Interventions

We included a total of 24 studies, which contain three studies with two intervention arms (Asai 2002 active; Asai 2002 passive; Prunet 2008 active; Prunet 2008 passive; Chambers 2015 hospitals; Chambers 2015 long‐term nursing care) and one study with three intervention arms (L'Ecuyer 1996 2wva; L'Ecuyer 1996 mbc; L'Ecuyer 1996 pbc), corresponding to 29 different comparisons of safety medical devices that we named as different studies to increase transparency of the meta‐analyses. We elaborated on the details of the devices in Table 1. Based on the information in the articles, we checked on the Internet if the devices were still for sale and if they still resembled the original description given in the article. Even though we could not be sure that the devices currently sold were exactly similar to those in the articles, we are confident that the main safety features are still the same.

The types of devices used in the various studies were:

• safe blood collection devices (n = 3) (Baskin 2014; Goldwater 1989; Rogues 2004);

• safe IV systems (n = 9) (Asai 1999 active; Asai 2002 active; Asai 2002 passive; Azar‐Cavanagh 2007; Cote 2003; L'Ecuyer 1996 2wva; L'Ecuyer 1996 mbc; L'Ecuyer 1996 pbc; Mendelson 1998; Prunet 2008 active; Prunet 2008 passive; Seiberlich 2016; Sossai 2010);

• safe injection device (n = 4) (Gaballah 2012; Goris 2015; van der Molen 2011; Zakrzewska 2001);

• multiple safety devices interventions (n = 5) (Chambers 2015 hospitals; Chambers 2015 long‐term nursing care; Phillips 2013; Reddy 2001; Valls 2007; Whitby 2008); and

• safe needle disposal boxes (n = 3) (Edmond 1988; Grimmond 2010; Richard 2001).

Safety engineered devices can be divided into two broad categories, passive and active devices. Passive devices have a safety function that is automatically activated without the user's interference. This type of safety device is supposed to offer better protection because the human factor is excluded. Active devices require one‐ or two‐handed activation by a health worker after use.

Four studies used a similar type of safe active IV system (Autoguard IV) (Asai 1999 active; Asai 2002 active; Cote 2003; Prunet 2008 active). The safety mechanism of this device is activated by pushing a button which retracts the needle. Two studies evaluated a passive and an active system (Asai 2002 active; Asai 2002 passive; Prunet 2008 active; Prunet 2008 passive). In addition to the Autoguard IV, Asai 2002 passive and Prunet 2008 passive used a passive device. Asai 2002 passive used the Protective Acuvance, which consists of two needles (one inside the other) where the tip of the needle is automatically changed to a blunt needle upon withdrawing. Prunet 2008 passive used the Introcan safety, which automatically shields the needle tip upon withdrawing. The Introcan safety IV system was also used by Sossai 2010. Whereas Seiberlich 2016 used a safe active IV system (ViaValve), which consisted of a valve to prevent blood flow back out of the catheter hub on initial venipuncture.

A needleless system refers to a device that does not use needles for the collection of body fluids or administration of medication or fluid after initial IV access is established (Mendelson 1998). L'Ecuyer 1996 2wva; L'Ecuyer 1996 mbc; L'Ecuyer 1996 pbc used three needleless IV systems. One, the safety needleless IV tubing system (blunt metal cannula), was replaced after four months by a blunt plastic cannula due to dissatisfaction of employees with the device. Mendelson 1998 evaluated a needleless IV system which is incompatible with a needle. All other studies had employed either a combination of the needleless system and insertion or evaluated the effects of safe insertion only.

In the five studies involving multiple safety devices, one study included safety‐engineered needles and needleless devices that were either passive or semi‐automatic (Chambers 2015 hospitals; Chambers 2015 long‐term nursing care). The study by Phillips 2013 used safety‐engineered sharps. Reddy 2001 used safety syringes and needleless IV systems. Valls 2007 used safety vacuum phlebotomy systems, blood‐gas syringes with a needle sheath, lancets with retractable single‐use puncture sticks, safe IV catheters (passive and active), and safe injection devices. Whitby 2008 used multiple passive safety‐engineered devices including retractable syringes, needle‐free IV systems and safety winged butterfly needles.

