Cephalomedullary nails versus extramedullary implants for extracapsular hip fractures in older adults

Summary of findings 1. Cephalomedullary nails compared to extramedullary implants for extracapsular hip fractures in adults

Outcomes	*Anticipated absolute effects^ (95% CI)**		Relative effect (95% CI)	Number of participants (studies)	Certainty of the evidence (GRADE)	Comments
Cephalomedullary nails compared to extramedullary implants for extracapsular hip fractures in adults
Population: older adults with stable or unstable extracapsular hip fractures Setting: hospitals; included studies were conducted in: Australia, Austria, Brazil, Canada, China, Denmark, Finland, France, Greece, Hong Kong, India, Iran, Israel, Italy, Japan, Mexico, New Zealand, Norway, Pakistan, South Korea, Spain, Sweden, Switzerland, The Netherlands, Turkey, USA, UK Intervention: cephalomedullary nails (Gamma nail, Gamma 3 nail, PFN, ultra‐short PFN, expandable PFN, PFNA, Targon PFN, TRIGEN INTERTAN nail, Holland nail, Küntscher‐Y nail) Comparison: extramedullary implants (SHS, DHS, ABMI hip screw, compression hip screw, LISS, Medoff sliding plate, blade plates, percutaneous compression plate, dynamic Condylar screw, locking compression plate)
Outcomes	Risk with extramedullary implants	Risk with cephalomedullary nails	Relative effect (95% CI)	Number of participants (studies)	Certainty of the evidence (GRADE)	Comments
Activities of daily living (ADL), early (≤ 4 months): using LEM (range from 0 to 100), FIM (range from 0 to 100), JOA (range from 0 to 20); higher scores indicate better performance in ADL Follow‐up: time points in the included studies were at 4 weeks and 3 months	‐	‐	‐	509 (4 studies)	Very low^a	We did not pool data because of high statistical heterogeneity.
Delirium (at end of follow‐up) Follow‐up: time points in the included studies were 4 months and 12 months	Study population		RR 1.22 (0.67 to 2.22)	1310 (5 studies)	Low^c
	30 per 1,000^b	37 per 1000 (20 to 67)
Functional status, early (≤ 4 months): using Zűckerman functional recovery score (0 to 44), and 100‐point functional recovery scale; in both scales, higher scores indicate better function Follow‐up: time points in the included studies were at 3 months and 4 months			SMD 0.02 higher (‐0.27 lower to 0.3 higher)	188 (2 studies)	Low^c	This effect did not indicate a clinically important difference, based on a 'rule of thumb' of: 0.2 for a small difference, 0.5 for a medium difference, and 0.8 for a large difference. Using the Zűckerman functional recovery score, this equates to a MD of 0.22 (this is unlikely to represent a clinically important difference on this 44‐point scale)
Health‐related quality of life, early (≤ 4 months)	‐	‐	‐			Inestimable
Mobility (≤ 4 months): assessed as number of participants with independent mobility Follow‐up: time points in the included studies were at 3 months and 4 months	Study population		RR 1.12 (1.01 to 1.23)	719 (7 studies)	Very low^d
	594 per 1,000^b	665 per 1000 (600 to 730)	RR 1.12 (1.01 to 1.23)	719 (7 studies)	Very low^d
Mortality, early (≤ 4 months) Follow‐up: time points in the included studies were during early postoperative period, within hospital, and at 1 month, 3 months, and 4 months	Study population		RR 0.96 (0.79 to 1.18)	4603 (30 studies)	Moderate^e
	83 per 1,000^b	80 per 1000 (66 to 98)	RR 0.96 (0.79 to 1.18)	4603 (30 studies)	Moderate^e
Mortality at 12 months Follow‐up: time points in the included studies were at 5 months, 6 months, 12 months, and 24 months	Study population		RR 0.99 (0.90 to 1.08)	7618 (47 studies)	Moderate^e
	204 per 1000^b	202 per 1000 (184 to 220)	RR 0.99 (0.90 to 1.08)	7618 (47 studies)	Moderate^e
Unplanned return to theatre (at end of follow‐up) Follow‐up: time points in the included studies were 3 months, 4 months, 5 months, 6 months, 12 months, and 24 months	Study population		RR 1.15 (0.89 to 1.50)	8398 (50 studies)	Low^f
	43 per 1,000^b	49 per 1000 (38 to 64)	RR 1.15 (0.89 to 1.50)	8398 (50 studies)	Low^f
*The risk in the intervention group (and its 95% confidence interval) is based on the assumed risk in the comparison group and the relative effect of the intervention (and its 95% CI). AMBI: manufacturer name for implant; CI: confidence interval; DHS: dynamic hip screw; FIM: functional independence measure; JOA: Japanese Orthopaedic Association;LEM: lower extremity measure; LISS: less invasive stabilisation system; MD: mean difference; PFN: proximal femoral nail; PFNA: proximal femoral nail antirotation; RR: risk ratio; SHS: sliding hip screw; SMD: standardised mean difference
GRADE Working Group grades of evidence High certainty: we are very confident that the true effect lies close to that of the estimate of the effect Moderate certainty: we are moderately confident in the effect estimate. The true effect is likely to be close to the estimate of the effect, but there is a possibility that it is substantially different Low certainty: our confidence in the effect estimate is limited. The true effect may be substantially different from the estimate of the effect Very low certainty: we have very little confidence in the effect estimate. The true effect is likely to be substantially different from the estimate of effect
^aDowngraded by three levels: one level for serious risks of bias and two levels for inconsistency owing to high levels of unexplained statistical heterogeneity ^bDerived from the pooled estimate of the cephalomedullary nails group ^cDowngraded by two levels: one level for serious risks of bias, and one level owing to imprecision denoted by the wide CI in this estimate. ^dDowngraded by two levels for serious risks of bias, and one level for inconsistency because this effect was not always apparent in other measures of early mobility (such as when measured using mobility scores) ^eDowngraded by one level for serious risks of bias ^fDowngraded by two levels: one level for serious risks of bias because all studies in this analysis were at high risk of detection bias, and one level for imprecision denoted by the wide CI in this estimate

Background

Description of the condition

Hip fracture is the general term for fracture of the proximal (upper) femur. These fractures can be subdivided into intracapsular fractures (those occurring within or proximal to the attachment of the hip joint capsule to the femur) and extracapsular (those occurring outside or distal to the hip joint capsule). Extracapsular hip fractures are defined as those fractures of the proximal femur within the area of bone from the attachment of the hip joint capsule to a level of five centimetres below the distal (lower) border of the lesser trochanter. Other terms used to describe these fractures include trochanteric, subtrochanteric, pertrochanteric and intertrochanteric fractures. These terms reflect the proximity of these fractures to the greater and lesser trochanters, which are two bony protuberances (bulges) at the upper end of the femur outside the joint capsule (Parker 2002).

Hip fractures occur predominantly in older people (aged over 65 years), especially women. In the UK, the mean age of a person with hip fracture is 83 years, and approximately two‐thirds occur in women (NHFD 2019). The relative proportion of extracapsular fractures also varies: 39% of hip fractures were extracapsular fractures in Bjorgul 2007, and 48% in Karagas 1996. A summary of the case‐mix for the 65,000 hip fractures occurring in 2018/19 in 175 hospitals in England, Wales and Northern Ireland was presented by an annual report of the National Hip Fracture Database (NHFD 2019). This showed that around three‐quarters of hip fractures (72.3%) occurred in women and over 91.1% of cases were aged over 70 years; around 40% of fractures were extracapsular.

Numerous subdivisions and classification methods exist for these fractures. The most practical classification, and that used for this review, is the basic division into stable trochanteric fractures (AO classification type A1) (Muller 1991) and unstable trochanteric fractures (AO classification type A2 and A3), with a separate category for subtrochanteric fractures. Stable trochanteric fractures are two‐part fractures in which the fracture line runs obliquely (at an angle) between the lesser and greater trochanter of the femur. Unstable trochanteric fractures again have an oblique fracture line running between the trochanters but in addition, there is comminution (multi‐fragmentation) of the fracture site. The comminution fragments may be the lesser trochanter, greater trochanter or both of these parts of the femur. Those fractures at the level of the lesser trochanter (AO A3, transtrochanteric) have a slightly more distally (lower) based fracture line which either runs transversely (across the bone) at the level of the lesser trochanter or in an oblique direction that is opposite (reverse) to that of stable and unstable trochanteric fractures. Transtrochanteric fractures may be two‐part or comminuted. This fracture pattern allows the femur to be displaced medially due to the pull of the abductor muscles. Subtrochanteric fractures are those fractures in which the fracture crossing the femur is predominately found within the five centimetres of bone immediately below the lesser trochanter. These fractures may be two‐part or comminuted and, in some instances, the fracture may extend proximally into the trochanteric region or distally into the shaft of the femur.

Description of the intervention

Operative treatment of extracapsular hip fractures was introduced in the 1950s using a variety of different implants. Implants may be either extramedullary or cephalomedullary in design. Worldwide, the most commonly used extramedullary implant is the sliding hip screw (SHS), which is synonymous with the term compression hip screw and equivalent models such as the Dynamic, Richards or AMBI hip screws. The SHS consists of a lag screw passed up the femoral neck to the femoral head. This lag screw is then attached to a plate on the side of the femur. These are considered 'dynamic' implants as they have the capacity for sliding at the plate/screw junction to allow for collapse at the fracture site, resulting in compression between the main fracture fragments. The Medoff plate (Medoff 1991) is a modification of the SHS. The difference is that the plate has an inner and outer sleeve, which can slide between each other. This creates additional capacity for sliding to occur at the level of the lesser trochanter as well as at the lag screw. Sliding at the lag screw can be prevented with a locking screw to create a 'one way' sliding Medoff instead of a 'two way' sliding Medoff. At a later date, the locking device on the lag screw can be removed to 'dynamise' the fracture. Another dynamic extramedullary device is the percutaneous compression plate (PCCP) (Orthofix), a minimally invasive device that is placed via two small incisions. It uses two smaller screws in the femoral head (as opposed to one large screw) to minimise damage to the lateral cortex and provide rotational stability.

Extramedullary devices may also be static devices; these do not allow collapse at the fracture site. These include pre‐contoured locking plates which allow placement of multiple screws in the femoral head that are locked into the plate, thereby preventing movement at the fracture site (e.g. the proximal femoral locking plate (PFLP)) and fixed nail plates such as the Jewett and the McLaughlin nail plates. Pre‐contoured locking plates designed for the distal (lower) femur may also be used as static fixed‐angle devices for extracapsular hip fractures by using them in a reverse position on the opposite proximal (upper) femur (e.g. reverse distal femoral less invasive stabilisation system (LISS) plates (rDF LISS) (DePuy Synthes) or the reverse distal femoral locking plate (rDFLP)). The 90‐ or 95‐degree blade plate is also a static extramedullary device. Though theoretically, the dynamic condylar screw plate has the capacity for sliding at the screw plate junction, it is more likely to act as a static device when used at the hip, with no slide occurring. Table 1 provides further details on the extramedullary devices assessed by the included trials in this review.

Table 1. Extramedullary devices evaluated by included trials

Name	Description
Sliding hip screw (SHS)	The SHS (DePuy Synthes) consists of a lag screw passed up the femoral neck to the femoral head. This lag screw is then attached to a plate on the side of the femur typically at 135º (130º to 150º available). These are considered 'dynamic' implants as they have the capacity for sliding at the plate/screw junction to allow for controlled collapse at the fracture site, thereby facilitating fracture healing.
Medoff sliding plate	The Medoff sliding plate (Swemac Ltd) is a modification of the sliding hip screw, the difference being the plate having an inner and outer sleeve, which can slide between each other. This creates an additional capacity for sliding to occur at the level of the lesser trochanter as well as at the lag screw. Sliding at the lag screw can be prevented with a locking screw to create a 'one way' sliding Medoff instead of a 'two way' sliding Medoff. At a later date the locking device on the lag screw can be removed to 'dynamise' the fracture.
Percutaneous compression plate (PCCP)	The PCCP (Orthofix) is an extramedullary device developed by Gotfried in the late 1990s. Similar to the SHS, it utilises a telescoping mechanism in the femoral neck to facilitate collapse of the fracture. It differs in that it is minimally invasive (inserted by 2 small incisions) and uses 2 small screws in the femoral head as opposed to one large screw (SHS). This design is to provide double axis fixation to prevent femoral neck rotation and also prevent damage to the lateral femoral cortex as 2 small screws are used.
Dynamic condylar screw (DCS)	The DCS (DePuy Synthes) device is similar to the SHS device described above. It consists of a lag screw placed in the femoral head that attaches to a plate on the side of the femur via a barrel. It differs however in the angle the lag screw is attached to the plate (95º). This acute angle means that the DCS is most likely to act as a static device with little or no movement taking place at the screw/barrel junction.
Proximal femoral locking plate (PFLP)	The PFLP (DePuy Synthes) device is a pre‐contoured fixed angle device where multiple screws (7.3 mm and 5 mm) are placed in the femoral head and fixed to a pre‐contoured 4.5 mm plate with a locking mechanism. This ensures it acts as a static device that does not allow movement at the fracture site.
Reverse distal femoral less invasive stabilisation system plate (rDF LISS)	The rDF LISS plate (DePuy Synthes) is a pre‐contoured fixed angle devices used for distal femoral fractures. It is essentially a locking plate that can be applied using a minimally invasive technique. It has been used for contralateral proximal femoral fractures by reversing the plate position and placing it on the proximal femur (Zhou 2012).
Reverse distal femoral locking compression plate (rDFLP)	The rDFLP (Greens Surgical, India) is a pre‐contoured fixed angle device designed for distal femoral fractures. It has combination holes in the area of the plate placed on the femoral shaft allowing locked and non‐locked screw placement. It can be used for contralateral proximal femoral fractures by reversing the plate position and placing it on the proximal femur (Haq 2014).
Blade plate	The blade plate is a fixed‐angle device where the blade (attached to a plate) is placed in the centre of the femoral head. The angle at the blade/plate junction is typically 95% with plate lengths of 50 mm to 80 mm.

Cephalomedullary nails used for internal fixation of extracapsular fractures can either be inserted from distal to proximal (condylocephalic nails; Parker 1998) or from proximal to distal (cephalocondylic nails). Cephalocondylic nails are inserted through the greater trochanter of the femur and secured by a screw which is passed through the proximal part of the nail (or vice versa), up the femoral neck into the femoral head. Theoretical biomechanical advantages of these cephalomedullary nails over screw‐and‐plate fixation are attributed to a reduced distance between the hip joint and the implant, which diminishes the bending moment across the implant/fracture construct.

Another potential biomechanical advantage is that fixation with cephalomedullary nails results in less femoral medialisation. The reason nails reduce femoral medialisation is that the proximal part of the nail acts as a lateral buttress that sits inside the proximal femur; this reduces the potential space for fractured osteoporotic bone to collapse into (Ong 2019). More femoral medialisation has been shown to result in inferior mobility because the hip abductor muscles are detensioned and so cannot work as well (Bretherton 2016).

Examples of cephalomedullary nails are the Gamma nail (Stryker‐Howmedica), the cephalomedullary hip screw (IMHS) (Smith & Nephew), the proximal femoral nail (PFN) (Synthes), the proximal femoral nail antirotation (PFNA) (Synthes), the Targon PF (proximal femoral) nail (B. Braun), the Holland nail and the Küntscher‐Y nail (Cuthbert 1976). Condylocephalic nails are inserted into the distal femur and passed up the cephalomedullary cavity across the fracture site and up into the femoral head; these nails are not included in this review. The best‐known type of this nail is the Ender nail. Table 2 presents further information on the cephalomedullary nails assessed by the included trials in this review. A review comparing different cephalomedullary nails for these fractures is available (Queally 2014).

See: Summary of findings 1 Cephalomedullary nails compared to extramedullary implants for extracapsular hip fractures in adults

Table 2. Cephalomedullary nails evaluated by the included trials

Name	Description
Gamma nail	The Gamma nail (Stryker) was introduced in the late 1980s for the treatment of extracapsular hip fractures. The implant consists of a sliding lag screw which passes through a short cephalomedullary nail. One or two screws may be passed through the nail tip to secure it to the femoral shaft (distal locking). Theoretical advantages of this implant are due to a percutaneous insertion technique and include reduced blood loss, reduced sepsis, minimal tissue trauma and reduced operating time. Modifications to the design of the Gamma nail and its instrumentation have occurred since its introduction. The long Gamma nail has a range of different lengths from 280 mm to 460 mm with two distal locking screws.
Gamma 3 nail	The Gamma 3 nail (Stryker) is the third generation of the gamma nail fixation system for proximal femoral fractures. It is a trochanteric entry nail with a reduced proximal nail diameter (15.5 mm versus 17 mm) to facilitate a shorter incision. Its length options range from 280 mm to 460 mm. Its neck‐shaft angle options include 120°, 125° and 130°. The lag screw shape has also been modified to provide superior cutting behaviour and greater resistance to cut‐out.
Intramedullary hip screw (IMHS)	The IMHS (Smith & Nephew), length 210 mm, was introduced in 1991 for the treatment of extracapsular femoral fractures. Like the Gamma nail, it consists of a nail inserted via the greater trochanter into the medullary cavity. It utilises a single screw in the femoral head that can slide through a barrel in the nail allowing fracture compression. Three different neck angles are available: 125°, 130° and 135°. Nail lengths are available from 195 mm to 440 mm.
Proximal femoral nail (PFN)	The PFN (DePuy Synthes), length 240 mm, was introduced in 1998 for the treatment of extracapsular fractures. Like the Gamma and IMHS, it consists of a nail inserted via the greater trochanter into the medullary cavity. Three lengths are available: 240 mm, 200 mm and an ultra‐short 180 mm. Two proximal lag screws are passed up the femoral neck to the head. Distal locking can performed in static or dynamic mode via two distal locking screws.
Proximal femoral nail antirotation (PFNA)	The PFNA (DePuy Synthes), length 170 mm, 200 mm or 240 mm, is a modification of the PFN. It is similar to the PFN apart from not having two proximal lag screws but instead a single helically‐shaped blade which is designed to provide increased angular and rotational stability. The helical blade is designed to avoid bone loss that occurs during drilling and insertion of a standard hip screw. It has 2 distal locking screw options for either dynamic or static locking. Blade‐shaft angle options include 125°, 130° and 135°.
Targon proximal femoral nail (PF)	The Targon PF (B. Braun), length 220 mm, is inserted into the intramedullary cavity via a trochanteric entry point. Proximally, this nail has a sliding lag screw and an antirotation pin. The Targon PF facilitates fracture dynamisation via a gliding screw that glides through a sleeve that is attached to the nail, thereby avoiding protrusion of the screw into peritrochanteric tissues.
Holland nail	The Holland nail (Zimmer Biomet) is like the Gamma and IMHS; it consists of a nail inserted via the greater trochanter in to the medullary cavity. Two proximal lag screws are passed up the femoral neck to the head.
Experimental nail (reported in Dujardin 2001)	An experimental mini‐invasive static intramedullary nail, which is not commercially available, is reported in Dujardin 2001. This consists of an intramedullary nail which is 170 mm long with a distal diameter of 12 mm and a proximal diameter of 13 mm. There are two five‐mm distal locking holes. The proximal hold of the femur is with two 7‐mm cannulated screws which diverge at a 30‐degree angle. Unlike the other proximal femoral nails, there is no sliding mechanism within the nail construct.
Kuntscher‐Y nail	The Kuntscher‐Y nail (Cuthbert 1976) is an early design of an intramedullary nail. It consists of a side arm and a separate slotted Kuntscher nail. The side arm is passed up the femoral neck, and then attached to an alignment jig to enable a slotted Kuntscher nail to be passed via the greater trochanter through a hole in the side arm and distally within the medullary cavity. The assembled implant construct has no capacity for sliding at the side arm and neither has it the capacity for distal locking.
Endovis nail	The Endovis nail (Citieffe) is available in 3 sizes (195 mm to 400 mm) and has a neck shaft angle of 130°. It has two cephalic screws for the femoral head to facilitate fracture compression. The distal section is slotted to produce a graduated variation of stiffness.
TRIGEN INTERTAN nail	The INTERTAN nail (Smith & Nephew) uses 2 cephalocervical screws in an integrated mechanism allowing intraoperative compression and rotational stability of the head‐neck fragments. It has a cannulated set screw mechanism that allows for the device to be used in fixed angle mode or in sliding/compression mode. Its length ranges from 18 cm to 46 cm (long nail option).
Russell‐Taylor Recon nail	The Russel‐Taylor Recon nail (Smith & Nephew) is an intramedullary nail that utilises a piriformis entry point. Two screws are available for fixation in the femoral head. It is a full length femoral nail with no short versions available for proximal femoral fixation only.
Trochanteric Fixation Nail (TFN)	The TFN nail (DePuy Synthes) is a titanium nail that utilises a helical blade for fixation in the femoral head instead of a lag screw. This design is intended to improve resistance to various collapse and improved rotational control of the medial fracture segment theoretically reducing the rate of cut‐out.

