Results of the search

We identified 2804 studies assessed from electronic searches and a further 240 studies from backward and forward citation searching in the 2014 search. We found three ongoing studies from clinical trial databases. We considered a total of 2831 unique titles and abstracts, and from these we assessed a further 139 full texts for eligibility, alongside the included studies in Walker 2009. See Figure 1.

Figure 1

---

---

Study flow diagram.

We reran the search in May 2015. We found seven studies of interest which we will incorporate when we next update the review. See also Studies awaiting classification.

Included studies

Thirty‐two studies with a total of 2844 participants met our inclusion criteria (see Characteristics of included studies). All included studies were randomized controlled trials. All 32 studies were from single centres, involved only adult participants, and included people of both sexes. Of these, 19 were new studies identified from the updated search in June 2014 and 13 were from the original review.

Twelve studies assessed axillary brachial plexus block (Bloc 2010; Casati 2007a; Chan 2007; Conceição 2009; Geiser 2011; Gurkan 2008; Liu 2005; Meierhofer 2014; Morros 2009; Shrestha 2011; Sites 2006; Strub 2011), four studies assessed infraclavicular brachial plexus block (Brull 2009; Dhir 2008; Sauter 2008; Trabelsi 2013) and three assessed supraclavicular brachial plexus block (Renes 2009; Williams 2003; Zaragoza‐Lemus 2012), all for surgery of the hand, wrist or forearm. One study assessed a wrist block (Macaire 2008) for carpal tunnel release. There were four studies assessing interscalene brachial plexus block (Danelli 2012; Kapral 2008; Liu 2009a; Salem 2012) and one coracoid infraclavicular brachial plexus block (Taboada 2009), all for surgery of the shoulder or upper arm. Soeding 2005 assessed both an interscalene brachial plexus block and an axillary brachial plexus block within the same study. Seidel 2013 assessed a sciatic nerve block for foot or ankle, Domingo‐Triado 2007 assessed a mid‐femoral sciatic block and three studies a popliteal block for foot or ankle surgery (Dufour 2008; Perlas 2008; Van Geffen 2009). One popliteal fossa block was for hallux vagus correction (Cataldo 2012).

Seventeen studies compared the use of ultrasound alone with peripheral nerve stimulation (Brull 2009; Casati 2007a; Chan 2007; Conceição 2009; Danelli 2012; Geiser 2011; Kapral 2008; Macaire 2008; Meierhofer 2014; Perlas 2008; Renes 2009; Sauter 2008; Seidel 2013; Taboada 2009; Trabelsi 2013; Van Geffen 2009; Zaragoza‐Lemus 2012). Nine studies compared the use of ultrasound combined with nerve stimulation (ultrasound + nerve stimulation) with nerve stimulation alone (Cataldo 2012; Chan 2007;Domingo‐Triado 2007;Dufour 2008;Gurkan 2008; Morros 2009; Salem 2012; Shrestha 2011; Williams 2003). Remaining studies compared ultrasound with landmark technique (Soeding 2005; Strub 2011) or transarterial approach (Sites 2006) or were three‐arm studies comparing ultrasound (in‐plane approach) with ultrasound (out‐of‐plane approach) and with peripheral nerve stimulation (Bloc 2010), ultrasound + nerve stimulation with nerve stimulation (with or without a stimulating catheter) (Dhir 2008), or ultrasound (single injection) with ultrasound (double injection) and with nerve stimulation (double injection) (Liu 2005).

There were three studies that included placement of a catheter in the nerve block procedure (Danelli 2012; Dhir 2008; Salem 2012). In Danelli 2012, the catheter was placed after the injection of local anaesthetic and was therefore comparable with non‐catheter‐placement studies. For Dhir 2008 and Salem 2012, however, the local anaesthetic was administered through the catheter. We included six studies that gave details of additional nerve blocks for tourniquet pain (Domingo‐Triado 2007; Dufour 2008; Meierhofer 2014; Perlas 2008; Seidel 2013; Van Geffen 2009).

Excluded studies

There were 18 studies in the original review (Walker 2009). Following the changes made to the review inclusion criteria in the 2014 update, we excluded five of these previously included studies. Four of them gave additional anaesthesia to all participants following the nerve block (Danelli 2009a; Dolan 2008; Marhofer 1997; Marhofer 1998) and one had used a MEAV study design (Casati 2007b).