In the studies on safe disposal boxes, Edmond 1988 evaluated a bedside needle disposal; Grimmond 2010 assessed a sharps container with enhanced safety features such as automatic lock‐out when full; and Richard 2001 introduced small containers in all patient areas combined with an educational program.

In studies focusing on safe blood collection, Rogues 2004 introduced two devices: re‐sheathable winged steel needles and Vacutainer blood‐collecting tubes with recapping sheaths. Goldwater 1989 used a shield on the needle cap to prevent the needle from injuring the worker. Baskin 2014 used a safety‐engineered blood gas syringe in which the cannula protection shield is activated with one hand after puncture and clicks irreversibly over the cannula.

Representing safe injection devices, Gaballah 2012 used safety dental syringes that did not require re‐sheating or removal of the needle from its syringe. Goris 2015 used passive subcutaneous retractable syringes that automatically and instantly retract the needle from the patient into the barrel of the syringe. van der Molen 2011 evaluated an injection needle with a safety feature shielding the needle after the injection, and Zakrzewska 2001 assessed one type of safety syringe for dentistry. The injection devices had an active safety mechanism that had to be activated by the workers.

A total of 17 studies reported introducing the safety devices together with training sessions (Azar‐Cavanagh 2007; Baskin 2014; Edmond 1988; Gaballah 2012; Goldwater 1989; Goris 2015; L'Ecuyer 1996 mbc; L'Ecuyer 1996 pbc; L'Ecuyer 1996 2wva; Mendelson 1998; Prunet 2008 active; Prunet 2008 passive; Richard 2001; Rogues 2004; Seiberlich 2016; Sossai 2010; Valls 2007; van der Molen 2011; Whitby 2008; Zakrzewska 2001). Goldwater 1989 briefly stated that staff completed an educational program. Two studies did not report on the integration of training or education as part of the study (Grimmond 2010; Reddy 2001).

Types of study designs

Study designs used to assess the effect of the intervention were:

• six RCTs (Asai 1999 active; Asai 2002 active; Asai 2002 passive; Baskin 2014; Cote 2003; Prunet 2008 active; Prunet 2008 passive; Seiberlich 2016);

• two cluster‐RCTs (L'Ecuyer 1996 2wva; L'Ecuyer 1996 mbc; L'Ecuyer 1996 pbc; van der Molen 2011);

• five CBAs (Gaballah 2012; Grimmond 2010; Mendelson 1998; Valls 2007; Zakrzewska 2001); and

• eleven ITS (Azar‐Cavanagh 2007; Chambers 2015 hospitals; Chambers 2015 long‐term nursing care; Edmond 1988; Goldwater 1989; Goris 2015; Phillips 2013; Reddy 2001; Richard 2001; Rogues 2004; Sossai 2010; Whitby 2008).

Participants

There were slight differences across studies in terms of selected participants for the study. In nine studies, researchers referred to the broad term of healthcare personnel or hospital workers as participants (Chambers 2015 hospitals; Chambers 2015 long‐term nursing care; Edmond 1988; Goris 2015; Grimmond 2010; Phillips 2013; Richard 2001; Rogues 2004; Sossai 2010; van der Molen 2011). Reddy 2001 included health personnel with the exception of physicians. Three studies included healthcare workers explicitly at risk of blood borne pathogen exposure from contaminated needles, referred to as house staff, physicians, medical students, nurses, nursing assistants, emergency medical technicians and environmental service workers (Azar‐Cavanagh 2007; Mendelson 1998; Whitby 2008). Three studies included nursing personnel only as participants (L'Ecuyer 1996 2wva; Seiberlich 2016; Valls 2007;). Two studies included anaesthesiologists (Cote 2003; Prunet 2008 active; Prunet 2008 passive). In two studies researchers and assistants were the persons handling the needles (Asai 1999 active; Asai 1999 active; Asai 2002 active). Dental clinic staff were the target group in one study (Zakrzewska 2001). One study included dental and nursing students (Gaballah 2012). One study included emergency department doctors (Baskin 2014). Another study included only laboratory staff (Goldwater 1989)

In one RCT the number of participants were 50 each in the intervention and control groups (Asai 1999 active; Asai 2002 active; Asai 2002 passive). In another RCT there were 254 and 251 participants in each of the intervention groups and 254 participants in the control group (Prunet 2008 active; Prunet 2008 passive). There were 119 participants in the control group and 211 in the intervention group in (Cote 2003) and 275 in each group in (Baskin 2014). In (Seiberlich 2016) there were 79 in the control group and 73 in the intervention group.