Why it is important to do this review

There is controversy over the choice of implant, especially the use of cephalomedullary nails versus sliding hip screws, for extracapsular hip fractures. Indeed, studies reporting a rapid increase in the use of cephalomedullary nails in the USA have pointed out, citing an earlier version of this review, that this phenomenon is not supported by the available evidence (Anglen 2008; Forte 2008; Forte 2010). The availability of new evidence — often on new implants that are aimed at avoiding the complications of cephalomedullary fixation (specifically, operative and later femoral fracture) — indicate a need to update this Cochrane Review (Parker 2010), which continues to compare different types of cephalomedullary nails with extramedullary implants.

The need for this review update was endorsed by a prioritisation process conducted as part of a National Institute for Health Research (NIHR)‐funded Cochrane Programme Grant on the management of hip fracture. This additionally provided the rationale for modifications to the review's protocol, together with the collection of additional context data and provision of additional results that might better inform current practice.

Objectives

To assess the relative effects of cephalomedullary nails versus extramedullary fixation implants for treating extracapsular proximal femoral (hip) fractures in older adults.

Methods

Criteria for considering studies for this review

Types of studies

We included randomised controlled trials (RCTs) and quasi‐RCTs comparing cephalocondylic intramedullary (cephalomedullary) nails with extramedullary implants in extracapsular hip fracture. Quasi‐RCTs are defined as trials in which the methods of allocating participants to an intervention are not random, but are intended to produce groups with similar future outcomes (Cochrane 2018). We included published papers and conference abstracts if they provided sufficient data relating to the methods and outcomes of interest.

Types of participants

We included older adults (at least 60 years of age) undergoing surgery in a hospital setting for an extracapsular proximal femoral fracture. We included trochanteric (stable or unstable) or subtrochanteric fractures which we expected to be caused by low‐energy trauma.

We expected trial populations to have a mean age of between 80 and 85 years, to include 70% women, 30% with chronic cognitive impairment, and 50% with an American Society of Anesthesiologists (ASA) score greater than two to indicate that a patient has no more than mild systemic disease without significant functional limitation (NHFD 2019; NICE 2011). This would be representative of the general hip‐fracture population.

We excluded studies that focused exclusively on the treatment of participants younger than 60 years of age, of participants with fractures caused by specific pathologies other than osteoporosis, and of participants with high‐energy fractures. However, we took a pragmatic approach to study inclusion criteria and included studies with mixed populations (fragility and other mechanisms, ages, or pathologies). We expected that the proportion of participants with standard fragility fractures was most likely to outnumber those with high‐energy or local pathological fractures; therefore, the results would be generalisable to the fragility‐fracture population. If the data were reported separately for fragility fractures, we planned to use these subgroup data for our main analyses. We considered it unlikely that participants under 60 years of age would have experienced a fragility hip fracture caused by low‐energy trauma.

Types of interventions

We included surgical fixation of the fracture with a cephalomedullary nail or with an extramedullary implant. In our categorisation of implants we noted the key design characteristics of the type of implant, as well as assessing their current use worldwide. For cephalomedullary nails, we considered short and long nails, and dynamic versus static implants. For extramedullary implants, we considered dynamic versus static devices. For descriptions of the cephalomedullary nails and extramedullary implants evaluated in the included trials, see Table 1 and Table 2.

Types of outcome measures

Depending on the length of follow‐up reported, we categorised the end points for outcomes into early (up to and including four months) or 12 months (prioritising 12‐month data, but in their absence including any data after four months). We selected four months as the definition of early because most of early recovery has been achieved at this time point (Griffin 2015). This is also in accordance with the core outcome set for hip fracture, which prioritises early outcome over late recovery (Haywood 2014). Although priority was given to early outcomes in the presentation of our data, we also included outcome data at late time points, and we therefore included all outcomes without a time limit.

Critical outcomes

We extracted information on the following seven 'critical' outcomes.

Activities of daily living (e.g. Barthel Index (BI), Functional Independence Measure (FIM)).
Delirium, using recognised assessment scores such as Mini‐mental state examination (MMSE) or the 4 'A's Tests (4AT) and the Abbreviated Mental Test Score (AMTS).
Functional status (region‐specific) (e.g. hip rating questionnaire, Harris Hip Score, Oxford Hip Score).
Health‐related quality of life (e.g. Short Form Health Survey (SF‐36), EuroQol‐ 5 Dimension (EQ‐5D)).
Mobility (e.g. indoor/outdoor walking status, Cumulated Ambulation Score, Elderly Mobility Scale score, Timed Up and Go test, Short Physical Performance Battery, Parker mobility score (Parker 1993), self‐reported walking scores (e.g. Mobility Assessment Tool — short form)).
Mortality.
Unplanned return to theatre: secondary procedure required for a complication resulting directly or indirectly from the index operation/primary procedure measured at the end of study follow‐up.

Other important clinical outcomes

We also reported the following 'important' outcomes. Where relevant, we categorised these into early (up to and including four months) and late (after four months).

Pain (verbal rating or visual analogue scale (VAS)).
Length of in‐hospital stay.
Discharge destination. We used study authors' definitions, which were variably defined in the included studies.
Adverse events.

We also grouped adverse events by relatedness to the implant or fracture, or both. We reported each adverse event type separately for maximum clarity. We anticipated that events may have included the following.

Related adverse events

Damage to a nerve, tendon or blood vessel
Intraoperative periprosthetic fracture
Postoperative periprosthetic fracture
Loosening of prosthesis
Screw cut‐out
Implant failure
Wound infection (we used study authors' definitions, which were often described as deep infection or superficial infection)

Unrelated adverse events

Acute kidney injury
Blood transfusion
Cerebrovascular accident
Chest infection/pneumonia
Decreased cognitive ability
Myocardial infarction/acute coronary syndrome
Sepsis
Urinary tract infection
Venous thromboembolic phenomena (deep vein thrombosis and pulmonary embolism)

Search methods for identification of studies

As well as developing a strategy for this review, we developed general search strategies for the large bibliographic databases to find records to feed into a number of Cochrane Reviews and review updates on hip fracture surgery (Lewis 2021; Lewis 2022a; Lewis 2022b; Lewis 2022c). We searched the main databases up to July 2020.

Electronic searches

We identified RCTs and quasi‐RCTs through literature searching with systematic and sensitive search strategies, as outlined in Chapter 4 of the Cochrane Handbook of Systematic Reviews of Interventions (Lefebvre 2019). We applied no restrictions on language, date, or publication status. We searched the following databases for relevant trials.

Cochrane Central Register of Controlled Trials (CENTRAL; CRS Web; 8 July 2020).
MEDLINE (Ovid; 1946 to 6 July 2020).
Embase (Ovid; 1980 to 7 July 2020).
Web of Science (SCI EXPANDED; 1900 to 8 July 2020).
Cochrane Database of Systematic Reviews (CDSR; the Cochrane Library; 7 July 2020).
Database of Abstracts of Reviews of Effects (DARE; www.crd.york.ac.uk/CRDWeb/; 17 December 2018).
Health Technology Assessment (HTA) database (www.crd.york.ac.uk/CRDWeb/; 17 December 2018).
Epistemonikos (www.epistemonikos.org/; 9 July 2020).
Proquest Dissertations and Theses (ProQuest; 1743 to 8 July 2020).
National Technical Information Service (NTIS, for technical reports; www.ntis.gov/; 10 July 2020).

We developed a subject‐specific search strategy in MEDLINE and other listed databases; we adapted strategies with consideration of differences between database interfaces as well as different indexing languages. In MEDLINE, we used the sensitivity‐maximising version of the Cochrane Highly Sensitive Search Strategy for identifying randomised trials (Lefebvre 2019). In Embase, we used the Cochrane Embase filter (www.cochranelibrary.com/central/central-creation) to focus on RCTs. The initial search was run in November 2018 and December 2018, and a top‐up search was run in July 2020 in all databases except for DARE and HTA, in which no new records have been added since the initial search. At the time of the search, CENTRAL was fully up‐to‐date with all records from the Cochrane Bone, Joint, and Muscle Trauma (BJMT) Group's Specialised Register, and so it was not necessary to search this separately. We developed the search strategy in consultation with Information Specialists (see Acknowledgements) and the Information Specialist for Cochrane BJMT. Search strategies can be found in Appendix 1.

We scanned ClinicalTrials.gov (www.clinicaltrials.gov/) for ongoing and unpublished trials on 10 July 2020. Details of the search strategies used for previous versions of the review are given in Parker 2010.

Searching other resources

We handsearched the following conference abstracts from 2016 to November 2018.

Fragility Fractures Network Congress.
British Orthopaedic Association Congress.
Orthopaedic World Congress (SICOT).
Orthopaedic Trauma Association Annual Meeting.
The Bone & Joint Journal Orthopaedic Proceedings.
American Academy of Orthopaedic Surgeons Annual Meeting.

In addition, one review author (MJP) kept updated records of all related publications which we used during interim work on this update.

Data collection and analysis

In order to reduce bias, we ensured that any review author who is also a study author, co‐applicant on the Cochrane Programme Grant on the management of hip fracture, or has had an advisory role on any potentially relevant study, remained independent of study selection decisions, risk of bias assessment and data extraction for their study.

Selection of studies

Two review authors screened titles and abstracts of all the retrieved bibliographic records in a web‐based systematic reviewing platform, Rayyan (Ouzzani 2016), and in the top‐up search using Covidence. Full texts of all potentially eligible records passing the title and abstract screening level were retrieved and examined independently by two review authors, using the eligibility criteria outlined in Criteria for considering studies for this review. Full‐text screening was conducted using Covidence. Disagreements were resolved by discussion or adjudication by a third review author. Duplicates were excluded and multiple reports of the same study collated so that each study, rather than each report, was the unit of interest in the review. We prepared a PRISMA flow diagram to outline the study selection process, numbers of records at each stage of selection, and reasons for exclusions of full‐text articles (Moher 2009). We reported in the review details of key excluded studies, rather than all studies that were excluded from consideration of full‐text articles.

Since publication of the previous review (Parker 2010), some additional review authors conducted interim searches for the review. Results were incorporated in a non‐published review file (see Acknowledgements).

Data extraction and management

All review authors conferred on the essential data for extraction, and a form was structured to align with default headings in the Characteristics of included studies (see Appendix 2). Two review authors piloted the form on five studies and compared results. We then made changes to the template following additional discussion with the author team. For the remaining data extraction, one review author independently extracted data and a second review author checked all the data for accuracy. We extracted the following data.

Study methodology: publication type; sponsorship/funding/notable conflicts of interest of trial authors; study design; number of centres and locations; size and type of setting; study inclusion and exclusion criteria; randomisation method; number of randomised participants, losses (and reasons for losses), and number analysed for each outcome. (Collecting information relating to the participant flow helped with the assessment of risk of attrition bias.)
Population: baseline characteristics of the participants by group and overall (age, gender, smoking history, medication, body mass index (BMI), comorbidities, functional status such as previous mobility, place of residence before fracture, cognitive status, American Society of Anesthesiologists (ASA) status, fracture type and stability).
Interventions: details of each intervention (number and type, manufacturer details); general surgical details (number of clinicians and their skills and experience, perioperative care such as use of prophylactic antibiotics or antithromboembolics, mobilisation or weight‐bearing protocols).
Outcomes: all outcomes measured or reported by study authors; outcomes relevant to the review (including measurement tools and time points of measure); extraction of outcome data into data and analysis tables or additional tables in Review Manager 2020.

As above, a previous review author team conducted interim data extraction, and we supplemented this with additional data extraction using these criteria (see Acknowledgements).

Assessment of risk of bias in included studies

We assessed risk of bias in the included studies using the Cochrane risk of bias tool (Higgins 2011). We assessed the following domains.

Random sequence generation (selection bias).
Allocation concealment (selection bias).
Blinding of participants and personnel (performance bias).
Blinding of outcome assessors (detection bias).
Incomplete outcome data (attrition bias).
Selective reporting (reporting bias).
Other risks of bias.

In addition, we also considered performance bias related to the experience of the clinicians (whether clinicians were equally experienced with the implants used in the study). We considered risk of detection bias separately for: subjective outcomes measured by clinicians, objective outcomes measured by clinicians, and participant‐reported outcomes (e.g. pain and health‐related quality of life). For each domain, two review authors judged whether study authors made sufficient attempts to minimise bias in their design. For each domain, we made judgements using three measures — high, low, or unclear risk of bias — and we recorded these judgements in risk of bias tables.

Measures of treatment effect

We calculated risk ratios (RRs) for dichotomous data outcomes with 95% confidence intervals (CIs); it was not appropriate to use Peto odds ratio (OR) to calculate effects because no outcomes had very low numbers of observed events. We expressed treatment effects for continuous data outcomes as mean differences (MDs) with 95% CIs; if the outcomes were measured using different scales, we planned to use standardised mean differences (SMDs) with 95% CIs.

In the event that studies reported dichotomous data using more than one category, we selected the following cut‐off points in the distribution of categories.

For functional status: we reported data for those with a score of excellent or good (using Harris Hip Score (HHS)) versus those with a score of moderate or poor.
For mobility: we reported data for those who were able to walk independently out of doors with no more than the use of one stick (NICE 2011), versus those who were more dependent.
For pain: we reported data for participants who reported no pain versus those who reported any category of pain.
For discharge destination: we reported data for participants who were discharged home versus those who were discharged to a care environment.

Unit of analysis issues

In preparation of the review, we encountered potential unit of analysis issues. We found that some studies reported the number of hip fractures (or cases) as well as the number of participants, with a very small number of participants having two fractured hips. Often, differentiating the denominators within a report was challenging. In such studies, depending on the outcome, the unit of analysis was either the participant (for example, for outcomes such as mortality, discharge destination, or some adverse events) or the hip (for example, for outcomes such as unplanned return to theatre). We noted this differentiation where applicable and used the unit of analysis (participants or case) that was appropriate for the outcome within these studies. One study included more than two interventions (Papasimos 2005); in the analysis, we combined data from the two cephalomedullary groups (trochanteric Gamma nails and proximal femoral nails) and compared these to the extramedullary intervention arm (AMBI hip screw).

Dealing with missing data

For each included study, we recorded the number of participant losses for each outcome. Unless reported otherwise, we assumed complete case data for mortality, unplanned return to theatre and adverse events. For outcomes that required participant assessment at end of follow‐up (such as health‐related quality of life), we prioritised intention‐to‐treat (ITT) data where these data were available. If ITT data were unavailable for these outcomes, and if study authors did not clearly report denominator figures for each group for the outcome, we reduced the denominator figure in each group to account for reported mortality. We did not impute missing data. We used the risk of bias tool to judge attrition bias. We judged studies to be at high risk of attrition bias if we noted large amounts of unexplained missing data, losses that could not be easily justified in the study population, or losses that were not sufficiently balanced between intervention groups. If we included a study with high attrition bias, we explored the effect during sensitivity analysis. We completed sensitivity analysis only for critical review outcomes and only considered attrition for outcomes that may be affected by these losses.

We attempted contact with study authors of more recently published trials when we noted that data for critical outcomes appeared to have been measured but not reported. For older studies, we used data collected by previous author teams; this included data from direct communication with study authors. Where standard deviations were not reported, we attempted to determine these from other reported data (such as standard errors, CIs, or exact P values). We noted in the Characteristics of included studies tables when we could not use outcome data because they were insufficiently reported or because numbers of losses in each group were not clearly specified.

Assessment of heterogeneity

We used the I² statistic, automatically calculated in Review Manager 2020, to quantify the possible degree of heterogeneity of treatment effects between trials. We assumed there to be moderate heterogeneity when the I² was between 30% and 60%; substantial heterogeneity when it was between 50% and 90%; and considerable heterogeneity when it was between 75% and 100%. We noted the importance of I² depending on: 1) magnitude and direction of effects; and 2) strength of evidence for heterogeneity. We investigated statistical heterogeneity using subgroup analysis in the event of at least 10 studies (Deeks 2021).

We assessed clinical and methodological diversity in terms of participants, interventions, outcomes, effect modifiers, and study characteristics for the included studies to determine whether a meta‐analysis was appropriate; we used the information collected during data extraction (Data extraction and management).

Assessment of reporting biases

We planned to investigate the potential for publication bias and explore possible small‐study biases using funnel plots. However, there were insufficient studies (fewer than 10) for most outcomes. For outcomes with 10 or more studies, we constructed a funnel plot and interpreted the plot using a visual inspection and the Harbord modified test in Stata; for the critical review outcomes we reported P values for the Harbord modified test or Egger's test. We incorporated this judgement into the assessment of publication bias within the GRADE assessment.

To assess outcome reporting bias, we screened clinical trials registers for protocols and registration documents of included studies that were prospectively published, and we sourced all clinical trials register documents that were reported in the study reports of included studies. We used evidence of prospective registration to judge whether studies were at risk of selective reporting bias.

Data synthesis

We conducted meta‐analyses only when meaningful, that is, when the treatments, participants, and the underlying clinical question were similar enough for pooling to make sense. We pooled results of comparable groups of trials using random‐effects models. This model was chosen after careful consideration of the extent to which any underlying effect could truly be thought to be fixed, given the complexity of the interventions included in this review. We presented 95% CIs throughout.

We found that some studies reported outcome data at more than one time point, and where possible, we reported data within two time point windows. Early data included data up to four months (with priority given to data closest to four months for studies that reported multiple time points within this window); 12‐month data included a window from later than four months and up to 24 months, but with priority being given to data at 12 months.

For studies that reported outcome data using more than one measurement tool, we selected the tool that was used most commonly by other studies in the comparison group, or which reported data for the most number of participants. For mobility, we prioritised data from mobility scores, followed by dichotomous data for independent mobility.

We considered the appropriateness or otherwise of pooling data where there was considerable heterogeneity (I² statistic value of greater than 75%) that could not be explained by the diversity of methodological or clinical features among trials. We presented data from these studies in the analyses and clearly reported these observations in the text for the critical outcomes in the review.

Subgroup analysis and investigation of heterogeneity

Although we aimed to explore possible sources of heterogeneity between studies (key effect modifiers such as age, gender, cognitive impairment, and functional status), we found insufficient studies reporting these data in a manner to allow for meaningful analysis. In addition, we noted that few studies sufficiently reported some of these possible effect modifiers.

We completed subgroup analysis on length of cephalomedullary nails (long and short nails). We found that some studies included both long and short nails; in other studies, the length of nail was not reported, and we included these in a subgroup for mixed or unknown nail lengths.

We also conducted subgroup analysis on fracture type (stable and unstable trochanteric fractures). We based the subclassification for fracture instability on either the trial authors' classification of unstable or stable fractures. However, if the study authors reported these data according to the AO classification system, we used this in preference to other classification systems: we considered that A1 were stable fractures and A2 (A2.1, 2.2 and 2.3) and A3 were unstable trochanteric fractures. We found several studies that included a mixed population of stable and unstable fractures or did not report the fracture subtypes, and we therefore included a third subgroup for 'mixed/unknown' fracture type. We did not include studies exclusively including subtrochanteric fractures in this subgroup analysis.