During the updated search we identified a further 18 studies that were excluded either as MEAV study designs, studies in which additional anaesthesia was given to all participants or studies in which participants were scheduled for surgery other than for lower/upper extremity procedures.

We excluded six abstracts which were potentially eligible but provided insufficient detail. We excluded two studies as ultrasound was used pre‐puncture in both groups, and one study due to lack of randomization details in the full text. Redborg 2009 had previously been in Characteristics of studies awaiting classification but was assessed for this update as not eligible due to the use of volunteer non‐surgical participants.

In total, we excluded 33 studies from the updated review. Details are in Characteristics of excluded studies.

Allocation

All 32 studies were described as randomized but only 20 provided sufficient detail of the methods used to be judged as being at low risk of bias (Brull 2009; Casati 2007a; Chan 2007; Conceição 2009; Danelli 2012; Domingo‐Triado 2007; Dufour 2008; Geiser 2011; Kapral 2008; Liu 2005; Liu 2009a; Meierhofer 2014; Morros 2009; Perlas 2008; Renes 2009; Sauter 2008; Seidel 2013; Shrestha 2011; Sites 2006; Strub 2011). There were only two studies that provided an adequate description of the methods used to conceal group allocation to participants and personnel (Kapral 2008; Perlas 2008).

Blinding

Due to the nature of the intervention, blinding of the practitioner was never going to be possible and it is therefore an unavoidable source of bias. It was necessary for us to judge performance bias across all outcomes in all studies as being at high risk of bias due to this lack of blinding. It was, however, possible for detection bias to be reduced by ensuring that observers/investigators collecting data for some of the outcomes were blinded to group allocation. Seventeen studies had reported sufficient information on whether outcome assessors were blinded to group allocation on at least one of the outcomes (Brull 2009; Casati 2007a; Cataldo 2012; Domingo‐Triado 2007; Dufour 2008; Gurkan 2008; Liu 2005; Liu 2009a;Meierhofer 2014; Perlas 2008; Sauter 2008; Seidel 2013; Shrestha 2011; Sites 2006; Taboada 2009; Trabelsi 2013; Van Geffen 2009). Only two trials made an attempt to blind the participants to the technique being used through the use of a sham ultrasound device (Chan 2007; Perlas 2008).

Incomplete outcome data

There were few losses of study participants overall and all but four of the studies (Brull 2009; Chan 2007; Dufour 2008; Sites 2006) were assessed as being at a low risk of attrition bias.

Brull 2009 had few losses for all outcomes except complications, for which only 49% of participants were contacted for follow‐up at one week postoperatively. We assessed this as being at unclear risk of bias. Chan 2007 lost 14% of its intended participants who were required to go to surgery before the end of 30 minutes post‐block. We judged that some outcome data could still have been collected from these participants and we therefore felt that this study was at high risk of attrition bias, along with Dufour 2008 which also had several losses. We judged Sites 2006 as being at high risk of bias, as this study stopped early due to a high number of failed blocks in the transarterial approach group.

Selective reporting

We were able to source the protocols for five of the studies from clinicaltrials.gov and compare the reported outcomes with protocol outcomes (Chan 2007; Danelli 2012; Perlas 2008; Sauter 2008; Seidel 2013). We judged these as being at low risk of selective outcome reporting bias, as all outcomes were reported as planned. However, we were unable to make a judgement on the remaining studies for high or low risks of bias.

Other potential sources of bias

There was one study that failed to report any baseline characteristics (Soeding 2005) and we were therefore unable to make a judgement on whether any bias could have been introduced. A further four studies reported baseline characteristics for which there were some discrepancies between groups: in Dhir 2008 there were more older participants in the nerve stimulation (with stimulating catheter) group; in Domingo‐Triado 2007 there were differences between groups in the types of surgery; in Geiser 2011 there were more women than men reported in the table, although the text stated that there was no difference; and in Dufour 2008 there was a difference in the ASA status between groups. We were unsure whether these differences could potentially introduce any bias into the results, and assessed them as being at unclear risk. The remaining 27 studies all had comparable baseline characteristics between participants.

We were interested in whether study authors had been provided with any funding for their research and therefore considered this in our assessment of risk of bias. There were five studies that declared that the ultrasound or nerve stimulator equipment had been provided by the manufacturer for the purpose of the study (Brull 2009; Chan 2007; Gurkan 2008; Sites 2006; Van Geffen 2009). We judged these studies to be at a higher risk of bias. All other studies either declared funding from departmental sources only, or did not make any funding declarations, and we assessed them as being at low risk of bias.