In the cluster‐RCTs, van der Molen 2011 reported on eight wards in each of the two intervention groups and the control group, representing approximately 265 workers in each of the these three groups during the initial phase. The authors adjusted for the cluster effect by means of a GEE‐analysis. L'Ecuyer 1996 2wva reported 19,436 patient‐days for the plastic two‐way valves, 3840 patient‐days for the metal blunt cannula (L'Ecuyer 1996 mbc) and 15,737 patient‐days for the plastic blunt needle (L'Ecuyer 1996 pbc). However, the study did not mention the number of wards that were randomised.

In the CBA studies, Grimmond 2010 recruited 14 hospitals in both the control and the intervention groups, approximating overall 19,880 full‐time equivalents (FTE) during the two‐year study period. Valls 2007 recruited seven wards for the intervention group and five wards for the control group from a hospital with 1000 workers. Zakrzewska 2001 had approximately 300 workers in both the intervention and control groups. Mendelson 1998 reported on eight medical units in both the intervention and control groups, corresponding to approximately 220 workers per group. Gaballah 2012 recruited three hospitals ‐ one for the control group and two for the intervention group. However, the authors did not report data relating to the number of participants.

In the ITS studies, Azar‐Cavanagh 2007 reported on 11,161 healthcare workers for the pre‐intervention period (18 months) and 12,851 healthcare workers for the post‐intervention period (18 months). Reddy 2001 reported on 3011 FTE for the pre‐intervention period (three years) and 3992 FTE for the post‐intervention period (three years). Rogues 2004 reported on 8500 FTE (2000 nurses) per year for the pre‐intervention period (four years) and post‐intervention period (three years). Edmond 1988 followed 278 nurses for the pre‐intervention period (eight months) but provided no information to determine if this number remained the same for the intervention period (four months). Richard 2001 did not report the number of participants in the one participating hospital during the seven‐year study period. Goldwater 1989 reported 127,000 venipunctures for the pre‐intervention period (six months), and 483,000 venipunctures with the device and 232,348 without the device during the intervention period (33 months). Sossai 2010 reported that the number of employees at the hospital fluctuated between 4447 and 4636 throughout the study period (two years pre‐intervention and three years post‐intervention). Chambers 2015 hospitals reported on an average of 325 000 FTE per year and included nine data points. Chambers 2015 long‐term nursing care also reported on an average of 325000 FTE per year and included nine data points. Goris 2015 reported on 857 895 employee productive hours for the pre‐intervention period and 237 202 employee productive hours for the post‐intervention period. Phillips 2013 reported on 184 years of cumulative data collected from 85 hospitals in the pre‐intervention period (six years) and 150 years of cumulative data collected from 85 hospitals in the post‐intervention period (five years). Whitby 2008 reported on 3053 FTE for the pre‐intervention period (12 months) and 6506 FTE for the post‐intervention period (24 months).

The average number of data points in the eleven ITS studies was 13.8 and ranged from six to 39.

Outcomes

Twenty‐one studies included self‐reported percutaneous injuries as their main outcome (Asai 1999 active; Asai 2002 active; Asai 2002 passive; Azar‐Cavanagh 2007; Chambers 2015 hospitals; Chambers 2015 long‐term nursing care; Cote 2003; Edmond 1988; Gaballah 2012; Goldwater 1989; Goris 2015; Grimmond 2010; L'Ecuyer 1996 2wva; L'Ecuyer 1996 mbc; L'Ecuyer 1996 pbc; Mendelson 1998; Phillips 2013; Reddy 2001; Richard 2001; Rogues 2004; Sossai 2010; Valls 2007; van der Molen 2011; Whitby 2008; Zakrzewska 2001). Seiberlich 2016 reported on incidence of blood leakage and blood exposure risk reduction. In two studies (Baskin 2014; Prunet 2008 active; Prunet 2008 passive) the main outcomes were both blood splashes and NSIs. In three studies researchers reported only blood splashes (Asai 1999 active; Asai 2002 passive; Cote 2003; Prunet 2008 active; Prunet 2008 passive). Three studies did not report NSIs as their main outcome as no injury was reported during the study (Asai 1999 active; Asai 2002 passive; Prunet 2008 active; Prunet 2008 passive). Cote 2003 reported that the study was underpowered to assess the difference in needlestick injuries between the groups.