We conducted a post hoc subgroup analysis for intraoperative and postoperative periprosthetic fractures. We noted that other reviews indicated that there may be fewer periprosthetic fractures in more recent studies because of improved implant designs (Bhandari 2009; Noris 2012). We therefore subgrouped these outcome data according to studies published before 2010 and from 2010 onwards.

We investigated whether the results of subgroups were significantly different by inspecting the overlap of CIs and performing the test for subgroup differences available in Review Manager 5 (Review Manager 2020).

Sensitivity analysis

We used sensitivity analysis to explore the effects of risks of bias on the review for critical outcomes. We performed analyses in which we excluded studies that met the following criteria.

Studies at high or unclear risk of selection bias for random sequence generation (this included studies that were described as quasi‐randomised, or that did not adequately describe methods used to randomise participants to intervention groups).
Studies at high risk of attrition bias (because studies reported a large number of losses that were unexplained or not justified for this population, or that were unbalanced between groups, and that we expected could influence outcome data).
Studies at high risk of performance bias (because the surgeons did not have comparable experience with both types of study implants).
Studies that used an extramedullary implant with static design.

Summary of findings and assessment of the certainty of the evidence

Two review authors used the GRADE system to assess the certainty of the body of evidence associated with the following seven critical outcomes in the review (Guyatt 2008).

Activities of daily living.
Delirium.
Functional status.
Health‐related quality of life.
Mobility.
Early mortality (measured within four months of surgery, and at 12 months).
Unplanned return to theatre.

For outcomes that were reported using more than one measurement tool, and that could not be combined in analysis, we assessed the certainty of the evidence for the outcome that used a measurement tool with the most participants. We only assessed the certainty of evidence when the evidence was supported by data with effect estimates. The GRADE approach assesses the certainty of a body of evidence based on the extent to which we can be confident that an estimate of effect or association reflects the item being assessed. Evaluation of the certainty of a body of evidence considers within‐study risk of bias, directness of the evidence (indirectness), heterogeneity of the data (inconsistency), precision of the effect estimates (imprecision), and risk of publication bias. The certainty of the evidence could be high, moderate, low or very low, being downgraded by one or two levels depending on the presence and extent of concerns in each of the five GRADE domains. We used footnotes to describe reasons for downgrading the certainty of the evidence for each outcome, and we used these judgements when drawing conclusions in the review.

We constructed a summary of findings table for the comparison of cephalomedullary nails versus extramedullary implants, using GRADE profiler software, to present the certainty of the evidence for these seven critical outcomes (GRADEpro GDT). We also assessed the certainty of the evidence for adverse event data related to the implant, fracture, or both, in which effect estimates clearly indicated an improvement or risk with one treatment over another.

Results

Description of studies

See Characteristics of included studies, Characteristics of excluded studies, and Characteristics of ongoing studies.

Results of the search

After the removal of duplicates from the search results, we screened 28,510 titles and abstracts, which included backward citation searches and searches of clinical trials registers. We excluded 27,426 irrelevant records. We reviewed the full text of 1171 records, and because of minor changes to the review criteria, this included studies in Parker 2010. We excluded 1029 records, and report the details of 10 key studies from these records. We included 76 studies (with 134 records) and identified two ongoing studies; we incorporated 34 new studies in the review. Four studies are awaiting classification. See Figure 1.

Figure 1

Flow diagram. Search conducted in November 2018 and December 2018, with a top‐up search in July 2020.

Included studies

Types of studies and setting

We included 76 studies (see Characteristics of included studies). Five studies were reported only as abstracts in which only limited study characteristics were reported (Benum 1994; Mehdi 2000; Michos 2001; Mott 1993; Raimondo 2012). Ten studies used methods to allocate participants to interventions which we assessed to be quasi‐randomised (Butt 1995; Goldhagen 1994; Guyer 1991; Hardy 1998; Leung 1992; Lopez 2002; Park 1998; Sharma 2018; Verettas 2010; Yamauchi 2014). The earliest study was reported in 1988 and the latest in 2020; 47% of the studies were completed from 2010 onwards.

Eleven studies were conducted across multiple centres (Ahrengart 1994; Andalib 2020; Baumgaertner 1998; Benum 1994; Davis 1988; Ekstrom 2007; Matre 2013; Mott 1993; Rahme 2007; Reindl 2015; Sanders 2017). Twelve studies were completed in the UK (Adams 2001; Barton 2010; Bridle 1991; Butt 1995; Davis 1988; Harrington 2002; Haynes 1996; Little 2008; Mehdi 2000; Parker 2012; Parker 2017; Radford 1993); twelve in China (Cai 2016; Chen 2018; Gou 2013; Han 2012; Li 2018; Song 2011; Tao 2013; Wang 2019; Xu 2010; Xu 2018; Zhou 2012; Zou 2009); five in Greece (Aktselis 2014; Kouvidis 2012; Michos 2001; Papasimos 2005; Verettas 2010); four in Switzerland (Guyer 1991; Pelet 2001; Sadowski 2002; Saudan 2002); three each in Canada (O'Brien 1995; Reindl 2015; Sanders 2017), India (Haq 2014; Singh 2017; Singh 2019), Spain (Lopez 2002; Utrilla 2005; Varela‐Egocheaga 2009), Sweden (Ahrengart 1994; Ekstrom 2007; Mehdi 2000) and the USA (Baumgaertner 1998; Goldhagen 1994; Mott 1993); and two each in Brazil (Guerra 2014; Sharma 2018), France (Dujardin 2001; Giraud 2005), Italy (Carulli 2017; Raimondo 2012), Japan (Kuwabara 1998; Yamauchi 2014), Norway (Benum 1994; Matre 2013), Pakistan (Adeel 2020; Akhtar 2016), South Korea (Hong 2011; Park 1998) and Turkey (Eceviz 2020; Zehir 2015). The remainder took place in European countries (Hardy 1998; Hoffmann 1999; Kukla 1997; Ovesen 2006; Pahlpatz 1993; Pajarinen 2005) or Australia (Rahme 2007), Hong Kong (Leung 1992), Iran (Andalib 2020), Israel (Chechik 2014), Mexico (Calderon 2013) or New Zealand (Hoffman 1996).

Types of participants

In total 10,979 participants with 10,998 hip fractures were recruited across the 76 studies. Of the included studies, 43 specified a lower age limit for participant inclusion; one only accepted participants older than 70 years (Verettas 2010); 13 used 65 years as the lower limit (Aktselis 2014; Cai 2016; Eceviz 2020; Guerra 2014; Harrington 2002; Kouvidis 2012; Kuwabara 1998; Leung 1992; Tao 2013; Utrilla 2005; Xu 2010; Xu 2018; Zehir 2015); 17 used 60 years (Bridle 1991; Calderon 2013; Chechik 2014; Chen 2018; Dujardin 2001; Gou 2013; Han 2012; Hardy 1998; Hong 2011; Kukla 1997; Li 2018; Matre 2013; Papasimos 2005; Radford 1993; Singh 2019; Song 2011; Varela‐Egocheaga 2009); four used 55 years (Reindl 2015; Sadowski 2002; Sanders 2017; Saudan 2002); two used 50 years (Davis 1988; Hoffman 1996) and 40 years (Adeel 2020; Akhtar 2016); four used 18 years (Barton 2010; Haq 2014; Sharma 2018; Singh 2017) and one used 16 years (Pelet 2001). The studies with 18 and 16 years as a lower cut‐off reported a mean age which reassured us that the study population was representative of the age group under investigation in this review. Three studies reported an upper age limit for participants; these were 70 years (Akhtar 2016), 75 years (Adeel 2020) and 90 years (Calderon 2013). The mean age for all participants was greater than 70 years of age in 82% of included studies. Three studies had a mean age less than 60 years of age (Akhtar 2016; Haq 2014; Singh 2017). Five studies did not report the age of participants (Ahrengart 1994; Goldhagen 1994; Han 2012; Pahlpatz 1993; Reindl 2015).

Gender was reported in 70 studies; overall, 72% of participants were female. Twelve studies specified in their inclusion criteria that participants should have been able to walk prior to surgery (Akhtar 2016; Andalib 2020; Cai 2016; Eceviz 2020; Guerra 2014; Kukla 1997; Papasimos 2005; Reindl 2015; Sanders 2017; Xu 2010; Yamauchi 2014; Zehir 2015). Nine studies excluded participants with cognitive impairment (Chechik 2014; Chen 2018; Eceviz 2020; Harrington 2002; Li 2018; Parker 2012; Reindl 2015; Wang 2019; Yamauchi 2014) and 55% of the studies excluded pathological fractures.

Most studies included participants with trochanteric fractures; 12 studies also included subtrochanteric fractures (Benum 1994; Butt 1995; Ekstrom 2007; Goldhagen 1994; Guyer 1991; Haynes 1996; Leung 1992; Matre 2013; Michos 2001; Miedel 2005; Mott 1993; Pahlpatz 1993). Rahme 2007 included only subtrochanteric fractures, and Eceviz 2020 included only basicervical fractures. Three studies included only stable fractures (Cai 2016; Eceviz 2020; Sharma 2018) and 17 studies investigated unstable fractures (Adeel 2020; Akhtar 2016; Aktselis 2014; Andalib 2020; Barton 2010; Calderon 2013; Ekstrom 2007; Haq 2014; Harrington 2002; Miedel 2005; Papasimos 2005; Reindl 2015; Sadowski 2002; Singh 2017; Verettas 2010; Xu 2010; Zehir 2015). Two studies did not report fracture subtypes (Michos 2001; Raimondo 2012), and the remaining studies included both stable and unstable fractures.

Three studies included participants with a preoperative waiting in excess of two weeks (Haq 2014; Zehir 2015; Zhou 2012), two studies included patients with a wait of up to two weeks (Hong 2011; Reindl 2015), two studies reported a wait of seven days (Akhtar 2016; Tao 2013), four studies reported a mean waiting time of five days (Eceviz 2020; Singh 2017; Wang 2019; Yamauchi 2014), four studies had a mean of three days (Cai 2016; Kouvidis 2012; Rahme 2007; Song 2011) and 12 studies reported a waiting time of less than 48 hours (Adams 2001; Calderon 2013; Chechik 2014; Dujardin 2001; Goldhagen 1994; Haynes 1996; Hoffman 1996; Kukla 1997; O'Brien 1995; Pajarinen 2005; Sanders 2017; Verettas 2010). The remaining 49 studies did not report the preoperative waiting time.

Types of interventions

All studies used two‐arm designs, except for Papasimos 2005 which compared two cephalomedullary nails and an extramedullary implant.

Cephalomedullary implants

We included a number of different cephalomedullary nails in this review. Twenty‐nine studies reported outcomes of the Gamma nail (Adams 2001; Ahrengart 1994; Aktselis 2014; Barton 2010; Benum 1994; Bridle 1991; Butt 1995; Goldhagen 1994; Guyer 1991; Han 2012; Haynes 1996; Hoffman 1996; Kukla 1997; Kuwabara 1998; Leung 1992; Lopez 2002; Michos 2001; Miedel 2005; Mott 1993; O'Brien 1995; Ovesen 2006; Pahlpatz 1993; Park 1998; Pelet 2001; Radford 1993; Reindl 2015; Song 2011; Utrilla 2005; Verettas 2010). One study specified a Gamma 3 nail (Varela‐Egocheaga 2009). A proximal femoral nail (PFN) was used in 12 studies (Adeel 2020; Calderon 2013; Ekstrom 2007; Guerra 2014; Haq 2014; Hong 2011; Pajarinen 2005; Rahme 2007; Sadowski 2002; Saudan 2002; Singh 2017; Singh 2019) and a further 13 used the proximal femoral nail antirotation (PFNA) (Akhtar 2016; Carulli 2017; Chen 2018; Gou 2013; Li 2018; Tao 2013; Wang 2019; Xu 2010; Xu 2018; Yamauchi 2014; Zehir 2015; Zhou 2012; Zou 2009). One study used an ultra‐short PFN (Sharma 2018) and one described using an expandable PFN (Chechik 2014). Three studies specifically used a Targon PFN (Giraud 2005; Parker 2012; Parker 2017) and two studies used the TRIGEN INTERTAN nail (Matre 2013; Sanders 2017). Five used an intramedullary hip screw (IMHS) (Baumgaertner 1998; Hardy 1998; Harrington 2002; Hoffmann 1999; Mehdi 2000). One study used a mixture of Gamma nails and PFNs (Papasimos 2005). Holland nails and Küntscher‐Y nails were used in one study each (Little 2008 and Davis 1988, respectively). Six studies reported a nonspecific intervention, describing the implant used as a cephalomedullary or intramedullary nail (Andalib 2020; Cai 2016; Dujardin 2001; Eceviz 2020; Kouvidis 2012; Raimondo 2012). In Dujardin 2001, the nail was described as an experimental device that is not commercially available.

Two studies used long cephalomedullary nails (Barton 2010; Little 2008), and 20 studies used mixed nail lengths or the length of the nail was unknown (Adeel 2020; Akhtar 2016; Calderon 2013; Chechik 2014; Davis 1988; Hong 2011; Kuwabara 1998; Li 2018; Lopez 2002; Matre 2013; Michos 2001; Mott 1993; O'Brien 1995; Pahlpatz 1993; Pelet 2001; Rahme 2007; Raimondo 2012; Sanders 2017; Singh 2017; Singh 2019). The remaining studies used short nails. Twelve studies reported using double femoral head screws (Dujardin 2001; Eceviz 2020; Ekstrom 2007; Giraud 2005; Haq 2014; Kouvidis 2012; Little 2008; Pajarinen 2005; Parker 2012; Parker 2017; Saudan 2002; Sharma 2018); three used a mixture of single and double femoral head screws (Andalib 2020; Papasimos 2005; Verettas 2010); one study used dual integrated screws (Sanders 2017); seven studies did not report the number of femoral head screws (Adeel 2020; Baumgaertner 1998; Calderon 2013; Guerra 2014; Rahme 2007; Sadowski 2002; Singh 2017); and the remaining studies used a single femoral head screw. Fourteen studies used blades rather than screws (Akhtar 2016; Carulli 2017; Gou 2013; Hong 2011; Li 2018; Singh 2019; Tao 2013; Wang 2019; Xu 2010; Xu 2018; Yamauchi 2014; Zehir 2015; Zhou 2012; Zou 2009) and two studies used a mixture of blades and screws (Andalib 2020; Reindl 2015). Distal locking was reported in 32% of studies, using one to two screws.

Nine studies reported using dynamic femoral head fixation (Bridle 1991; Eceviz 2020; Goldhagen 1994; Kouvidis 2012; Little 2008; Parker 2012; Parker 2017; Reindl 2015; Varela‐Egocheaga 2009) and six reported static fixation (Chechik 2014; Davis 1988; Dujardin 2001; Singh 2019; Tao 2013; Wang 2019). The remaining studies did not report whether femoral head screw fixation was static or dynamic. One study described the implant as an experimental nail (Dujardin 2001).

Extramedullary implants

Seven studies reported using static extramedullary plates (Han 2012; Haq 2014; Pelet 2001; Rahme 2007; Singh 2017; Tao 2013; Zhou 2012); the remainder all used dynamic plates. The implants were described as either dynamic hip screws (Adeel 2020; Bridle 1991; Butt 1995; Calderon 2013; Carulli 2017; Giraud 2005; Guerra 2014; Guyer 1991; Haynes 1996; Hoffmann 1999; Hong 2011; Kukla 1997; Leung 1992; O'Brien 1995; Ovesen 2006; Pahlpatz 1993; Pajarinen 2005; Radford 1993; Reindl 2015; Saudan 2002; Sharma 2018; Singh 2019; Song 2011; Verettas 2010; Wang 2019; Xu 2010; Xu 2018; Yamauchi 2014; Zehir 2015; Zou 2009), sliding hip screws (Barton 2010; Baumgaertner 1998; Davis 1988; Dujardin 2001; Eceviz 2020; Lopez 2002; Mehdi 2000; Michos 2001; Mott 1993; Parker 2012; Parker 2017; Sanders 2017), AMBI hip screws (Hoffman 1996; Kouvidis 2012; Papasimos 2005), compression hip screws (Adams 2001; Ahrengart 1994; Aktselis 2014; Benum 1994; Chechik 2014; Goldhagen 1994; Hardy 1998; Harrington 2002; Kuwabara 1998; Little 2008; Park 1998; Utrilla 2005), Less Invasive Stabilization System plate (LISS) (Tao 2013; Zhou 2012), Medoff sliding plate (Ekstrom 2007; Mehdi 2000), blade plates (Li 2018; Pelet 2001; Rahme 2007), percutaneous compression plates (Gou 2013; Singh 2017), dynamic condylar screws (Akhtar 2016; Sadowski 2002) or locking compression plates (Han 2012; Singh 2017). One study used a mixture of dynamic hip screws and dynamic condylar screws (Andalib 2020). In another study, the type of extramedullary device was not explicitly stated but from information within the report we assume that a dynamic hip screw was used (Cai 2016).

Types of outcome measures

Three studies reported no review outcomes (Akhtar 2016; Song 2011; Wang 2019). All other studies reported data contributing to the critical outcomes in the review, except Hong 2011 and Mehdi 2000; these two studies reported adverse events related to the implant, index fracture, or both.

Sources of funding and declarations of interest

Study authors reported no conflicts of interest in 45% of studies. Five studies received industry funding (Hardy 1998; Haynes 1996; Matre 2013; Miedel 2005; Sanders 2017). The remaining studies did not report sources of funding or any potential conflicts of interest.

Excluded studies

Studies previously excluded are reported in Parker 2010. Here, we report the details of 10 key excluded studies (see Characteristics of excluded studies). Lee 2007 included only participants younger than 55 years of age. This study was included in a previous version of the review (Parker 2010); we have since changed the review criteria to include adults older than 60 years, and therefore Lee 2007 is no longer eligible (see Differences between protocol and review). We excluded Stern 2011 because this study was designed to compare screws and helical blades and the cephalomedullary nails and extramedullary implants were used in both intervention groups. We excluded two studies because they were reported only as abstracts with insufficient detail to allow inclusion (Ahmad 2011; Gupta 2012). We excluded six clinical trial reports. Two of these were terminated early and have not published findings (ACTRN12608000162314; NCT03065101). Four were completed in 2011/2012, according to the clinical trials register; we excluded these because we expect publication of findings is now unlikely (NCT00686023; NCT00736684; NCT01173744; NCT01238068).

Studies awaiting classification

We received confirmation that three studies have been completed but have not yet published and data were not currently available; these have been categorised as awaiting classification (NCT02788994; NCT01380444; NCT03849014). We also identified a fourth study which appears to be the pilot study of NCT01380444 (REGAIN 2008). It is anticipated that these studies will have an estimated number of participants totalling 856. They are investigating the Endovis intermedullary nail, PFN and Gamma 3 nail, in comparison to SHS. See Characteristics of studies awaiting classification.

Ongoing studies

We found two ongoing studies (IRCT20141209020258N80; NCT03906032). Both studies compare a PFN and dynamic hip screw (DHS). These studies have an estimated enrolment of 388 participants. See Characteristics of ongoing studies.

Risk of bias in included studies

We only completed risk of bias assessments for studies that reported outcome data of interest to this review. We assessed detection bias separately for subjective and objective measures. Blank spaces in the risk of bias figure indicate that risk of bias assessment was not completed for the study or for the particular domain. See Figure 2.

Figure 2

'Risk of bias' summary: review authors' judgements about each risk of bias item for each included study. Blank spaces in the figure indicate that 'Risk of bias' judgements were not made because study authors did not report data for these outcomes.