The experience of practitioners in both ultrasound and control techniques, as well as the number of practitioners involved, varied across studies. There were 12 studies that described the person giving the block as experienced (Bloc 2010; Brull 2009; Casati 2007a; Danelli 2012; Kapral 2008; Meierhofer 2014; Morros 2009; Perlas 2008; Salem 2012; Sauter 2008; Soeding 2005; Taboada 2009). However, even for those that stated that the practitioners had experience in both techniques, it was not clear whether this experience was equivalent for each technique. For this reason we were unable to judge whether any bias had been introduced by the practitioners in these studies. Some procedures were performed by anaesthesia residents under supervision (Chan 2007; Williams 2003), and we felt that could be likely to introduce bias, particularly for block performance time, and therefore assessed them as being at high risk. We also rated studies at high risk of bias if it was clear that the practitioner had more experience in one technique than the other, or that different procedures were intentionally performed by different practitioners.

Primary outcomes

1. Block success ‐ predicted adequacy of block

There were 17 studies with 1346 participants comparing ultrasound guidance with either nerve stimulation (15 studies: Bloc 2010; Brull 2009; Chan 2007; Geiser 2011; Kapral 2008; Liu 2009a; Macaire 2008; Meierhofer 2014; Perlas 2008; Renes 2009; Sauter 2008; Taboada 2009; Trabelsi 2013; Van Geffen 2009; Zaragoza‐Lemus 2012), anatomical landmark technique (one study: Soeding 2005) or a transarterial approach (one study: Sites 2006) and reporting on predicted adequacy of the block. This outcome was often described by the authors as "block success" and was evaluated using appropriate sensory and motor testing at intervals following the procedure, using a scale to determine the degree of block. We combined data described as "complete", "successful" of "sufficient" block. If studies separated results for adequacy of sensory and motor success, we used data from the sensory block.

For the purpose of this analysis we combined the two ultrasound groups in both Bloc 2010 and Liu 2005. We also included data for Chan 2007 for the ultrasound alone versus nerve stimulation group. The analysis demonstrated a statistically significant difference between the ultrasound versus nerve stimulation groups, with a higher rate of predicted adequacy of the block in the ultrasound group than the comparison group (Mantel‐Haenszel (M‐H) odds ratio (OR) 3.01 (95% confidence interval (CI) 2.11 to 4.31), 1250 participants, P value < 0.00001). When data for Soeding 2005 (anatomical landmark technique) and Sites 2006 (transarterial approach) were also included the result remained statistically significant in favour of ultrasound (M‐H OR 3.06 (95% CI 2.18 to 4.30), 1346 participants, P value < 0.00001). This result shows a low level of heterogeneity (I² = 13%). Considering the potential effect of bias on this result we graded this to be moderate‐quality evidence of an effect in the 'Summary of findings' table. A funnel plot did not suggest publication bias for this outcome. See Analysis 1.1, summary of findings Table for the main comparison and Figure 4.

Figure 4

---

---

Funnel plot of comparison: 1 Ultrasound technique versus other technique, outcome: 1.1 Predicted adequacy of block.

There were nine studies with 719 participants comparing ultrasound + nerve stimulation versus nerve stimulation technique and reporting on predicted adequacy of the block (Cataldo 2012; Chan 2007; Dhir 2008; Domingo‐Triado 2007; Dufour 2008; Gurkan 2008; Morros 2009; Salem 2012; Williams 2003). For the purpose of this analysis, we combined data in Dhir 2008 for the two nerve stimulation groups. The analysis demonstrated a statistically significant difference between the two groups, again with a higher rate of predicted adequacy of the block in the ultrasound group, than the comparison group (M‐H OR 3.33 (95% CI 2.13 to 5.20), P value < 0.00001). This result shows a low level of heterogeneity (I² = 26%). See Analysis 2.1.