The denominators for the self‐reported NSIs included: the number of procedures (Baskin 2014; Goldwater 1989; Rogues 2004), medical devices (Prunet 2008 active; Prunet 2008 passive; Sossai 2010), FTE (Chambers 2015 hospitals; Chambers 2015 long‐term nursing care; Grimmond 2010; Phillips 2013; Reddy 2001; Whitby 2008), health workers (Azar‐Cavanagh 2007; Edmond 1988; van der Molen 2011), patient‐days and productive hours worked (L'Ecuyer 1996 2wva; L'Ecuyer 1996 mbc; L'Ecuyer 1996 pbc), study weeks (Mendelson 1998), hours worked (Zakrzewska 2001), patients‐days and patients (Valls 2007), employee productive hours (Goris 2015). Richard 2001 reported the number of percutaneous injuries and the proportion of injuries due to improper disposal of sharps, which was defined by the authors as an NSI to worker assisting with a procedure, or NSI located on the non‐dominant hand while removing the needle. The denominators for the blood splashes were patients (Asai 1999 active; Asai 2002 active; Asai 2002 passive; Prunet 2008 active; Prunet 2008 passive) and number of procedures (Baskin 2014; Cote 2003). In one study the denominator for NSIs was not reported (Gaballah 2012).

Researchers reported the ease of use of the devices in six studies (Asai 1999 active; Asai 2002 active; Asai 2002 passive; Baskin 2014; Mendelson 1998; Prunet 2008 active; Prunet 2008 passive; Seiberlich 2016). Five studies included a cost analysis of the intervention (Goris 2015; Mendelson 1998; Valls 2007; Whitby 2008; Zakrzewska 2001).

To be able to estimate the absolute effect of an intervention it was important to know what the control group injury rate or the baseline rate was. The NSI rate varied from 5.0 percutaneous injuries (PIs) per 1000 person‐years for Azar‐Cavanagh 2007 to 1.03 per 1000 FTE‐years for Reddy 2001. Rogues 2004 reported a rate of 17.0 phlebotomy related PIs per 100,000 devices purchased. Sossai 2010 had a baseline rate of 9.67 per 100,000 catheters used per year. Goldwater 1989 reported a rate of about 49 per 100,000 venipuncture‐years.

Geographical location

The included studies originated from nine different countries. Nine studies were from the USA (Azar‐Cavanagh 2007; Cote 2003; Edmond 1988; Goris 2015; Grimmond 2010; L'Ecuyer 1996 2wva; L'Ecuyer 1996 mbc; L'Ecuyer 1996 pbc; Mendelson 1998; Phillips 2013; Reddy 2001), two from Japan (Asai 1999 active; Asai 2002 active; Asai 2002 passive), two from France (Prunet 2008 active; Prunet 2008 passive; Rogues 2004), two from Canada (Chambers 2015 hospitals; Chambers 2015 long‐term nursing care; Seiberlich 2016), two from the UK (Gaballah 2012; Zakrzewska 2001) and one each from New Zealand (Goldwater 1989), India (Richard 2001), Italy (Sossai 2010), Spain (Valls 2007), the Netherlands (van der Molen 2011), Turkey (Baskin 2014) and Australia (Whitby 2008).

