Perioperative care bundles for the prevention of surgical-site infections: meta-analysis

Abstract

Background

Care bundles are used widely to prevent surgical-site infections (SSIs). Recent systematic reviews suggested larger effects from bundles with more interventions. These reviews were largely based on uncontrolled before–after studies and did not consider their biases. The aim of this meta-analysis was to determine the effectiveness of care bundles to prevent SSIs and explore characteristics of effective care bundles.

Methods

A systematic review, reanalysis, and meta-analysis of available evidence were undertaken. RCTs, controlled before–after studies, and uncontrolled before–after studies with sufficient data for reanalysis as interrupted time series studies (ITS) were eligible. Studies investigating the use of a care bundle, with at least one intraoperative intervention, compared with standard care were included.

Results

Four RCTs, 1 controlled before–after study, and 13 ITS were included. Pooled data from RCTs were heterogeneous. Meta-analysis of ITS resulted in a level change of −1.16 (95 per cent c.i.−1.78 to −0.53), indicating a reduction in SSI. The effect was larger when the care bundle comprised a higher proportion of evidence-based interventions. Meta-regression analyses did not show statistically significant associations between effect estimates and number of interventions, number of evidence-based interventions, or proportion of evidence-based interventions.

Conclusion

Meta-analysis of ITS indicated that perioperative care bundles prevent SSI. This effect is inconsistent across RCTs. Larger bundles were not associated with a larger effect, but the effect may be larger if the care bundle contains a high proportion of evidence-based interventions. No strong evidence for characteristics of effective care bundles was identified.

This meta-analysis reports on the effectiveness of care bundles for the prevention of surgical-site infections stratified by study type. Data from RCTs were heterogeneous and data from interrupted time series showed a preventive effect. Meta-regression did not show significant associations between preventive effect and number of (evidence-based) interventions.

Introduction

Surgical-site infections (SSIs) increase morbidity, mortality, and costs^1–4. The WHO⁵, Centers for Disease Control and Prevention (CDC)⁶, and National Institute for Health and Care Excellence (NICE)⁷ have published guidelines on interventions for the prevention of SSI that are applicable to all surgical subspecialties. Interventions are often combined into care bundles to improve compliance and enforce their potential cumulative effect^8,9. After success with bundles for the prevention of ventilator-associated pneumonia¹⁰ and catheter-related bloodstream infections¹¹, the Institute for Healthcare Improvement (IHI)⁸ defined a care bundle as a straightforward set of three to five evidence-based practices that, when performed collectively, improve patient outcomes. The aim of care bundles is to prevent a specific complication such as SSI or anastomotic leakage, and thereby differs from that of general safety procedures such as the WHO Surgical Safety Checklist or the SURPASS checklist, which is to improve patient outcomes in a much broader sense^12,13.

Previous meta-analyses^14–17 found a preventative effect of care bundles for colorectal and obstetric surgery on SSI. In addition, a meta-regression analysis^18,19 showed a larger preventative effect for bundles with a larger number of interventions. However, the conclusions were largely based on uncontrolled before–after studies. These studies are prone to biases such as secular trends and may overestimate treatment effects²⁰; their inclusion in meta-analyses is discouraged²⁰. However, when sufficient data points are available, uncontrolled before–after studies can be reanalysed as interrupted time series studies (ITS). This allows control for secular trends and more reliably estimates treatment effects. Published meta-analyses so far have not applied this method. The true effect and optimal design of care bundles remains unclear.

A systematic review and reanalysis was performed according to the Cochrane Effective Practice and Organization of Care (EPOC) guidelines²¹, with the aim of evaluating the effects of implementation of a care bundle on the incidence of SSI among patients undergoing any type of surgery. There are no arguments for substantial biological or pathophysiological variation between surgical subspecialties that warrant subgroup analysis according to type of surgery. In addition, proposed bundle characteristics, such as the total number of interventions, number of evidence-based interventions, and proportion of evidence-based bundles relative to total number of interventions, were analysed^18,19.

Methods

The PRISMA²² statement was followed, and the protocol was registered in the international prospective register of systematic reviews (PROSPERO) (CRD42019126578).