2. Block success ‐ supplementation requirement

There were 18 studies with 1807 participants comparing ultrasound guidance with either nerve stimulation (15 studies: Bloc 2010; Casati 2007a; Chan 2007; Conceição 2009; Geiser 2011; Kapral 2008; Liu 2005; Liu 2009a; Macaire 2008; Meierhofer 2014; Perlas 2008; Renes 2009; Sauter 2008; Seidel 2013; Taboada 2009), anatomical landmark technique (two studies: Soeding 2005; Strub 2011) or a transarterial approach (one study: Sites 2006) and reporting on supplementation rates. Authors sometimes described this outcome as "block failure" and defined it as the need for participants to be given either supplementary block, local anaesthetic, supplementary intraoperative analgesics or conversion to general anaesthesia. For the purpose of this outcome, we combined all supplementary anaesthesia/analgesia together. As above, we combined the two ultrasound groups in Liu 2005. There was a statistically significant difference between groups with fewer participants in the ultrasound group requiring additional supplementation (Mantel‐Haenszel (M‐H) OR 0.31 (95% CI 0.21 to 0.46),1570 participants, P value < 0.00001). When we included data for Soeding 2005 and Strub 2011 (anatomical landmark technique) and Sites 2006 (transarterial approach), the results remained statistically significant (M‐H OR 0.28 (95% CI 0.20 to 0.39), P value < 0.00001). Again, there was a low level of heterogeneity in this result (I² = 16%), and we graded it as moderate‐quality evidence of an effect in the 'Summary of findings' table. See Analysis 1.2 and summary of findings Table for the main comparison.

There were nine studies (Chan 2007; Dhir 2008; Domingo‐Triado 2007; Dufour 2008; Gurkan 2008; Morros 2009; Salem 2012; Shrestha 2011; Williams 2003) with 712 participants comparing ultrasound guidance + nerve stimulation versus nerve stimulation technique and reporting on supplementation rates, as above. We combined data for the two nerve stimulation groups in Dhir 2008. The results were statistically significantly different, again with a lesser need for supplementation in the ultrasound group (M‐H OR 0.34 (95% CI 0.21 to 0.56), 1807 participants, P value < 0.00001, I² = 0%. See Analysis 2.2.

3. Complications

Complication rates were recorded in 23 trials There were no reported incidences of major complications (cardiorespiratory arrest, seizures, pneumothorax, nerve injury) in any included study. Paraesthesia and vascular puncture were the most frequently reported complications and we included data for these in meta‐analysis.

There were six studies (Bloc 2010; Brull 2009; Conceição 2009; Liu 2005; Sauter 2008; Soeding 2005) with 471 participants that reported data for paraesthesia for the groups ultrasound versus nerve stimulation (five studies) and anatomical landmark technique (one study). This analysis showed more incidences of paraesthesia in the nerve stimulation group (M‐H OR, 0.42 (95% CI 0.23 to 0.76)). However there was a high level of heterogeneity in this analysis (I² = 75%) and it is clear that the result is influenced by Brull 2009 with a very large number of events (22 events, 45%) in the nerve stimulation group. We downgraded the quality of this evidence to very low on account of the relatively few events reported in studies and the high level of heterogeneity. See Analysis 1.4 and summary of findings Table for the main comparison. There were only three studies (Dhir 2008; Dufour 2008; Shrestha 2011) with 178 participants that reported data for paraesthesia for the groups ultrasound + nerve stimulation versus nerve stimulation alone. There was no significant difference in this analysis (M‐H OR, 0.97 (95% CI 0.30 to 3.12), P value = 0.95). See Analysis 2.4.

There were five studies with 387 participants that reported data for vascular puncture for the groups ultrasound versus nerve stimulation (four studies: Bloc 2010; Brull 2009; Conceição 2009; Taboada 2009) and transarterial approach (one study: Sites 2006). The result (M‐H OR 0.19 (95% CI 0.07 to 0.57)) showed that there were fewer incidences of vascular puncture in the ultrasound groups and we graded this as low level of evidence in the 'Summary of findings' table. See Analysis 1.5 and summary of findings Table for the main comparison. There were only two studies (Gurkan 2008; Shrestha 2011; 143 participants) that reported data for vascular puncture from our comparison groups of ultrasound + nerve stimulation versus nerve stimulation alone, and again this analysis showed a statistically significant effect of fewer incidences of vascular puncture in the ultrasound + nerve stimulation group (M‐H OR 0.22 (95% CI 0.05 to 0.90)). See Analysis 2.5.

One study reported data for axillary vessels puncture (Liu 2005, three incidences in the nerve stimulation group). Meierhofer 2014 gave data for venous puncture (one incidence in the nerve stimulation group) and for arterial puncture (two incidences in both groups) and Morros 2009 also for arterial puncture (one incidence in the ultrasound group and two in the nerve stimulation group).

