size and duration of follow-up. Second, we lacked robust data from the systematic reviews on treatment effectiveness for removal of the CVC, antibiotic locks for treatment and early stopping of treatment.

Instead, we present simple balance sheet tabulations to illustrate the potential consequences of alternative treatment strategies for different clinical subgroups. The purpose is to show which strategies yield the greatest potential gains in order to guide further research. These illustrative analyses are based on optimistic assumptions about the effectiveness of different interventions, and take no account of the uncertainty in the parameters, which is particularly problematic given the number of zero cells (see Tables 19 and 20). The data used in the analyses are shown for outcome at 28 days in Table 19 and for outcomes at 6 months in Table 25 in Appendix 9. Our inferences are based on outcomes at 28 days because of potential selection biases and the scarcity of data in the 6-month follow-up.

Methods

Results

The distribution of outcomes according to clinical subgroup is shown in Table 19 for the cohort followed up to 28 days. The outcomes for the whole cohort, given different treatment strategies, and differences compared with standard care are summarised in the clinical balance sheet in Table 20. Tables for the 6-month follow-up are shown in Appendix 11.

The differences in outcomes between the test–treatment strategies and standard care show that the largest potential gains are associated with early stopping of i.v. treatment in low-risk children (see Table 20). These analyses are only illustrative and cannot distinguish between strategies when the differences are small. For example, we have no measure of the uncertainty around the 352 days per 1000 children admitted (see Table 20) saved by stopping after a negative DNA test compared with 2 days later after a negative blood culture result, and cannot therefore infer whether these savings would justify the additional cost of DNA testing for all admitted children. The strategy of lock treatment for any child with a positive test would result in a moderate reduction in i.v. treatment days. The benefits would vary only marginally with or without the inclusion of DNA testing.

Least benefits are to be gained from strategies involving early removal of the CVC, in terms of i.v. treatment days saved. These estimates are limited by a low event rate for recurrent i.v. treatment and CVC removal in high-risk children, and do not take into account the very high values placed by parents, children and clinicians on avoiding complications of CVC infection. There is some evidence that early CVC removal on day 1 would save more i.v. treatment days than removal on day 3, but at a cost of unnecessary CVC removal. Clinicians and patients would need to decide whether benefits in the order of 23 additional i.v. treatment days saved per additional CVC removed unnecessarily (385/17) would be an acceptable trade-off.

Conclusions

These crude analyses suggest that the largest reduction in adverse outcomes would result from a strategy of early discharge (≤ 48 hours post admission) for low-risk children. Moderate potential benefits would result from antibiotic lock treatment for all children with a positive test result, and least benefit would be derived from a strategy of early CVC removal. Bacterial DNA testing would add marginal benefits to these strategies that might not justify the costs of testing. These analyses are based on sparse data and on assumptions of treatment effectiveness that need to be evaluated in RCTs.

Main findings

The diagnostic accuracy study (see Chapter 2) showed that bacterial DNA testing had limited accuracy for predicting CVC-associated bloodstream infection. Raised DNA values (> 0.125 pg/µl) detected two-thirds (66%) of probable CVC-associated infections but were not specific (88%). These results are comparable with other diagnostic tests for CVC-associated infection. 15

In the prognostic analyses (see Chapter 3), we found that high bacterial DNA was associated with CVC removal and with an increased duration of i.v. antibiotic treatment (i.v. days). However, DNA levels did not improve the prediction of these outcomes over and above other characteristics that would be available to clinicians, such as clinical signs of CVC-associated infection and blood culture results. However, DNA was predictive of CVC removal and duration of i.v. treatment, in combination with clinical signs, when we assumed that blood culture results were not yet available. These findings suggest that DNA testing should not be added to the baseline test work-up for children with cancer who are admitted with suspected infection.

To determine which treatments would be most effective for children at different levels of risk of CVC-associated infection, we undertook a series of systematic reviews (see Chapter 5):

• We found no trials that evaluated early removal of the CVC compared with treating infection with the CVC in situ. Findings from observational studies that compared removal with retention and treatment were confounded by deferred removal in the sickest patients.

• We found only one trial and two before–after studies, which provided no clear evidence that antibiotic lock treatment reduced the risk of recurrent infection or CVC removal. However, we could not exclude a small to moderate benefit.