Year of study

Of the 24 included studies, 19 had been published after the year 2000 (Asai 2002 active; Asai 2002 passive; Azar‐Cavanagh 2007; Baskin 2014; Chambers 2015 hospitals; Chambers 2015 long‐term nursing care; Cote 2003; Gaballah 2012; Goris 2015; Grimmond 2010; Phillips 2013; Prunet 2008 active; Prunet 2008 passive; Reddy 2001; Richard 2001; Rogues 2004; Seiberlich 2016; Sossai 2010; Valls 2007; van der Molen 2011; Whitby 2008; Zakrzewska 2001), whereas three studies had been published in the 1990s (Asai 1999 active; L'Ecuyer 1996 2wva; L'Ecuyer 1996 mbc; L'Ecuyer 1996 pbc; Mendelson 1998) and two studies in the 1980s (Edmond 1988; Goldwater 1989).

Adequate sequence generation

We judged one of the six RCTs to have a low risk of bias for sequence generation because the researchers used a ballot box to randomise patients (Prunet 2008 active; Prunet 2008 passive). One RCT used randomisation by week (Cote 2003) and we judged it to have a high risk of bias due to the predictability of the randomisation. In one RCT (Seiberlich 2016) randomisation was done on a 1:1 basis by the participating clinicians and hence we judged it to have a high risk of bias. We judged three of the six RCTs to have an unclear risk of bias because the authors did not report specific information on the method used for randomisation (Asai 1999 active; Asai 2002 active; Asai 2002 passive; Baskin 2014).

Neither of the two cluster‐RCTs provided sufficient information about their randomisation process and therefore we judged them to have an unclear risk of bias (L'Ecuyer 1996 2wva; L'Ecuyer 1996 mbc; L'Ecuyer 1996 pbc; van der Molen 2011).

Allocation concealment

We judged three of the six RCTs to have a low risk of bias for allocation concealment because the researchers used sealed opaque envelopes or a single‐blinded envelope (Asai 2002 active; Asai 2002 passive; Baskin 2014; Prunet 2008 active; Prunet 2008 passive). We judged three RCTs and two cluster‐RCTs (Asai 1999 active; Cote 2003; L'Ecuyer 1996 2wva; L'Ecuyer 1996 mbc; L'Ecuyer 1996 pbc; Seiberlich 2016; van der Molen 2011) to have an unclear risk of bias because the authors reported no information about allocation concealment.

Similar recruitment of groups

Among the six RCTs and two cluster‐RCTs, we judged Baskin 2014; Prunet 2008 passive and van der Molen 2011 to have a low risk of recruitment bias. According to our judgment, four studies had an unclear risk of recruitment bias because they did not report information related to the recruitment of study groups (Asai 1999 active; Asai 2002 active; Cote 2003; Seiberlich 2016). We judged one study to be at high risk of recruitment bias due to a difference in the recruitment process for the intervention and control groups (L'Ecuyer 1996 2wva; L'Ecuyer 1996 mbc; L'Ecuyer 1996 pbc).

Among the five CBA studies, we judged two studies to have a low risk of recruitment bias (Grimmond 2010; Mendelson 1998). The study by Grimmond 2010 reported a small difference in staff full‐time equivalents (FTE) (< 1%) and the study by Mendelson 1998 was completed within a relatively short period of time (six months). We judged one study to have an unclear risk of recruitment bias due to the lack of information related to the recruitment of groups (Zakrzewska 2001). We judged two studies to be at high risk of recruitment bias because in one the researchers self‐assigned control and intervention hospital wards (Valls 2007) and in the other study the authors recruited control and intervention groups from different hospitals (Gaballah 2012).

Adjustment for baseline differences

For an RCT, any baseline difference should be due to chance if the randomisation process was appropriately completed. According to our judgment Asai 1999 active; Asai 2002 active; Asai 2002 passive; Cote 2003; L'Ecuyer 1996 2wva; L'Ecuyer 1996 mbc; L'Ecuyer 1996 pbc and Seiberlich 2016 had an unclear risk of bias due to baseline imbalance as they provided no information about the participants in the intervention and control groups. We judged Baskin 2014; Prunet 2008 active; Prunet 2008 passive and van der Molen 2011 to have a low risk of bias as they had adequately adjusted for baseline differences.

Among the five CBAs, we judged four studies to have an unclear risk of bias due to baseline imbalance as they reported no information regarding the adjustment for baseline difference (Gaballah 2012; Grimmond 2010; Mendelson 1998; Valls 2007). We judged Zakrzewska 2001 to have a low risk of bias in this domain because both groups were similar.