Studies gave differing terms for nerve damage, but it was not always clear whether these effects were immediate, medium‐ or long‐term and whether they were equivalent between studies. It was therefore not possible to combine them in meta‐analysis. Studies that reported on such complications were Domingo‐Triado 2007; Kapral 2008; Perlas 2008; Salem 2012; Sites 2006; and Strub 2011. Of these, Kapral 2008 and Sites 2006 reported no events and Salem 2012 reported that such effects were equivalent between groups. Perlas 2008 reported more numbness at 24 hours postoperatively in the ultrasound group (eight versus four events) and more weakness at this time point in the ultrasound group (10 versus two events). Strub 2011 reported two events of neuralgia in the hand in the traditional nerve block technique and no events in the ultrasound group. Domingo‐Triado 2007 reported that one participant had neuropathic pain at one week in the nerve stimulation group which resolved within 10 days.

Other effects reported were tachycardia (Brull 2009, one event in the nerve stimulation group), subcutaneous haematoma (Liu 2005, one event in the nerve stimulation group), haematoma requiring additional manual compression (Sites 2006, two events in the transarterial approach group), axillary haematoma (Strub 2011, five events in the traditional nerve block technique and two in the ultrasound group), prolonged pain in axilla (Strub 2011, three in traditional group and one in ultrasound group) and respiratory discomfort (Williams 2003, one participant in each group).

Secondary outcomes

1a.Time to perform block

There were 25 studies that reported data for time to perform block (Bloc 2010; Brull 2009; Cataldo 2012; Chan 2007; Conceição 2009; Danelli 2012; Dhir 2008; Domingo‐Triado 2007; Dufour 2008; Geiser 2011; Gurkan 2008; Liu 2005; Macaire 2008; Meierhofer 2014; Morros 2009; Perlas 2008; Salem 2012; Sauter 2008; Shrestha 2011; Sites 2006; Strub 2011; Taboada 2009; Trabelsi 2013; Van Geffen 2009; Williams 2003). However, some of these data were presented as median and interquartile range, or presented means without standard deviations, and it was therefore not possible to include them in the meta‐analysis (Bloc 2010; Brull 2009; Conceição 2009; Danelli 2012; Domingo‐Triado 2007; Macaire 2008; Sauter 2008; Strub 2011). We did not combine data for Cataldo 2012 in our analysis which had given time data by each blocked nerve. We also decided that those studies which included catheter placement should not be included in this analysis, due to the increased length of time involved in this procedure. We therefore excluded Dhir 2008 and Salem 2012. Whilst there was some variation in definition of performance time, we felt that these were similar enough to warrant meta‐analysis. We present definitions, where available, in Characteristics of included studies.

Our first meta‐analysis was conducted using only 10 studies of ultrasound versus nerve stimulation (nine studies: Chan 2007; Conceição 2009; Geiser 2011; Liu 2005; Meierhofer 2014; Perlas 2008; Taboada 2009; Trabelsi 2013; Van Geffen 2009) and transarterial approach (one study: Sites 2006) with a total of 690 participants. For Liu 2005, it was not possible to combine the data for the two ultrasound groups and we therefore only compared the ultrasound (double‐injection) group against the nerve stimulation (double‐injection) group.

The analysis showed a statistically significant difference, with performance time being less in the ultrasound group (mean difference (MD), IV fixed‐effect, ‐1.06 (95% CI ‐1.41 to ‐0.72), P value < 0.00001). See Analysis 1.3.

For performance time, seven studies with a total of 587 participants (Chan 2007; Dufour 2008; Gurkan 2008; Morros 2009; Salem 2012; Shrestha 2011; Williams 2003) compared ultrasound guidance + nerve stimulation with nerve stimulation technique. There was a statistically significant difference between groups, with performance time being less in the nerve stimulation group (MD, IV, fixed‐effect, 0.76 (95% CI 0.55 to 0.98)). See Analysis 2.3.

There was a high level of heterogeneity for both these analyses (I² = 88% in Analysis 1.3 and I² = 80% in Analysis 2.3). We subsequently downgraded the quality of evidence for this result to very low, taking into account the high level of heterogeneity and potential differences in experience of personnel which could affect this result. See summary of findings Table for the main comparison.