• We found 24 trials published since 1994 on prophylactic use of antibiotic or antiseptic locks. Overall, antimicrobial locks halved the risk of bloodstream infection in a variety of patient groups, and we found weak evidence to suggest that antibiotic locks were more effective than antiseptic locks. Contrary to the available evidence, our national survey of paediatric oncology centres found that locks are being used for treatment rather than prevention and that problems related to formulation of lock solutions currently impede a shift to prophylactic use in children.

• We found no relevant studies for slow infusion compared with bolus administration of antibiotics. However, we found one systematic review of 17 RCTs that compared continuous infusion with intermittent administration of antibiotics. 40 Clinical failure was lower, albeit with equivocal statistical significance, in patients randomised to continuous infusion,40 and the length of time with drug concentration above the minimum inhibitory concentration was higher with continuous infusion. 137

• We found no studies that compared early stopping of antibiotics with a standard duration of i.v. therapy of at least 5 days in children with cancer.

In the clinical effectiveness analysis (see Chapter 5), we made optimistic assumptions about the effectiveness of interventions to illustrate where the greatest potential benefits (measured by i.v. treatment days avoided) might be gained from changes to standard care. Most i.v. treatment days would be saved by early stopping of treatment for children at low risk of infection who had no positive baseline tests for CVC-associated bloodstream infection. We assumed that stopping treatment had no adverse effects. Further i.v. treatment days could be saved by using lock treatment for children with any positive result at baseline, assuming a 30% reduction in i.v. treatment days associated with recurrent infection. Relatively few i.v. treatment days would be saved by early removal of the CVC, and this could incur a penalty of unnecessary removal of the CVC. The analyses did not take into account the uncertainty in any of the estimates used, and were not designed to distinguish between the marginal benefits of different test–treatment strategies. There have been few studies of the economic and other costs of the various CVC-associated infection management strategies. A formal economic analysis would be required to quantify the relative effectiveness of alternative test–treatment strategies, taking into account the uncertainty of the parameter estimates and the severity of the various outcomes (see Recommendations for research).

Study limitations

The accuracy study was limited primarily by the lack of an adequate reference standard. We used criteria for a CVC-associated bloodstream infection that combined blood culture results with clinical signs of CVC-associated infection and response to treatment, based on the judgement of the clinician. These judgements may have been strongly influenced by the blood culture result, which could have biased results in favour of underestimating the accuracy of bacterial DNA testing. In addition, the type of treatment and treatment response were poorly documented. This may have led to overinclusion of patients in the probable and possible categories of CVC-associated infection, thereby underestimating the sensitivity of DNA testing.

A second problem arose because the reference standard was not related to the type of treatment required. We endeavoured to address this issue in the clinical effectiveness analysis (see Chapter 5), by using clinical opinion to determine which treatments would be offered to patients with different combinations of clinical signs and test results. A third issue is that 79 patients were counted more than once (there were 260 admissions and CVCs for 181 children). Such clustering of patients would mean that the reported CIs overestimate the certainty of the estimates of sensitivity and specificity. Accuracy also changed slightly owing to clustering of patients with repeated admissions when we restricted analyses to the first admission only. Strengths of the study were the clearly defined and clinically relevant criteria for inclusion of patients in the study and prospective collection of data on clinical signs, blood culture, bacterial DNA and clinician opinion, in all patients.

The prognostic study provided information on the risks of outcomes related to CVC-associated infection, such as recurrent bloodstream infection requiring i.v. antibiotics and CVC removal, according to clinical signs and test results at admission. These analyses allowed us to determine the effectiveness of bacterial DNA testing over and above other test results that would be available to the clinician. However, there were several limitations. First, when we restricted analyses to one infection event per child, there were few outcome events (e.g. only 10 children had their CVC removed in the 28 days after admission with fever). Using the outcome total i.v. treatment days improved the power to detect an effect. We sought further follow-up data to 6 months post admission, but these data were complete for only 99 of the 181 children in the cohort. Second, the clinically important outcomes, CVC removal and i.v. treatment days, were affected by local practice and are partly conditional on the clinical signs and blood culture results at admission. More objective measures of signs and symptoms reflecting resolution and recurrence of infection would be ideal, but are complex to record and analyse. Parents contributing to the CCLG have highlighted the need for research into repeated measures of patient-reported quality of life in order to capture subtle morbidity associated with persistent CVC infection and its management.

