Types of surgery

The types of surgeries performed were reported as: elective hip replacement or arthroplasty (Aksoy 2014; Cucereanu Badica 2010; Cucereanu Badica 2010; Frassanito 2008; Gelmanas 2010; Kendrisic 2013; Kratz 2015; Murray 2005; Souron 2003): partial hip replacement (Türker 2003); total hip arthroplasty or replacement (Asano 2010; Becchi 2008; Bhatia 2008; Biboulet 2004; Bichel 1998; Borisov 2012; Celidonio 2008; Chen 2015; Duarte 2009; Forget 2009; Fouad 2010; Fournier 1998; Fredrickson 2015; Ginz 2000; Goytizolo 2016; Green 2014; Jensen 2012; Kai 2010; Kearns 2011; Köroğlu 2008; Marino 2009; Marshall 2008; Nicholson 2002; Nishio 2014; Nohel 2011; Saksena Shrivastava 2011; Shariat 2013; Siddiqui 2007; Singelyn 2005; Stevens 2000; Stevens 2007; Striebel 1993; Thomas 2009; Thybo 2016; Twyman 1990; Uhrbrand 1992; Utebey 2009; Van Herreweghe 2015); or hip surgery (Anis 2011).

Comparators

Peripheral nerve blocks were compared to no block (Anis 2011; Biboulet 2004; Chen 2015; Cucereanu Badica 2010; Ginz 2000; Goytizolo 2016; Green 2014; Kratz 2015; Marino 2009; Murray 2005; Nicholson 2002; Saksena Shrivastava 2011 (this trial also contained a group who received epidural analgesia, but these data were not retained for analysis), Siddiqui 2007; Stevens 2000; Striebel 1993; Twyman 1990; Uhrbrand 1992; Van Herreweghe 2015; Wiesmann 2014); sham block (Fournier 1998; Jensen 2012; Köroğlu 2008; Shariat 2013; Stevens 2007; Thybo 2016), systemic analgesia (Becchi 2008; Kendrisic 2013; Nishio 2014 (this trial contained a group who received caudal analgesia, but these data were not retained in the analysis), Nohel 2011); systemic analgesia or epidural analgesia (Singelyn 2005; Utebey 2009); epidural analgesia (Asano 2010; Bichel 1998; Borisov 2012; Celidonio 2008; Duarte 2009; Forget 2009; Fouad 2010; Gelmanas 2010; Kai 2010; Marshall 2008; Türker 2003); spinal anaesthesia (Aksoy 2014); spinal analgesia (Bhatia 2008; Frassanito 2008; Fredrickson 2015; Kearns 2011; Souron 2003); or intravenous lidocaine (Thomas 2009).

Because trials where peripheral nerve blocks were compared to no block, sham block or systemic analgesia all reverted to systemic analgesia as the main modality for postoperative pain treatment (except Goytizolo 2016), these were treated as one comparison. Goytizolo 2016 studied the addition of single injection psoas compartment block compared to epidural analgesia.

Types of peripheral nerve blocks

Peripheral nerve blocks were psoas compartment block (single injection: Anis 2011; Frassanito 2008; Goytizolo 2016; Green 2014; Kai 2010; Souron 2003; Stevens 2000; Twyman 1990; repeated doses: Bhatia 2008; Utebey 2009; continuous block: Asano 2010 (this trial also contained a group who received continuous psoas compartment block plus local anaesthetic infiltration, but data were not retained in the analysis); Becchi 2008; Duarte 2009; Fouad 2010; Fredrickson 2015; Gelmanas 2010; Kendrisic 2013; Marshall 2008; Türker 2003), psoas compartment block plus sciatic nerve block plus iliac crest infiltration (single injection: Aksoy 2014), psoas compartment block or femoral nerve block (single injection: Biboulet 2004; continuous nerve block: Marino 2009), femoral nerve block (or 3‐in‐1 block) (single injection: Fournier 1998; Köroğlu 2008; Kratz 2015; Nicholson 2002; Wiesmann 2014; repeated doses: Striebel 1993; continuous block: Bichel 1998; Chen 2015; Nishio 2014; Nohel 2011; Saksena Shrivastava 2011; Singelyn 2005; Thomas 2009), femoral nerve block with sciatic nerve block for one group and without sciatic nerve block for another group (continuous block: Celidonio 2008), fascia iliaca block (single injection: Cucereanu Badica 2010; Forget 2009; Kearns 2011; Murray 2005; Shariat 2013; Stevens 2007; Van Herreweghe 2015; continuous block: Borisov 2012), femoral nerve block plus lateral femoral cutaneous nerve block (single injection: Uhrbrand 1992), femoral nerve block plus obturator nerve block plus lateral femoral cutaneous nerve block (single injection: Jensen 2012), obturator nerve block (single injection: Ginz 2000) or lateral femoral cutaneous nerve block (single injection: Thybo 2016). For fascia iliaca blocks, two studies punctured the skin above the inguinal ligament (Stevens 2007; Van Herreweghe 2015).

Block technique

Psoas compartment, femoral or 3‐in‐1 nerve blocks were performed using a nerve stimulator in 23 studies (Anis 2011; Becchi 2008; Bhatia 2008; Biboulet 2004; Bichel 1998; Duarte 2009; Fournier 1998; Frassanito 2008; Fredrickson 2015; Ginz 2000; Goytizolo 2016; Köroğlu 2008; Kratz 2015; Marino 2009; Nicholson 2002; Siddiqui 2007; Singelyn 2005; Souron 2003; Stevens 2000; Striebel 1993; Türker 2003; Uhrbrand 1992; Wiesmann 2014). Three studies reported using ultrasound guidance alone (Kearns 2011; Shariat 2013; Thybo 2016) and two studies used ultrasound guidance plus nerve stimulator (Aksoy 2014; Nishio 2014). One study used fluoroscopic guidance plus nerve stimulator (Utebey 2009). Landmarks were used in four studies (Borisov 2012; Forget 2009; Stevens 2007; Twyman 1990). One study used injection under direct vision during surgery (Green 2014). Techniques were unspecified in 10 studies (Celidonio 2008; Chen 2015; Cucereanu Badica 2010; Gelmanas 2010; Jensen 2012; Murray 2005; Nohel 2011; Thomas 2009; Van Herreweghe 2015; Türker 2003).

Timing of injection

Local anaesthetics in peripheral nerve blocks were injected before surgical incision in 30 studies (Aksoy 2014; Bhatia 2008; Bichel 1998; Cucereanu Badica 2010; Duarte 2009; Forget 2009; Fouad 2010; Fournier 1998; Frassanito 2008; Fredrickson 2015; Ginz 2000; Goytizolo 2016; Kai 2010; Kearns 2011; Kendrisic 2013; Köroğlu 2008; Kratz 2015; Murray 2005; Nicholson 2002; Siddiqui 2007; Singelyn 2005; Souron 2003; Stevens 2000; Striebel 1993; Thomas 2009; Türker 2003; Twyman 1990; Uhrbrand 1992; Utebey 2009; Wiesmann 2014). Local anaesthetic was injected after surgical incision in nine studies (Anis 2011; Becchi 2008; Biboulet 2004; Borisov 2012; Chen 2015; Green 2014; Marino 2009; Shariat 2013; Thybo 2016). Timing of block performance was unclear in 12 studies (Asano 2010; Bakalov 2016; Celidonio 2008; Gelmanas 2010; Jensen 2012; Lončar 2016; Marshall 2008; Nishio 2014; Nohel 2011; Saksena Shrivastava 2011; Stevens 2007; Van Herreweghe 2015).

Local anaesthetics in peripheral nerve blocks

Local anaesthetics administered as single injection blocks (psoas, femoral or 3‐in‐1) included bupivacaine (12 studies: Aksoy 2014; Anis 2011; Biboulet 2004; Fournier 1998; Ginz 2000; Goytizolo 2016; Green 2014; Köroğlu 2008; Kratz 2015; Stevens 2000; Twyman 1990; Wiesmann 2014), levobupivacaine (3 studies: Forget 2009; Kearns 2011; Murray 2005), ropivacaine (7 studies: Cucereanu Badica 2010; Frassanito 2008; Jensen 2012; Shariat 2013; Souron 2003; Thybo 2016; Van Herreweghe 2015) or a mixture of lidocaine and bupivacaine (2 studies: Nicholson 2002; Uhrbrand 1992). Volumes varied from 8 mL to 40 mL.