Eligibility criteria

Studies evaluating surgical patients of all ages were eligible (Population). The studies had to compare the effects of a care bundle (Intervention) versus standard care (Comparator) on the incidence of SSI (Outcome). The care bundles had to be a set of interventions designed to prevent SSI, structurally applied in a perioperative care setting, and had to include at least one intervention in the intraoperative phase (no postoperative enhanced recovery after surgery protocols). The complete bundle had to be implemented at a fixed point in time. Studies that made other changes to standard care after implementation of the bundle were included, but analysed until the change in standard care.

(Non-)randomized trials, (un)controlled before–after studies, ITS, and repeated–measures studies were eligible. Uncontrolled before–after studies were included only if the data permitted reanalysis as ITS. As a minimum requirement, at least three data points before and after intervention were required, as recommended by the Cochrane EPOC group^23,24. Methods of data collection had to be consistent throughout the study. Databases were searched without restriction on language or publication date.

Search strategy

On 6 June 2020, a clinical librarian searched MEDLINE (PubMed), Excerpta Medica Database (Embase), Cumulative Index to Nursing and Allied Health Literature (CINAHL), and the Cochrane Central Register of Controlled Trials (CENTRAL). Search terms and details are provided in Appendix S1.

Outcome measures

The primary outcome was incidence of SSI as defined by the authors in the original articles. Secondary outcomes included interventions in the care bundle, compliance with the bundle, adverse events, and bundle characteristics, including the number of interventions and number of evidence-based interventions. Interventions were deemed evidence-based if these had been suggested for consideration or recommended by WHO global, CDC or NICE guidelines for the prevention of SSI^5–7. If an intervention was supported by a systematic review more recent than the guidelines, or if it concerned an intervention outside the scope of the guidelines, it was also considered evidence-based. A systematic review was considered supportive if it reported a meta-analysis with a statistically significant pooled risk reduction (95 per cent confidence interval below 1 in favour of the intervention). Multiple interventions aimed at optimizing the same preventative measure were counted as a single intervention (for example, correct dosing and timing of surgical antibiotic prophylaxis).

Study selection and data extraction

Two authors independently screened titles and abstracts, and subsequently assessed full-text articles. Disagreements were resolved by discussion with a third author. Data were extracted independently by two authors using a standardized form. If required, SSI rates were retrieved from graphs using PlotDigitizer^® (J.A. Huwaldt, USA, http://plotdigitizer.sourceforge.net/). If incomplete data were reported, the authors were contacted for complementary data. Articles were excluded if the authors could not provide the relevant data.

Risk of bias

The Cochrane risk-of-bias tool for randomized trials (RoB 2) was used for randomized trials²⁵. The Risk Of Bias In Non-randomised Studies—of Interventions (ROBINS-I) tool was used for non-randomized studies²⁶. The ROBINS-I tool evaluates confounding bias based on three aspects of confounding: overall potential for confounding, baseline confounding, and time-varying confounding. Selection bias was assessed based on the baseline variables collected after the start of the study and whether the start of follow-up and intervention coincided²⁶. Publication bias was assessed using a funnel plot²⁷.

Statistical analysis

Uncontrolled before–after studies were reanalysed as ITS. For this, different methods are available and appropriate²⁸. Following the Cochrane EPOC guidelines, segmented regression analysis with autoregressive integrated moving average models were used²⁴. To avoid inaccuracies, adjustment for autocorrelation was performed, using a Lag-1 autocorrelation model, if more than 10 observations were reported²⁹. To account for differences in scale (such as infections per 100 admissions or per 1000 patient-days), effect estimates were standardized by dividing the results by the standard deviation of the preintervention slope.

Figure 1 shows a graphical representation of interrupted time series analysis. A change in level between the interval before and after the intervention indicates a direct effect of the intervention on SSI risk. A change in slope between the interval before and after the intervention indicates whether the intervention changed the underlying secular trend in infections and may consequently reduce or increase SSI risk over time. This underlying trend is influenced by all factors such as other changes in perioperative care and is not limited to the intervention alone. The finding that the intervention demonstrates a direct effect, without a subsequent upward change in underlying trend, strengthens the idea that the intervention causes a lasting change.

Fig. 1.