Risk of bias in ITS studies

See Table 2 for an overview of our judgment of all 11 included studies' risk of bias in all seven risk of bias domains relevant to the ITS design, and the consequent level of evidence provided. Among the 11 included ITS studies, five studies fulfilled the criterion that the intervention was independent of other changes (Azar‐Cavanagh 2007; Chambers 2015 hospitals; Chambers 2015 long‐term nursing care; Goris 2015; Phillips 2013; Rogues 2004). None of the studies reported a repeated measures analysis nor tested for trend, but this was overcome by our re‐analysis of the data. Six studies (Azar‐Cavanagh 2007; Edmond 1988; Goldwater 1989; Reddy 2001; Rogues 2004; Whitby 2008) used a data collection method which was sustained throughout the study and thus was unlikely to have affected the data collection. Three studies reported information to help determine if blind outcome assessment was used (Chambers 2015 hospitals; Chambers 2015 long‐term nursing care; Phillips 2013; Goris 2015). For the criterion of the completeness of the data set, five studies reported outcome data adequately (Azar‐Cavanagh 2007; Goldwater 1989; Goris 2015; Sossai 2010; Whitby 2008). We assessed the outcome measures of nine studies to be reliable because they used a consistent reporting system for NSI throughout the study period or they sourced data from a reliable source such as administrative data (Azar‐Cavanagh 2007; Chambers 2015 hospitals; Chambers 2015 long‐term nursing care; Edmond 1988; Goris 2015; Phillips 2013; Reddy 2001; Rogues 2004; Sossai 2010; Whitby 2008). One ITS study had an additional risk of bias due to participating health workers having access to conventional needles during the intervention period (Reddy 2001).

RCT

Outcome: needlestick injuries (NSIs)

One RCT (Baskin 2014) randomised patients to two types of syringes and evaluated the effect of safety engineered blood gas syringes on NSI compared to a conventional heparinised syringe group in the physicians who drew the blood samples. Both intervention (n = 275) and control groups (n = 275) included patients who visited the emergency department. After an immediate follow up, there was a statistically non‐significant decrease in the NSI following the intervention (RR 0.20, 95% CI 0.01 to 4.15) (Analysis 1.1).

Outcome: blood splashes

The same study (Baskin 2014) also examined contact with blood. There was a statistically non‐significant decrease in the incidence of blood splashes (RR 0.14, 95% CI 0.02 to 1.15) (Analysis 1.2).

ITS

Outcome: needlestick injuries (NSIs)

The two included ITS studies evaluated very different interventions. Therefore, we did not combine the studies in a meta‐analysis. One study evaluated a shield on the needle cap that should prevent the needle from injuring the worker when the cap is put back on the needle (Goldwater 1989). There was a non‐significant trend towards a decrease of injuries in this study (Analysis 2.1). The other used a needle sheath (Rogues 2004). In this study the level of injuries decreased substantially (effect size (ES) ‐6.88, 95% CI ‐9.53 to ‐4.23) but the trend over time showed a non‐significant decrease (Analysis 2.2).

RCT

Outcome: needlestick injuries (NSIs)

One trial evaluated the effect of three different safe IV systems to prevent NSI, which resulted in a non‐significant reduction of reported NSIs with a RR of 0.62 (95% CI 0.27 to 1.41) (Analysis 3.1) (L'Ecuyer 1996 mbc; L'Ecuyer 1996 2wva; L'Ecuyer 1996 pbc).

Outcome: incidence of blood contamination

Seven trials with 1641 participants studied if safe IV systems resulted in a change in blood contamination compared to the usual systems. There was a statistically non‐significant increased risk of blood contamination with the safe systems with a RR of 1.38 (95% CI 1.00 to 1.92). Active systems, which had to be activated by health workers, displayed a statistically significant increase in blood splashes (RR 1.60, 95% CI 1.08 to 2.36). Passive systems, which don't have to be activated, displayed a similar incidence in blood splashes in both the intervention and control groups (RR 0.94, 95% CI 0.50 to 1.75) (Analysis 3.2).