If potency equivalences were assumed as: lidocaine = 1, mepivacaine = 0.8, ropivacaine = 3, levobupivacaine = 3.9 and bupivacaine = 4 (Berde 2009), doses in lidocaine equivalents varied from 75 mg to 800 mg and concentrations from 8 mg/mL to 22.5 mg/mL.

Local anaesthetics used as infusions in continuous blocks included bupivacaine 5 studies: (Bichel 1998; Gelmanas 2010; Siddiqui 2007; Singelyn 2005; Türker 2003), lidocaine (Thomas 2009), ropivacaine (6 studies: Becchi 2008; Borisov 2012; Duarte 2009; Fredrickson 2015; Marino 2009; Nishio 2014) or were unspecified (Chen 2015; Nohel 2011). Concentrations in lidocaine equivalents varied from 4.5 to 6 mg/mL. Doses varied form 14 to 66 mg/hour (lidocaine equivalents) and volumes from 3 to 11 mL/hour. Infusions were maintained for 24 hours (Thomas 2009), 36 hours (Siddiqui 2007), 48 hours (7 studies: Becchi 2008; Bichel 1998; Borisov 2012; Celidonio 2008; Duarte 2009; Gelmanas 2010; Singelyn 2005), 72 hours (Chen 2015) or an unspecified duration (Nohel 2011). Local anaesthetics were administered as patient‐controlled analgesia in three studies (Chen 2015; Duarte 2009; Fredrickson 2015). Twelve studies reported that local anaesthesia was administered by continuous infusion (Becchi 2008; Bichel 1998; Borisov 2012; Celidonio 2008; Gelmanas 2010; Marino 2009; Nishio 2014; Nohel 2011; Siddiqui 2007; Singelyn 2005; Thomas 2009; Türker 2003).

Local anaesthetics injected for repeated doses included bupivacaine (Striebel 1993; Utebey 2009) and lidocaine (Bhatia 2008). Trials re‐injected only one dose varying from 4.9 to 20 mg/mL in lidocaine equivalent concentrations with volumes between 13 mL and 30 mL for doses between 130 mg and 600 mg (in lidocaine equivalents).

Adjuvants in peripheral nerve blocks

For single injection blocks, adjuvant drugs added to local anaesthetics included epinephrine (4 studies: Aksoy 2014; Fournier 1998; Stevens 2000; Uhrbrand 1992), and clonidine (6 studies: Anis 2011; Biboulet 2004; Bichel 1998; Celidonio 2008; Kratz 2015; Wiesmann 2014). For one trial (Anis 2011), clonidine was added for one group while no adjuvant was added for another group. Epinephrine plus clonidine was administered in two studies (Forget 2009; Stevens 2007). No adjuvants were added to the local anaesthetic blocks in 16 studies (Anis 2011; Cucereanu Badica 2010; Goytizolo 2016; Green 2014; Jensen 2012; Frassanito 2008; Ginz 2000; Kearns 2011; Köroğlu 2008; Murray 2005; Nicholson 2002; Shariat 2013; Souron 2003; Thybo 2016; Twyman 1990; Van Herreweghe 2015).

For continuous blocks, fentanyl was added in two studies (Gelmanas 2010; Türker 2003), and no adjuvants were added in 11 studies (Becchi 2008; Borisov 2012; Chen 2015; Duarte 2009; Fredrickson 2015; Marino 2009; Nishio 2014; Nohel 2011; Siddiqui 2007; Singelyn 2005; Thomas 2009).

There was no adjuvant added in trials where a repeated dose was administered (Bhatia 2008; Striebel 1993; Utebey 2009).

1.1.1 Pain at rest and with movement on arrival in the postoperative care unit

Nine trials (429 participants) evaluated the effect of peripheral nerve blocks on pain at rest on arrival in the postoperative care unit: psoas compartment block (Biboulet 2004; Goytizolo 2016; Utebey 2009) or femoral nerve block (Köroğlu 2008; Kratz 2015; Nishio 2014 (2 subgroups depending on comparison versus opioids or non‐opioids); Stevens 2000; Striebel 1993; Wiesmann 2014). Random sequence generation was rated as at low risk of bias for Goytizolo 2016; Köroğlu 2008; Kratz 2015; Nishio 2014; Striebel 1993 and Wiesmann 2014; and this domain was judged as unclear for Biboulet 2004; Stevens 2000 and Utebey 2009. Allocation concealment was rated as at low risk of bias for Goytizolo 2016; Kratz 2015; Nishio 2014 and Wiesmann 2014; and as unclear for Biboulet 2004; Köroğlu 2008; Stevens 2000; Striebel 1993 and Utebey 2009. Blinding of participants or personnel taking care of the participants was rated as at low risk of bias for Köroğlu 2008 and Stevens 2000; as unclear for Biboulet 2004 and Goytizolo 2016; and as at high risk for Kratz 2015; Nishio 2014; Striebel 1993; Utebey 2009 and Wiesmann 2014. Blinding of outcome assessor was rated as at low risk of bias for Biboulet 2004; Goytizolo 2016; Kratz 2015; Köroğlu 2008 and Stevens 2000; as unclear for Nishio 2014; Striebel 1993; Utebey 2009 and Wiesmann 2014; and at high risk for Kratz 2015. Incomplete outcome data (attrition bias) was rated as at low risk of bias for all trials included in this analysis. Other risks were judged as at low risk of bias for Goytizolo 2016; Köroğlu 2008; Nishio 2014; Stevens 2000; Striebel 1993; Utebey 2009 and Wiesmann 2014. Biboulet 2004 and Kratz 2015 were judged as unclear because data were not analysed according to the intention‐to‐treat principle. Furthermore, for Kratz 2015, although this was not statistically significant, pain scores measured before the intervention were higher for participants included in the femoral nerve block group (4.2 ± 2.9 versus 2.9 ± 3.1).

Peripheral nerve blocks decreased pain at rest on arrival in the postoperative care unit (SMD ‐1.12, 95% CI ‐1.67 to ‐0.56; I² 84%). Egger's regression intercept showed no evidence of small‐study effect. Analysis using Duval and Tweedie's trim and fill technique indicated imbalance of two studies missing to left of mean for an adjusted point estimate (SMD ‐1.44, 95% CI ‐2.09 to ‐0.80; random‐effects model). The effect was similar for psoas compartment versus femoral nerve block (I² 0%; Analysis 1.1). Volume of solution injected as a loading dose was unknown for Nishio 2014. Excluding this study, the effect was proportional to the volume of local anaesthetic injected (Figure 4; P < 0.00001) and to the dose in lidocaine equivalents (Figure 5; P = 0.0001). Taking a study at low risk of bias for which the standard deviation (SD) is known and with a typical SD in the control group (Wiesmann 2014; SD = 2.2), the difference would be equivalent to 3.2 on a scale from 0 to 10. For a large trial and using the same SD, 68 participants would be required to decrease VAS scores from 5 to 3.5 (Inference for means calculator) (alpha 0.05; beta 0.2; 2‐sided test).

Figure 4

---

---

Comparison peripheral nerve block versus no block or sham block or systemic analgesia

Meta‐regression analysis of effect of peripheral nerve block on pain on arrival in postoperative care unit versus volume of the local anaesthetic injected as the loading dose. The effect size was proportional to the concentration used (P < 0.00001).

The largest effect size (standardized mean difference (SMD) ‐4.56, 95% CI ‐5.91 to ‐3.20) was obtained by Köroğlu 2008 where a 3‐in‐1 block with 40 mL 0.25% bupivacaine was compared to a sham block and postoperative analgesia was completed with IV patient‐controlled analgesia with tramadol plus IV meperidine on request.

Figure 5

---

---

Comparison peripheral nerve blocks versus no block or sham block or systemic analgesia.