Graphical representation of interrupted time series

This figure represents the interrupted time series analysis before and intervention. Line A to B represents the incidence of surgical-site infection (SSI) before intervention, and line C to D the incidence of SSI after intervention. The slope of these lines is estimated based on the data. The lines are extrapolated to the point of intervention. This change in level at the point of intervention is considered the direct change attributable to the intervention. The slope of the lines before and after intervention represents the underlying secular trend in SSI incidence. This trend is sensitive to changes in all factors (including other changes in perioperative care) and not limited the intervention only. The change in slope (s) represents the change in underlying trend. A change in level suggests that the change in SSI incidence is occurring irrespective of factors other than the implementation of the intervention.

Meta-analyses were stratified by study design, using a random-effects model. For ITS, pooled standardized effect sizes and corresponding 95 per cent confidence intervals were estimated. Potential statistical heterogeneity was expressed using the I² statistic. An I² value exceeding 75 per cent indicated considerable heterogeneity³⁰.

If 10 or more studies were included in the meta-analyses, a predefined meta-regression analysis and subgroup analysis was performed²¹. Meta-regression was used to explore the association between the effect size and the total number of interventions, the number of evidence-based interventions, and the proportion of evidence-based interventions with respect to the total number of interventions in a bundle. An illustrative subgroup analysis comparing the proportion of evidence-based interventions categorized per 25 per cent was undertaken. Treatment effects were calculated using SPSS^® version 26 (IBM, Armonk, NY, USA) and statistical analysis was undertaken in Stata^® version 15.1 (StataCorp, College Station, TX, USA).

Results

Study selection

Some 4446 titles and abstracts were screened and 334 full texts were reviewed. Eighteen studies^31–48 were included. The selection process, including reasons for exclusion, is described in Fig. S1 (Appendix S2).

Study characteristics

Study characteristics are summarized in Table 1. Four RCTs^31–34, 1 CBA study³⁵, and 13 uncontrolled before–after studies (reanalysed as ITS)^36–48 were included. Colorectal^{31,33,34,37,39,43–45,47}, general^32,33, gynaecological^33,38,42, orthopaedic^35,48, paediatric^41,44,46, and (cardio)vascular^36,40,41 surgical procedures were included. Studies were published between 2011 and 2020.

Table 1.

Study characteristics

Reference	Study design*	Study Setting	Type of surgery	CDC wound class	SSI definition†	Compliance reported	Adverse events reported	Follow-up
Anthony et al.31	RCT	USA, single centre	Colorectal surgery	II/III	CDC	Yes	No	30 days
Beldi et al.32	RCT	Switzerland, single centre	General surgery	II/III	CDC	No	No	30 days
Kwaan et al.33	RCT	USA, single centre	Colorectal, gynaecological, urology, general surgery	II	CDC	No	Yes	30 days
Ruiz- et al.34	RCT	Spain, single centre	Colorectal surgery	II/III	CDC	No	No	30 days
Calderwood et al.35	CBA	USA, multicentre	Hip and knee arthroplasty	I	CDC	No	No	90 days
Al Salmi et al.36	ITS	Saudi Arabia, single centre	Cardiovascular surgery	I	CDC	No	No	90 days
Cima et al.37	ITS	USA, single centre	Colorectal surgery	II/III	ASC NSQIP	No	No	30 days
Davidson et al.38	ITS	USA, single centre	Caesarean section	II	CDC	Yes	No	30 days
Dean et al.39	ITS	UK, single centre	Colorectal surgery	II/III	Public Health England SSI surveillance service‡	Yes	No	30 days
Dua et al.40	ITS	USA, multicentre	Vascular surgery	I	No formal definition given	No	No	n.s.
Hodge et al.41	ITS	USA, single centre	Paediatric cardiothoracic surgery	I	CDC	Yes	No	n.s.
Johnson et al.42	ITS	USA, single centre	Gynaecological surgery	II/III	CDC	No	No	30 days
Lutfiyya et al.43	ITS	USA, single centre	Colorectal surgery	II/III	ASC NSCIP	No	No	30 days
Nordin et al.44	ITS	USA, single centre	Paediatric gastrointestinal surgery	II/III	CDC	Yes	No	30 days
Rumberger et al.45	ITS	USA, single centre	Colorectal surgery	II/III	NHSN‡	No	No	30 days
Toltzis et al.46	ITS	USA, multicentre	Paediatric surgery	I	CDC/NHSN	No	No	n.s.
Weiser et al.47	ITS	USA, single centre	Colorectal surgery and combined colorectal and liver resections	II/III	CDC	No	No	30 days
Yamada et al.48	ITS	Japan, single centre	Spinal surgery	I	CDC	No	No	1 year

^*All interrupted time series studies (ITS) were originally uncontrolled before–after studies but reanalysed as ITS.