Allocation

Twenty‐nine studies described adequate methods to randomise participants to treatment groups, and we judged these studies to be at low risk of selection bias for sequence generation (Andalib 2020; Barton 2010; Cai 2016; Chechik 2014; Davis 1988; Eceviz 2020; Ekstrom 2007; Giraud 2005; Guerra 2014; Haq 2014; Hoffman 1996; Hong 2011; Li 2018; Little 2008; Matre 2013; Mott 1993; Ovesen 2006; Pajarinen 2005; Parker 2012; Pelet 2001; Reindl 2015; Sadowski 2002; Sanders 2017; Saudan 2002; Singh 2019; Varela‐Egocheaga 2009; Xu 2010; Zehir 2015; Zhou 2012). Of these, 11 studies also reported an adequate method of concealment, and we judged these to also have a low risk of selection bias for allocation concealment (Chechik 2014; Davis 1988; Eceviz 2020; Ekstrom 2007; Hoffman 1996; Matre 2013; Ovesen 2006; Pajarinen 2005; Parker 2012; Sanders 2017; Singh 2019). Five studies reported an adequate method of allocation concealment but did not report methods for randomisation (Aktselis 2014; Adams 2001; Ahrengart 1994; Baumgaertner 1998; Parker 2017).

We judged 10 quasi‐randomised studies to be at high risk of selection bias (sequence generation) owing to the methods used to allocate participants to treatment groups (Butt 1995; Goldhagen 1994; Guyer 1991; Hardy 1998; Leung 1992; Lopez 2002; Park 1998; Sharma 2018; Verettas 2010; Yamauchi 2014). Similarly, we also judged allocation concealment to be at high risk of bias in these studies. Although Haynes 1996 reported an appropriate method of sequence generation (described as using "randomisation cards"), which could be adequately concealed, we judged the risk of selection bias for sequence generation to be high; the study reports that some surgeons may have omitted participants from the study if a card was drawn for 'Gamma nails', due to unfamiliarity with intramedullary nailing technique.

The remaining studies did not report methods for randomisation or methods used to conceal allocation. We therefore judged the risk of bias as unclear in both domains.

Blinding

It is not possible to blind clinicians to the types of surgical interventions reported in this review. However, we did not expect that surgeons' performance would be influenced by the lack of blinding, and we judged all studies to be at low risk of performance bias related to blinding.

We expected, however, that surgeons' experience in using the implants could influence their performance. We extracted descriptions in the study report that either directly described that surgeons did not have comparable experience with both types of implants in their study (Baumgaertner 1998; Guyer 1991; Harrington 2002; Haynes 1996; Leung 1992; Pelet 2001; Tao 2013), or that indirectly inferred evidence of a learning curve or similar (Benum 1994; Goldhagen 1994; Hardy 1998; Hoffman 1996; Little 2008; Mott 1993; O'Brien 1995; Papasimos 2005; Zhou 2012); we judged these 16 studies to be at high risk of performance bias related to surgeon experience. We judged 24 studies to be at low risk of performance bias related to surgeon experience because surgeons were equally experienced with each type of implant under investigation (Adams 2001; Ahrengart 1994; Aktselis 2014; Barton 2010; Bridle 1991; Cai 2016; Dujardin 2001; Eceviz 2020; Gou 2013; Haq 2014; Kouvidis 2012; Kukla 1997; Matre 2013; Mehdi 2000; Pajarinen 2005; Parker 2012; Parker 2017; Radford 1993; Sadowski 2002; Saudan 2002; Singh 2019; Utrilla 2005; Varela‐Egocheaga 2009; Xu 2010). The remaining studies reported insufficient detail and the risk of performance bias related to surgeon experience was unclear.

For detection bias, we considered whether outcomes were assessed by clinicians or participants, and whether assessment of these measures was likely to involve a subjective decision. We judged mortality to be an objective measure, and judged risk of detection bias to be low for all studies that measured this outcome. Although studies mostly did not describe whether participants were aware of treatment allocation, we judged the risk of detection bias to be low for subjective outcomes that were participant‐reported. However, we expected that all other clinically‐assessed outcomes were at high risk of detection bias because clinicians or other outcome assessors were likely to be aware of the type of treatment used.

Incomplete outcome data

For attrition bias, we considered whether study authors clearly reported participant losses, whether losses were balanced between study groups, and whether the reasons for losses seemed acceptable. We noted that most losses were caused by death and, because of the typical age of participants in these studies, we were not concerned by these losses.

In nine studies, we noted that a high number of losses were not clearly explained or were explained for reasons other than death, for example, because of loss to follow‐up (Ahrengart 1994; Benum 1994; Ekstrom 2007; Guyer 1991; Matre 2013; Pahlpatz 1993; Pajarinen 2005; Papasimos 2005; Sanders 2017). We judged these studies to be at high risk of attrition bias. Risk of attrition bias was unclear in three studies, and this was because of limited information reported in the abstract (Mehdi 2000; Raimondo 2012), and because the number of participants randomised to each group was not reported (Tao 2013).

Selective reporting

We assessed only one study to be at low risk of selective reporting bias (Sanders 2017); this study was prospectively registered with a clinical trials register and the outcomes reported in the study report were consistent with those listed in the register. Five studies were retrospectively registered with a clinical trials register, and it was not possible to use these register documents to effectively assess risk of selective reporting bias (Barton 2010; Cai 2016; Eceviz 2020; Parker 2017; Reindl 2015). We identified one clinical trials register report and could not be certain whether the report was linked to one of our included studies because of some discrepancies in the report, and we judged risk of selective reporting bias for this study to be also unclear (Chechik 2014).

Because the remaining studies did not report clinical trials registration or a prepublished protocol, it was not possible to assess risk of selective reporting bias, and we therefore judged risk of selective reporting bias in these studies to also be unclear.

Other potential sources of bias

We judged three studies to be at high risk of bias because they were reported only as abstracts which we expected were not peer‐reviewed; in addition, we could not be certain of other potential sources of bias because of the limited detail in the reports (Benum 1994; Mehdi 2000; Raimondo 2012). We noted differences in patient management between study groups in Tao 2013 and Park 1998, in particular related to the time before weight‐bearing was allowed; because this could influence the data we judged the risk of other bias to be high in these studies. We identified no other potential sources of bias in the remaining studies.

Effects of interventions

We summarise which studies are included in each analysis in Appendix 3. For outcomes measured with scales, we present the range of scores and direction of effect for each scale in Appendix 4.

We used GRADE to assess the certainty of the evidence for the critical outcomes measured within four months of surgery (activities of daily living (ADL), functional status, health‐related quality of life, and mobility), within four months and at 12 months for mortality, and at the end of follow‐up for delirium and unplanned return to theatre). For outcomes assessed using more than one measurement, we graded the evidence for the outcome with most studies or participants. See summary of findings Table 1.

We summarise the effects of other important review outcomes in a table and report the results here only when there was evidence of a difference between the interventions. No subgroup or sensitivity analyses are reported for these outcomes. We have presented GRADE assessments for adverse events that clearly favoured one treatment; we did not complete GRADE assessments for other important outcomes.

Critical outcomes

Activities of daily living

Within four months of surgery, we found the following.

We did not pool studies for the performance of ADL within four months because statistical heterogeneity was substantial (I² = 91%); see Analysis 1.1 for data from these individual studies. This outcome was measured using the Lower Extremity Measure (LEM), the Functional Independence Measure (FIM), and the Japanese Orthopaedic Association (JOA) score; higher scores in all scales indicate better performance of ADL. The studies reported these data at four weeks (Yamauchi 2014), and three months (Andalib 2020; Reindl 2015; Sanders 2017). The certainty of this evidence was very low; we downgraded by one level for serious risks of bias and by two levels for inconsistency owing to substantial levels of statistical heterogeneity.
Miedel 2005 reported the number of participants who were independent in the performance of ADL; the estimate was imprecise but suggested little evidence of a difference between interventions (RR 0.82, 95% CI 0.62 to 1.08, favours extramedullary implants; 1 study, 168 participants; Analysis 1.2).
Pahlpatz 1993 reported change in levels of independence using the Broos scale at three months. These data are reported in Appendix 5.
In addition, Aktselis 2014 reported early performance in ADL using the Barthel Index. We did not calculate an effect estimate because the number of analysed participants was unclear. See Appendix 6 for mean scores as reported by study authors.

At 12 months after surgery, we found the following.

The effect estimate for the performance of ADL was imprecise but provided evidence of little difference between interventions (SMD 0.01, 95% CI ‐0.26 to 0.27, favours cephalomedullary implants; 8 studies, 835 participants; I² = 70%; Analysis 1.4). The outcome was measured using the Barthel Index, FIM, LEM, and Jensen's scoring system; we inverted the data for the Jensen's score so that higher scores in all scales in the analysis indicate better performance in ADL. All data were reported at 12 months.
Miedel 2005 also reported the number of participants who were independent in the performance of ADL at 12 months. Again, the estimate was imprecise but suggested little evidence of a difference between interventions (RR 0.90, 95% CI 0.70 to 1.16, favours extramedullary implants; 1 study, 156 participants; Analysis 1.5).
Pahlpatz 1993 reported change in levels of independence using the Broos scale at six months and we reported these data in Appendix 5.

Delirium

The data for delirium indicated little evidence of a difference between implants, but this estimate was imprecise (RR 1.22, 95% CI 0.67 to 2.22, favours extramedullary implants; 5 studies, 1310 participants; I² = 0%; low‐certainty evidence; Analysis 1.7). Delirium was described in the studies as acute psychosis (Hoffmann 1999), mental disturbances (Papasimos 2005), confusion/delirium (Parker 2012; Parker 2017) and disorientation (Varela‐Egocheaga 2009). Time points were not clearly specified in studies; overall study follow‐up ranged from four months to 12 months. We downgraded the GRADE assessment by one level for serious risks of bias, and one level owing to imprecision denoted by the wide CI in this estimate.

Functional status

Within four months of surgery, we found the following.

We found little evidence of a difference in functional status, although the estimate was imprecise (SMD 0.02, 95% CI ‐0.27 to 0.30; 2 studies, 188 participants, favours cephalomedullary implants; I² = 0%; low‐certainty evidence; Analysis 1.8). This outcome was measured using Zűckerman functional recovery scores and a 100‐point functional recovery score; for both scales, higher scores indicate better functional status. Using the Zűckerman functional recovery, this effect estimate equates to a MD of 0.22, which is unlikely to be a clinically important difference. The studies reported these data at three months (Guerra 2014) and four months (Kouvidis 2012). We downgraded the certainty of the evidence by one level for serious risks of bias and one level for imprecision as the CI included both clinically relevant benefits and harms.
We noted similar findings when this outcome was measured as the proportion of participants with excellent or good functional status (RR 1.04, 95% CI 0.96 to 1.13, favours cephalomedullary implants; 2 studies, 188 participants; I² = 0%; Analysis 1.9). This was measured using the Harris Hip Score (HHS) and the scoring system by D'Aubigne 1954; see Appendix 5 for all categories of these scoring systems in these two studies. This was reported at three months (Xu 2018), and three to four months (Hoffmann 1999).
In addition, Raimondo 2012 reported early functional status using the HHS. We did not calculate an effect estimate because the number of analysed participants was unclear. See Appendix 6 for mean scores as reported by study authors.

At 12 months after surgery, we found the following.

We did not pool studies for functional status at 12 months because statistical heterogeneity was substantial (I² = 94%); see Analysis 1.10 for data from these individual studies. This outcome was measured using the Zűckerman functional recovery score, HHS and modified HHS, Oxford Hip Score (OHS), and a 100‐point functional recovery score which is not defined. For all scales, higher scores indicate better function. Data were reported at 16 months (Gou 2013), 18 months (Li 2018), 24 months (Singh 2017), and at 12 months in all the other studies.
This outcome was also measured as the number of participants with excellent or good functional status using the HHS score, the Sanders scoring system and the Salvati and Wilson scoring system. We found little evidence of a difference between intervention groups and the estimate was imprecise, including clinically relevant benefits and harms (RR 1.06, 95% CI 0.89 to 1.27, favours cephalomedullary implants; 3 studies, 257 participants; I² = 67%; Analysis 1.11). The data for other categories of these scoring systems in these studies is in Appendix 5.
In addition, Raimondo 2012 reported functional status at 12 months using the HHS. We did not calculate an effect estimate because the number of analysed participants was unclear. See Appendix 6 for mean scores as reported by study authors.

Health‐related quality of life

Within four months of surgery, we found the following.

Aktselis 2014 reported health‐related quality of life using EQ‐5D at three months, but we did not calculate an effect estimate because the number of analysed participants was unclear. See Appendix 6 for mean scores as reported by study authors; study authors reported a P value of 0.483 for their data.

At 12 months after surgery, we found the following.

We found little evidence of a difference in health‐related quality of life measured at 12 months in all studies using the physical component score (PCS) of SF‐12 and using EQ‐5D. The effect estimate included clinically relevant benefits and harms (SMD 0.28, 95% CI ‐0.15 to 0.71, favours cephalomedullary implants; 4 studies, 279 participants; I² = 65%; Analysis 1.12).

Mobility

Within four months of surgery, we found the following.

We found that more people had independent mobility when a cephalomedullary implant was used (RR 1.12, 95% CI 1.01 to 1.23, favours cephalomedullary implants; 7 studies, 719 participants; I² = 0%; Analysis 1.13). This was measured at three months in three studies (Carulli 2017; Guyer 1991; Park 1998), and at four months in the remaining studies. The certainty of this evidence was deemed to be very low (for reasons, see below).
We found little evidence of a difference in mobility scores when measured using the Parker 1993 mobility scale at three months (Parker 2012; Parker 2017) (MD 0.16, 95% CI ‐0.15 to 0.48, favours cephalomedullary implants; 2 studies, 695 participants; I² = 0%; Analysis 1.14); in this scale, higher scores indicate better mobility. In addition, two studies reported Parker mobility scores at six weeks (Eceviz 2020) and three months (Aktselis 2014). We did not calculate an effect estimate for these studies because the number of analysed participants was unclear and distribution values were not available. See Appendix 6 for mean scores as reported by study authors.
We also found that performance in a 10‐metre walking speed test, 14 days postoperatively, was improved for participants with a cephalomedullary implant in Li 2018 (MD 0.70, 95% CI 0.63 to 0.77, favours cephalomedullary implants; 1 study, 80 participants; Analysis 1.15).
Sanders 2017 reported this outcome as the proportion of participants who had sufficient ambulation to perform a Timed Up and Go test (TUG) at three months, and found little or no difference between interventions (RR 1.15, 95% CI 0.95 to 1.38, favours cephalomedullary implants; 1 study, 249 participants; Analysis 1.16).
Reindl 2015 reported the time to complete a TUG at three months, with no evidence of a difference in number of seconds to complete this test (MD 0.00, 95% CI ‐5.93 to 5.93, favours cephalomedullary implants; 1 study, 167 participants; Analysis 1.17).
For Analysis 1.13, we downgraded the evidence by three levels to very low certainty. We downgraded by two levels for serious risks of bias because all studies were at unclear risk of bias in at least domain, and in Park 1998 risk of other bias was high because of patient management differences between groups which could influence this outcome. We also downgraded by one level for inconsistency because we noted that effects were not consistent across the different measures of mobility at this time point; we therefore could not confidently draw conclusions about early mobility from these data.

At 12 months after surgery, we found the following.

We found that participants with cephalomedullary implants had more improvement in mobility when measured using the Parker 1993 mobility scale (MD 0.48, 95% CI 0.10 to 0.87, favours cephalomedullary implants; 14 studies, 1746 participants; I² = 63%; Analysis 1.18). This outcome was measured at 10 months (Han 2012), 16 months (Gou 2013), 24 months (Singh 2017), and at 12 months in the remaining studies. We generated a funnel plot (Figure 3), and we found no statistical evidence of small‐study effects (using Egger's test, P = 0.718).
Barton 2010 measured this outcome using a five‐point mobility scale according to the number of walking aids used, and reported as a change‐from‐baseline score. We found some evidence of a difference between intervention groups at 12 months, but the estimate was imprecise and included the possibility of little or no clinically relevant difference (MD 0.34, 95% CI ‐0.25 to 0.93, favours extramedullary implants; 1 study, 151 participants; Analysis 1.19).
We found little evidence of a difference in the proportion of people who had independent mobility (RR 1.07, 95% CI 0.94 to 1.22, favours cephalomedullary implants; 12 studies, 1524 participants; I² = 33%; Analysis 1.20). Data were reported at six months in Goldhagen 1994, Haynes 1996, Kuwabara 1998 and Zehir 2015, and at 12 months in the remaining studies. We generated a funnel plot (Figure 4), and we found no statistical evidence of small‐study effects (using the Harbord modified test, P = 0.656).
Two studies reported the proportion of people who failed to regain their pre‐fracture mobility, with little evidence of a difference between groups (RR 1.12, 95% CI 0.85 to 1.46, favours extramedullary implants; 2 studies, 246 participants; Analysis 1.23).
Matre 2013 and Sanders 2017 reported this outcome as the proportion of participants who had sufficient ambulation to perform a TUG at 12 months. However, we did not pool this data because we noted substantial statistical heterogeneity (I² = 90%); see Analysis 1.21 for data from these individual studies.
Reindl 2015 reported the time to complete a TUG at 12 months, with little evidence of a difference in the number of seconds to complete this test (MD ‐1.00, 95% CI ‐6.91 to 4.91, favours cephalomedullary implants; 1 study, 167 participants; Analysis 1.22).
Kouvidis 2012 reported the number of participants who remained in bed, or in a wheelchair, with little evidence of a difference between interventions (RR 1.61, 95% CI 0.40 to 6.45, favours extramedullary implants; 1 study, 122 participants; Analysis 1.24).

Figure 3

Figure 4

Mortality

Within four months of surgery, we found the following.

We found little evidence of a difference in early mortality between the interventions, although the estimate was imprecise, including clinically relevant benefits and harms (RR 0.96, 95% CI 0.79 to 1.18, favours cephalomedullary implants; 30 studies, 4603 participants; I² = 0%; moderate‐certainty evidence; Analysis 1.25). This outcome includes data reported during the early postoperative period, within hospital, and at one month, three months, and four months after surgery. We generated a funnel plot (Figure 5), and we found no statistical evidence of small‐study effects (using the Harbord modified test, P = 0.390).
We downgraded the evidence by one level because the evidence included studies with unclear and high risks of bias. We recognise that any benefit in this outcome is clinically meaningful for individuals who gain that benefit, such that a minimal clinically important difference for mortality is nonsensical. We also recognise that the estimate is based on data from 30 studies and 4603 participants; therefore we did not downgrade for imprecision.

Figure 5

At 12 months after surgery, we found the following.

We found a similar estimate at the later time point (RR 0.99, 95% CI 0.90 to 1.08, favours cephalomedullary implants; 47 studies, 7618 participants; I² = 0%; moderate‐certainty evidence; Analysis 1.26). Most studies reported this outcome at 12 months, but this analysis also includes data reported at five months (Butt 1995), six months (Ahrengart 1994; Bridle 1991; Dujardin 2001; Goldhagen 1994; Harrington 2002; Haynes 1996; Hoffman 1996; Kukla 1997; Leung 1992; Pahlpatz 1993; Singh 2019; Zhou 2012), 16 months (Zehir 2015), and 24 months (Gou 2013; Sharma 2018). We generated a funnel plot (Figure 6), and we found no statistical evidence of small‐study effects (using the Harbord modified test, P = 0.817).
As for the evidence for early mortality, we downgraded the certainty of this evidence by one level for risks of bias, and we did not downgrade for imprecision.

Figure 6

Unplanned return to theatre

We found little evidence of a difference in unplanned return to theatre at the end of study follow‐up according to the type of implant. The estimate was imprecise and included large clinically relevant benefits and harms (RR 1.15, 95% CI 0.89 to 1.50, favours extramedullary implants; 50 studies, 8398 participants; I² = 20%; low‐certainty evidence; Analysis 1.27). Most studies reported this outcome at 12 months, but this analysis also included data reported at three months (Giraud 2005; Guyer 1991), four months (Hoffmann 1999; Pajarinen 2005), five months (Butt 1995), six months (Ahrengart 1994; Benum 1994; Goldhagen 1994; Haynes 1996; Hoffman 1996; Kukla 1997; Leung 1992), and approximately 24 months (Sharma 2018; Singh 2017; Zhou 2012). We generated a funnel plot (Figure 7), and found no statistical evidence of small‐study effects (using the Harbord modified test, P = 0.372).