A third limitation was that treatments given in response to baseline characteristics altered prognostic outcomes. We assumed that the cohort represented ‘standard care’, as no centres routinely used targeted treatment for CVC infection, and only 24 of 260 infection episodes had any record of such a targeted intervention (see Table 2, Chapter 2). However, we did not attempt to analyse details of the type, intensity and changes in treatment that would have been given to patients who failed to respond to the initial treatment. As a result, the prognostic analyses probably underestimate the discrimination of baseline characteristics for patients who could benefit from interventions for CVC-associated infection. This bias increases with length of follow-up (from 28 days to 6 months). Fourth, our focus was on baseline test characteristics that can be used to inform early decisions about treatment. In practice, clinicians also use repeated observations to determine changes in signs, symptoms or test results, in response to treatment. These may be more accurate than baseline characteristics, especially for patients at high risk of complications who may benefit from CVC removal. Repeated measures were not recorded in our data and are more complex to analyse. Despite these limitations, these prognostic analyses provide useful information on outcomes in a setting where targeted interventions for CVC-associated infection were rarely used.

A fifth limitation of this study was the requirement to limit blood sampling and additional manipulations in this vulnerable population. A larger blood volume could potentially give improved test sensitivity (through extraction of bacterial DNA from a larger volume of blood). A series of blood samples from each CVC lumen would have allowed us to investigate the hypothesis that microbial colonisation is not homogenous along the intra luminal pathway.

The main limitation of the series of systematic reviews was the poor quality of included studies and lack of RCTs. An exception was the review of antimicrobial locks for preventing CVC-associated infection, which included 24 RCTs. Using the incidence rate ratio as the effect measure, we were unable to use standard metaregression techniques to explore sources of heterogeneity between these trials, but instead presented results for subgroups. A key question for future studies is to quantify the effectiveness of antibiotic compared with antiseptic lock solution for preventing CVC-associated infection.

Implications for practice

The bacterial DNA test reported in this study is recommended for diagnosing CVC-associated infection by the most recent guidelines from the Infectious Disease Society of America. 13 This recommendation is based on a small accuracy study of patients receiving total parenteral nutrition. 28 Our findings do not support routine use of DNA testing in children with cancer who are admitted from the community. However, we recommend that repeated DNA testing be evaluated (as a marker of microbial load and response to treatment) to identify children who might benefit from CVC removal. This approach is analogous to the use of viral load to assist in the assessment of treatment efficacy in patients with cytomegalovirus or human immunodeficiency virus (HIV) infection.

We found strong evidence to support prophylactic use of locks for prevention of CVC-associated infection in children with cancer, and weak evidence to suggest that antibiotic locks are more effective than antiseptic locks. We found no RCTs that evaluated the effectiveness of ethanol locks. Further research is urgently needed to address issues related to formulation and administration of antibiotic locks to facilitate implementation.

As reported by others, our study highlighted variation in the management of children with cancer and fever who are admitted from home. 29 In view of the lack of evidence to support the treatment options considered in this study, there is no evidence to recommend approaches to management other than the use of antibiotic locks for prophylaxis.

Recommendations for research

We list the research priorities in order of the number of patients likely to benefit from changes in practice:

1. The option of stopping i.v. treatment early compared with the standard duration of i.v. treatment (5–7 days) requires evaluation for children at low risk of systemic infection, for example those with no signs or test results indicative of CVC-associated infection (approximately 77% of children with cancer and a CVC who are admitted with fever; see Table 20, Chapter 5). Bacterial DNA testing would only marginally improve the identification of this low-risk group and may not justify the costs of testing. Further work is required to define the comparator arms. For example, should ‘early stopping’ mean no i.v. treatment at all and no admission, or discharge after 48 hours of i.v. antibiotics once blood culture results are known? What additional monitoring should be included for the early discharge arm to ensure readmission if symptoms persist? Similarly, what type, dosage and duration of i.v. treatment should define the standard treatment arm, before and after blood culture results are known? The fever and neutropenia study group of the CCLG is currently seeking funding to evaluate strategies for early discharge. 138