†Resources: Centers for Disease Control and Prevention (CDC)⁴⁹, American College of Surgeons National Surgical Quality Improvement Program (ASC NSCIP)⁵⁰, Public Health England surgical-site infection (SSI) surveillance service⁵¹, CDC/National Health Safety Network (NHSN)⁵³

‡Only superficial SSI are described. CBA, controlled before–after; n.s., not stated.

Care bundles

There was wide variation in both the number and types of interventions in the care bundles. The total number of interventions ranged from 3³⁴ to 17³⁷. The number of evidence-based interventions ranged from 0^32,33,44 to 10⁴³. The median proportion of evidence-based interventions among bundle interventions was 50 (range 0–100) per cent. One RCT withheld some evidence-based interventions from the intervention group while applying them in the control group. The most frequently reported interventions were surgical-site preparation with alcohol-based chlorhexidine gluconate; selection, timing, and redosing of surgical antibiotic prophylaxis; hair removal with clippers; and use of a separate closing tray with surgical instruments for wound closure. Tables 2–4 list the bundle components and estimated (standardized) effect size of each care bundle. Details of the bundles and the assessment as evidence-based are described in Appendix S3.

Table 2.

Included RCTs: outcomes and care bundles

Reference	Composition of care bundle	No. of interventions	No. of EB interventions	IHI criteria	SSI rate*		P
					Control group	Intervention group
Anthony et al.31	Preoperative: omission of mechanical bowel preparation and oral antibiotics Intraoperative: fluid restriction, wound protector† Perioperative: normothermia†, supplementary oxygen (FiO2 80%)†	5	3	No	24 of 97 (24)	45 of 100 (45.0)	0.003
Beldi et al.32	Perioperative: caps cover ear and neck Intraoperative: double gloves, glove change, iodine-impregnated incision drape, replacement of surgical instruments, abdominal lavage with Ringer’s solution, drape change before closure, irrigation of surgical site with Ringer’s solution	8	0	No	64 of 469 (13.6)	74 of 492 (15.0)	0.581
Kwaan et al.33	Intraoperative, before closure: clean up surgical field, glove change, (optional gown change), drape change, closing tray	4	0	No	15 of 121 (12.4)	13 of 112 (11.6)	0.830
Ruiz-Tovar et al.34	Intraoperative: abdominal lavage with antibiotic solution, fascial closure with triclosan-coated sutures†, skin staples with mupirocin ointment	3	1	No	20 of 99 (20)	2 of 99 (2)	0.007

^*Values are n (%).

†These interventions are considered evidence-based (EB) in recent WHO, Centers for Disease Control and Prevention, or National Institute for Health and Care Excellence guidelines^5–7. IHI, Institute for Healthcare Improvement; SSI, surgical-site infection; FiO₂, fraction inspired oxygen.

Table 4.