We also considered those studies for which we did not conduct meta‐analysis for this outcome. Of those studies comparing ultrasound with nerve stimulation, five reported a statistically significantly shorter time to perform block in the ultrasound groups (Bloc 2010: P value < 0.05; Brull 2009: P value < 0.001; Danelli 2012: P value = 0.01; Macaire 2008: P value = 0.02; Sauter 2008: P value = 0.003). Conceição 2009 reported a shorter time in the ultrasound group but this was not statistically significant. Cataldo 2012 reported a statistically significantly shorter time in the ultrasound + nerve stimulation group (P value = 0.02). Dhir 2008 reported a statistically significantly shorter time in the nerve stimulation compared to other groups (P value < 0.0001). Strub 2011 reported no significant differences between groups in time to perform block. We were unable to extract data for Salem 2012, due to their methods of presentation of results.

1b. Onset time of block

There were 15 studies (Casati 2007a; Cataldo 2012; Danelli 2012; Domingo‐Triado 2007; Gurkan 2008; Kapral 2008; Macaire 2008; Meierhofer 2014; Salem 2012; Sauter 2008; Seidel 2013; Shrestha 2011; Strub 2011; Taboada 2009; Trabelsi 2013) that evaluated onset time of block. However, the studies reported data in different ways, sometimes reporting median (range or interquartile range) and sometimes mean and standard deviation. There were also differences in whether results were presented for each nerve separately or combined, and whether or not sensory or motor block onset time was reported separately. We therefore did not combine these data in meta‐analysis. Six studies reported a statistically significant difference between groups, favouring a shorter onset time in the ultrasound group with P value less than 0.05 (Casati 2007a (for sensory block only); Kapral 2008; Seidel 2013; Shrestha 2011; Strub 2011; Trabelsi 2013 (for the sensory block only)). Nine studies reported no significant differences between groups in onset time (Casati 2007a (for motor block only); Cataldo 2012; Danelli 2012; Domingo‐Triado 2007; Meierhofer 2014; Salem 2012; Sauter 2008; Taboada 2009; Trabelsi 2013 (for motor block only)). Macaire 2008 reported a significantly shorter onset time in the nerve stimulation group (P value < 0.02).

1c. Duration of block

There were four studies that evaluated duration of block. Kapral 2008 reported a statistically significant difference between groups, with blocks in the ultrasound group having a longer duration than the nerve stimulation group (P value < 0.05). The remaining studies reported that there was no significant difference between groups for block duration time (Dhir 2008; Domingo‐Triado 2007; Soeding 2005).

2. Number of attempts

There were seven studies (Casati 2007a; Cataldo 2012; Danelli 2012; Dufour 2008; Sauter 2008; Shrestha 2011; Van Geffen 2009) that reported on the number of attempts, defined as needle/skin punctures or needle passes (forward movement preceded by retraction of needle). Individual study definitions are given in Characteristics of included studies. Results were reported as mean and standard deviation or median and range, and it was not possible to pool data. Three studies reported that there were significantly fewer needle passes or skin punctures in the ultrasound group (Danelli 2012: P value = 0.01; Sauter 2008: P value < 0.001; Van Geffen 2009: P value = 0.029). Shrestha 2011 reported fewer attempts in the ultrasound + nerve stimulation group than the nerve stimulation group. Cataldo 2012 reported significantly more needle punctures in the intervention group (P value = 0.004). Dufour 2008 , whilst reporting significantly more needle passes to locate the first nerve in the comparison, reported with Casati 2007a that the difference in the number of skin punctures was not significant. We graded this evidence as being of low quality in the summary of findings Table for the main comparison.

3. Participant discomfort

There were seven studies (Bloc 2010; Casati 2007a; Dufour 2008; Macaire 2008; Meierhofer 2014; Sauter 2008; Van Geffen 2009) in which participants reported discomfort. Five of these reported responses on a visual analogue scale or numerical rating score for satisfaction with the procedure or discomfort/pain during procedure (Dufour 2008; Macaire 2008; Meierhofer 2014; Sauter 2008; Van Geffen 2009). All five studies reported that there was no statistically significant difference between groups. Casati 2007a asked if participants would accept the same procedure again and there were no statistically significant differences between groups. Only Bloc 2010 reported a statistically significant difference for this outcome, with fewer participants describing the procedure as unpleasant in the ultrasound (out‐plane approach) than the ultrasound (in‐plane approach) or nerve stimulation group. We graded evidence for this outcome as being of low quality in the summary of findings Table for the main comparison.