2. We have concluded that there is strong evidence in favour of the use of antibiotic locks for the prevention of CVC-associated bloodstream infection. On the other hand, the systematic review found relatively few studies in children with cancer (four studies). 85–87,100 None of the trials involved children undergoing care in a UK cancer centre. A large proportion of the trials of use of antimicrobial locks for prevention have involved renal patients:

   1. There are significant differences in treatment and preventive practices both between specialty groups and across national boundaries, so it can be argued that a UK trial of the use of antimicrobial locks is still required to provide evidence relevant to UK paediatric oncology practice. This type of trial should include a cost-effectiveness analysis. It should also include methods that would detect the emergence of antibiotic resistance and drug toxicities associated with long-term exposure to antimicrobial lock solutions.

   2. In addition, laboratory studies are required to determine the optimal formulation (e.g. concentration and diluent) of lock solutions for home use and storage conditions.

   3. When locks are expected to be used for longer than 18 months, for example in patients with total parenteral nutrition, surveillance studies are needed to evaluate the emergence of antimicrobial resistance.

   4. There is a need for head-to-head clinical trials to determine the optimum type of antibiotic, lock dwell time and frequency of administration. Important potential adverse effects include the impact of antibiotic lock prophylaxis on organism selection and resistance and on antibiotic use for treating symptomatic infections.

3. Antibiotic lock treatment should be evaluated for children with cancer who are admitted with one or more positive signs of CVC-associated infection (e.g. FRC, positive blood culture, or, if available, raised levels of bacterial DNA). Our analyses (see Chapter 6) suggest that this group comprises 23% of children admitted with fever and suspected CVC-associated infection. A placebo-controlled randomised trial is required to determine the effectiveness and cost-effectiveness of early i.v. antibiotic lock therapy compared with deferred antibiotic lock treatment in children with cancer. Follow-up should be continued for at least 3 months to detect late recurrence of bloodstream infection. 36,139 The study should determine the effect of lock treatment on clinical outcome measures reflecting signs and symptoms of persisting or recurrent infection and on patient-reported measures of quality of life. Reliance on microbiological outcomes (i.e. blood culture), or CVC removal, which is strongly influenced by blood culture results, may not adequately distinguish the effect of lock treatment on morbidity.

4. Prior to a trial of locks for treatment, preliminary research is required to define the type of antibiotics and the appropriate formulation, according to blood culture results. There is also a need to determine how long the lock solution should dwell in the CVC before the lumen is aspirated and used for other treatments. Currently dwell time is very variable. The recent Infectious Diseases Society of America guidelines recommend that antibiotic locks be left in each of the venous access device lumens for a minimum of 1 hour in any 24-hour period and a maximum of 24 hours. 13

5. There is controversy about the benefits of early CVC removal compared with in situ treatment of CVC infection in children with proven and persistent CVC-associated bloodstream infection. Paediatric oncologists in the CCLG reference group reported that CVC removal is not routine for any children with CVC-associated infection. On the other hand, recent international guidelines recommend CVC removal if CVC-associated infection is due to S. aureus, P. aeruginosa, fungi or mycobacteria. 13 An observational study by Raad et al. 139 showed that, in adults with cancer and CVC-associated infection due to coagulase-negative staphylococci, the rate of recurrent infection was 6.6 times higher with continued treatment with the CVC in situ than with CVC removal. Further uncertainties remain with respect to a number of other CVC-associated infections (both Gram-positive and -negative). 42,47,53,140 An RCT is needed to compare early CVC removal with treatment in situ of specific CVC-associated bloodstream infections, but should be confined to children with repeatedly positive tests for CVC-associated bloodstream infection. A diagnostic accuracy study could be integrated into this trial to determine whether repeated DNA testing improves the prediction of children likely to benefit from early CVC removal. This trial would involve around 5% of children admitted to hospital with cancer and fever (estimate based on 10/179 patients requiring CVC removal within 28 days), but would also be suitable for a much larger number of hospitalised children with suspected CVC infection. An additional element to a prospective study could be to compare bacterial DNA results in febrile and non-febrile children. This might allow determination of the natural history of DNA levels in children with and without CVC-associated infection. However, this type of study would raise ethical issues about sampling in children without a clinical indication.