Included observational cohort studies: outcomes and care bundles

Reference	Composition of care bundle	No. of interventions	No. of EB interventions	IHI criteria	Standardized level change	Standardized slope change
Al Salmi et al.36	Preoperative: MRSA screening and decolonization†, supervised CHG showering† Postoperative: patient education Perioperative: glucose control†	4	3	Yes	−4.00 (−7.91, −0.10)	−0.50 (−2.74, 1.74)
Cima et al.37	Preoperative: CHG cloths, CHG shower†, patient education Intraoperative: SAP following SCIP†,*, SAP redosing†, CHG skin preparation†, closing tray, glove change Postoperative: good hand hygiene†, hand hygiene signage, hand-cleansing agent available, dressing removal within 48 h, CHG shower after dressing removal, hand wipes for patients, dismissal with CHG, patient education, follow-up telephone calls	17	5	No	−0.692 (−1.69, 0.31)	0.07 (−0.06, 0.20)
Davidson et al.38	Preoperative: patient education (including smoking cessation)†, CHG shower†, CHG surgical-site preparation† Intraoperative: weight-based SAP, SAP redosing†, CHG skin preparation, iodine vaginal preparation, hair removal with clippers†, double gloves, closing tray, glove change Postoperative: euglycaemia for diabetic patients†, dressing removal POD 2, patient education, follow-up telephone call	14	6	No	−1.57 (−3.29, 0.16)	0.09 (−0.29, 0.47)
Dean et al.39	Intraoperative: CHG skin preparation†, wound protector†, SAP redosing†, triclosan-coated suture†	4	4	Yes	−1.25 (−3.18,0.69)	0.88 (0.02, 1.73)
Dua et al.40	Intraoperative: SAP following SCIP†,*, hair removal with clippers† Postoperative: euglycaemia for cardiac patients†, urinary catheter POD 1 Perioperative: normothermia†, patients on beta-blocker receive beta-blocker, appropriate venous thromboembolism prophylaxis	7	4	No	0.08 (−1.35, 1.50)	−1.33 (−1.89, −0.78)
Hodge et al.41	Preoperative: MRSA screening and decolonization† Intraoperative: CHG skin preparation†, cefazolin administration cardiopulmonary bypass prime	3	2	No	−0.43 (−1.25, 0.40)	0.07 (−0.03, 0.17)
Johnson et al.42	Preoperative: patient education, CHG shower†, CHG cloths Intraoperative: SAP following SCIP†,*, SAP redosing†, CHG skin preparation†, closing tray, glove change (optional gown change) Postoperative: good hand hygiene†, hand-cleansing agent available, dressing removal POD 2, CHG shower after dressing removal, patient education, discharge with CHG, follow-up telephone call	15	5	No	−0.24 (−1.20, 0.73)	−0.06 (−0.17, 0.05)
Lutfiyya et al.43	Preoperative: patient education, smoking cessation†, CHG cloths, MBP and oral non-absorbable antibiotics†, diabetic screening (HbA1c), hair removal with clippers† Intraoperative: SAP following SCIP†,*, weight-based SAP, SAP redosing†, CHG skin preparation†, double gloves, wound irrigation† Postoperative: silver dressings POD 5 Perioperative: normothermia†, perioperative blood glucose control†, supplementary oxygen (FiO2, 80%) until 4 h after surgery†	16	10	No	−0.85 (−1.83, 0.14)	−0.01 (−0.09, 0.08)
Nordin et al.44	Intraoperative: glove change, redrape surgical field, closing tray	3	0	No	−1.25 (−3.01, 0.60)	0.14 (−0.16, 0.44)
Rumberger et al.45	Preoperative: hair removal with clippers† Intraoperative: SAP timing†, CHG skin preparation†, normothermia†, local bupivacaine anaesthesia Postoperative: ambulate evening of surgery, regular diet, removal urinary catheters POD 1, minimal narcotic regimen Perioperative: glucose control†, supplementary oxygen (intraoperative FiO2 100%, 6 h postoperative FiO2 80%)†	11	5	No	−0.11 (−2.28, 2.06)	−0.32 (0.67, 0.03)
Toltzis et al.46	Preoperative: hair removal with clippers† Intraoperative: SAP timing†, CHG skin preparation†	3	3	Yes	−1.64 (−1.25, −0.04)	-0.14 (-0.42, 0.14)
Weiser et al.47	Preoperative: screening HbA1c and consultation if raised, MBP and oral non-absorbable antibiotics†, CHG shower†, preoperative risk assessment using MSK SSI prediction tool Intraoperative: SAP following SCIP†,*, SAP redosing†, hair removal with clippers†, closing tray, normothermia (measured in postoperative unit only)† Postoperative: shower POD 2	10	6	No	−4.79 (−6.90, −2.68)	−0.33 (−1.00, 0.34)
Yamada et al.48	Preoperative: SAP with additional vancomycin† Intraoperative: wound irrigation† Perioperative: nasal mupirocin and CHG body decontamination (5 days)	3	2	No	−3.93 (−6.79, −1.08)	0.99 (−0.69, 2.67)

Values in parentheses are 95 per cent confidence intervals.

*Surgical Care Improvement Project (SCIP) compliance: correct antibiotics; administration 60 min before incision; postoperative discontinuation within 24 h. IHI, Institute for Healthcare Improvement; MRSA, methicillin-resistant Staphylococcu aureus.

