Types of patients

The five trials involved a total of 2277 participants, with the totals in individual trials ranging from 32 to 1653. One study involved children (Benhamou 2002), and the remaining four included adults only (Engervall 1995; Rasero 2000; Timsit 2009; Vokurka 2009). Two studies were set in a bone marrow transplant unit (Benhamou 2002; Rasero 2000), two studies consisted of patients undergoing treatment for a haematological malignancy (Engervall 1995; Vokurka 2009), and one study recruited patients receiving treatment in intensive care (Timsit 2009). Four countries were represented (two studies from France and one each from Italy, Sweden, and the Czech Republic). All studies were conducted in acute in‐patient settings. Patients were excluded from the studies if their skin was already damaged (Benhamou 2002; Rasero 2000; Vokurka 2009); were having treatment that would make them more susceptible to skin damage, such as the chemotherapeutic drug busulphan‐thiotepa (Benhamou 2002), or radiation to the chest (Vokurka 2009); or if they had allergies to polyurethane dressings (Rasero 2000; Timsit 2009; Vokurka 2009), chlorhexidine (Timsit 2009), or disinfectant (Vokurka 2009).

Types of interventions

Time frames for dressing changes varied between 2 and15 days. One study planned to compare 15‐day and 4‐day dressing changes for tunnelled catheters (Benhamou 2002). CVAD dressings were monitored on a daily basis in all trials and patients were followed up until the CVAD was removed or until discharge as a minimum. However, in this study there were a large number of protocol violations, that is, dressings were changed on days other than the day indicated in the protocol. In the 15‐day group, only 67 (17%) of the 365 dressing were changed on day 15 and, in the 4‐day group, 516 (76%) of the 678 dressings were changed on the correct day. This meant that dressing changes in the 15‐day group were actually changed, on average, every eight days and, in the 4‐day group, every four days. Reasons for the protocol violations included soiled and dislodged dressings and problems with the catheter that required the dressing to be removed. Two studies compared once versus twice‐weekly dressing changes for tunnelled CVADs (Engervall 1995; Timsit 2009), two studies compared once versus twice‐weekly dressing changes for non‐tunnelled CVADs (Timsit 2009; Vokurka 2009), and one study compared 5‐day versus 10‐day dressing changes for tunnelled CVADs and 2‐day versus 5‐day dressing changes for non‐tunnelled CVADs (Rasero 2000). Again however, reflecting the reality of pragmatic research in clinical settings, many of the dressings were not changed according to the group schedule. In the tunnelled CVAD 10‐day group 9.6% were not changed on the correct day, while in the tunnelled CVAD 5‐day group the proportion was 8.0%; in addition 6.8% of non‐tunnelled CVCs in the 5‐day group were not changed as scheduled and in the non‐tunnelled CVC 2‐day group the rate was 12.5%.

The dressings were applied under controlled conditions in all groups. Three studies stated that nurses were responsible for the dressing changes (Benhamou 2002; Engervall 1995; Rasero 2000). Four studies used Tegaderm (3M, St Paul, USA) dressings (Benhamou 2002; Engervall 1995; Rasero 2000; Timsit 2009), and one study used Bioclusive (Johnson and Johnson, New Jersey, USA; Vokurka 2009). One study used chlorhexidine gluconate‐impregnated sponges (Biopatch, Ethicon, New Jersey, USA) around the entry or exit site of the CVAD under the dressings (Timsit 2009).

Three studies used the same dressings throughout the period of observation. Two studies used different dressings that depended upon skin damage. One study used a Tegaderm (3M) covering a sterile gauze for grade 0 to 1 skin damage (48/56; 85% in the 15‐day group and 32/56; 57% in the 4‐day group), sterile gauze with Mefix for grade 2 to 3 skin damage (7/56; 13% in the 15‐day group and 23/56; 41% in the 4‐day group) and sterile gauze with tape for grade 4 skin damage (1/56; 2% in both the 15‐ and 4‐day groups; Benhamou 2002). The other study used a Tegaderm (3M) for undamaged skin or an exit site with mild erythema, but if the exit site had extensive erythema or other signs of local infection then the dressings were changed daily using a gauze dressing moistened with 10% ethanol with aluminium acetotartrate 10% until the erythema had disappeared, at which point the patient was returned to the allocated group (Engervall 1995). Patients in the once‐weekly group had more extra dressings due to erythema compared to the twice‐weekly group (3%; 0 to 91% once‐weekly group; 0%; 0 to 17% twice‐weekly group; P value 0.08 expressed as extra dressings days per CVAD days).