Meta‐regression of pain on arrival in postoperative care unit versus local anaesthetic dose in lidocaine equivalent injected as loading dose. The effect size was proportional to the dose injected (P = 0.0001).

The largest effect size (SMD ‐4.56, 95% CI ‐5.91 to ‐3.20) was obtained by Köroğlu 2008 where a 3‐in‐1 block with 40 mL 0.25% bupivacaine was compared to a sham block and postoperative analgesia was completed with IV patient‐controlled analgesia with tramadol plus IV meperidine on request.

For pain at rest on arrival in the postoperative care unit, we downgraded evidence quality by one level for risk of bias because 50% or more of the included trials were judged at unclear or high risk of bias for allocation concealment, blinding of the outcome assessor or both. We did not downgrade evidence quality for inconsistency because reasonable explanation was found to explain heterogeneity. We did not downgrade for indirectness because all trials were direct comparisons performed with the population of interest and this was not a surrogate marker. We did not downgrade for imprecision because the optimal information size was achieved. We did not downgrade for publication bias because applying a correction would not change the conclusion. We rated the quality of the body of evidence as moderate.

Köroğlu 2008 provided data for pain with movement at arrival in the postoperative care unit (SMD ‐3.62, 95% CI ‐4.78 to ‐2.46).

1.1.2 Pain at rest and with movement from 0.5 to 2 hours after surgery

Nine trials (438 participants) evaluated pain at rest from 0.5 to 2 hours after surgery: femoral nerve block (Kratz 2015; Lončar 2016; Striebel 1993; Wiesmann 2014), fascia iliaca block (Lončar 2016; Shariat 2013) or psoas compartment block (Anis 2011; Green 2014; Siddiqui 2007; Stevens 2000) (SMD ‐0.67, 95% CI ‐1.06 to ‐0.29; I² 71%; Analysis 1.2). In Shariat 2013, participants were recruited only if pain scores remained at ≥ 3 (scale from 0 to 10) despite intravenous patient‐controlled analgesia with morphine for one hour in the postoperative care unit. Egger's regression intercept showed that a small‐study effect could participate to the high I² statistic value (P < 0.05; 2‐tailed). Analysis using Duval and Tweedie's trim and fill method indicated imbalance of two studies missing to right of mean for an adjusted point estimate (SMD ‐0.50, 95% CI ‐0.91 to ‐0.09; random‐effects model). Psoas compartment block decreased pain at rest from 0.5 to 2 hours after surgery (SMD ‐0.91, 95% CI ‐1.20 to ‐0.62; I² 0%) but we did not find a difference for a femoral nerve block: (SMD ‐0.28, 95% CI ‐0.85 to 0.28; I² 68%) or fascia iliaca block (SMD ‐1.21, 95% CI ‐3.34 to 0.92); heterogeneity (I²) between subgroups 49% (P value = 0.14). When Siddiqui 2007 was considered (control group SD = 1.81), the difference in pain score for psoas compartment blocks was equivalent to 1.7 (on a scale from 0 to 10) and 46 participants would be required in a large trial to achieve decrease from 5 to 3.5 in VAS scores (alpha 0.05; beta 0.2; 2‐sided test).

We found no data for pain with movement from 0.5 to 2 hours after surgery or block placement for trials where the block was performed for postoperative analgesia.

1.1.3 Pain at rest and with movement from 4 to 6 hours after surgery

We included 13 trials (599 participants) that evaluated pain at rest between 4 and 6 hours after surgery (SMD ‐0.62, 95% CI ‐0.92 to ‐0.32; I² 66%; Analysis 1.3). Egger's regression intercept showed no evidence of small‐study effect (2‐sided test). Analysis using Duval and Tweedie's trim and fill method indicated imbalance of one trial missing to right of mean for an adjusted point estimate (SMD ‐0.56, 95% CI ‐0.86 to ‐0.27; random‐effects model).

Femoral nerve block was used in five studies (223 participants) (Kratz 2015; Nishio 2014; Singelyn 2005; Striebel 1993; Wiesmann 2014) (SMD ‐0.30, 95% CI ‐0.98 to 0.38; I² 81%). Two studies (Kratz 2015; Wiesmann 2014) added clonidine to the loading dose of local anaesthetic. Singelyn 2005 added epinephrine, Striebel 1993 added no adjuvant, and the information was not reported by Nishio 2014. Fascia iliaca with a puncture site above (Stevens 2007), or below the inguinal ligament (Cucereanu Badica 2010) was used for two trials (SMD ‐0.36, 95% CI ‐1.06 to 0.33; I² 54%).

Psoas compartment block was used in six trials (303 participants) (SMD ‐0.88, 95% CI ‐1.12 to ‐0.64; I² 0%); I² for heterogeneity between subgroups was 50%. The difference for psoas compartment block was equivalent to 1.4 (on a scale from 0 to 10) (Siddiqui 2007: control group SD = 2.2). Further analysis from data provided by Siddiqui 2007 indicated that 68 participants would be required in a simple trial to decrease VAS scores from 5 to 3.5 (alpha 0.05; beta 0.2; 2‐sided test).

Three trials (Becchi 2008; Cucereanu Badica 2010; Singelyn 2005, 120 participants) evaluated pain scores with movement from 4 to 6 hours (SMD ‐0.46, 95% CI ‐1.12 to 0.20; I² 64%; Analysis 1.4). Egger's regression intercept showed no evidence of small‐study effect. Analysis using Duval and Tweedie's trim and fill method showed no evidence of publication bias. Femoral nerve block (Singelyn 2005) (SMD 0.00, 95% CI ‐0.86 to 0.86), fascia iliaca block with the puncture site below the inguinal ligament (Cucereanu Badica 2010) (SMD ‐0.18, 95% CI ‐0.91 to 0.55) or psoas compartment block (Becchi 2008) (SMD ‐1.00, 95% CI ‐1.51 to ‐0.49) were used (I² for heterogeneity between subgroups 64%).

1.1.4 Pain at rest and with movement at 24 hours after surgery

Six trials (Becchi 2008; Kratz 2015; Siddiqui 2007; Singelyn 2005; Stevens 2007; Wiesmann 2014, 303 participants) evaluated pain scores at rest after single injection block with clonidine (Kratz 2015; Stevens 2007; Wiesmann 2014) or continuous peripheral nerve block (Becchi 2008; Siddiqui 2007; Singelyn 2005) (SMD ‐0.66, 95% CI ‐1.05 to ‐0.28; I² 61%; Analysis 1.5). Egger's regression intercept showed no evidence of small‐study effect. Analysis using Duval and Tweedie's trim and fill method showed no evidence of publication bias. Femoral nerve block (Kratz 2015; Singelyn 2005; Wiesmann 2014) (SMD ‐0.68, 95% CI ‐1.27 to ‐0.10; I² 63%), fascia iliaca block (SMD 0.00, 95% CI ‐0.59 to 0.59) or psoas compartment block (SMD ‐0.97, 95% CI ‐1.37 to ‐0.56; I² 0%) were used; I² for heterogeneity between subgroups 72%. Considering a trial with low risk of bias and typical SD (Shariat 2013; control group SD = 2.2), the difference would be equivalent to 1.5 on a scale from 0 to 10.

Three trials with 317 participants (Becchi 2008; Marino 2009 (2 subgroups); Singelyn 2005), evaluated pain scores with movement after continuous peripheral nerve block (there were no data for single injection block for this outcome) at 24 hours after surgery (SMD ‐0.71, 95% CI ‐1.26 to ‐0.17; I² 78%; Analysis 1.6). Egger's regression intercept showed no evidence of small‐study effect. Analysis using Duval and Tweedie's trim and fill method showed no evidence of publication bias (random‐effects model). The effect was seen with psoas compartment block (SMD ‐1.13, 95% CI ‐1.46 to ‐0.80; I² 0%) but we did not find a difference with femoral nerve block (SMD ‐0.24, 95% CI ‐0.59 to 0.12; I² 0%; I² for heterogeneity for subgroups 92%). Taking a trial at low risk of bias and with a typical SD (Marino 2009; control group SD = 2.3), the difference would be equivalent to 1.6 on a scale from 0 to 10.