^†These interventions are considered evidence-based (EB) in recent WHO, Centers for Disease Control and Prevention, or National Institute for Health and Care Excellence guidelines^5–7. ‡CHG, chlorhexidine gluconate; SAP, surgical antibiotic prophylaxis; POD, postoperative day; MBP, mechanical bowel preparation; FiO₂, fraction inspired oxygen; MSK, Memorial Sloan Kettering Cancer Center; SSI, surigcal-site infection.

Table 3.

Included controlled before–after study: outcomes and care bundle

Reference		No. of interventions	No. of EB interventions	IHI criteria	Surgery type	OR*	P
Calderwood et al.35	Preoperative: MRSA screening and decolonization†, CHG shower/bath† Intraoperative: SAP timing†, alcohol-based skin preparation†, hair removal with clippers†	5	5	Yes	Knee surgery	0.88 (0.78, 0.99)	0.04
					Hip surgery	0.85 (0.75, 0.96)	0.01

Values in parentheses are 95 per cent confidence intervals.

^*Risk-adjusted OR.

†These interventions are considered evidence-based (EB) in recent WHO, Centers for Disease Control and Prevention, or National Institute for Health and Care Excellence guidelines^5–7. IHI, Institute for Healthcare Improvement; MRSA, methicillin-resistant Staphylococcus aureus; CHG, chlorhexidine gluconate; SAP, surgical antibiotic prophylaxis.

Effect of care bundles and meta-analyses

Owing to the large variation in results, and in particular the inconsistency in direction of effect, it was felt inappropriate to perform a meta-analysis of these studies and pool their results into a single weighted average. One RCT³⁴ found a lower SSI risk, 1 RCT³¹ an increased SSI risk, and 2^32,33 reported no effect.

One CBA study³⁵ demonstrated a favourable effect for the intervention in knee and hip surgery, with adjusted ORs of 0.88 (95 per cent c.i. 0.78 to 0.99) and 0.85 (95 per cent c.i. 0.75 to 0.96) respectively.

Table 4 describes the results of the uncontrolled before–after studies. Of the 13 studies, 12^{36–39,41–48} originally reported a significant decrease in SSI incidence after bundle implementation, and 1⁴⁰ reported an increase in SSI incidence. After reanalysis, only 4^36,46–48 of the 12 studies still showed a statistically significant level change after bundle implementation, indicating a direct effect of the care bundle. There was no significant change in slope, indicating that there were no changes in the underlying secular trend, and there was no further gradual effect following bundle implementation^36,46–48. Hence, the effect was most likely attributed to the intervention. The remaining 9 studies^37–45 did not demonstrate a significant change in level. A significant change in slope was found in 2 studies^39,40. One study³⁹ demonstrated a positive change in slope, indicating an increase in underlying trend, and the other⁴⁰ a negative change, indicating the opposite. Meta-analysis of the ITS demonstrated a pooled effect estimate of the level change of −1.16 (95 per cent c.i. −1.78 to −0.53) (Fig. 2), indicating a significant decrease in SSI incidence after bundle implementation. The pooled slope change of −0.04 (95 per cent c.i. −0.16 to 0.07) indicated no change in underlying SSI trend. Additional forest plots of slope change of ITS are presented in Fig. S2 (Appendix S4).

Fig. 2.

Forest plot of level change of interrupted time series

Standardized effect sizes are shown with 95 per cent confidence intervals.

Meta-regression

There were insufficient data for meta-regression and subgroup analyses among the RCTs and the CBA study. Meta-regression of ITS showed no evidence of an association between the estimated effect and the total number of interventions (level coefficient per intervention 0.05, 95 per cent c.i. −0.10 to 0.20; slope coefficient per intervention 0.00, 95 per cent c.i. −0.04 to 0.04), the number of evidence-based interventions (level coefficient per additional evidence-based intervention 0.16, −0.36 to 0.40; slope coefficient per evidence-based intervention −0.01, −0.19 to 0.16), or proportion of evidence-based interventions (level coefficient −0.12 (−0.44 to 0.21) and slope coefficient −0.01 (−0.1 to 0.1) per 10 per cent more evidence-based interventions). Plots of the meta-regression are shown in Figs S3–S8 (Appendix S4).