Subgroup analysis

The outcomes 'Time to perform block' and 'Paraesthesia' both had a high level of statistical heterogeneity (I² = 88% in Analysis 1.3; I² = 80% in Analysis 2.3; I²= 75% in Analysis 1.4).

1.Different types of comparisons

We performed subgroup analysis according to the comparison group, i.e. nerve stimulation, anatomical landmark or transarterial approach. For 'Time to perform block', all but one study compared with nerve stimulation and there remained a high level of heterogeneity in this group, I² = 89%. See Analysis 3.1. Similarly for 'Paraesthesia', we were not able to explain heterogeneity by subgroup analysis, again with only one study not comparing against nerve stimulation and statistical heterogeneity remaining high for this group (I² = 79%). See Analysis 5.1.

2. Different types of nerve blocks

We performed subgroup analysis according to the type of nerve block and block approach. For 'Time to perform block' results remained statistically significant and with a high level of heterogeneity in the axillary and infraclavicular brachial plexus block, but for those studies which used the popliteal fossa sciatic block there was no difference in the time to perform the block between ultrasound and nerve stimulation use (MD, IV, fixed‐effect ‐1.00 (95% CI ‐2.43 to 0.44), I² = 0%). However there were only two studies using this block in this analysis (see Analysis 3.2). For those studies combining ultrasound with peripheral nerve stimulation, there were only single studies in the nerve block approaches other than for axillary plexus block, and for this there was no difference in the result and I² remained similarly high (see Analysis 4.1).

For the outcome 'Paraesthesia', subgroup analysis showed that for the infraclavicular brachial plexus block there were still significantly fewer events of paraesthesia with ultrasound use, although statistical heterogeneity remained high (I² = 87%). But for the axillary brachial plexus block there was no longer any statistical difference between block technique, with only moderate heterogeneity (I² = 40%). See Analysis 5.2.

3. Studies with catheter placement

Three of our studies (Danelli 2012; Dhir 2008; Salem 2012) had included catheter placement as part of the nerve block procedure and whilst we had included Dhir 2008 and Salem 2012 in the main analysis which had a low I² value, we separated Salem 2012 from the 'Time to perform block' outcome. This outcome remained statistically significantly in favour of the control. See Analysis 4.2.

4. Experience of practitioners

Although experience of practitioners is an important consideration for this review, we did not perform subgroup analysis. Several studies did not provide details on experience, and for those that did it was often unclear whether the experience was equivalent between techniques. Subgroup analysis would not have provided a reliable result.

5. Other heterogeneity

It is likely that heterogeneity for the outcome of 'Time to perform block' may be as a result of the variety of definitions used by study authors for this outcome measure. However, these outcomes varied such that it was not feasible to perform subgroup analysis and provide a reliable result.

Sensitivity analysis

In sensitivity analysis, we considered the effect of bias on our primary outcome only.

We removed those studies that had not reported clearly on their methods of sequence generation (Macaire 2008; Taboada 2009; Trabelsi 2013; Van Geffen 2009; Zaragoza‐Lemus 2012) and this did not affect the results. As it was feasible for outcomes to be assessed by blinded observers, we removed those studies that we had judged as being at either unclear or high risk of bias for this domain (Chan 2007; Geiser 2011; Kapral 2008; Macaire 2008; Renes 2009; Taboada 2009; Van Geffen 2009; Zaragoza‐Lemus 2012) and again this did not affect the results.

We similarly removed studies at high or unclear risk of attrition bias (Brull 2009; Chan 2007), with no difference to meta‐analysis results.

We had reported on whether studies had received any funding assistance and for sensitivity analysis removed those studies that we had assessed as being at high risk of bias due to the supplying of study equipment (Brull 2009; Chan 2007; Gurkan 2008; Sites 2006; Van Geffen 2009). For Analysis 1.1, we removed Brull 2009, Chan 2007, Sites 2006 and Van Geffen 2009, and for Analysis 2.1, we removed Gurkan 2008. This did not make any difference to our statistically significant result in favour of ultrasound.

In sensitivity analysis, we chose to remove those studies which we had judged as having a high risk of bias for practitioner experience (Cataldo 2012; Chan 2007; Dufour 2008; Sites 2006; Strub 2011; Van Geffen 2009; Williams 2003).This did not make any difference to our results in favour of ultrasound guidance for success of the block.