1.1.5 Pain at rest and with movement at 48 hours after surgery

Data were available for two trials (Becchi 2008; Singelyn 2005; 93 participants) that evaluated pain at rest at 48 hours after surgery for participants who benefited from continuous peripheral nerve block (SMD ‐0.80, 95% CI ‐1.35 to ‐0.25; I² 29%). Singelyn 2005 used continuous femoral nerve block and Becchi 2008 used continuous psoas compartment block (I² for heterogeneity between subgroups 29%; Analysis 1.7). Taking a trial at low risk of bias and with a typical SD (Duarte 2009; control group SD = 1.80), the difference would be equivalent to 1.44 on a scale from 0 to 10.

Three trials (Becchi 2008; Marino 2009; Singelyn 2005; 317 participants) evaluated pain with movement at 48 hours after surgery in participants with continuous peripheral nerve block (SMD ‐0.62, 95% CI ‐1.13 to ‐0.11; I² 75%; Analysis 1.8). Egger's regression intercept showed no evidence of small‐study effect. Analysis using Duval and Tweedie's trim and fill method showed no evidence of publication bias (random‐effects model). Continuous compartment block reduced pain scores with movement at 48 hours after surgery (SMD ‐0.93, 95% CI ‐1.25 to ‐0.61; I² 0%). We did not find a difference with continuous femoral nerve block (SMD ‐0.16, 95% CI ‐0.60 to 0.28; I² 17%; I² for heterogeneity between subgroups 87%). Taking a trial at low risk of bias and with a typical SD (Singelyn 2005; SD of the control group 1.80), the difference would be equivalent to 1.12 on a scale from 0 to 10.

1.3.1 Complications from peripheral nerve blocks (damage to surrounding structures at site of nerve block)

Becchi 2008 reported one instance of catheter dislodgment. Marino 2009 reported that catheters were not functional (N = 2) or were occluded (N = 2). Epidural extension of psoas compartment block was reported by Biboulet 2004 (N = 4), Marino 2009 (N = 5), Stevens 2000 (N = 3) and Utebey 2009 (N = 2). Uhrbrand 1992 reported one local haematoma. One persistent (at 3 months) delayed paresis was reported in Siddiqui 2007 (see details above).

1.3.2 Complications from systemic analgesia or general anaesthesia (other complications as detailed in each study)

Siddiqui 2007 reported that one participant developed respiratory depression from systemic analgesia.

1.4.a Gastrointestinal: nausea and vomiting, constipation, ileus

We included five trials (282 participants) that reported results for nausea (Goytizolo 2016; Jensen 2012), nausea and vomiting (Stevens 2007; Wiesmann 2014) or vomiting (Bakalov 2016) for the first 24 hours (RR 0.86, 95% CI 0.53 to 1.41; I² 17%; Analysis 1.10). Egger's regression intercept showed no evidence of small‐study effect. Analysis using Duval and Tweedie's trim and fill method indicated imbalance of one trial missing to right of mean for an adjusted point estimate (RR 0.96, 95% CI 0.59 to 1.58).

We could not obtain data suitable for analysis for constipation or ileus.

1.4.b Pulmonary: pneumonia, bronchitis

No included studies reported this outcome.

1.4.c Cardiovascular: hypotension, myocardial infarction, blood loss and blood transfusion

Four trials (150 participants) reported on the risk of hypotension (RR 0.95, 95% CI 0.50 to 1.81; I² 0%) (Becchi 2008; Siddiqui 2007; Singelyn 2005; Utebey 2009). Egger's regression intercept showed no evidence of small‐study effect. Analysis using Duval and Tweedie's trim and fill method showed no evidence of publication bias.

We could not obtain data suitable for analysis for myocardial infarction.

Based on data from eight trials (323 participants), we did not find differences for intra‐operative blood loss (MD ‐52.47 mL, 95% CI ‐170.88 to 65.95; I² 79%; Analysis 1.12) (Cucereanu Badica 2010; Kratz 2015; Nishio 2014; Siddiqui 2007; Stevens 2000; Twyman 1990; Utebey 2009; Wiesmann 2014). Egger's regression intercept showed no evidence of small‐study effect. Data for amount of blood transfused were available for three trials (Nicholson 2002; Singelyn 2005; Utebey 2009; 82 participants) (MD ‐0.59 units, 95% CI ‐1.25 to 0.08; I² 48%; Analysis 1.13).

1.4.d Neurological: acute confusional state, drowsiness, cerebrovascular accident, postoperative cognitive dysfunction

Only Marino 2009 provided results for acute confusional state (RR 0.10, 95% CI 0.02 to 0.54). Marino 2009 was judged as at low risk of bias for random sequence generation, incomplete outcome data (attrition bias), selective reporting (reporting bias) and other bias; at unclear risk of bias for allocation concealment and blinding of outcome assessment; and as at high risk of bias for blinding of participant or blinding of personnel taking care of the participant, The number needed to treat for an additional beneficial outcome (NNTB) = 12 (95% CI 11 to 22). In a large trial, 3158 participants (1579 per group) would be required to eliminate a 25% difference (alpha 0.05; beta 0.2; 1‐sided test).

For grade of the evidence for reduced risk of confusional state, we downgraded by two levels because the included trial was at unclear risk of bias for allocation concealment and blinding of outcome assessor and as at high risk for blinding of participant or blinding of personnel taking care of the participant. Inconsistency and publication bias could not be assessed. The trial was a direct comparison. We downgraded the quality by two levels because the optimal information was not achieved (low number of trials/participants). We rated the level as very low.

Three trials (Jensen 2012; Köroğlu 2008; Nishio 2014; 86 participants) provided results for drowsiness (sedated) up to 48 hours (RR 0.40, 95% CI 0.13 to 1.18; I² 0%; Analysis 1.14). Egger's regression intercept showed no evidence of small‐study effect. Analysis using Duval and Tweedie's trim and fill method indicated imbalance of one trial missing to left of mean for an adjusted point estimate (RR 0.39, 95% CI 0.14 to 1.06). Based on an incidence of 26% of excessive sedation (drowsiness) in the control groups, 1304 (652 per group) would be required in a large trial to eliminate a 25% difference (alpha 0.05; beta 0.2; 2‐sided test).

We found no data suitable for analysis for cerebrovascular accident or postoperative cognitive dysfunction.

1.4.e Thromboembolic complications: deep venous thrombosis or pulmonary embolism

We could not obtain data suitable for analysis for deep venous thrombosis.

Chen 2015 reported no pulmonary emboli in 88 participants.

1.4.f Other medical complications

1.4.f.1 Risk of pruritus

Four trials (Goytizolo 2016; Köroğlu 2008; Marino 2009; Siddiqui 2007; 379 participants) reported on the risk of pruritus during the first 48 hours after surgery (RR 0.34, 95% CI 0.09 to 1.27; I² 66%; Analysis 1.15). Random sequence generation was judged as at low risk for Goytizolo 2016; Köroğlu 2008; Marino 2009 and Siddiqui 2007. Allocation concealment was judged as at low risk for Goytizolo 2016 and Siddiqui 2007; and as at unclear risk for Köroğlu 2008 and Marino 2009. Blinding of participant and/or personnel taking care of the participant was judged as at low risk of bias for Köroğlu 2008; as at unclear risk for Goytizolo 2016 and as at high risk of bias for Marino 2009 and Siddiqui 2007. Blinding of outcome assessment was judged as at low risk for Goytizolo 2016 and Köroğlu 2008; as at unclear risk for Marino 2009; and as at high risk for Siddiqui 2007. Incomplete outcome data (attrition bias) was judged as at low risk for Goytizolo 2016; Köroğlu 2008; Marino 2009 and Siddiqui 2007. Selective reporting (reporting bias) was judged as at low risk for Goytizolo 2016; Köroğlu 2008; Marino 2009 and Siddiqui 2007. Other bias were judged as at low risk for Goytizolo 2016; Köroğlu 2008; Marino 2009 and Siddiqui 2007. Egger's regression intercept showed no significant evidence of small‐study effect (2‐sided test). Analysis using Duval and Tweedie's trim and fill method indicated imbalance of two trials missing to right of mean for an adjusted point estimate (RR 0.85, 95% CI 0.25 to 2.81; random‐effects model). The risk was reduced when continuous block was used (RR 0.16, 95% CI 0.04 to 0.70; I² 31%; 2 trials, 259 participants) but not when single injection block was used (RR 0.92, 95% CI 0.50 to 1.68; I² 0%). Considering a basal rate of 20% (control group), the NNTB for continuous nerve blocks would be 4 (95% CI 4 to 8). Considering a basal rate of 20%, 1428 participants (714 per group) would be required in a large trial to eliminate a 25% difference (alpha 0.05; beta 0.2; 1‐sided test).