Figure 7

We downgraded the certainty of the evidence by one level for serious risks of bias (all studies in this analysis were at high risk of detection bias) and one level for imprecision. The absolute risk of return to theatre was low in both groups (approximately 5%) and so despite a large sample of 8398 participants, the CI was wide.

Other important outcomes

We report the summary effects of important outcomes in Table 3. We found little or no difference in measures of pain scores or those experiencing pain within four months of surgery, and little or no difference in the number of people experiencing pain at 12 months. We did not pool data for measures of pain at 12 months because of substantial statistical heterogeneity which we could not explain. We also noted little or no difference in length of hospital stay or in discharge destination to own home or previous residence.

Table 3. Effects of other important outcomes

Outcome	Number of studies	Studies	Participants	Effect estimate
Pain, early (≤ 4 months) Mean scores, using VAS, Salvati and Wilson scores, JOA scores; we inverted data in analysis where appropriate so that lower scores indicating less pain Follow‐up: at 4 weeks, 6 weeks, and 3 months	4	Dujardin 2001; Matre 2013; Parker 2017; Yamauchi 2014	832	SMD ‐0.13, 95% CI ‐0.43 to 0.17, favours cephalomedullary implants; I² = 67%; Analysis 1.28
Pain, early (≤ 4 months) Number of people experiencing pain Follow‐up: during postoperative period, and at 3 and 4 months	4	Aktselis 2014; Guyer 1991; Hoffmann 1999; Zehir 2015	417	RR 0.79, 95% CI 0.42 to 1.46, favours cephalomedullary implants; I² = 63%; Analysis 1.29
Pain at 12 months Mean scores, using VAS, HHS subscore; we inverted data in analysis where appropriate so that lower scores indicate less pain Follow‐up: at 12 months and 18 months	6	Chechik 2014; Li 2018; Matre 2013; Parker 2017; Sadowski 2002; Saudan 2002	1025	We did not pool these data because of substantial statistical heterogeneity (I² = 96%)
Pain at 12 months Number of people experiencing pain Follow‐up: at 6 months and 12 months	10	Ahrengart 1994; Aktselis 2014; Baumgaertner 1998; Calderon 2013; Carulli 2017; Hardy 1998; Leung 1992; Parker 2012; Pelet 2001; Utrilla 2005	952	RR 1.00, 95% CI 0.75 to 1.36, favours extramedullary implants; I² = 26%; Analysis 1.31
Length of hospital stay	26	Aktselis 2014; Barton 2010; Baumgaertner 1998; Carulli 2017; Chechik 2014; Chen 2018; Dujardin 2001; Gou 2013; Harrington 2002; Hoffman 1996; Kouvidis 2012; Kukla 1997; Leung 1992; O'Brien 1995; Ovesen 2006; Pajarinen 2005; Parker 2012; Parker 2017; Sadowski 2002; Saudan 2002; Singh 2017; Tao 2013; Varela‐Egocheaga 2009; Xu 2010; Xu 2018; Zehir 2015	3647	MD ‐0.52 days, 95% CI ‐1.23 to 0.18, favours cephalomedullary; I² = 79%; Analysis 1.32
Discharge destination Number of people discharged to own home or to previous residence	14	Baumgaertner 1998; Carulli 2017; Haynes 1996; Hoffmann 1999; Miedel 2005; Pajarinen 2005; Parker 2012; Parker 2017; Pelet 2001; Sadowski 2002; Sanders 2017; Saudan 2002; Varela‐Egocheaga 2009; Zehir 2015	2451	RR 1.00, 95% CI 0.96 to 1.04, favours extramedullary implants; I² = 0%; Analysis 1.32

CI: confidence interval; HHS: Harris Hip Score; JOA: Japanese Orthopaedic Association; MD: mean difference; RR: risk ratio: SMD: standardised mean difference: VAS: visual analogue scale

We report the summary effects of adverse effects related to the implant, index fracture, or both, in Table 4. We found fewer intraoperative periprosthetic fractures when extramedullary implants were used (RR 2.94, 95% CI 1.65 to 5.24; 35 studies, 4872 participants; I² = 0; moderate‐certainty evidence), as well as fewer postoperative periprosthetic fractures (RR 3.62, 95% CI 2.07 to 6.33; 46 studies, 7021 participants; I² = 0%; moderate‐certainty evidence). We noted that participants had fewer superficial infections with cephalomedullary implants (RR 0.71, 95% CI 0.53 to 0.96; 35 studies, 5087 participants; I² = 0%; moderate‐certainty evidence), and there were fewer non‐unions (RR 0.55, 95% CI 0.32 to 0.96; 40 studies, 4959 participants; I² = 0%; moderate‐certainty evidence). For other adverse events related to the implant, fracture or both (loosening, cut‐out, implant failure, and deep infection), we found little or no difference between interventions. See Table 4 and Analysis 1.34.

Table 4. Adverse events related to implant, fracture, or both

Outcome	Number of studies	Studies	Participants	Effect estimate; Analysis 1.34
Intraoperative periprosthetic fracture	35	Adams 2001; Ahrengart 1994; Aktselis 2014; Barton 2010; Baumgaertner 1998; Benum 1994; Bridle 1991; Ekstrom 2007; Goldhagen 1994; Guyer 1991; Hardy 1998; Harrington 2002; Hoffman 1996; Hoffmann 1999; Hong 2011; Kouvidis 2012; Kukla 1997; Kuwabara 1998; Leung 1992; Lopez 2002; Mehdi 2000; Miedel 2005; Mott 1993; O'Brien 1995; Ovesen 2006; Papasimos 2005; Park 1998; Pelet 2001; Radford 1993; Saudan 2002; Sharma 2018; Utrilla 2005; Verettas 2010; Xu 2010; Zhou 2012	4872	RR 2.94, 95% CI 1.65 to 5.24, favours extramedullary implants; I² = 0%
Postoperative periprosthetic fracture	46	Adams 2001; Ahrengart 1994; Aktselis 2014; Barton 2010; Baumgaertner 1998; Benum 1994; Bridle 1991; Butt 1995; Calderon 2013; Chechik 2014; Ekstrom 2007; Giraud 2005; Goldhagen 1994; Gou 2013; Guyer 1991; Hardy 1998; Harrington 2002; Hoffman 1996; Hoffmann 1999; Hong 2011; Kouvidis 2012; Kukla 1997; Kuwabara 1998; Leung 1992; Little 2008; Lopez 2002; Matre 2013; Michos 2001; Miedel 2005; Mott 1993; O'Brien 1995; Ovesen 2006; Pajarinen 2005; Papasimos 2005; Park 1998; Parker 2012; Parker 2017; Radford 1993; Sanders 2017; Sharma 2018; Singh 2019; Utrilla 2005; Xu 2010; Xu 2018; Zhou 2012; Zou 2009	7021	RR 3.62, 95% CI 2.07 to 6.33, favours extramedullary implants; I² = 0%
Loosening of prosthesis	3	Li 2018; Raimondo 2012; Singh 2017	195	RR 0.57, 95% CI 0.12 to 2.76, favours cephalomedullary implants; I² = 0%
Cut‐out	49	Adams 2001; Ahrengart 1994; Aktselis 2014; Barton 2010; Baumgaertner 1998; Benum 1994; Bridle 1991; Chechik 2014; Davis 1988; Ekstrom 2007; Giraud 2005; Goldhagen 1994; Guyer 1991; Hardy 1998; Harrington 2002; Haynes 1996; Hoffman 1996; Hong 2011; Kouvidis 2012; Kukla 1997; Kuwabara 1998; Leung 1992; Little 2008; Lopez 2002; Matre 2013; Mehdi 2000; Michos 2001; Miedel 2005; Mott 1993; O'Brien 1995; Ovesen 2006; Pajarinen 2005; Papasimos 2005; Park 1998; Parker 2012; Parker 2017; Pelet 2001; Radford 1993; Reindl 2015; Sadowski 2002; Saudan 2002; Singh 2019; Utrilla 2005; Varela‐Egocheaga 2009; Xu 2010; Xu 2018; Zehir 2015; Zhou 2012; Zou 2009	7843	RR 0.93, 95% CI 0.71 to 1.22, favours cephalomedullary implants; I² = 0%
Implant failure	24	Adams 2001; Adeel 2020; Aktselis 2014; Andalib 2020; Barton 2010; Butt 1995; Cai 2016; Carulli 2017; Chechik 2014; Davis 1988; Kukla 1997; Little 2008; O'Brien 1995; Pelet 2001; Radford 1993; Sadowski 2002; Sanders 2017; Saudan 2002; Sharma 2018; Utrilla 2005; Xu 2010; Xu 2018; Zhou 2012; Zou 2009	3190	RR 0.81, 95% CI 0.55 to 1.20, favours cephalomedullary implants; I² = 0%
Deep infection	35	Adams 2001; Ahrengart 1994; Aktselis 2014; Andalib 2020; Barton 2010; Cai 2016; Davis 1988; Giraud 2005; Guyer 1991; Hardy 1998; Hoffman 1996; Hoffmann 1999; Kukla 1997; Leung 1992; Little 2008; Matre 2013; Mehdi 2000; Miedel 2005; Mott 1993; O'Brien 1995; Ovesen 2006; Pajarinen 2005; Park 1998; Parker 2012; Parker 2017; Pelet 2001; Radford 1993; Reindl 2015; Sadowski 2002; Saudan 2002; Singh 2017; Utrilla 2005; Zehir 2015; Zhou 2012; Zou 2009	6184	RR 0.76, 95% CI 0.41 to 1.38. favours cephalomedullary implants; I² = 0%
Superficial infection	35	Adams 2001; Adeel 2020; Andalib 2020; Bridle 1991; Butt 1995; Cai 2016; Carulli 2017; Chechik 2014; Davis 1988; Eceviz 2020; Ekstrom 2007; Gou 2013; Kouvidis 2012; Kuwabara 1998; Lopez 2002; Little 2008; Miedel 2005; O'Brien 1995; Pajarinen 2005; Papasimos 2005; Parker 2012; Parker 2017; Radford 1993; Rahme 2007; Raimondo 2012; Sharma 2018; Singh 2017; Singh 2019; Utrilla 2005; Verettas 2010; Xu 2010; Xu 2018; Zehir 2015; Zhou 2012; Zou 2009	5087	RR 0.71, 95% CI 0.53 to 0.96, favours cephalomedullary implants; I² = 0%
Non‐union	40	Adeel 2020; Ahrengart 1994; Aktselis 2014; Andalib 2020; Barton 2010; Baumgaertner 1998; Cai 2016; Calderon 2013; Dujardin 2001; Ekstrom 2007; Giraud 2005; Goldhagen 1994; Gou 2013; Haq 2014; Hardy 1998; Harrington 2002; Hong 2011; Kouvidis 2012; Kukla 1997; Leung 1992; Li 2018; Little 2008; Michos 2001; Ovesen 2006; Papasimos 2005; Park 1998; Parker 2012; Parker 2017; Pelet 2001; Radford 1993; Rahme 2007; Sadowski 2002; Saudan 2002; Sharma 2018; Singh 2017; Singh 2019; Tao 2013; Xu 2010; Zhou 2012; Zou 2009	4959	RR 0.55, 95% CI 0.32 to 0.96, favours cephalomedullary implants ; I² = 0%

CI: confidence interval; RR: risk ratio

For adverse events unrelated to the implant, fracture, or both (acute kidney injury, blood transfusion, cerebrovascular accident, pneumonia, myocardial infarction, urinary tract infection, deep vein thrombosis, and pulmonary embolism), we found little or no difference between types of implants. See Table 5 and Analysis 1.35.

Table 5. Adverse events unrelated to implant, fracture, or both

Outcome	Number of studies	Studies	Number of participants	Effect estimate; Analysis 1.35
Acute kidney injury	2	Parker 2012; Parker 2017	1000	RR 1.19, 95% CI 0.34 to 4.19, favours extramedullary implants; I² = 0%
Blood transfusion	17	Adams 2001; Barton 2010; Harrington 2002; Little 2008; Matre 2013; Ovesen 2006; Parker 2012 Parker 2017; Raimondo 2012; Sadowski 2002; Saudan 2002; Sharma 2018; Utrilla 2005; Verettas 2010; Xu 2010; Yamauchi 2014	3726	RR 0.87, 95% CI 0.74 to 1.03, favours cephalomedullary implants; I² = 76%
Cerebrovascular accident	11	Bridle 1991; Butt 1995; Chechik 2014; Gou 2013; Hoffman 1996; Parker 2012; Parker 2017; Sadowski 2002; Varela‐Egocheaga 2009; Xu 2010; Zhou 2012	2000	RR 1.41, 95% CI 0.61 to 3.24, favours cephalomedullary implants; I² = 0%
Chest infection/pneumonia	25	Bridle 1991; Butt 1995; Cai 2016; Carulli 2017; Davis 1988; Giraud 2005; Gou 2013; Hardy 1998; Hoffman 1996; Hoffmann 1999; Kukla 1997; Little 2008; Lopez 2002; Mott 1993; O'Brien 1995; Papasimos 2005; Parker 2012; Parker 2017; Sadowski 2002; Saudan 2002; Singh 2019; Tao 2013; Varela‐Egocheaga 2009; Xu 2010; Zehir 2015	3657	RR 1.05, 95% CI 0.80 to 1.39, favours extramedullary implants; I² = 0%
Myocardial infarction/acute coronary syndrome	11	Butt 1995; Chechik 2014; Gou 2013; Hardy 1998; Hoffman 1996; Parker 2012; Parker 2017; Sadowski 2002; Saudan 2002; Varela‐Egocheaga 2009; Zhou 2012	1800	RR 0.77, 95% CI 0.44 to 1.35, favours cephalomedullary implants; I² = 0%
Urinary tract infection	16	Butt 1995; Cai 2016; Carulli 2017; Davis 1988; Hardy 1998; Hoffman 1996; Lopez 2002; O'Brien 1995; Papasimos 2005; Sadowski 2002; Saudan 2002; Tao 2013; Varela‐Egocheaga 2009; Xu 2010; Zehir 2015	1943	RR 1.06, 95% CI 0.79 to 1.41, favours extramedullary implants; I² = 11%
Deep vein thrombosis	30	Adams 2001; Ahrengart 1994; Butt 1995; Carulli 2017; Davis 1988; Giraud 2005; Gou 2013; Hardy 1998; Hoffman 1996; Hoffmann 1999; Kukla 1997; Li 2018; Little 2008; Lopez 2002; Mott 1993; Pajarinen 2005; Papasimos 2005; Parker 2012; Parker 2017; Radford 1993; Sadowski 2002; Saudan 2002; Sharma 2018; Singh 2019; Tao 2013; Utrilla 2005; Verettas 2010; Zehir 2015; Zhou 2012; Zou 2009	4589	RR 1.07, 95% CI 0.76 to 1.49, favours extramedullary implants; I² = 0%
Pulmonary embolism	14	Bridle 1991; Carulli 2017; Hardy 1998; Kukla 1997; Little 2008; O'Brien 1995; Papasimos 2005; Parker 2012; Parker 2017; Pelet 2001; Sadowski 2002; Saudan 2002; Xu 2010; Zehir 2015	2434	RR 1.27, 95% CI 0.54 to 3.03, favours extramedullary implants; I² = 0%

CI: confidence interval; RR: risk ratio

Subgroup analyses

We only conducted relevant subgroup analyses for outcomes with at least 10 studies. Overall, our analyses provided no evidence of subgroup effects between the length of cephalomedullary nail used, the stability of the fracture, or between newer and older designs of cephalomedullary nail. For a summary of the subgroup analyses, see Appendix 7. Subgroup analysis according to fracture stability for unplanned return to theatre is presented in Figure 8, and subgroup analysis according to the date of study publication for postoperative periprosthetic fractures is presented in Figure 9.

Figure 8

Figure 9

$Postoperative periprosthetic fractures: subgrouped according to date of publication$

Postoperative periprosthetic fractures: subgrouped according to date of publication

Sensitivity analysis

We excluded studies from the primary analyses of our critical outcomes that had high or unclear risks of selection bias for random sequence generation; high risk of attrition bias; high risk of performance bias because surgeons were not equally experienced with both implants; or in which the extramedullary implant had a static design. Overall, these analyses provided no evidence that decisions regarding the approach in the primary analysis influenced the inferences made. See Appendix 8.

Discussion

Summary of main results

We included 76 studies (66 randomised controlled trials (RCTs), 10 quasi‐RCTs) with a total of 10,979 participants with 10,988 extracapsular hip fractures. The majority of the studies included trochanteric fractures; 12 of these also included subtrochanteric fractures, one included only basicervical fractures and one included only subtrochanteric fractures. Three studies included only stable fractures, 17 included only unstable fractures and the remaining studies reported a mixed or unknown sample. We also identified two ongoing studies with an estimated recruitment of 388 participants.

We found little evidence to suggest that there was any difference between the interventions across the totality of our critical outcomes; see summary of findings Table 1. We collected data at two time points: within four months of surgery; and after four months of surgery, prioritising data at the 12 month time point whenever possible. We found little evidence of a difference between cephalomedullary nails and extramedullary implants in mortality within four months and 12 months of surgery; we judged this evidence to be of moderate certainty. Similarly, we found little evidence indicating any difference in unplanned return to theatre; we judged this evidence to be low‐certainty (despite a large sample size, the absolute risk of reoperation was low and the effect estimate was imprecise). The evidence for functional status at four months, and delirium, was derived from few studies and was imprecise including clinically relevant benefits and harms. We judged the certainty of the evidence for mobility at four months to be very low. Studies reported mobility using different measures (such as the number of people with independent mobility and scores on different mobility scales) and the findings from these measures were not consistent. Evidence for independent mobility was presented in most studies reporting this outcome, but these included studies at unclear risks of bias; this potential bias, alongside the inconsistency between different measures, meant that we could not be confident in the findings for early mobility. We were also very uncertain of the findings for performance of activities of daily living (ADL) at four months; we did not pool the data from the four studies because of substantial heterogeneity. Only one small study reported health‐related quality of life at four months, from which we were unable to calculate an effect estimate.

For these same outcomes but reported at 12 months, we found little evidence of any difference in the performance of ADL, in measures of health‐related quality of life, or functional status. Whilst with some instruments we found little or no difference in mobility, we noted that for one commonly used instrument, the Parker Mobility Scale, there was evidence of a benefit in mobility at 12 months with cephalomedullary nails.

In terms of other important outcomes, we identified no evidence of differences in pain, length of hospital stay or the number of people discharged to their own home or previous residence. For adverse events related to the implant or fracture, we found fewer superficial infections and non‐union when a cephalomedullary nail was used, but an increased risk of intraoperative and postoperative implant‐related fractures. The absolute risk of these events was low, and the certainty of the evidence was moderate; the difference between event risks equates to a number needed to treat for an additional harmful outcome of 67 for fracture risk, and a number needed to treat for an additional beneficial outcome of 303 for superficial infection risk when using a cephalomedullary nail. In the previous version of this review, it was noted that an evolution in nail design may reduce the implant‐related fracture risk; a subgroup analysis exploring this demonstrated no evidence to support such a hypothesis.

We performed further subgroup analyses which showed little evidence of a difference according to whether a short or long cephalomedullary nail was used, or amongst patients with stable or unstable fractures. However, many of the studies included a mix of nail lengths and fracture stabilities, thus limiting the certainty that there was no true difference between subgroups.

Overall completeness and applicability of evidence

The evidence is applicable to older adults with extracapsular fragility hip fractures sustained following low‐energy trauma. Where reported, we noted a range of mean ages from 54 to 85 years, and 72% of participants were female. We expected that most studies would include some participants with cognitive impairment; although this was often not reported, only nine studies excluded people with cognitive impairment. Studies did not consistently report American Society of Anesthesiologists (ASA) status scores to indicate participants' fitness for surgery. In general, we assess that the review includes participants that are largely representative of the general hip fracture population.