6. We do not recommend an RCT involving DNA testing as a single test on admission for children with cancer admitted from the community with fever. DNA testing may add more information for inpatients on i.v. antibiotics, in whom blood cultures are unreliable. In addition, DNA testing is a developing area, and it is likely that test performance will improve. There is also a need for further evaluation of sampling strategies, for example to determine the optimal blood volume required and whether the discard sample can be used.

7. Variation in practice between centres could be defined more clearly, systematised, and then examined in observational studies to provide information on the effectiveness of alternative practices. 29,138 Specification of management protocols, and linkage of routine data on individual patient admissions and blood culture results over time are now feasible and should be considered as a potentially efficient approach to evaluating the impact of variation in practice.

Contribution of authors

Mike Millar conceived, designed and co-ordinated the overall study, and co-ordinated the accuracy study. He is also guarantor for the report. He developed the design for the synthesis of evidence together with Ruth Gilbert. Enid Hennessy carried out the statistical analyses for the accuracy study. Weiwei Zhou carried out all statistical analyses in this report and wrote the methods sections. Mark Wilks, Roderick Skinner and Barry Pizer contributed to the overall design of the study and Mark Wilks contributed to the laboratory analyses. Ruth Gilbert designed and carried out the synthesis of evidence following on from the accuracy study and wrote the report together with Mike Millar. All authors contributed to the drafting of the report.

Disclaimers

The views expressed in this publication are those of the authors and not necessarily those of the HTA programme or the Department of Health.

Synopsis

A complete copy is available from the CCLG website (www.cclg.org.uk) or from the principal investigator, Mike Millar.

The diagnosis or exclusion of a central venous catheter (CVC)-associated infection may carry considerable importance in the management of a child with cancer or leukaemia but is frequently difficult. Improvement in diagnostic techniques may allow genuine CVC-associated infections to be treated earlier and more effectively, and may reduce the rate of unnecessary CVC removal in patients who do not have CVC-associated infection.

There are two parts to this study and it is expected to run for a total of 3 years and 6 months. Patients will be recruited and informed consent obtained separately for each part of the study.

The aim of Part One is to determine the optimum threshold values for a molecular test for the diagnosis or exclusion of CVC-associated infection in children and adolescents (0–18 years inclusive) undergoing treatment at a collaborating UKCCSG centre. Part One simply requires observation of all febrile episodes in recruited patients, with blood samples being taken on one occasion at the time of fever from all CVC lumens for the quantitative 16S rDNA test method (1 ml EDTA anticoagulated) as well as for the standard culture method. Simple clinical information will be collected concerning the febrile episode and its treatment and outcome. Patients will be followed for 4 weeks from its presentation.

It is anticipated that any UKCCSG centre will be able to enter patients into Part One, even though some may not take part subsequently in Part Two (see below). The optimum threshold values will be derived from the results of Part One before patient recruitment commences for Part Two of the study. Part One requires 1000 febrile episodes and should be completed in the first year of the study. It is anticipated that there will be a hold on the study in between Part One and Part Two while the data from Part One are analysed. This may be up to 2 months.

The aim of Part Two of the study is to test the hypothesis that a test for CVC-associated infection based on quantitative bacterial 16S rDNA analysis will improve the management of suspected CVC-associated infections in patients being treated for cancer. For logistical reasons, Part Two may be performed on a limited centre basis. Patients (0–18 years inclusive) will be eligible whether or not they have previously participated in Part One, and patients will be randomised to availability of the 16S rDNA test (Arm A) or not (Arm B). All other aspects of management, including investigation by standard culture techniques, will remain unaltered in both randomisation groups. The 16S rDNA result should be made available to the doctor with responsibility for patient care within 48 hours, and will incorporate the likelihood ratios (both negative and positive) of probable CVC-associated infection for a known test value. A repeat sample(s) can be tested during the same episode if the clinician considers that it/they will help patient management. Simple clinical information will be collected, as in Part One. The primary outcome measure is CVC survival; secondary outcome measures include duration of antibiotic treatment and hospitalisation for fever, mortality and an economic analysis. Part Two requires randomisation of 330 patients and should be completed within 24 months. This allows 18 months for patient recruitment and a minimum of 6 months’ follow-up after each episode.