For grade of evidence for reduced risk of pruritus for continuous nerve blocks, we downgraded by two levels for risk of bias because 75% of the included trials were judged as at unclear/high risk for allocation concealment and/or blinding of outcome assessors. There was no significant inconsistency for this subgroup (I² < 50%). We included direct comparisons only. We downgraded the level by one because the optimal information size was not achieved. Publication bias could not be assessed. We rated the level of evidence as very low.

1.4.f.2 Risk of respiratory depression

Three trials (Becchi 2008; Siddiqui 2007; Wiesmann 2014; 184 participants) reported risk of respiratory depression (RD ‐0.01, 95% CI ‐0.05 to 0.03; I² 0%). Egger's regression intercept showed no significant evidence of small‐study effect (2‐sided test). Analysis using Duval and Tweedie's trim and fill method showed no evidence of publication bias.

1.5.a Hospital length of stay

Three trials (Marino 2009; Singelyn 2005; Thybo 2016; 348 participants (the forest plot (Analysis 1.17) displays 349 participants due to subgroup division) reported length of hospital stay (SMD ‐0.56, 95% CI ‐0.91 to ‐0.21; I² 54%; Analysis 1.17) (Marino 2009 (2 subgroups); Singelyn 2005; Thybo 2016). Random sequence generation was judged as at low risk for Marino 2009; Singelyn 2005 and Thybo 2016. Allocation concealment was judged as at low risk for Thybo 2016; and as at unclear risk for Marino 2009 and Singelyn 2005. Blinding of participant and/or personnel taking care of the participants was judged as at low risk for Thybo 2016 and as at high risk for Marino 2009 and Singelyn 2005. Blinding of outcome assessment for judged as at low risk for Thybo 2016; and as at unclear risk of bias for Marino 2009 and Singelyn 2005. Incomplete outcome data, selective reporting and other bias were judged as at low risk for Marino 2009; Singelyn 2005; and as at unclear risk for Thybo 2016. Egger's regression intercept showed no evidence of small‐study effect. Analysis using Duval and Tweedie's trim and fill method indicated imbalance of one trial missing to right of mean for an adjusted point estimate (SMD ‐0.46, 95% CI ‐0.78 to ‐0.13; random‐effects model). Continuous femoral nerve or psoas compartment blocks reduced length of hospital stay (SMD ‐0.75, 95% CI ‐1.02 to ‐0.48: I² 0%). We found no effect from single injection femoral lateral cutaneous nerve block (SMD ‐0.13, 95% CI ‐0.52 to 0.26). Taking one study (Marino 2009; SD of control 1 day), the difference would be equivalent to 0.75 day.

We assessed evidence quality of continuous block only for hospital length of stay. We downgraded the level by two because 75% or more of the included trials were rated as unclear/high risk for allocation concealment and/or blinding of outcome assessor. There was no inconsistency for this subgroup. We included direct comparisons only. We downgraded by one level for low number of trials included. Publication bias could not be assessed. We rated the level of evidence as very low.

1.5.b Costs of treatment

We could not obtain data suitable for analysis for this outcome.

1.5.c Re‐hospitalization due to pain

We could not obtain data suitable for analysis for this outcome.

1.5.d Re‐hospitalization due to any other reason, including re‐operation

We could not obtain data suitable for analysis for this outcome.

1.7.1 Time to start rehabilitation, e.g. time to sit up in bed

Thybo 2016 (60 participants) provided results for time to first mobilization (SMD ‐0.30, 95% CI ‐0.81 to 0.22).

1.7.2. Time to achieve rehabilitation milestones, e.g. transfer unassisted in and out of bed and ability to walk unassisted with a walker on a level surface

1.7.2.1 Walking

Based on data from two trials (Green 2014; Marino 2009; 272 participants) we did not find a difference in numbers of participants who walked on postoperative day 1: (RD 0.01, 95% CI ‐0.03 to 0.05; I² 0%; Analysis 1.18). Random sequence generation, incomplete outcome data, selective reporting and other bias were judged as at low risk for Green 2014 and for Marino 2009. Allocation concealment was judged as at unclear risk for Green 2014 and for Marino 2009. Blinding of participant and/or personnel taking care of participant was judged as unclear risk for Green 2014 and as at high risk for Marino 2009. Blinding of outcome assessment was judged as at low risk for Green 2014 and at as unclear risk for Marino 2009. Marino 2009 also reported results for numbers of participants who could walk more than 12.2 m at 48 hours (RR 0.51, 95% CI 0.03 to 7.88) for continuous femoral nerve block and (RR 11.15, 95% CI 0.67 to 184.61) for continuous psoas compartment block (P value for heterogeneity between subgroups 0.12). Given that an overall percentage of 11% of participants were able to walk more than 12.2 m at 48 hours, 618 participants (309 per group) would be required in a large trial to eliminate a difference of 50% (alpha 0.05; beta 0.2; 1‐sided test).

We rated the level of evidence for number of participants who could walk on postoperative day 1. We downgraded the level by two for risk of bias because 75% of included trials were at unclear/high risk for allocation concealment and/or blinding of outcome assessor. There was no inconsistency. We included direct comparisons only. We downgraded by one level for imprecision. Publication bias could not be assessed. We rated the evidence as very low.

1.7.2.2 Hip flexion

Two trials (Biboulet 2004 ‐ two subgroups; Singelyn 2005; 68 participants) reported results for hip flexion at seven days after surgery (MD ‐1.40°, 95% CI ‐9.04° to 6.25°; I² 43 %; Analysis 1.19.

1.7.2.3 Delayed rehabilitation

Nishio 2014 reported delayed rehabilitation due to drowsiness for 1/10 participants receiving continuous femoral nerve blocks versus 5/18 participants for systemic analgesia (RR 0.28, 95% CI 0.04 to 2.06).

2.1.1 Pain at rest and with movement on arrival in the postoperative care unit

Fredrickson 2015 (50 participants) compared single injection psoas compartment block with 40 mL 0.5% bupivacaine to single injection spinal injection of 2 mL bupivacaine 0.5% with morphine 0.1 mg. Worst pain scores at rest in the postoperative care unit were higher for participants who received psoas compartment block (SMD 0.99, 95% CI 0.41 to 1.58).

Four trials (Bichel 1998; Duarte 2009; Türker 2003; Utebey 2009; 118 participants) compared continuous femoral nerve block (Bichel 1998), continuous psoas compartment block (Duarte 2009; Türker 2003), or repeated dose psoas compartment block (Utebey 2009) with epidurals. Random sequence generation and allocation concealment were judged as at low risk of bias for Duarte 2009 and as at unclear risk for Bichel 1998; Türker 2003 and Utebey 2009. Blinding of participant and/or personnel taking care of the participants was judged as at unclear risk for Duarte 2009 and Türker 2003; and as at high risk of bias for Bichel 1998 and Utebey 2009. Blinding of outcome assessment was judged as at low risk for Duarte 2009; Türker 2003; and as at unclear risk for Bichel 1998 and Utebey 2009. Incomplete outcome data was judged as at low risk for Bichel 1998; Duarte 2009; Türker 2003 and Utebey 2009. Selective reporting was judged as at low risk for Bichel 1998; Türker 2003 and Utebey 2009; and as at high risk for Duarte 2009. Other bias were judged as at low risk for Bichel 1998; Türker 2003 and Utebey 2009; and as at unclear risk for Duarte 2009. We found no difference for pain at rest on arrival in the postoperative care unit (MD 0.39 95% CI ‐0.15 to 0.94; I² 0%; Analysis 2.1). There was no evidence of small‐study effect. Analysis using Duval and Tweedie's trim and fill method indicated imbalance of one trial missing to the right of mean for an adjusted point estimate (MD 0.40, 95% CI ‐0.14 to 0.94). In Türker 2003, participants had partial hip replacement. There was no mention concerning hip fracture in the report but this intervention is often performed for patients with hip fracture. We were unable to contact the authors to establish with certainty if participants had hip fracture. When Türker 2003 was excluded from the analysis, MD was 0.41 (95% CI ‐0.47 to 1.29; I² 0%). Based on results reported byDuarte 2009 (control group SD = 3.3), 76 participants (38 per group), would be required in a large trial to decrease pain scores from 5.5 to 4.