The included studies were conducted between 1988 and 2020, and more than half were conducted before 2010. Owing to limitations in the quality of reporting, we could not easily judge whether patient care pathways in these older studies were comparable to current standards of care. It is certainly possible that important developments have been made in cointerventions, such as the introduction of orthogeriatric care in some parts of the world, that have yielded improved outcomes for patients. We are unable to comment about whether such cointerventions may have changed the estimates of the relative benefits and harms between treatments reported here, or the absolute risks following treatment for extracapsular hip fractures.

The studies reported interventions that are generally available for worldwide use; only one study used a cephalomedullary implant described as an experimental design (Dujardin 2001). An evolution in nail design has occurred across the period of time that these studies have been conducted, which raises the possibility that some of the earliest data are no longer applicable to practice. However, our subgroup analysis showed no statistical evidence of a difference between studies published before and after 2010. Overall adverse events were infrequent, and a larger sample would be required to properly evaluate any temporal trends that may reflect improvement in design.

We found that few studies reported outcomes such as ADL or health‐related quality of life. These are key components of the core outcome set for hip fracture and yet our ability to draw inferences on the effect of interventions on these outcomes was limited. However, mortality and unplanned return to theatre were generally well‐reported, and these outcomes are valued by patients and clinicians in determining the effectiveness of the interventions. We note that this review does not include four studies that were completed in 2011 and 2012 which have not published their findings.

Quality of the evidence

We used GRADE to formally assess the certainty of the evidence for the critical outcomes in this review, with a particular focus on early patient‐reported outcome measures (PROMS). We judged several studies to have an unclear risk of selection bias because they did not provide information about randomisation methods; several other studies were deemed to be at high risk of selection bias because they used quasi‐randomised methods to allocate participants to groups. We used sensitivity analysis to explore this and found that re‐analysing the data without these studies sometimes influenced the direction of the effect, but this rarely changed our inferences. For most outcomes, we downgraded the certainty of the evidence for risk of selection bias. We downgraded the evidence for unplanned return to theatre because all studies for this outcome were at high risk of detection bias.

As with other hip fracture‐related Cochrane Reviews (Lewis 2021; Lewis 2022a), PROMS were reported less frequently; approximately two‐thirds of the studies predated the publication of the core outcome set which guided the selection of the critical outcomes in this review (Haywood 2014). Where estimates were imprecise, as demonstrated by a wide confidence interval or few study participants, we downgraded for imprecision.

We also downgraded for inconsistency because we were unable to pool data for performance of ADL, owing to substantial statistical heterogeneity. Although we attempted to explore this in the sensitivity analysis, we had insufficient studies to confidently ascertain the reason for this heterogeneity. We did not downgrade the evidence for indirectness as the study populations and types of interventions were consistent with our protocol. We evaluated the risk of publication bias in only six analyses (in which we had more than 10 studies) and found no reason to downgrade the evidence for this potential limitation.

Potential biases in the review process

The review authors conducted a thorough search and independently assessed study eligibility, extracted data, and assessed risk of bias in the included studies before reaching consensus together or with one other review author. This is an update of a previous Cochrane Review from 2010 (Parker 2010), and we have made minor changes to the review in order to meet current methodological expectations in Cochrane intervention reviews (MECIR). The review forms part of a series of Cochrane Reviews of surgery for hip fractures (Lewis 2021; Lewis 2022a; Lewis 2022b; Lewis 2022c). In addition to methodological changes, we made changes to the review in response to guidance resulting from the prioritisation process underpinning this project.

We included only older adults in this review update, in order to better reflect the general population with low‐energy fragility hip fractures. This resulted in the exclusion of just one study. We captured outcome data at an additional earlier time point (within four months of surgery); previously, the review included data only at 12 months. There is increasing loss to follow‐up over the first year after surgery and some evidence of consistency between quality of life and 'poor outcome' (dead or deterioration in residential status) at four months and 12 months (Griffin 2015). We judged that the earlier time point would provide valuable data. We also restructured the outcomes, bringing them in line with those identified during the prioritisation process and introducing seven critical outcomes consistent with the recommendations from the core outcome set for hip fracture (Haywood 2014). This restructuring resulted in the loss of a small number of outcomes from the review, however the data are still available in Parker 2010. We note that the data for most of the removed outcomes were sparse and typically heterogenous.

The review includes cephalomedullary nails and extramedullary implants from different manufacturers, and there is inevitable variation in the precise detail of their design. We made the assumption that this variation was unlikely to be clinically relevant and chose to group implants from different manufacturers in the analyses. Following consensus discussions with clinicians, we subgrouped the data according to the length of cephalomedullary nails, fracture stability, and (in order to explore newer and older designs of cephalomedullary nals) also by date of reporting; we used sensitivity analysis to remove static designs from the evidence set. These approaches were, however, very limited in explaining variation between the studies because most studies reported using mixed types of implants or only short nails, or included a mixed population of fractures.

We used GRADE only to assess the certainty of the evidence for the critical outcomes in this review that were included in our summary of findings table, as well as for adverse events that indicated a clear improvement or risk with one treatment. We did not report any judgements of certainty for the remaining review outcomes.

Agreements and disagreements with other studies or reviews

The previous version of this review indicated that the sliding hip screw (SHS) appeared to be superior to cephalomedullary nails (Parker 2010); the evidence indicated a lower complication rate for the SHS and an absence of outcome data to support the use of the cephalomedullary nail. Bhandari 2009, which only included studies published up to 2005, also reported findings suggesting that previous concerns about the risk of increased femoral shaft fracture with Gamma nails may have been resolved with improved implant design and improved learning curves with the devices. Another review that specifically focused on the impact of different generations of Gamma nails included studies up to and including 2010, however not all studies were randomised or had a comparator (Noris 2012). The findings of the review by Noris and colleagues also suggested a reduced risk of postoperative fracture but did not address functional and mobility outcomes.

The findings of a more recent review and meta‐analysis reported the effectiveness of different implants for trochanteric fractures (Arirachakaran 2017); these included the dynamic hip screw, Medoff sliding plate, percutaneous compression plating, proximal femoral nails, Gamma nails, and Less Invasive Stabilisation System. However, the key outcomes in the work by Ariachakaran and colleagues were operative time, blood loss and hospital stay, which differ from our critical outcomes.

Other reviews in this area focused on specific types of implants such as short or long nails (Bovbjerg 2019), single or double screws (Cipollaro 2019), whether to use distal locking (Li 2020), or whether reaming was necessary (Clark 2021). Although we explored the length of cephalomedullary nails in subgroup analysis, our analyses included few studies of only long nails and we could not confidently report differences between the two lengths.

Our review included two large multicentre studies (of over 500 participants) published within the last ten years, the findings of which are consistent with our review (Matre 2013; Parker 2012). A further large, multicentre study is due to be published soon (NCT01380444); this may influence the results of our review and will be included in future updates.

Figure 1

Flow diagram. Search conducted in November 2018 and December 2018, with a top‐up search in July 2020.

Figure 2

Figure 3

Figure 4

Figure 5

Figure 6

Figure 7

Figure 8

$Postoperative periprosthetic fractures: subgrouped according to date of publication$

Figure 9

Postoperative periprosthetic fractures: subgrouped according to date of publication

Analysis 1.1

Comparison 1: Cephalomedullary nails versus extramedullary implants, Outcome 1: ADL, early (≤ 4 months)

Analysis 1.2

Comparison 1: Cephalomedullary nails versus extramedullary implants, Outcome 2: ADL (≤ 4 months; independent in performance of ADL)

Analysis 1.3

Comparison 1: Cephalomedullary nails versus extramedullary implants, Outcome 3: ADL, early (≤ 4 months; change in social dependency scale)

Analysis 1.4

Comparison 1: Cephalomedullary nails versus extramedullary implants, Outcome 4: ADL at 12 months

Analysis 1.5

Comparison 1: Cephalomedullary nails versus extramedullary implants, Outcome 5: ADL (12 months; independent in performance of ADL)

Analysis 1.6

Comparison 1: Cephalomedullary nails versus extramedullary implants, Outcome 6: ADL at 12 months (change scores in social dependency scale

Analysis 1.7

Comparison 1: Cephalomedullary nails versus extramedullary implants, Outcome 7: Delirium

Analysis 1.8

Comparison 1: Cephalomedullary nails versus extramedullary implants, Outcome 8: Functional status, early (≤ 4 months)

Analysis 1.9

Comparison 1: Cephalomedullary nails versus extramedullary implants, Outcome 9: Functional status, early (≤ 4 months; excellent or good)

Analysis 1.10

Comparison 1: Cephalomedullary nails versus extramedullary implants, Outcome 10: Functional status at 12 months (mean scores)

Analysis 1.11

Comparison 1: Cephalomedullary nails versus extramedullary implants, Outcome 11: Functional status (12 months; excellent or good using HHS)

Analysis 1.12

Comparison 1: Cephalomedullary nails versus extramedullary implants, Outcome 12: HRQoL at 12 months

Analysis 1.13

Comparison 1: Cephalomedullary nails versus extramedullary implants, Outcome 13: Mobility (≤ 4 months; independent mobility)

Analysis 1.14

Comparison 1: Cephalomedullary nails versus extramedullary implants, Outcome 14: Mobility, early (≤ 4 months; mobility scales, mean scores)

Analysis 1.15

Comparison 1: Cephalomedullary nails versus extramedullary implants, Outcome 15: Mobility (≤ 4 months; 10 metre walking speed test)

Analysis 1.16

Comparison 1: Cephalomedullary nails versus extramedullary implants, Outcome 16: Mobility (≤ 4 months; able to complete TUG)

Analysis 1.17

Comparison 1: Cephalomedullary nails versus extramedullary implants, Outcome 17: Mobility, early (≤ 4 months; TUG, mean scores)

Analysis 1.18

Comparison 1: Cephalomedullary nails versus extramedullary implants, Outcome 18: Mobility at 12 months (mobility scales, mean scores)

Analysis 1.19

Comparison 1: Cephalomedullary nails versus extramedullary implants, Outcome 19: Mobility (at 12 months; change from baseline)

Analysis 1.20

Comparison 1: Cephalomedullary nails versus extramedullary implants, Outcome 20: Mobility (12 months; independent mobility)

Analysis 1.21

Comparison 1: Cephalomedullary nails versus extramedullary implants, Outcome 21: Mobility (12 months; able to complete TUG)

Analysis 1.22

Comparison 1: Cephalomedullary nails versus extramedullary implants, Outcome 22: Mobility at 12 months (TUG, mean scores)

$Comparison 1: Cephalomedullary nails versus extramedullary implants, Outcome 23: Failure to regain pre‐fracture mobility (at 12 months)$

Analysis 1.23

Comparison 1: Cephalomedullary nails versus extramedullary implants, Outcome 23: Failure to regain pre‐fracture mobility (at 12 months)

Analysis 1.24

Comparison 1: Cephalomedullary nails versus extramedullary implants, Outcome 24: Mobility at 12 months (remained in bed or wheelchair)

Analysis 1.25

Comparison 1: Cephalomedullary nails versus extramedullary implants, Outcome 25: Mortality, early (≤ 4 months)

Analysis 1.26

Comparison 1: Cephalomedullary nails versus extramedullary implants, Outcome 26: Mortality at 12 months

Analysis 1.27

Comparison 1: Cephalomedullary nails versus extramedullary implants, Outcome 27: Unplanned return to theatre

Analysis 1.28

Comparison 1: Cephalomedullary nails versus extramedullary implants, Outcome 28: Pain, early (≤ 4 months; pain scales, mean scores)

Analysis 1.29

Comparison 1: Cephalomedullary nails versus extramedullary implants, Outcome 29: Experiencing pain (≤ 4 months)

Analysis 1.30

Comparison 1: Cephalomedullary nails versus extramedullary implants, Outcome 30: Pain at 12 months (pain scales, mean scores)

Analysis 1.31

Comparison 1: Cephalomedullary nails versus extramedullary implants, Outcome 31: Experiencing pain (at 12 months)

Analysis 1.32

Comparison 1: Cephalomedullary nails versus extramedullary implants, Outcome 32: Length of hospital stay (days)

Analysis 1.33

Comparison 1: Cephalomedullary nails versus extramedullary implants, Outcome 33: Discharge destination (to own home/previous residence)

$Comparison 1: Cephalomedullary nails versus extramedullary implants, Outcome 34: Adverse event related to implant, fracture, or both$

Analysis 1.34

Comparison 1: Cephalomedullary nails versus extramedullary implants, Outcome 34: Adverse event related to implant, fracture, or both

$Comparison 1: Cephalomedullary nails versus extramedullary implants, Outcome 35: Adverse events unrelated to implant, fracture, or both$

Analysis 1.35

Comparison 1: Cephalomedullary nails versus extramedullary implants, Outcome 35: Adverse events unrelated to implant, fracture, or both

Analysis 2.1

Comparison 2: Cephalomedullary nails versus extramedullary implants: subgrouped by short or long intramedullary nails, Outcome 1: Functional status at 12 months (mean scores)

Analysis 2.2

Comparison 2: Cephalomedullary nails versus extramedullary implants: subgrouped by short or long intramedullary nails, Outcome 2: Mobility at 12 months (mobility scales, mean scores)

Analysis 2.3

Comparison 2: Cephalomedullary nails versus extramedullary implants: subgrouped by short or long intramedullary nails, Outcome 3: Mobility (12 months; independent mobility)

Analysis 2.4

Comparison 2: Cephalomedullary nails versus extramedullary implants: subgrouped by short or long intramedullary nails, Outcome 4: Early mortality

Analysis 2.5

Comparison 2: Cephalomedullary nails versus extramedullary implants: subgrouped by short or long intramedullary nails, Outcome 5: Mortality at 12 months

Analysis 2.6

Comparison 2: Cephalomedullary nails versus extramedullary implants: subgrouped by short or long intramedullary nails, Outcome 6: Unplanned return to theatre

$Comparison 3: Cephalomedullary nails versus extramedullary implants: subgrouped by stable and unstable fractures, Outcome 1: Functional status at 12 months (mean scores)$

Analysis 3.1

Comparison 3: Cephalomedullary nails versus extramedullary implants: subgrouped by stable and unstable fractures, Outcome 1: Functional status at 12 months (mean scores)

$Comparison 3: Cephalomedullary nails versus extramedullary implants: subgrouped by stable and unstable fractures, Outcome 2: Mobility at 12 months (mobility scales, mean scores)$

Analysis 3.2

Comparison 3: Cephalomedullary nails versus extramedullary implants: subgrouped by stable and unstable fractures, Outcome 2: Mobility at 12 months (mobility scales, mean scores)

$Comparison 3: Cephalomedullary nails versus extramedullary implants: subgrouped by stable and unstable fractures, Outcome 3: Mobility (12 months; independent mobility)$

Analysis 3.3

Comparison 3: Cephalomedullary nails versus extramedullary implants: subgrouped by stable and unstable fractures, Outcome 3: Mobility (12 months; independent mobility)

$Comparison 3: Cephalomedullary nails versus extramedullary implants: subgrouped by stable and unstable fractures, Outcome 4: Early mortality$

Analysis 3.4

Comparison 3: Cephalomedullary nails versus extramedullary implants: subgrouped by stable and unstable fractures, Outcome 4: Early mortality

$Comparison 3: Cephalomedullary nails versus extramedullary implants: subgrouped by stable and unstable fractures, Outcome 5: Mortality at 12 months$

Analysis 3.5

Comparison 3: Cephalomedullary nails versus extramedullary implants: subgrouped by stable and unstable fractures, Outcome 5: Mortality at 12 months

$Comparison 3: Cephalomedullary nails versus extramedullary implants: subgrouped by stable and unstable fractures, Outcome 6: Unplanned return to theatre$

Analysis 3.6

Comparison 3: Cephalomedullary nails versus extramedullary implants: subgrouped by stable and unstable fractures, Outcome 6: Unplanned return to theatre

$Comparison 4: Intraoperative and postoperative periprosthetic fractures: subgrouped by year of publication, Outcome 1: Intraoperative periprosthetic fracture$

Analysis 4.1

Comparison 4: Intraoperative and postoperative periprosthetic fractures: subgrouped by year of publication, Outcome 1: Intraoperative periprosthetic fracture

$Comparison 4: Intraoperative and postoperative periprosthetic fractures: subgrouped by year of publication, Outcome 2: Postoperative periprosthetic fracture$

Analysis 4.2

Comparison 4: Intraoperative and postoperative periprosthetic fractures: subgrouped by year of publication, Outcome 2: Postoperative periprosthetic fracture

Summary of findings 1. Cephalomedullary nails compared to extramedullary implants for extracapsular hip fractures in adults

Outcomes	*Anticipated absolute effects^ (95% CI)**		Relative effect (95% CI)	Number of participants (studies)	Certainty of the evidence (GRADE)	Comments
Cephalomedullary nails compared to extramedullary implants for extracapsular hip fractures in adults
Population: older adults with stable or unstable extracapsular hip fractures Setting: hospitals; included studies were conducted in: Australia, Austria, Brazil, Canada, China, Denmark, Finland, France, Greece, Hong Kong, India, Iran, Israel, Italy, Japan, Mexico, New Zealand, Norway, Pakistan, South Korea, Spain, Sweden, Switzerland, The Netherlands, Turkey, USA, UK Intervention: cephalomedullary nails (Gamma nail, Gamma 3 nail, PFN, ultra‐short PFN, expandable PFN, PFNA, Targon PFN, TRIGEN INTERTAN nail, Holland nail, Küntscher‐Y nail) Comparison: extramedullary implants (SHS, DHS, ABMI hip screw, compression hip screw, LISS, Medoff sliding plate, blade plates, percutaneous compression plate, dynamic Condylar screw, locking compression plate)
Outcomes	Risk with extramedullary implants	Risk with cephalomedullary nails	Relative effect (95% CI)	Number of participants (studies)	Certainty of the evidence (GRADE)	Comments
Activities of daily living (ADL), early (≤ 4 months): using LEM (range from 0 to 100), FIM (range from 0 to 100), JOA (range from 0 to 20); higher scores indicate better performance in ADL Follow‐up: time points in the included studies were at 4 weeks and 3 months	‐	‐	‐	509 (4 studies)	Very low^a	We did not pool data because of high statistical heterogeneity.
Delirium (at end of follow‐up) Follow‐up: time points in the included studies were 4 months and 12 months	Study population		RR 1.22 (0.67 to 2.22)	1310 (5 studies)	Low^c
	30 per 1,000^b	37 per 1000 (20 to 67)
Functional status, early (≤ 4 months): using Zűckerman functional recovery score (0 to 44), and 100‐point functional recovery scale; in both scales, higher scores indicate better function Follow‐up: time points in the included studies were at 3 months and 4 months			SMD 0.02 higher (‐0.27 lower to 0.3 higher)	188 (2 studies)	Low^c	This effect did not indicate a clinically important difference, based on a 'rule of thumb' of: 0.2 for a small difference, 0.5 for a medium difference, and 0.8 for a large difference. Using the Zűckerman functional recovery score, this equates to a MD of 0.22 (this is unlikely to represent a clinically important difference on this 44‐point scale)
Health‐related quality of life, early (≤ 4 months)	‐	‐	‐			Inestimable
Mobility (≤ 4 months): assessed as number of participants with independent mobility Follow‐up: time points in the included studies were at 3 months and 4 months	Study population		RR 1.12 (1.01 to 1.23)	719 (7 studies)	Very low^d
	594 per 1,000^b	665 per 1000 (600 to 730)	RR 1.12 (1.01 to 1.23)	719 (7 studies)	Very low^d
Mortality, early (≤ 4 months) Follow‐up: time points in the included studies were during early postoperative period, within hospital, and at 1 month, 3 months, and 4 months	Study population		RR 0.96 (0.79 to 1.18)	4603 (30 studies)	Moderate^e
	83 per 1,000^b	80 per 1000 (66 to 98)	RR 0.96 (0.79 to 1.18)	4603 (30 studies)	Moderate^e
Mortality at 12 months Follow‐up: time points in the included studies were at 5 months, 6 months, 12 months, and 24 months	Study population		RR 0.99 (0.90 to 1.08)	7618 (47 studies)	Moderate^e
	204 per 1000^b	202 per 1000 (184 to 220)	RR 0.99 (0.90 to 1.08)	7618 (47 studies)	Moderate^e
Unplanned return to theatre (at end of follow‐up) Follow‐up: time points in the included studies were 3 months, 4 months, 5 months, 6 months, 12 months, and 24 months	Study population		RR 1.15 (0.89 to 1.50)	8398 (50 studies)	Low^f
	43 per 1,000^b	49 per 1000 (38 to 64)	RR 1.15 (0.89 to 1.50)	8398 (50 studies)	Low^f
*The risk in the intervention group (and its 95% confidence interval) is based on the assumed risk in the comparison group and the relative effect of the intervention (and its 95% CI). AMBI: manufacturer name for implant; CI: confidence interval; DHS: dynamic hip screw; FIM: functional independence measure; JOA: Japanese Orthopaedic Association;LEM: lower extremity measure; LISS: less invasive stabilisation system; MD: mean difference; PFN: proximal femoral nail; PFNA: proximal femoral nail antirotation; RR: risk ratio; SHS: sliding hip screw; SMD: standardised mean difference
GRADE Working Group grades of evidence High certainty: we are very confident that the true effect lies close to that of the estimate of the effect Moderate certainty: we are moderately confident in the effect estimate. The true effect is likely to be close to the estimate of the effect, but there is a possibility that it is substantially different Low certainty: our confidence in the effect estimate is limited. The true effect may be substantially different from the estimate of the effect Very low certainty: we have very little confidence in the effect estimate. The true effect is likely to be substantially different from the estimate of effect
^aDowngraded by three levels: one level for serious risks of bias and two levels for inconsistency owing to high levels of unexplained statistical heterogeneity ^bDerived from the pooled estimate of the cephalomedullary nails group ^cDowngraded by two levels: one level for serious risks of bias, and one level owing to imprecision denoted by the wide CI in this estimate. ^dDowngraded by two levels for serious risks of bias, and one level for inconsistency because this effect was not always apparent in other measures of early mobility (such as when measured using mobility scores) ^eDowngraded by one level for serious risks of bias ^fDowngraded by two levels: one level for serious risks of bias because all studies in this analysis were at high risk of detection bias, and one level for imprecision denoted by the wide CI in this estimate