Subgroup analysis

The effect estimate of level change was −1.71 (95 per cent c.i. −2.89 to −0.53; I² = 0 per cent) (slope coefficient 0.18, 95 per cent c.i. −0.62 to 0.98; I² = 60.5 per cent) for the subgroup of bundles that consisted of at least 75 per cent evidence-based interventions, −1.61 (−3.01 to –0.21; I² = 0 per cent) (slope coefficient −0.18, −0.43 to 0.09; I² = 85.0 per cent) for bundles that consisted of 50–75 per cent evidence-based interventions, and −0.64 (−1.22 to −0.05; I² =70.4 per cent) (slope coefficient −0.01, −0.12 to 0.11; I² = 38.0 per cent) for bundles with less than 50 per cent evidence-based interventions The third and fourth categories were merged because of the limited number of studies in each category.

Compliance and adverse events

Six studies^{31,38,39,41,44,47} reported details of compliance with the individual interventions of the bundle. Davidson et al.³⁸ reported a compliance rate of more than 90 per cent for all interventions except for a postdischarge telephone call (more than 53 per cent). Other studies^31,39,47 reported compliance with the complete bundle. Two studies^41,44 did not report specific compliance rates but used these for implementation purposes. The other studies refrained from reporting detailed compliance rates. Only one study³³ mentioned potential adverse events attributable to the intervention.

Risk of bias

Three RCTs^31,33,34 had some concerns about bias due to absence of blinding. One RCT³² had a high risk of bias. The CBA study³⁵ was scored as having a moderate risk of bias. Almost all ITS were judged as having a moderate risk of bias. One ITS³⁶ was judged to have a serious risk of bias. Risk-of-bias evaluation is summarized in Fig. 2 and in more detail in Appendix S5.

The funnel plots were plotted by study type (Figs S9 and S10, Appendix S4). For ITS, there was some publication bias favouring effective studies. Publication bias was observed, but post hoc trim-and-fill analysis indicated that this did not affect the results of the meta-analysis. There were too few RCTs available for assessment of publication bias.

Discussion

This systematic review and meta-regression evaluated the effect of a perioperative care bundle on the incidence of SSI. There was no consistent benefit across study types. The results of RCTs were heterogeneous, but meta-analysis of ITS did demonstrate a reduction in SSI after implementation of perioperative care bundles. There was some evidence of publication bias, but trim-and-fill analysis indicated that this did not affect the result of the meta-analysis. Meta-regression analyses indicated no association between the number of interventions or the number of evidence-based interventions and the effect estimate. There was a larger effect estimate when the proportion of evidence-based interventions among the bundle interventions increased, but this was inseparable from chance.

Previous systematic reviews^14–17,19 found a large benefit for care bundles in patients undergoing colorectal surgery and caesarean delivery, and suggested that large bundles, including 11 interventions and more, are more effective than smaller bundles, regardless of their composition. This contrasts with earlier design principles to keep care bundles small, simple and evidence-based⁸, and creates vulnerability to the introduction of disputed, ineffective, or even harmful interventions that may impair compliance and outcomes⁵². The conclusions of these systematic reviews are based on crude analyses of unadjusted before–after studies. This practice ignores secular trends and risks overestimating the effect⁵⁴. The present review followed Cochrane EPOC methodology, and strictly included evidence from RCTs, CBA studies, ITS, and uncontrolled before–after studies with sufficient data for proper reanalysis as ITS to account for these biases. Only 4^37,43,45,47 of the 37 cohort studies included in the latest published systematic review¹⁷ on SSI prevention bundles allowed proper reanalysis (as ITS) to be included in the present review. For the other 33 before–after studies¹⁷, it was impossible to reliably attribute causative effect to the interventions. Of the 12 before–after studies^{36–39,41–48} that originally reported statistically significant improvement after bundle implementation, only 4^36,46–48 showed a sustained significant improvement after adjusting for secular trends by reanalysis as ITS. This suggests that the results of some plain before–after studies should be interpreted with caution. Furthermore, the present findings contrast with the suggestion that more interventions translate into more effectiveness. No increase in effectiveness was found when more interventions were combined.