For the level of evidence for continuous block or repeated dose, we downgraded by one level for risk of bias because 50% or more of the included trials were judged as being unclear/high risk for allocation concealment and/or blinding of outcome assessor. The optimal information size was achieved. Correcting for the possibility of publication bias would not change the conclusion. We rated the quality of evidence as moderate.

Two trials compared continuous femoral nerve (Bichel 1998) or psoas compartment block (Türker 2003) with continuous epidural for pain with movement. We did not find a difference for pain with movement on arrival in the postoperative care unit (MD 0.08, 95% CI ‐0.46 to 0.62; I² 23%; fixed‐effect model; Analysis 2.2). When Türker 2003 was excluded from the analysis, MD was 0.43 (95% CI ‐0.37 to 1.23).

2.1.2 Pain at rest and with movement from 0.5 to 2 hours after surgery

Two trials (Celidonio 2008; Forget 2009; 66 participants) compared single injection fascia iliaca block with single injection epidural block (Forget 2009), or continuous femoral nerve block with or without the addition of single parasacral sciatic nerve block with continuous epidural (Celidonio 2008). Pain at rest from 0.5 to 2 hours after surgery was higher when femoral nerve or fascia iliaca block without parasacral nerve block was used (MD 3.45, 95% CI 2.01 to 4.90; I² 30%; Analysis 2.3).

Pain with movement at two hours was higher regardless of whether parasacral sciatic nerve block was provided (MD 1.54, 95% CI 0.56 to 2,52) or not (3.50; 95% CI 1.99 to 5.01) (Celidonio 2008).

2.1.3 Pain at rest and with movement from 4 to 6 hours after surgery

Nine trials (398 participants) reported pain at rest from 4 to 6 hours after surgery. Four trials studied a continuous femoral nerve block (alone: Celidonio 2008 1 subgroup; Bichel 1998; Singelyn 2005 or with a parasacral sciatic nerve block: Celidonio 2008 (1 subgroup)). Two trials studied a fascia iliaca block (Borisov 2012; Forget 2009). Three trials studied a psoas compartment block (Duarte 2009; Souron 2003; Türker 2003). Peripheral nerve blocks were compared with an epidural (single injection: Forget 2009; continuous: Bichel 1998; Borisov 2012; Celidonio 2008; Duarte 2009; Singelyn 2005; Türker 2003) in six trials. Two trials compared peripheral nerve blocks with an intrathecal injection. The intrathecal injection contained opioids only in one trial (Souron 2003) or opioids plus a local anaesthetic (Kearns 2011).

Pain was higher with peripheral nerve blocks (SMD 0.41, 95% CI 0.11 to 0.71; I² 49%; Analysis 2.4). There was no statistically significant evidence of small‐study effect, nor significant evidence of publication bias (random‐effects model). When Türker 2003 was excluded from the analysis MD was 0.43 (95% CI 0.10 to 0.75; I² 53%). Excluding Forget 2009 (where piritramide consumption at 1 hour was higher for the peripheral nerve block group: 7.3 ± 3.5 versus 1.8 ± 3.7 mg), SMD was 0.55 (95% CI 0.34 to 0.76; I² 0%). Taking Singelyn 2005 (control group SD = 2.4), this would be equivalent to 1.3.

Seven trials (325 participants) reported results for pain with movement at 4 to 6 hours for continuous femoral nerve block (Bichel 1998; Celidonio 2008; Singelyn 2005), continuous femoral nerve block plus single injection parasacral nerve block (Celidonio 2008), fascia iliaca block (single injection: Kearns 2011; continuous block: Borisov 2012) or continuous psoas compartment block (Duarte 2009; Türker 2003) compared with continuous epidural (Bichel 1998; Borisov 2012; Celidonio 2008; Duarte 2009; Singelyn 2005; Türker 2003) or single injection intrathecal injection of an opioid and a local anaesthetic (Kearns 2011).

Pain with movement at 4 to 6 hours after surgery was higher when a peripheral nerve block was used (SMD 0.53, 95% CI 0.20 to 0.86; I² 48%; Analysis 2.5. There was no evidence of small‐study effect nor publication bias (random‐effects model). When Türker 2003 was excluded from the analysis, MD was 0.63 (95% CI 0.33 to 0.93; I² 29%).The effect size was inversely proportional to the concentration of the local anaesthetic injected as the loading dose, i.e. the difference between the two treatment groups was lower when higher concentrations were used (Figure 6; P = 0.003). Taking Duarte 2009 as a measure of variance (control group SD = 3.0), the difference would be equivalent to 1.6.

Figure 6

---

---

Comparison peripheral nerve block versus neuraxial block.

Meta‐regression analysis of the concentration of local anaesthetic in the initial loading dose in lidocaine equivalent versus pain with movement at 4 to 6 hours after surgery. The higher the concentration, the lower was the difference between the treatment groups (P = 0.003)

2.1.4 Pain at rest and with movement at 24 hours after surgery

Eight trials (328 participants) evaluated pain at rest at 24 hours with single injection fascia iliaca (Forget 2009; Kearns 2011), psoas compartment block (Frassanito 2008), continuous femoral nerve block (Bichel 1998; Celidonio 2008; Singelyn 2005), continuous femoral nerve block plus single injection parasacral nerve block (Celidonio 2008) or continuous psoas compartment block (Duarte 2009; Türker 2003) compared with single injection epidural (Forget 2009) or intrathecal injection of an opioid plus a local anaesthetic (Kearns 2011; Frassanito 2008) or continuous epidural (Bichel 1998; Celidonio 2008; Duarte 2009; Singelyn 2005; Türker 2003). At 24 hours, we did not find a difference between the analgesic techniques for pain at rest (SMD 0.10, 95% CI ‐0.22 to 0.42; I² 45%; Analysis 2.6). There was no evidence of small‐study effect. Analysis using Duval and Tweedie's trim and fill method indicated imbalance of one trial missing to left of mean for an adjusted point estimate 0.04 (95% CI ‐0.27 to 0.35; random‐effects model). When Türker 2003 was excluded from the analysis MD would be 0.08 (95% CI ‐0.28 to 0.44; I² 51%). Results for pain were given on a scale from 0 to 4 where 0 = no pain; 1 = light pain; 2 = moderate pain; 3 = significant pain; 4 = unbearable level of pain in Bichel 1998. Analysis excluding Bichel 1998 indicated SMD 0.20 (95% CI ‐0.03 to 0.43; I² 0%).

Seven trials (217 participants) reported results for pain with movement at 24 hours for continuous femoral nerve block (Bichel 1998; Celidonio 2008; Singelyn 2005), continuous femoral nerve block plus single injection parasacral sciatic nerve block (Celidonio 2008), continuous psoas compartment block (Duarte 2009; Fredrickson 2015; Türker 2003) or single injection fascia iliaca block (Kearns 2011) compared with continuous epidural (Bichel 1998; Celidonio 2008; Duarte 2009; Singelyn 2005; Türker 2003) or with single intrathecal injection of a local anaesthetic plus an opioid (Fredrickson 2015; Kearns 2011). We did not find a difference between peripheral nerve blocks and neuraxial blocks for pain with movement at 24 hours (SMD 0.28, 95% CI ‐0.06 to 0.62); I² 51%; Analysis 2.7) and no statistically significant evidence of small‐study effect nor publication bias. When Türker 2003 was excluded from the analysis, MD was 0.37 (95% CI 0.03 to 0.71; I² 43%).