Outcome: incidence of blood leakage

One RCT study (Seiberlich 2016) evaluated the effect of a passive safe IV system on the reduction of blood leakage events during insertion of the catheter, withdrawal of the needle and connection of the luer. The study showed a significant reduction in the incidence of blood leakage events with safe IV systems (RR 0.21, 95% CI 0.11 to 0.37) (Analysis 3.3).

CBA

Outcome: needlestick injuries (NSIs)

One CBA study (Mendelson 1998) evaluated the effect of safe IV systems to prevent NSI, which resulted in a non‐significant reduction of reported NSIs with a RR of 0.06 (95% CI 0.0 to 1.09) (Analysis 4.1).

ITS

Outcome: needlestick injuries (NSIs)

In two ITS studies (Azar‐Cavanagh 2007; Sossai 2010) the results were statistically very heterogenous (I² = 79% for level and I² = 99% for trend) and therefore we did not combine them in a meta‐analysis. The level in both studies decreased with a big effect size (Analysis 5.1). The trend over time decreased substantially in one study but not in the other (Analysis 5.2).

RCT

Outcome: needlestick injuries (NSIs)

One RCT (van der Molen 2011) evaluated the effect of a workshop on NSI combined with the introduction of safety engineered injection needles in seven wards (n = 267) compared to a non‐intervention control group (eight wards, n = 266) and to a workshop on the prevention of NSIs only control group in eight wards (n = 263). NSIs were measured by questionnaires and by the hospital reporting system.

At six‐months follow‐up, there was a statistically non‐significant decrease in NSI based on the questionnaires (RR 0.49, 95% CI 0.16 to 1.56), but based on the hospital records there was a statistically non‐significant increase in NSI (RR 1.20, 95% CI 0.42 to 3.39) (Analysis 6.1; Analysis 6.2).

At 12‐months follow‐up, based on the questionnaire results there was a statistically significant reduction of NSI with RR of 0.20 (95% CI 0.04 to 0.96), but based on the hospital recording system there was a statistically non‐significant reduction of NSI with RR 0.72 (95% CI 0.28 to 1.81) (Analysis 6.3; Analysis 6.4).

CBA

In one study among dentists (Zakrzewska 2001) the risk of NSI was smaller with safe syringes compared to traditional ones but the difference was not significant (RR 0.34, 95% CI 0.04 to 3.28) (Analysis 7.1). Another study which was carried out among dental students (Gaballah 2012) evaluated the risk of NSI with safety dental syringes compared to conventional dental syringes. The authors did not report complete data regarding the type of syringe system causing NSIs for the departments in the intervention and control groups and therefore we did not analyse the results.

ITS

Outcome: needlestick injuries (NSIs) change in level

One study among healthcare workers (Goris 2015) evaluated the effect of a trial with passive safety‐engineered injection systems compared to active safety‐engineered injection systems on the incidence of NSI. There was no considerable effect on the level of NSI following the introduction of the intervention (ES 0.23, 95% CI ‐1.89 to 2.35) (Analysis 8.1).

Outcome: needlestick injuries (NSIs) change in slope

The same study showed a statistically non‐significant long term trend of a decrease in NSI (ES ‐0.74, 95% CI ‐1.66 to 0.18) (Analysis 8.2).

CBA

Outcome: needlestick injuries (NSIs)

One study that compared hospital level injury rates (Valls 2007) found a decrease in NSI in the hospitals that introduced safety devices compared to those that did not (RR 0.11, 95% CI 0.01 to 0.81) (Analysis 10.1).

ITS

Outcome: needlestick injuries (NSIs) change in level

In one ITS study (Reddy 2001) there was a statistically non‐significant increase in the level of injuries following the introduction of the safety syringes and needleless IV system (ES 0.43, 95% CI ‐0.30 to 1.16) (Analysis 9.1). Another ITS study (Whitby 2008) showed a statistically non‐significant decrease in the level of NSI following the introduction of safety syringes, needless IV systems and safety‐engineered needles (ES ‐1.04, 95% CI ‐2.20 to 0.12) (Analysis 9.1).