Summary of findings 1. Cephalomedullary nails compared to extramedullary implants for extracapsular hip fractures in adults

Table 1. Extramedullary devices evaluated by included trials

Name	Description
Sliding hip screw (SHS)	The SHS (DePuy Synthes) consists of a lag screw passed up the femoral neck to the femoral head. This lag screw is then attached to a plate on the side of the femur typically at 135º (130º to 150º available). These are considered 'dynamic' implants as they have the capacity for sliding at the plate/screw junction to allow for controlled collapse at the fracture site, thereby facilitating fracture healing.
Medoff sliding plate	The Medoff sliding plate (Swemac Ltd) is a modification of the sliding hip screw, the difference being the plate having an inner and outer sleeve, which can slide between each other. This creates an additional capacity for sliding to occur at the level of the lesser trochanter as well as at the lag screw. Sliding at the lag screw can be prevented with a locking screw to create a 'one way' sliding Medoff instead of a 'two way' sliding Medoff. At a later date the locking device on the lag screw can be removed to 'dynamise' the fracture.
Percutaneous compression plate (PCCP)	The PCCP (Orthofix) is an extramedullary device developed by Gotfried in the late 1990s. Similar to the SHS, it utilises a telescoping mechanism in the femoral neck to facilitate collapse of the fracture. It differs in that it is minimally invasive (inserted by 2 small incisions) and uses 2 small screws in the femoral head as opposed to one large screw (SHS). This design is to provide double axis fixation to prevent femoral neck rotation and also prevent damage to the lateral femoral cortex as 2 small screws are used.
Dynamic condylar screw (DCS)	The DCS (DePuy Synthes) device is similar to the SHS device described above. It consists of a lag screw placed in the femoral head that attaches to a plate on the side of the femur via a barrel. It differs however in the angle the lag screw is attached to the plate (95º). This acute angle means that the DCS is most likely to act as a static device with little or no movement taking place at the screw/barrel junction.
Proximal femoral locking plate (PFLP)	The PFLP (DePuy Synthes) device is a pre‐contoured fixed angle device where multiple screws (7.3 mm and 5 mm) are placed in the femoral head and fixed to a pre‐contoured 4.5 mm plate with a locking mechanism. This ensures it acts as a static device that does not allow movement at the fracture site.
Reverse distal femoral less invasive stabilisation system plate (rDF LISS)	The rDF LISS plate (DePuy Synthes) is a pre‐contoured fixed angle devices used for distal femoral fractures. It is essentially a locking plate that can be applied using a minimally invasive technique. It has been used for contralateral proximal femoral fractures by reversing the plate position and placing it on the proximal femur (Zhou 2012).
Reverse distal femoral locking compression plate (rDFLP)	The rDFLP (Greens Surgical, India) is a pre‐contoured fixed angle device designed for distal femoral fractures. It has combination holes in the area of the plate placed on the femoral shaft allowing locked and non‐locked screw placement. It can be used for contralateral proximal femoral fractures by reversing the plate position and placing it on the proximal femur (Haq 2014).
Blade plate	The blade plate is a fixed‐angle device where the blade (attached to a plate) is placed in the centre of the femoral head. The angle at the blade/plate junction is typically 95% with plate lengths of 50 mm to 80 mm.

Table 1. Extramedullary devices evaluated by included trials

Table 2. Cephalomedullary nails evaluated by the included trials

Name	Description
Gamma nail	The Gamma nail (Stryker) was introduced in the late 1980s for the treatment of extracapsular hip fractures. The implant consists of a sliding lag screw which passes through a short cephalomedullary nail. One or two screws may be passed through the nail tip to secure it to the femoral shaft (distal locking). Theoretical advantages of this implant are due to a percutaneous insertion technique and include reduced blood loss, reduced sepsis, minimal tissue trauma and reduced operating time. Modifications to the design of the Gamma nail and its instrumentation have occurred since its introduction. The long Gamma nail has a range of different lengths from 280 mm to 460 mm with two distal locking screws.
Gamma 3 nail	The Gamma 3 nail (Stryker) is the third generation of the gamma nail fixation system for proximal femoral fractures. It is a trochanteric entry nail with a reduced proximal nail diameter (15.5 mm versus 17 mm) to facilitate a shorter incision. Its length options range from 280 mm to 460 mm. Its neck‐shaft angle options include 120°, 125° and 130°. The lag screw shape has also been modified to provide superior cutting behaviour and greater resistance to cut‐out.
Intramedullary hip screw (IMHS)	The IMHS (Smith & Nephew), length 210 mm, was introduced in 1991 for the treatment of extracapsular femoral fractures. Like the Gamma nail, it consists of a nail inserted via the greater trochanter into the medullary cavity. It utilises a single screw in the femoral head that can slide through a barrel in the nail allowing fracture compression. Three different neck angles are available: 125°, 130° and 135°. Nail lengths are available from 195 mm to 440 mm.
Proximal femoral nail (PFN)	The PFN (DePuy Synthes), length 240 mm, was introduced in 1998 for the treatment of extracapsular fractures. Like the Gamma and IMHS, it consists of a nail inserted via the greater trochanter into the medullary cavity. Three lengths are available: 240 mm, 200 mm and an ultra‐short 180 mm. Two proximal lag screws are passed up the femoral neck to the head. Distal locking can performed in static or dynamic mode via two distal locking screws.
Proximal femoral nail antirotation (PFNA)	The PFNA (DePuy Synthes), length 170 mm, 200 mm or 240 mm, is a modification of the PFN. It is similar to the PFN apart from not having two proximal lag screws but instead a single helically‐shaped blade which is designed to provide increased angular and rotational stability. The helical blade is designed to avoid bone loss that occurs during drilling and insertion of a standard hip screw. It has 2 distal locking screw options for either dynamic or static locking. Blade‐shaft angle options include 125°, 130° and 135°.
Targon proximal femoral nail (PF)	The Targon PF (B. Braun), length 220 mm, is inserted into the intramedullary cavity via a trochanteric entry point. Proximally, this nail has a sliding lag screw and an antirotation pin. The Targon PF facilitates fracture dynamisation via a gliding screw that glides through a sleeve that is attached to the nail, thereby avoiding protrusion of the screw into peritrochanteric tissues.
Holland nail	The Holland nail (Zimmer Biomet) is like the Gamma and IMHS; it consists of a nail inserted via the greater trochanter in to the medullary cavity. Two proximal lag screws are passed up the femoral neck to the head.
Experimental nail (reported in Dujardin 2001)	An experimental mini‐invasive static intramedullary nail, which is not commercially available, is reported in Dujardin 2001. This consists of an intramedullary nail which is 170 mm long with a distal diameter of 12 mm and a proximal diameter of 13 mm. There are two five‐mm distal locking holes. The proximal hold of the femur is with two 7‐mm cannulated screws which diverge at a 30‐degree angle. Unlike the other proximal femoral nails, there is no sliding mechanism within the nail construct.
Kuntscher‐Y nail	The Kuntscher‐Y nail (Cuthbert 1976) is an early design of an intramedullary nail. It consists of a side arm and a separate slotted Kuntscher nail. The side arm is passed up the femoral neck, and then attached to an alignment jig to enable a slotted Kuntscher nail to be passed via the greater trochanter through a hole in the side arm and distally within the medullary cavity. The assembled implant construct has no capacity for sliding at the side arm and neither has it the capacity for distal locking.
Endovis nail	The Endovis nail (Citieffe) is available in 3 sizes (195 mm to 400 mm) and has a neck shaft angle of 130°. It has two cephalic screws for the femoral head to facilitate fracture compression. The distal section is slotted to produce a graduated variation of stiffness.
TRIGEN INTERTAN nail	The INTERTAN nail (Smith & Nephew) uses 2 cephalocervical screws in an integrated mechanism allowing intraoperative compression and rotational stability of the head‐neck fragments. It has a cannulated set screw mechanism that allows for the device to be used in fixed angle mode or in sliding/compression mode. Its length ranges from 18 cm to 46 cm (long nail option).
Russell‐Taylor Recon nail	The Russel‐Taylor Recon nail (Smith & Nephew) is an intramedullary nail that utilises a piriformis entry point. Two screws are available for fixation in the femoral head. It is a full length femoral nail with no short versions available for proximal femoral fixation only.
Trochanteric Fixation Nail (TFN)	The TFN nail (DePuy Synthes) is a titanium nail that utilises a helical blade for fixation in the femoral head instead of a lag screw. This design is intended to improve resistance to various collapse and improved rotational control of the medial fracture segment theoretically reducing the rate of cut‐out.

Table 2. Cephalomedullary nails evaluated by the included trials

Table 3. Effects of other important outcomes

Outcome	Number of studies	Studies	Participants	Effect estimate
Pain, early (≤ 4 months) Mean scores, using VAS, Salvati and Wilson scores, JOA scores; we inverted data in analysis where appropriate so that lower scores indicating less pain Follow‐up: at 4 weeks, 6 weeks, and 3 months	4	Dujardin 2001; Matre 2013; Parker 2017; Yamauchi 2014	832	SMD ‐0.13, 95% CI ‐0.43 to 0.17, favours cephalomedullary implants; I² = 67%; Analysis 1.28
Pain, early (≤ 4 months) Number of people experiencing pain Follow‐up: during postoperative period, and at 3 and 4 months	4	Aktselis 2014; Guyer 1991; Hoffmann 1999; Zehir 2015	417	RR 0.79, 95% CI 0.42 to 1.46, favours cephalomedullary implants; I² = 63%; Analysis 1.29
Pain at 12 months Mean scores, using VAS, HHS subscore; we inverted data in analysis where appropriate so that lower scores indicate less pain Follow‐up: at 12 months and 18 months	6	Chechik 2014; Li 2018; Matre 2013; Parker 2017; Sadowski 2002; Saudan 2002	1025	We did not pool these data because of substantial statistical heterogeneity (I² = 96%)
Pain at 12 months Number of people experiencing pain Follow‐up: at 6 months and 12 months	10	Ahrengart 1994; Aktselis 2014; Baumgaertner 1998; Calderon 2013; Carulli 2017; Hardy 1998; Leung 1992; Parker 2012; Pelet 2001; Utrilla 2005	952	RR 1.00, 95% CI 0.75 to 1.36, favours extramedullary implants; I² = 26%; Analysis 1.31
Length of hospital stay	26	Aktselis 2014; Barton 2010; Baumgaertner 1998; Carulli 2017; Chechik 2014; Chen 2018; Dujardin 2001; Gou 2013; Harrington 2002; Hoffman 1996; Kouvidis 2012; Kukla 1997; Leung 1992; O'Brien 1995; Ovesen 2006; Pajarinen 2005; Parker 2012; Parker 2017; Sadowski 2002; Saudan 2002; Singh 2017; Tao 2013; Varela‐Egocheaga 2009; Xu 2010; Xu 2018; Zehir 2015	3647	MD ‐0.52 days, 95% CI ‐1.23 to 0.18, favours cephalomedullary; I² = 79%; Analysis 1.32
Discharge destination Number of people discharged to own home or to previous residence	14	Baumgaertner 1998; Carulli 2017; Haynes 1996; Hoffmann 1999; Miedel 2005; Pajarinen 2005; Parker 2012; Parker 2017; Pelet 2001; Sadowski 2002; Sanders 2017; Saudan 2002; Varela‐Egocheaga 2009; Zehir 2015	2451	RR 1.00, 95% CI 0.96 to 1.04, favours extramedullary implants; I² = 0%; Analysis 1.32
CI: confidence interval; HHS: Harris Hip Score; JOA: Japanese Orthopaedic Association; MD: mean difference; RR: risk ratio: SMD: standardised mean difference: VAS: visual analogue scale

Table 3. Effects of other important outcomes

Table 4. Adverse events related to implant, fracture, or both

Outcome	Number of studies	Studies	Participants	Effect estimate; Analysis 1.34
Intraoperative periprosthetic fracture	35	Adams 2001; Ahrengart 1994; Aktselis 2014; Barton 2010; Baumgaertner 1998; Benum 1994; Bridle 1991; Ekstrom 2007; Goldhagen 1994; Guyer 1991; Hardy 1998; Harrington 2002; Hoffman 1996; Hoffmann 1999; Hong 2011; Kouvidis 2012; Kukla 1997; Kuwabara 1998; Leung 1992; Lopez 2002; Mehdi 2000; Miedel 2005; Mott 1993; O'Brien 1995; Ovesen 2006; Papasimos 2005; Park 1998; Pelet 2001; Radford 1993; Saudan 2002; Sharma 2018; Utrilla 2005; Verettas 2010; Xu 2010; Zhou 2012	4872	RR 2.94, 95% CI 1.65 to 5.24, favours extramedullary implants; I² = 0%
Postoperative periprosthetic fracture	46	Adams 2001; Ahrengart 1994; Aktselis 2014; Barton 2010; Baumgaertner 1998; Benum 1994; Bridle 1991; Butt 1995; Calderon 2013; Chechik 2014; Ekstrom 2007; Giraud 2005; Goldhagen 1994; Gou 2013; Guyer 1991; Hardy 1998; Harrington 2002; Hoffman 1996; Hoffmann 1999; Hong 2011; Kouvidis 2012; Kukla 1997; Kuwabara 1998; Leung 1992; Little 2008; Lopez 2002; Matre 2013; Michos 2001; Miedel 2005; Mott 1993; O'Brien 1995; Ovesen 2006; Pajarinen 2005; Papasimos 2005; Park 1998; Parker 2012; Parker 2017; Radford 1993; Sanders 2017; Sharma 2018; Singh 2019; Utrilla 2005; Xu 2010; Xu 2018; Zhou 2012; Zou 2009	7021	RR 3.62, 95% CI 2.07 to 6.33, favours extramedullary implants; I² = 0%
Loosening of prosthesis	3	Li 2018; Raimondo 2012; Singh 2017	195	RR 0.57, 95% CI 0.12 to 2.76, favours cephalomedullary implants; I² = 0%
Cut‐out	49	Adams 2001; Ahrengart 1994; Aktselis 2014; Barton 2010; Baumgaertner 1998; Benum 1994; Bridle 1991; Chechik 2014; Davis 1988; Ekstrom 2007; Giraud 2005; Goldhagen 1994; Guyer 1991; Hardy 1998; Harrington 2002; Haynes 1996; Hoffman 1996; Hong 2011; Kouvidis 2012; Kukla 1997; Kuwabara 1998; Leung 1992; Little 2008; Lopez 2002; Matre 2013; Mehdi 2000; Michos 2001; Miedel 2005; Mott 1993; O'Brien 1995; Ovesen 2006; Pajarinen 2005; Papasimos 2005; Park 1998; Parker 2012; Parker 2017; Pelet 2001; Radford 1993; Reindl 2015; Sadowski 2002; Saudan 2002; Singh 2019; Utrilla 2005; Varela‐Egocheaga 2009; Xu 2010; Xu 2018; Zehir 2015; Zhou 2012; Zou 2009	7843	RR 0.93, 95% CI 0.71 to 1.22, favours cephalomedullary implants; I² = 0%
Implant failure	24	Adams 2001; Adeel 2020; Aktselis 2014; Andalib 2020; Barton 2010; Butt 1995; Cai 2016; Carulli 2017; Chechik 2014; Davis 1988; Kukla 1997; Little 2008; O'Brien 1995; Pelet 2001; Radford 1993; Sadowski 2002; Sanders 2017; Saudan 2002; Sharma 2018; Utrilla 2005; Xu 2010; Xu 2018; Zhou 2012; Zou 2009	3190	RR 0.81, 95% CI 0.55 to 1.20, favours cephalomedullary implants; I² = 0%
Deep infection	35	Adams 2001; Ahrengart 1994; Aktselis 2014; Andalib 2020; Barton 2010; Cai 2016; Davis 1988; Giraud 2005; Guyer 1991; Hardy 1998; Hoffman 1996; Hoffmann 1999; Kukla 1997; Leung 1992; Little 2008; Matre 2013; Mehdi 2000; Miedel 2005; Mott 1993; O'Brien 1995; Ovesen 2006; Pajarinen 2005; Park 1998; Parker 2012; Parker 2017; Pelet 2001; Radford 1993; Reindl 2015; Sadowski 2002; Saudan 2002; Singh 2017; Utrilla 2005; Zehir 2015; Zhou 2012; Zou 2009	6184	RR 0.76, 95% CI 0.41 to 1.38. favours cephalomedullary implants; I² = 0%
Superficial infection	35	Adams 2001; Adeel 2020; Andalib 2020; Bridle 1991; Butt 1995; Cai 2016; Carulli 2017; Chechik 2014; Davis 1988; Eceviz 2020; Ekstrom 2007; Gou 2013; Kouvidis 2012; Kuwabara 1998; Lopez 2002; Little 2008; Miedel 2005; O'Brien 1995; Pajarinen 2005; Papasimos 2005; Parker 2012; Parker 2017; Radford 1993; Rahme 2007; Raimondo 2012; Sharma 2018; Singh 2017; Singh 2019; Utrilla 2005; Verettas 2010; Xu 2010; Xu 2018; Zehir 2015; Zhou 2012; Zou 2009	5087	RR 0.71, 95% CI 0.53 to 0.96, favours cephalomedullary implants; I² = 0%
Non‐union	40	Adeel 2020; Ahrengart 1994; Aktselis 2014; Andalib 2020; Barton 2010; Baumgaertner 1998; Cai 2016; Calderon 2013; Dujardin 2001; Ekstrom 2007; Giraud 2005; Goldhagen 1994; Gou 2013; Haq 2014; Hardy 1998; Harrington 2002; Hong 2011; Kouvidis 2012; Kukla 1997; Leung 1992; Li 2018; Little 2008; Michos 2001; Ovesen 2006; Papasimos 2005; Park 1998; Parker 2012; Parker 2017; Pelet 2001; Radford 1993; Rahme 2007; Sadowski 2002; Saudan 2002; Sharma 2018; Singh 2017; Singh 2019; Tao 2013; Xu 2010; Zhou 2012; Zou 2009	4959	RR 0.55, 95% CI 0.32 to 0.96, favours cephalomedullary implants ; I² = 0%
CI: confidence interval; RR: risk ratio