Search for papers on prognosis

An initial search was carried out for papers reporting prognosis for subsequent infection, infection complications or death in children with CVC-associated infection.

We searched MEDLINE using PubMed and the following terms. The search was repeated for EMBASE (see Figure 5, Chapter 4).

Our search included synonyms for text words in all fields and MeSH terms for: central venous, intravascular, port, or indwelling AND catheter or device AND removal.

Cochrane Central Register of Controlled Trials

We searched CENTRAL for any RCTs that included text words or MeSH terms relating to CVC and infection for patients of any age.

(central ven* cathe*):ti,ab,kw OR MeSH descriptor Catheterization, Central Venous

AND

(infec*):ti,ab,kw OR MeSH descriptor Infection

Early central venous catheter removal versus treatment in situ to reduce infection complications

We considered any potentially eligible studies found from scanning the abstracts generated by the searches above. In addition, we conducted specific searches of PubMed for comparative studies of CVC removal versus treatment in situ.

The search terms used were: terms for CVC, removal included as a text word, date limit for articles published from 1995 onwards.

Antimicrobial locks for treatment or prevention (any age group)

We considered any potentially eligible studies found from scanning the abstracts generated by the searches above. In addition, we conducted specific searches of PubMed for reviews, meta-analyses or RCTs of antimicrobial locks.

Slow infusion versus bolus injection of antibiotics

This search considered any potentially eligible studies found from scanning the abstracts generated by the searches above. In addition, we conducted specific searches using terms for CVC AND RCTs (as above) AND synonyms for infusion, parenteral, intravenous or bolus.

Antimicrobial locks used for the treatment or prevention of central venous catheter-associated infections in children with cancer

Questionnaire objective To identify concerns with respect to the use of antimicrobial locks for the treatment or prevention of CVC-associated infections in children undergoing treatment for cancer.

An antimicrobial lock is defined as the ‘locking’ of a solution containing an antimicrobial substance into the lumen of a central vascular access device in order to treat or prevent infection associated with that device. These antimicrobial locks can be used for variable periods but usually in excess of 2 hours. Antimicrobial substances that have been reported for prevention or treatment can be any one or more of a wide range of chemicals (including chelating agents, alcohols and acids), antibiotics, enzymes and disinfectants. Antimicrobial locks used for prevention are almost always both antimicrobial and anticoagulant. An example of antimicrobial locks for prevention is the use of vancomycin with heparin. In this case the vancomycin/heparin solution is used as a direct replacement for heparin and may be left in the line for variable durations depending upon the requirements for line usage.

Nine systematic reviews have shown that lock solutions prevent CVC infection (see, for example, Yahav et al. 106 and Safdar and Maki77). Most of the RCTs were of adult patients undergoing haemodialysis through a CVC: the pooled relative risk of CVC-associated infection found in the meta-analyses is 0.3–0.6 in patients in whom antimicrobial locks were used as a routine locking solution compared with patients given placebo or heparin locks. To understand what this means, imagine that the relative risk is 0.5. This means that the frequency of infection would be halved, for example from a rate of 10/1000 CVC days with heparin to 5/1000 CVC days with a lock solution.

There are limited published data on the use of locks for treatment, but a major potential benefit is prevention of recurrent infection (see Rijnders et al. 36). In the only RCT that has examined treatment of CVC infection using antibiotic locking solutions (heparin with vancomycin or ceftazidine), probably because of the small sample size the study found no significant difference (HR 0.55, p = 0.10). 36 It might be that further studies will show that locks can be used to treat CVC-associated infections that respond poorly to current treatment regimes such as those caused by fungi, S. aureus and resistant Gram-negative bacteria, but there are insufficient data to answer this question at the moment.

In summary, there is evidence for the effectiveness of locks for prevention of CVC infection, particularly in patients undergoing haemodialysis.

We are trying to find out whether locks are being used in paediatric oncology practice and what are the concerns and perceived disadvantages, including feasibility issues, disadvantages, contraindications and potential costs.

Could you please answer the following questions in as much detail as possible.

Questions

Health Technology Assessment programme

1. Director, NIHR HTA programme, Professor of Clinical Pharmacology, University of Liverpool

2. Professor of Dermato-Epidemiology, Centre of Evidence-Based Dermatology, University of Nottingham