Skin decontamination varied between the groups. Two studies used the same antiseptic solution to clean the skin before the insertion of the CVAD and at dressing changes; one study used an alcohol‐based povidone‐iodine solution (Timsit 2009); and one used povidone‐iodine solution; whether the antiseptic was alcohol‐based was unclear (Vokurka 2009). One study used a 0.5% alcohol based chlorhexidine solution during insertion and at dressing changes, but changed to aqueous‐based povidone‐iodine if the skin became damaged (Benhamou 2002). One study did not describe skin decontamination that occurred prior to CVAD insertion but used 70% ethanol at dressing changes (Engervall 1995). One study did not mention which antimicrobial solution was used for skin decontamination (Rasero 2000).

Types of outcomes

Only one trial used a standard definition for confirmed CRBSI (Timsit 2009), two trials reported blood culture results (Benhamou 2002; Vokurka 2009), and one study reported blood culture and CVAD‐tip culture results separately (Engervall 1995). Blood cultures were performed on clinical suspicion of infection or determined by a temperature threshold stipulated by each study author. Three studies provided information about suspected CRBSI but these studies used different definitions (Benhamou 2002; Engervall 1995; Timsit 2009): Benhamou 2002 did not provide a definition; Engervall 1995 defined suspected CRBSI as not responding to antibiotics; and Timsit 2009 had an investigator blinded to the study group review the patient's case including the medical chart in order to perform an independent blinded review. In all five studies, catheter‐site infection was defined by skin colonisation and additionally in three studies by local signs of catheter‐site infection such as the presence of inflammation, erythema, tenderness, swelling or discharge (Benhamou 2002; Engervall 1995; Timsit 2009). Two studies measured pain: Benhamou 2002 used categories of none, moderate or severe; and Vokurka 2009 used a visual analogue score ranging from 0 to 10 (0: no pain, 5: moderate pain, 10: severe pain). No study measured quality of life. Two studies measured cost (Rasero 2000; Timsit 2009).

In the Timsit 2009 trial, a 2 x 2 factorial design was used, in which participants were randomised to a 3‐ or 7‐day dressing change and to a dressing alone or with chlorhexidine gluconate‐impregnated sponge (CHGIS). They also combined arterial and central catheters in their analysis. The authors were contacted and provided information based on central catheters only and reported separately for the CHGIS and non‐CHGIS groups. For our analysis we have included only the non‐CHGIS group, to maintain consistency with other trials.

Random sequence generation

Three studies used computer‐generated lists to generate the allocation sequence (Benhamou 2002; Timsit 2009; Vokurka 2009). One study used manually mixed envelopes (Engervall 1995). One study did not describe how the random sequence was generated (Rasero 2000).

Allocation concealment

Vokurka 2009 used computer software to conceal the allocation of trial patients into individual groups. We contacted the trialists of Engervall 1995 who stated that they had used randomisation envelopes. The other three studies did not describe how the allocation was concealed (Benhamou 2002; Rasero 2000; Timsit 2009).

Blinding of participants and personnel

None of the studies was able to blind participants or staff involved in direct care from identifying the allocated intervention due to the nature of the intervention.

Blinding of outcome assessment

Timsit 2009 mentioned blinding of outcome assessment, stating that staff involved in analysing catheter cultures and reviewing CRBSI were blinded to the study groups. It was unclear in the remaining four trials whether outcome assessors were blinded (Benhamou 2002; Engervall 1995; Rasero 2000; Vokurka 2009).

Confirmed catheter‐related bloodstream infection (995 central venous catheters)

Only one study (Timsit 2009), that had uncertain risk of bias for allocation concealment, reported confirmed CRBSI as per our protocol. There was no clear evidence of a difference between groups for this outcome (RR 1.42; 95% CI 0.40 to 4.98; Analysis 1.1). LOW QUALITY EVIDENCE (downgraded for risk of bias and imprecision): (summary of findings Table for the main comparison).

Suspected catheter‐related bloodstream infection (151 participants)

We were able to extract data from two studies that reported suspected CRBSI (Benhamou 2002; Engervall 1995). Benhamou 2002 stated that no CVADs were removed due to suspicion of CRBSI. In the Engervall 1995 trial 6/20 (30%) of CVADs were removed in the once‐weekly group and 4/19 (21%) in the twice‐weekly group due to suspected CRBSI (RR 0.70; 95% CI 0.23 to 2.10; Analysis 1.2). Both studies were at uncertain risk of bias for allocation concealment, blinding of outcome assessment and selective reporting. There was no clear evidence of a difference between the groups for this outcome. LOW QUALITY EVIDENCE (downgraded for risk of bias and imprecision).