Table 4. Adverse events related to implant, fracture, or both

Table 5. Adverse events unrelated to implant, fracture, or both

Outcome	Number of studies	Studies	Number of participants	Effect estimate; Analysis 1.35
Acute kidney injury	2	Parker 2012; Parker 2017	1000	RR 1.19, 95% CI 0.34 to 4.19, favours extramedullary implants; I² = 0%
Blood transfusion	17	Adams 2001; Barton 2010; Harrington 2002; Little 2008; Matre 2013; Ovesen 2006; Parker 2012 Parker 2017; Raimondo 2012; Sadowski 2002; Saudan 2002; Sharma 2018; Utrilla 2005; Verettas 2010; Xu 2010; Yamauchi 2014	3726	RR 0.87, 95% CI 0.74 to 1.03, favours cephalomedullary implants; I² = 76%
Cerebrovascular accident	11	Bridle 1991; Butt 1995; Chechik 2014; Gou 2013; Hoffman 1996; Parker 2012; Parker 2017; Sadowski 2002; Varela‐Egocheaga 2009; Xu 2010; Zhou 2012	2000	RR 1.41, 95% CI 0.61 to 3.24, favours cephalomedullary implants; I² = 0%
Chest infection/pneumonia	25	Bridle 1991; Butt 1995; Cai 2016; Carulli 2017; Davis 1988; Giraud 2005; Gou 2013; Hardy 1998; Hoffman 1996; Hoffmann 1999; Kukla 1997; Little 2008; Lopez 2002; Mott 1993; O'Brien 1995; Papasimos 2005; Parker 2012; Parker 2017; Sadowski 2002; Saudan 2002; Singh 2019; Tao 2013; Varela‐Egocheaga 2009; Xu 2010; Zehir 2015	3657	RR 1.05, 95% CI 0.80 to 1.39, favours extramedullary implants; I² = 0%
Myocardial infarction/acute coronary syndrome	11	Butt 1995; Chechik 2014; Gou 2013; Hardy 1998; Hoffman 1996; Parker 2012; Parker 2017; Sadowski 2002; Saudan 2002; Varela‐Egocheaga 2009; Zhou 2012	1800	RR 0.77, 95% CI 0.44 to 1.35, favours cephalomedullary implants; I² = 0%
Urinary tract infection	16	Butt 1995; Cai 2016; Carulli 2017; Davis 1988; Hardy 1998; Hoffman 1996; Lopez 2002; O'Brien 1995; Papasimos 2005; Sadowski 2002; Saudan 2002; Tao 2013; Varela‐Egocheaga 2009; Xu 2010; Zehir 2015	1943	RR 1.06, 95% CI 0.79 to 1.41, favours extramedullary implants; I² = 11%
Deep vein thrombosis	30	Adams 2001; Ahrengart 1994; Butt 1995; Carulli 2017; Davis 1988; Giraud 2005; Gou 2013; Hardy 1998; Hoffman 1996; Hoffmann 1999; Kukla 1997; Li 2018; Little 2008; Lopez 2002; Mott 1993; Pajarinen 2005; Papasimos 2005; Parker 2012; Parker 2017; Radford 1993; Sadowski 2002; Saudan 2002; Sharma 2018; Singh 2019; Tao 2013; Utrilla 2005; Verettas 2010; Zehir 2015; Zhou 2012; Zou 2009	4589	RR 1.07, 95% CI 0.76 to 1.49, favours extramedullary implants; I² = 0%
Pulmonary embolism	14	Bridle 1991; Carulli 2017; Hardy 1998; Kukla 1997; Little 2008; O'Brien 1995; Papasimos 2005; Parker 2012; Parker 2017; Pelet 2001; Sadowski 2002; Saudan 2002; Xu 2010; Zehir 2015	2434	RR 1.27, 95% CI 0.54 to 3.03, favours extramedullary implants; I² = 0%
CI: confidence interval; RR: risk ratio

Table 5. Adverse events unrelated to implant, fracture, or both

Comparison 1. Cephalomedullary nails versus extramedullary implants

Outcome or subgroup title	No. of studies	No. of participants	Statistical method	Effect size
1.1 ADL, early (≤ 4 months) Show forest plot	4		Std. Mean Difference (IV, Random, 95% CI)	Totals not selected

1.2 ADL (≤ 4 months; independent in performance of ADL) Show forest plot	1		Risk Ratio (M‐H, Fixed, 95% CI)	Totals not selected

1.3 ADL, early (≤ 4 months; change in social dependency scale) Show forest plot	1		Mean Difference (IV, Fixed, 95% CI)	Totals not selected

1.4 ADL at 12 months Show forest plot	8	835	Std. Mean Difference (IV, Random, 95% CI)	0.01 [‐0.26, 0.27]

1.5 ADL (12 months; independent in performance of ADL) Show forest plot	1		Risk Ratio (M‐H, Fixed, 95% CI)	Totals not selected

1.6 ADL at 12 months (change scores in social dependency scale Show forest plot	1		Mean Difference (IV, Fixed, 95% CI)	Totals not selected

1.7 Delirium Show forest plot	5	1310	Risk Ratio (M‐H, Random, 95% CI)	1.22 [0.67, 2.22]

1.8 Functional status, early (≤ 4 months) Show forest plot	2	188	Std. Mean Difference (IV, Random, 95% CI)	0.02 [‐0.27, 0.30]

1.9 Functional status, early (≤ 4 months; excellent or good) Show forest plot	2	188	Risk Ratio (M‐H, Random, 95% CI)	1.04 [0.96, 1.13]

1.10 Functional status at 12 months (mean scores) Show forest plot	12		Std. Mean Difference (IV, Random, 95% CI)	Totals not selected

1.11 Functional status (12 months; excellent or good using HHS) Show forest plot	3	257	Risk Ratio (M‐H, Random, 95% CI)	1.06 [0.89, 1.27]

1.12 HRQoL at 12 months Show forest plot	4	279	Std. Mean Difference (IV, Random, 95% CI)	0.28 [‐0.15, 0.71]

1.13 Mobility (≤ 4 months; independent mobility) Show forest plot	7	719	Risk Ratio (M‐H, Random, 95% CI)	1.12 [1.01, 1.23]

1.14 Mobility, early (≤ 4 months; mobility scales, mean scores) Show forest plot	2	695	Mean Difference (IV, Random, 95% CI)	0.16 [‐0.15, 0.48]

1.15 Mobility (≤ 4 months; 10 metre walking speed test) Show forest plot	1		Mean Difference (IV, Fixed, 95% CI)	Totals not selected

1.16 Mobility (≤ 4 months; able to complete TUG) Show forest plot	1		Risk Ratio (M‐H, Fixed, 95% CI)	Totals not selected

1.17 Mobility, early (≤ 4 months; TUG, mean scores) Show forest plot	1		Mean Difference (IV, Fixed, 95% CI)	Totals not selected

1.18 Mobility at 12 months (mobility scales, mean scores) Show forest plot	14	1746	Mean Difference (IV, Random, 95% CI)	0.48 [0.10, 0.87]

1.19 Mobility (at 12 months; change from baseline) Show forest plot	1		Mean Difference (IV, Fixed, 95% CI)	Totals not selected

1.20 Mobility (12 months; independent mobility) Show forest plot	12	1524	Risk Ratio (M‐H, Random, 95% CI)	1.07 [0.94, 1.22]

1.21 Mobility (12 months; able to complete TUG) Show forest plot	2		Risk Ratio (M‐H, Random, 95% CI)	Totals not selected

1.22 Mobility at 12 months (TUG, mean scores) Show forest plot	1		Mean Difference (IV, Fixed, 95% CI)	Totals not selected

1.23 Failure to regain pre‐fracture mobility (at 12 months) Show forest plot	2	246	Risk Ratio (M‐H, Random, 95% CI)	1.12 [0.85, 1.46]

1.24 Mobility at 12 months (remained in bed or wheelchair) Show forest plot	1		Risk Ratio (M‐H, Fixed, 95% CI)	Totals not selected

1.25 Mortality, early (≤ 4 months) Show forest plot	30	4603	Risk Ratio (M‐H, Random, 95% CI)	0.96 [0.79, 1.18]

1.26 Mortality at 12 months Show forest plot	47	7618	Risk Ratio (M‐H, Random, 95% CI)	0.99 [0.90, 1.08]

1.27 Unplanned return to theatre Show forest plot	50	8398	Risk Ratio (M‐H, Random, 95% CI)	1.15 [0.89, 1.50]

1.28 Pain, early (≤ 4 months; pain scales, mean scores) Show forest plot	4	832	Std. Mean Difference (IV, Random, 95% CI)	‐0.13 [‐0.43, 0.17]

1.29 Experiencing pain (≤ 4 months) Show forest plot	4	417	Risk Ratio (M‐H, Random, 95% CI)	0.79 [0.42, 1.46]

1.30 Pain at 12 months (pain scales, mean scores) Show forest plot	6		Std. Mean Difference (IV, Random, 95% CI)	Totals not selected

1.31 Experiencing pain (at 12 months) Show forest plot	10	1552	Risk Ratio (M‐H, Random, 95% CI)	1.00 [0.75, 1.32]

1.32 Length of hospital stay (days) Show forest plot	26	3647	Mean Difference (IV, Random, 95% CI)	‐0.52 [‐1.23, 0.18]

1.33 Discharge destination (to own home/previous residence) Show forest plot	14	2451	Risk Ratio (M‐H, Random, 95% CI)	1.00 [0.96, 1.04]

1.34 Adverse event related to implant, fracture, or both Show forest plot	68		Risk Ratio (M‐H, Random, 95% CI)	Totals not selected

1.34.1 Intra‐operative periprosthetic fracture	35		Risk Ratio (M‐H, Random, 95% CI)	Totals not selected
1.34.2 Postoperative periprosthetic fracture	46		Risk Ratio (M‐H, Random, 95% CI)	Totals not selected
1.34.3 Loosening of prosthesis	3		Risk Ratio (M‐H, Random, 95% CI)	Totals not selected
1.34.4 Screw cut out	49		Risk Ratio (M‐H, Random, 95% CI)	Totals not selected
1.34.5 Implant failure	24		Risk Ratio (M‐H, Random, 95% CI)	Totals not selected
1.34.6 Deep infection	35		Risk Ratio (M‐H, Random, 95% CI)	Totals not selected
1.34.7 Superficial infection	35		Risk Ratio (M‐H, Random, 95% CI)	Totals not selected
1.34.8 Non‐union	40		Risk Ratio (M‐H, Random, 95% CI)	Totals not selected
1.35 Adverse events unrelated to implant, fracture, or both Show forest plot	44		Risk Ratio (M‐H, Random, 95% CI)	Totals not selected

1.35.1 Acute kidney injury	2		Risk Ratio (M‐H, Random, 95% CI)	Totals not selected
1.35.2 Blood transfusion	17		Risk Ratio (M‐H, Random, 95% CI)	Totals not selected
1.35.3 Cerebrovascular accident	11		Risk Ratio (M‐H, Random, 95% CI)	Totals not selected
1.35.4 Chest infection/pneumonia	25		Risk Ratio (M‐H, Random, 95% CI)	Totals not selected
1.35.5 Myocardial infarction/acute coronary syndrome	11		Risk Ratio (M‐H, Random, 95% CI)	Totals not selected
1.35.6 Urinary tract infection	16		Risk Ratio (M‐H, Random, 95% CI)	Totals not selected
1.35.7 Venous thromboembolic phenomena (DVT)	30		Risk Ratio (M‐H, Random, 95% CI)	Totals not selected
1.35.8 Venous thromboembolic phenomena (PE)	14		Risk Ratio (M‐H, Random, 95% CI)	Totals not selected

Comparison 1. Cephalomedullary nails versus extramedullary implants

Comparison 2. Cephalomedullary nails versus extramedullary implants: subgrouped by short or long intramedullary nails

Outcome or subgroup title	No. of studies	No. of participants	Statistical method	Effect size
2.1 Functional status at 12 months (mean scores) Show forest plot	10	775	Std. Mean Difference (IV, Random, 95% CI)	0.31 [‐0.44, 1.05]

2.1.1 Short nail	5	351	Std. Mean Difference (IV, Random, 95% CI)	0.07 [‐0.22, 0.36]
2.1.2 Unknown nail lengths	5	424	Std. Mean Difference (IV, Random, 95% CI)	0.53 [‐1.25, 2.31]
2.2 Mobility at 12 months (mobility scales, mean scores) Show forest plot	14	1746	Mean Difference (IV, Random, 95% CI)	0.48 [0.10, 0.87]

2.2.1 Short nail	11	1493	Mean Difference (IV, Random, 95% CI)	0.34 [0.02, 0.65]
2.2.2 Long nail	1	156	Mean Difference (IV, Random, 95% CI)	2.10 [1.32, 2.88]
2.2.3 Unknown nail length	2	97	Mean Difference (IV, Random, 95% CI)	0.26 [‐0.67, 1.20]
2.3 Mobility (12 months; independent mobility) Show forest plot	12	1524	Risk Ratio (M‐H, Random, 95% CI)	1.07 [0.94, 1.22]

2.3.1 Short nail	10	1455	Risk Ratio (M‐H, Random, 95% CI)	1.07 [0.94, 1.21]
2.3.2 Mixed or unknown nail length	2	69	Risk Ratio (M‐H, Random, 95% CI)	1.17 [0.33, 4.16]
2.4 Early mortality Show forest plot	30	4603	Risk Ratio (M‐H, Random, 95% CI)	0.96 [0.79, 1.18]

2.4.1 Short nail	22	2953	Risk Ratio (M‐H, Random, 95% CI)	0.95 [0.76, 1.20]
2.4.2 Long nail	2	400	Risk Ratio (M‐H, Random, 95% CI)	1.80 [1.02, 3.18]
2.4.3 Mixed or unknown nail lengths	6	1250	Risk Ratio (M‐H, Random, 95% CI)	0.64 [0.36, 1.14]
2.5 Mortality at 12 months Show forest plot	47	7618	Risk Ratio (M‐H, Random, 95% CI)	0.99 [0.90, 1.08]

2.5.1 Short nail	34	5374	Risk Ratio (M‐H, Random, 95% CI)	0.97 [0.87, 1.08]
2.5.2 Long nail	2	400	Risk Ratio (M‐H, Random, 95% CI)	1.28 [0.89, 1.85]
2.5.3 Mixed or unknown nail lengths	11	1844	Risk Ratio (M‐H, Random, 95% CI)	0.98 [0.82, 1.16]
2.6 Unplanned return to theatre Show forest plot	50	8398	Risk Ratio (M‐H, Random, 95% CI)	1.15 [0.89, 1.50]

2.6.1 Short nail	36	6266	Risk Ratio (M‐H, Random, 95% CI)	1.12 [0.79, 1.57]
2.6.2 Long nail	2	400	Risk Ratio (M‐H, Random, 95% CI)	1.15 [0.24, 5.40]
2.6.3 Mixed and unknown nail lengths	12	1732	Risk Ratio (M‐H, Random, 95% CI)	1.16 [0.81, 1.67]

Comparison 2. Cephalomedullary nails versus extramedullary implants: subgrouped by short or long intramedullary nails

Comparison 3. Cephalomedullary nails versus extramedullary implants: subgrouped by stable and unstable fractures

Outcome or subgroup title	No. of studies	No. of participants	Statistical method	Effect size
3.1 Functional status at 12 months (mean scores) Show forest plot	12	899	Std. Mean Difference (IV, Random, 95% CI)	0.27 [‐0.35, 0.88]

3.1.1 Stable fractures	2	254	Std. Mean Difference (IV, Random, 95% CI)	0.04 [‐0.21, 0.29]
3.1.2 Unstable fractures	3	144	Std. Mean Difference (IV, Random, 95% CI)	0.38 [‐0.04, 0.79]
3.1.3 Mixed stable and unstable	7	501	Std. Mean Difference (IV, Random, 95% CI)	0.30 [‐0.87, 1.47]
3.2 Mobility at 12 months (mobility scales, mean scores) Show forest plot	14	1746	Mean Difference (IV, Random, 95% CI)	0.48 [0.10, 0.87]

3.2.1 Unstable fractures	5	265	Mean Difference (IV, Random, 95% CI)	0.73 [0.19, 1.26]
3.2.2 Mixed stable and unstable fractures	9	1481	Mean Difference (IV, Random, 95% CI)	0.42 [‐0.06, 0.90]
3.3 Mobility (12 months; independent mobility) Show forest plot	12	1524	Risk Ratio (M‐H, Random, 95% CI)	1.07 [0.94, 1.22]

3.3.1 Unstable fractures	2	318	Risk Ratio (M‐H, Random, 95% CI)	1.34 [0.64, 2.82]
3.3.2 Mixed stable and unstable fractures	10	1206	Risk Ratio (M‐H, Random, 95% CI)	1.02 [0.92, 1.14]
3.4 Early mortality Show forest plot	30	4603	Risk Ratio (M‐H, Random, 95% CI)	0.96 [0.79, 1.18]

3.4.1 Unstable fractures	8	1112	Risk Ratio (M‐H, Random, 95% CI)	1.05 [0.54, 2.07]
3.4.2 Mixed stable and unstable fractures	22	3491	Risk Ratio (M‐H, Random, 95% CI)	0.95 [0.76, 1.19]
3.5 Mortality at 12 months Show forest plot	46	7558	Risk Ratio (M‐H, Random, 95% CI)	0.98 [0.90, 1.07]

3.5.1 Stable fractures	3	322	Risk Ratio (M‐H, Random, 95% CI)	0.81 [0.29, 2.23]
3.5.2 Unstable fractures	10	1464	Risk Ratio (M‐H, Random, 95% CI)	1.01 [0.82, 1.24]
3.5.3 Mixed stable and unstable fractures	33	5772	Risk Ratio (M‐H, Random, 95% CI)	0.98 [0.89, 1.08]
3.6 Unplanned return to theatre Show forest plot	49	8338	Risk Ratio (M‐H, Random, 95% CI)	1.19 [0.93, 1.53]

3.6.1 Stable fractures	2	116	Risk Ratio (M‐H, Random, 95% CI)	2.81 [0.31, 25.48]
3.6.2 Unstable fractures	12	1549	Risk Ratio (M‐H, Random, 95% CI)	0.78 [0.38, 1.61]
3.6.3 Mixed stable and unstable fractures	35	6673	Risk Ratio (M‐H, Random, 95% CI)	1.30 [1.03, 1.65]

Comparison 3. Cephalomedullary nails versus extramedullary implants: subgrouped by stable and unstable fractures

Comparison 4. Intraoperative and postoperative periprosthetic fractures: subgrouped by year of publication

Outcome or subgroup title	No. of studies	No. of participants	Statistical method	Effect size
4.1 Intraoperative periprosthetic fracture Show forest plot	35	4872	Risk Ratio (M‐H, Random, 95% CI)	2.94 [1.65, 5.24]

4.1.1 Published before 2010	27	4049	Risk Ratio (M‐H, Random, 95% CI)	3.19 [1.72, 5.93]
4.1.2 Published from 2010 onwards	8	823	Risk Ratio (M‐H, Random, 95% CI)	1.67 [0.34, 8.35]
4.2 Postoperative periprosthetic fracture Show forest plot	46	7021	Risk Ratio (M‐H, Random, 95% CI)	3.62 [2.07, 6.33]

4.2.1 Published before 2010	30	4059	Risk Ratio (M‐H, Random, 95% CI)	4.43 [2.12, 9.26]
4.2.2 Published from 2010 onwards	16	2962	Risk Ratio (M‐H, Random, 95% CI)	2.77 [1.18, 6.51]

Comparison 4. Intraoperative and postoperative periprosthetic fractures: subgrouped by year of publication