2.1.5 Pain at rest and with movement at 48 hours after surgery

Seven trials (289 participants) evaluated pain at rest at 48 hours and compared continuous femoral nerve block (Bichel 1998; Celidonio 2008; Singelyn 2005),continuous femoral nerve block with the addition of single injection parasacral sciatic nerve block (Celidonio 2008), continuous psoas compartment block (Duarte 2009; Türker 2003) or single injection fascia iliaca block (Duarte 2009; Türker 2003) with continuous epidural (Bichel 1998; Celidonio 2008; Duarte 2009; Singelyn 2005; Türker 2003), or with single injection epidural (Forget 2009) or with single injection intrathecal injection of an opioid plus local anaesthetic (Kearns 2011). We did not find a difference between the two analgesic techniques (SMD 0.07, 95% CI ‐0.16 to 0.31; I² 22%; Analysis 2.8) and there was no statistically significant evidence of small‐study effect. Analysis using Duval and Tweedie's trim and fill method indicated imbalance of two trials missing to left of mean for an adjusted point estimate (SMD ‐0.01, 95% CI ‐0.23 to 0.21; fixed‐effect model). When Türker 2003 was excluded from the analysis, MD was 0.04 (95% CI ‐0.21 to 0.29; I² 26%).

Seven trials evaluated pain with movement at 48 hours after surgery. Of these, only five (164 participants) were continuous peripheral nerve blocks: femoral nerve block with (Celidonio 2008) or without the addition of single injection parasacral sciatic nerve block (Bichel 1998; Celidonio 2008; Singelyn 2005) or psoas compartment block (Duarte 2009; Türker 2003) with continuous epidural as the comparator. We did not find a difference between the modalities (SMD 0.33, 95% CI ‐0.13 to 0.79; I² 50%; Analysis 2.9) nor statistically significant evidence of small‐study effect. Analysis using Duval and Tweedie's trim and fill method indicated imbalance of one trial missing to left of mean for an adjusted point estimate (SMD 0.21, 95% CI ‐0.24 to 0.67; random‐effects model). When Türker 2003 was excluded from the analysis, MD was 0.41 (95% CI ‐0.12 to 0.94; I² 54%). Neuraxial block was however associated with better pain control when a ropivacaine infusion was used in the peripheral nerve block (SMD 0.80, 95% CI 0.36 to 1.24; I² 0%; P value for heterogeneity between subgroups (ropivacaine versus bupivacaine) = 0.003).

2.3.1 Complications from peripheral nerve blocks

Five non functional catheters were reported: no analgesia with a fascia iliaca block (Borisov 2012: n = 1), kinked catheters with psoas compartment bocks (Türker 2003: n = 2) and psoas compartment block catheter occlusions (blood clots) (Marino 2009: n = 2). Epidural extension of psoas compartment block was reported in Biboulet 2004 (n = 4), Frassanito 2008 (n = 1), Marino 2009 (n = 5), Stevens 2000 (n = 3) and Utebey 2009 (n = 2). Two vascular punctures during psoas compartment blocks were reported by Souron 2003. Six transient neurological symptom events suggestive of local irritation were reported by Fredrickson 2015 after psoas compartment blocks. Kearns 2011 reported a transient femoral nerve palsy that resolved at three months following single injection ultrasound guided (out‐of‐plane technique) fascia iliaca block.

2.3.2 Complications from neuraxial blocks

Fredrickson 2015 reported transient neurological symptoms suggestive of local irritation (n = 8 for spinal anaesthesia). Six epidural catheters problems were reported: five contralateral analgesia (Singelyn 2005: n = 3 and Türker 2003: n = 2) and one kinked catheter (Türker 2003). Fredrickson 2015 also reported two inpatient falls with intrathecal morphine.

2.4.a Gastrointestinal: nausea and vomiting, constipation, ileus

Data were provided in two trials (Fredrickson 2015; Souron 2003; 30 participants). We did not find a difference for vomiting (Souron 2003) or postoperative nausea and vomiting (Fredrickson 2015) (RR 0.79, 95% CI 0.44 to 1.39; I² 0%; Analysis 2.12).

2.4.b Pulmonary: pneumonia, bronchitis

None of the included studies reported these outcomes.

2.4.c Cardiovascular: hypotension, myocardial infarction, blood loss and blood transfusion

Eight trials (Aksoy 2014; Bhatia 2008; Frassanito 2008; Kearns 2011; Singelyn 2005; Souron 2003; Türker 2003; Utebey 2009; 429 participants) reported on hypotension. We did not find a difference in risk of hypotension (RD ‐0.09, 95% CI ‐0.22 to 0.03; I² 83%; Analysis 2.13). Egger's regression intercept showed no statistical evidence of small‐study effect (2‐sided test). Analysis using Duval and Tweedie's trim and fill method indicated imbalance of two trials missing to right of mean for an adjusted point estimate (RD ‐0.02, 95% CI ‐0.13 to 0.10; random‐effects model). When Türker 2003 was excluded from the analysis, RD was ‐0.05 (95% CI ‐0.16 to 0.06; I² 78%). Hypotension risk decreased when peripheral nerve blocks were compared to continuous neuraxial blocks (RD ‐0.34, 95% CI ‐0.48 to ‐0.20; 3 trials, 113 participants; I² 0%; NNTB = 4, 95% CI 3 to 6).

Aksoy 2014 (70 participants) reported no myocardial infarctions.

Five trials (Aksoy 2014; Bichel 1998; Duarte 2009; Türker 2003; Utebey 2009; 188 participants) reported intra‐operative blood loss (MD ‐8.84 mL, 95% CI ‐80.82 to 63.14 mL; I² 83%; Analysis 2.14) with no statistically significant evidence of small‐study effect (2‐sided test). Analysis using Duval and Tweedie's trim and fill method indicated imbalance of one trial missing to left of mean for an adjusted point estimate ‐0.04 (95% CI ‐0.71 to 0.63; random‐effects model). When Türker 2003 was excluded from the analysis, MD was ‐22.75 (95% CI ‐109.88 to 64.38; I² 87%), and when Utebey 2009 was excluded, MD was 39.94 mL (95% CI 18.05 to 61.84 mL; I² 0%).

Three trials with 87 participants provided results for intra‐operative (Utebey 2009) or total number of units transfused (Duarte 2009; Singelyn 2005) (MD ‐0.23 units, 95% CI ‐0.65 to 0.19 units; I² 33%; Analysis 2.15). There was no statistically significant evidence of small‐study effect. Analysis using Duval and Tweedie's trim and fill method indicated imbalance of two trials missing to right of mean for an adjusted point estimate (MD 0.10 unit, 95% CI ‐0.36 to 0.55 unit; random‐effects model).

2.4.d Neurological: acute confusional state, drowsiness, cerebrovascular accident, postoperative cognitive dysfunction

Fredrickson 2015 (50 participants) evaluated the number of participants with confusion at 48 hours (RR 0.29, 95% CI 0.01 to 6.69). For Fredrickson 2015, random sequence generation, allocation concealment, incomplete outcome data, selective reporting and other bias were judged as at low risk of bias and blinding of participant, personnel taking care of the participant and outcome assessment were judged as at high risk of bias.

We downgraded the quality of evidence for acute confusional state by one level for risk of bias for absence of blinding of outcome assessor. Inconsistency and publication bias could not be assessed. The trial is a direct comparison. We downgraded the quality by two levels for imprecision. We rated the quality as very low.

Three trials (176 participants) evaluated drowsiness at 1 hour (Forget 2009) or up to 48 hours after surgery (Kearns 2011; Souron 2003) (RD 0.04, 95% CI ‐0.07 to 0.15; I² 75%; Analysis 2.16). There was no statistically significant evidence of small‐study effect nor publication bias (random‐effects model).

Aksoy 2014 (70 participants) measured mini mental tests 24 hours after surgery (SMD 0.05, 95% CI ‐0.42 to 0.52).