Outcome: needlestick injuries (NSIs) change in slope

In the study by (Reddy 2001) the ES for the change in long‐term time trend showed an increase in the number of reported NSIs (ES 0.56, 95% CI 0.23 to 0.89) (Analysis 9.2). In the other ITS study (Whitby 2008) there was a statistically non‐significant decrease in the trend of reported NSI (ES ‐0.01, 95% CI ‐0.15 to 0.13) (Analysis 9.2).

CBA

Outcome: needlestick injuries (NSIs)

In one CBA study (Grimmond 2010), the NSI rate decreased following the introduction of sharps containers compared to departments where these were not introduced with a RR of 0.88 (95% CI 0.78 to 0.99) (Analysis 12.1). This reduction was statistically significant when only container‐related NSIs were counted with a RR of 0.22 (95% CI 0.11 to 0.41) (Analysis 12.2).

ITS

Two ITS studies (Edmond 1988; Richard 2001) showed an increased level of NSI immediately after the introduction of sharps containers and a contradictory effect in the long‐term trend which prevented the synthesis of these studies in a meta‐analysis (Analysis 11.1; Analysis 11.2).

Outcome: needlestick injuries (NSIs) change in level

One ITS study had two intervention arms. One arm comprised of long‐term nursing care (Chambers 2015 long‐term nursing care) and the other comprised of hospitals (Chambers 2015 hospitals). According to the results the level of NSI decreased in long‐term nursing care after the introduction of legislation. However, the intervention arm comprising of hospitals showed an increase in the level of NSI. Another ITS study (Phillips 2013) also showed a decrease in the level of NSI following the introduction of legislation. Since these results were very heterogenous we did not combine them in a meta‐analysis (Analysis 13.1).

Outcome: needlestick injuries (NSIs) change in slope

In one ITS study the NSI trend over time decreased in one of the intervention arms comprising of long‐term nursing care (Chambers 2015 long‐term nursing care) and increased in the other arm which included hospitals (Chambers 2015 hospitals). The other ITS study (Phillips 2013) showed a decrease in the long term trend of NSI (Analysis 13.2).

Secondary outcomes

1. Cost

A total of five studies reported information regarding the cost of the intervention. Valls 2007 reported that the direct cost of the use of safety devices was an additional USD 19,417 (USD 0.75 per patient) for the emergency department and USD 16,336 (USD 0.56 per patient‐day) for the hospital wards compared to the pre‐intervention period. Zakrzewska 2001 reported that the price of the safety syringes was comparable to the non‐disposable syringes, approximately USD 0.33 per item. Mendelson 1998 reported that the estimated incremental hospital‐wide cost was USD 82,822 (in 1991) but the cost of injury prevented was USD 1593. Whitby 2008 reported that the overall increased cost for provision of safety‐engineered retractable syringes in the 800‐bed hopsital was USD 46,000 per annum, USD 14 for each at‐risk healthcare worker per year or USD 2 per occupied bed‐day per annum. Goris 2015 reported a net annual increase of USD 20,708.42 on conversion of ASED to PSED at the Barnes‐Jewish Hospital. The study also reported that the total cost avoidance of a conversion from ASED to PSED was USD 68,768.28.

2. Ease of use

Asai 1999 active reported no difference between the safety devices and the conventional devices in terms of ease of insertion. However, the authors reported statistically higher ease of handling for the safety device compared to the conventional one. Asai 1999 active, Asai 2002 active, and Asai 2002 passive reported that the Autoguard IV was significantly easier to insert and handle compared to the other safety device and the conventional catheter needle. Mendelson 1998 reported that 94% of the individuals who completed the survey (approximately 52% response rate) were comfortable using the safe IV system after five or less trials. Prunet 2008 active and Prunet 2008 passive reported that the Insyte Autoguard device was significantly more difficult to insert when compared to conventional devices and the passive devices. With both safety devices the needle was significantly more difficult to withdraw in comparison to the conventional catheter. Baskin 2014 reported that there was no significant difference between a conventional heparinised insulin syringe and safety‐engineered blood gas syringe in terms of ease of use. Seiberlich 2016 reported that the blood control PIVC and standard PIVC were similar in terms of ease of use.