All‐cause mortality (896 participants)

Three studies at uncertain risk of bias, included information about all‐cause mortality (Benhamou 2002; Engervall 1995; Timsit 2009). It was possible to combine the data from all these studies; the studies were homogenous so the fixed‐effect model was used for data synthesis (I² = 0%). There was no clear difference in all‐cause mortality between longer (5‐15 days) and shorter (2‐5 days) time intervals between dressing changes (RR 1.06; 95% CI 0.90 to 1.25; Analysis 1.3). LOW QUALITY EVIDENCE (downgraded for risk of bias and imprecision): (summary of findings Table for the main comparison).

Catheter‐site infection (371 participants)

All five studies reported catheter‐site infection but in a variety of different ways. Benhamou 2002 and Rasero 2000 reported the proportions of participants developing a catheter‐site infection. Engervall 1995 reported the rate of exit site infections per 100 CVAD days; Vokurka 2009 reported positive skin swabs and Timsit 2009 reported rates of skin colonisation, Data from the two studies (Benhamou 2002; Rasero 2000) that reported risk of catheter‐site infection in a similar way were pooled using a fixed effect model (I² = 0%). There was no clear evidence of a difference in the risk of catheter‐site infection rate between longer (5‐15 days) and shorter (2‐5 days) time intervals between dressing changes (RR 1.07; 95% CI 0.71 to 1.63; Analysis 1.4). LOW QUALITY EVIDENCE (downgraded for risk of bias and imprecision).

Engervall 1995 reported 1.6 exit site infections per 100 CVAD days (median, range 0 to 13.3) in the longer interval (less frequent) group compared with 0 per 100 CVAD days (median, range 0 to 9.1) in the short interval (more frequent) group. Vokurka 2009 reported 13 positive skin swabs across both treatment groups but did not report by group. We contacted the trialists of Timsit 2009 but they were unable to provide per patient data. The Timsit 2009 study reported catheter‐site infection rates per catheter rather than by patient in their published paper.

Consequently it remains unclear whether longer or shorter intervals between dressing changes for CVADs influences the risk of catheter‐site infection.

Skin damage (1587 participants)

Skin damage was reported in four studies (Benhamou 2002; Rasero 2000; Timsit 2009; Vokurka 2009). Data from two trials were included in the forest plot (Benhamou 2002; Vokurka 2009). Results were highly heterogenous (I² = 78%), probably due to different scales being used to assess skin damage and dissimilar time frames for assessment, so we did not pool the data. One of these trials (Benhamou 2002) included only children and showed that fewer participants in the longer interval group (8/56) scored grade ≥ 2 on the skin damage scale compared with 24/56 in the shorter interval group (RR 0.33; 95% CI 0.16 to 0.68; P value 0.012; Analysis 1.5). There was no clear evidence of a difference in rates of skin damage between long and short intervals in adult patients (Vokurka 2009) (RR 1.26; 95% CI 0.46 to 3.41; Analysis 1.5. VERY LOW QUALITY EVIDENCE (downgraded for risk of bias, imprecision and heterogeneity).

Two trials could not be included in the skin damage forest plot. In the Rasero 2000 trial toxicity was graded on a 5‐point scale, but there were no reported differences between groups. We are unable to use the data from the Timsit 2009 trial due to the 2 x 2 factorial design.

Pain

Pain was assessed on a daily basis in two studies (Benhamou 2002; Vokurka 2009). The maximum intensity of pain reported was analysed. When data from the two studies were combined there was no clear evidence of a difference in pain however this comparison is underpowered (RR 0.80; 95% CI 0.46 to 1.38; Analysis 1.6; Benhamou 2002; Vokurka 2009). This was rated as LOW QUALITY EVIDENCE (downgraded for risk of bias and imprecision): (summary of findings Table for the main comparison). The pain classification systems used are detailed in the Characteristics of included studies. The pain data were dichotomised on the basis of a judgement that any pain experienced and reported by the patient was clinically significant.

Quality of life

None of the studies reported quality of life.

Cost

Rasero 2000 reported the costs of nursing time and dressings and stated that less frequent dressing changes would reduce costs by 400% in the tunnelled CVAD group and by 50% in the non‐tunnelled CVAD group when compared to the standard practice of changing dressings every second day. The monetary figures presented in the text and the table were different. Several attempts have been made to contact the authors for clarification but without success.