2.4.e Thromboembolic complications: deep venous thrombosis or pulmonary embolism

Two trials (Aksoy 2014; Türker 2003) with 100 participants reported no deep venous thrombosis.

Aksoy 2014 (70 participants) assessed pulmonary embolism and reported no events.

2.4.f Other medical complications

2.4.f.1 Risk of pruritus

From six trials (Frassanito 2008; Fredrickson 2015; Kearns 2011; Souron 2003; Türker 2003; Utebey 2009; 299 participants, peripheral nerve blocks reduced the risk of pruritus compared to neuraxial blocks (RR 0.33, 95% CI 0.19 to 0.58; I² 0%; Analysis 2.18). Random sequence generation was judged as at low risk of bias for Fredrickson 2015; Kearns 2011 and Souron 2003; and as at unclear risk for Frassanito 2008; Türker 2003 and Utebey 2009. Allocation concealment was judged as at low risk of bias for Fredrickson 2015; Kearns 2011; Souron 2003; at unclear risk for Frassanito 2008; Türker 2003 and Utebey 2009. Blinding of participant and personnel taking care of the participants was judged as at low risk for Türker 2003; as at unclear risk for Frassanito 2008; Kearns 2011; Souron 2003; and as at high risk for Fredrickson 2015 and Utebey 2009. Blinding of outcome assessment was judged as at low risk for (Frassanito 2008; Kearns 2011; Souron 2003 and Türker 2003; as at unclear risk for Utebey 2009; and as at high risk for Fredrickson 2015. Incomplete outcome data, selective reporting and others bias were judged as at low risk of bias for Frassanito 2008; Fredrickson 2015; Kearns 2011; Souron 2003; Türker 2003 and Utebey 2009.

There was no statistically significant evidence of small‐study effect nor publication bias (fixed‐effect model). When Türker 2003 was excluded from the analysis, RR was 0.33, 95% CI 0.18 to 0.61; I² 0%). Based on an incidence rate of 26% the NNTB would be 6 (95% CI 5 to 9) and 1028 participants (514 per group) would be required in a large trial to eliminate a 25% difference (alpha 0.05; beta 0.2; 1‐sided test).

For level of quality of the body of evidence for reduced pruritus, we did not downgrade for risk of bias because less than 50% of the trials were judged as at unclear/high risk of bias for allocation concealment and/or blinding of outcome assessor. There was no inconsistency, We included trials with direct comparisons only. We downgraded the level by one for imprecision because the optimal information size was not achieved. There was no evidence of publication bias. We rated the level as moderate.

2.4.f.2 Risk of respiratory depression

Based on data from four trials (Frassanito 2008; Fredrickson 2015; Kearns 2011; Souron 2003; 213 participants), we did not find a difference in the risk of respiratory depression between treatments (RD 0.01, 95% CI ‐0.03 to 0.05; I² 0%; Analysis 2.19). There was no statistically significant evidence of small‐study effect. Analysis using Duval and Tweedie's trim and fill method indicated imbalance of one trial missing to right of mean for an adjusted point estimate (RD 0.00, 95% CI ‐0.02 to 0.03; fixed‐effect model). Considering an incidence rate of 1%, 34,318 participants (17,159 per group) would be required to eliminate a 25% difference (alpha 0.05; beta 0.2; 1‐sided test).

2.5.a Hospital length of stay

From two trials (Duarte 2009; Singelyn 2005) with 64 participants we did not find a difference in hospital length of stay (MD 0.19 days, 95% CI ‐0.39 to 0.77 days; I² 0%; Analysis 2.20). Random sequence generation was judged as at low risk of bias for Duarte 2009 and Singelyn 2005. Allocation concealment was judged as at low risk for Duarte 2009 and as at unclear risk for Singelyn 2005. Blinding of participant and/or personnel taking care of the participant was judged as at unclear risk for Duarte 2009 and as at high risk for Singelyn 2005. Blinding of outcome assessment for judged as at low risk of bias for Duarte 2009 and as at unclear risk for Singelyn 2005. Duarte 2009 was judged as at low risk for incomplete outcome data, at high risk for selective reporting and as at unclear risk for other bias. Singelyn 2005 was judged as at low risk for all those three domains.

We downgraded the level of evidence for hospital length of stay by one level for risk of bias because 50% or more of the trials were judged as at unclear/high risk for allocation concealment and/or blinding of outcome assessor. There was no inconsistency and trials were direct comparisons. We downgraded by one level for low number of trials. Publication bias could not be assessed. We rated the quality as low.

2.5.b Cost of treatment

We found no data suitable for analysis for this outcome.

2.7.1 Time to start rehabilitation, e.g. time to sit up in bed

Two trials evaluated time from bed to chair while comparing a single‐injection block to morphine intrathecal injection (Kearns 2011; SMD 0.41, 95% CI 0.02 to 0.80) or continuous psoas compartment block to epidural local anaesthetic and opioid infusion (Duarte 2009; SMD ‐0.14, 95% CI ‐0.75 to 0.47; Analysis 2.21).

Fredrickson 2015 provided results for numbers of participants in chairs at 24 hours (20/27 versus 19/23) or at 48 hours (26/27 versus 22/23) for peripheral nerve block versus neuraxial block, respectively.

Kearns 2011 reported results for numbers of participants becoming mobile at first attempt (38/52 versus 44/51) for peripheral nerve block versus neuraxial block, respectively.

2.7.2 Walking

From three trials (Duarte 2009; Singelyn 2005; Türker 2003) with 94 participants we did not find a difference in time to first walk (SMD ‐0.41, 95% CI ‐1.09 to 0.27; I² 69%; Analysis 2.22'). Random sequence generation was judged as at low risk of bias for Duarte 2009 and Singelyn 2005; and as at unclear risk for Türker 2003. Allocation concealment was judged as at low risk for Duarte 2009 and as at unclear risk for Singelyn 2005 and Türker 2003. Blinding of participant and personnel taking care of participant was judged as at low risk for Türker 2003; as at unclear risk for Duarte 2009; and as at high risk for Singelyn 2005. Blinding of outcome assessment was judged as at low risk for Duarte 2009 and Türker 2003; and as at unclear risk for Singelyn 2005. Incomplete outcome data was judged as at low risk for Duarte 2009; Singelyn 2005 and Türker 2003. Selective reporting was judged as at low risk for Singelyn 2005 and Türker 2003; and at high risk for Duarte 2009. Other bias were judged as at low risk for Singelyn 2005 and Türker 2003; and as at unclear risk for Duarte 2009. There was no statistically significant evidence of small‐study effect (2‐sided test) nor evidence of publication bias. When Türker 2003 was excluded from the analysis, MD was ‐0.08 (95% CI ‐0.49 to 0.32; I² 0%). The comparison favoured peripheral nerve blocks when larger doses were used (P value for the meta‐regression 0.03).

We downgraded the level by one because 50% or more of the trials were judged as at unclear/high risk for allocation concealment and/or blinding of outcome assessor. We did not downgrade for inconsistency because reasonable explanation was found for heterogeneity. All trials included were direct comparisons. There was no evidence of publication bias. We rated the quality as low.

Fredrickson 2015 reported numbers of participants walking at 24 hours (15/27 versus 21/23) and 48 hours (24/27 versus 21/23) for peripheral nerve block versus neuraxial block, respectively.

Duarte 2009 reported time required to wean off the walker (MD 0.00, 95% CI ‐1.75 to 1.75).

2.7.3 Hip flexion

Singelyn 2005 reported degree of hip flexion from postoperative days 1 to 10. Mean differences were: MD 2.00° (95% CI ‐12.16° to 16.16°) and 0.00° (95% CI ‐6.59° to 6.59°) at postoperative days three and seven, respectively.

2.7.4 Duration of physiotherapy and daily life activity

Duarte 2009 reported duration of functional physiotherapy required (MD ‐0.30 day. 95% CI ‐1.13 to 0.53 day).

Duarte 2009 also reported time required to perform activities of daily life (MD 0.30 day, 95% CI ‐1.20 to 1.80 day).