The effect was larger when a bundle comprised a higher proportion of evidence-based interventions. Although this finding was inseparable from chance, this bundle characteristic showed the largest association with SSI risk reduction. Bundles with a higher proportion of evidence-based interventions typically had fewer interventions in general. This supports the notion that small evidence-based bundles may be most efficient and corresponds with recommendations by the IHI that care bundles should consist of 3–5 evidence-based interventions.

Data from RCTs were inconsistent. One RCT³¹ noted an increase in SSI in the intervention group, in contrast to the result of the meta-analysis of ITS. This bundle included a potentially harmful intervention (restrictive fluid regimen⁵⁵) and omission of a potential beneficial intervention that was applied in the control group (mechanical bowel preparation and oral antibiotics). It may be the choice of interventions rather than the bundling of interventions that led to this negative effect and consequent heterogeneity in the meta-analysis. If so, this would also explain the contradiction with the aggregated ITS results.

In contrast to previous systematic reviews, this study was not limited to a particular type of surgery and as such this is an important extension to current literature. The biological rationale of effectiveness in SSI risk for most interventions, for example surgical antibiotic prophylaxis, is not limited to a single surgical subspecialty but rather addresses generic SSI risks that affect all types of surgery. Differences in a priori SSI risk may result in different absolute effects between surgical subspecialties, but the relative effectiveness is generally comparable⁵⁶. To splinter the evidence between surgical subspecialties without a clear biological rationale unnecessarily undermines statistical power.

There are several limitations of this study to consider. The number of available RCTs and CBA studies was small, rendering ITS important for the present findings. There was some evidence of publication bias, but post hoc trim-and-fill analysis indicated that this was irrelevant. The interpretation of standardized effect sizes for level and slope is challenging. These measures are not easily translated into a direct SSI risk reduction. However, with studies measuring SSI incidence on different scales, this is the most accurate procedure to adjust for underlying trends, and pool and compare the available data.

With only 13 studies available in the largest meta-analysis, and an uneven distribution of studies across co-variates, the statistical power of the meta-regression analyses was limited, risking type II error²¹. The analyses should therefore be considered explorative and interpreted with caution.

The present analysis investigated the effect of care bundles on SSI incidence. It was not possible to assess the attributive effect of one, two and three interventions, as the definition of a bundle required a minimum of three interventions. Similarly, it was not possible to study the effectiveness of single interventions or the attributable effect of an intervention within a certain bundle. This limits the ability to guide design of future perioperative care bundles at the intervention level. Data at the individual-participant level with information on compliance of each intervention would aid such an analysis.

Finally, most studies lacked reports on bundle compliance, challenging the interpretation of the effects. Compliance is crucial for the effect of any intervention. A recent systematic review⁵⁷ investigated multimodal strategies for implementation of clinical interventions to prevent SSI. This is beyond the scope of the present systematic review, but is important context to put the clinical effects into perspective.

This review highlights the potential preventative effect of care bundles specifically aimed at reducing SSIs. Care bundles to reduce SSI should be implemented alongside a general surgical safety checklist aimed at preventing all types of perioperative complication, such as the WHO surgical safety checklist or SURPASS checklist^12,13.

The data did not allow analysis of the effectiveness of individual interventions as elements of care bundles. Therefore, one cannot advise on specific interventions or types of intervention. Interventions are based on the principle of improving hygiene (such as preoperative showering or chlorhexidine gluconate alcoholic skin preparation) or the principle of improving patient physiology (for example, perioperative warming, normoglycaemia or hyperoxygenation). Current guidelines^5–7 report evidence for interventions based on either of these two principles.

The present study suggests that bundles comprising a higher proportion of evidence-based interventions may be more beneficial for patients. Furthermore, there was no evidence of a larger effect from bundles consisting of a larger number of interventions. The IHI⁸ recommends the use of small evidence-based bundles, as smaller bundles are easier to implement. Future care bundles should focus on evidence-based interventions and research on these bundles should include measurement of compliance. Existing care bundles may be revised to exclude non-evidence-based interventions, and more attention should be given to adequate implementation of the evidence-based components of the bundle.

Funding

The authors have no funding to declare.

Supplementary material

Supplementary material is available at BJS online.

Data Availability

All data are published either in this manuscript, the included articles or the supplementary material.