Extramedullary versus intramedullary fixation of stable trochanteric femoral fractures: a systematic review and meta-analysis

Embase, Pubmed/Medline, Web of Science databases, and the Cochrane Central Register of Controlled Trials (CENTRAL) were initially searched on 22 March 2021 for studies comparing extramedullary and intramedullary fixation of AO type trochanteric fractures. The initial search was updated on 26 September 2022. Online Resource 1 shows the search string used and the search results. After deduplication, two reviewers (LHTN and ACP) independently screened titles and abstracts for eligibility for inclusion. Any disagreement was resolved by consensus. The same two reviewers subsequently conducted the full-text screening of eligible articles.

Studies were included when they presented data (a) published after 1990 of (b) acute (c) AO-OTA 31-A1 trochanteric fractures, (d) comparing intramedullary and extramedullary fixation (e) in patients > 50 years, (f) using currently available devices (an overview of included devices per study is provided in Online Resource 2). Studies were excluded when they (a) presented no original data, (b) did not mention relevant outcomes (e.g. only incidence of fixation with specific devices), (c) were biomechanical, in vitro or cadaveric studies, (d) pathological fractures, (e) bilateral fractures, (f) periprosthetic fractures, (g) were case reports, or (h) did not make distinction between type of fracture or treatment.

All baseline characteristics were independently extracted by two reviewers (LHTN and ACP) using a predefined data extraction sheet and included: first author, publication year, setting, study design, study period, follow-up time, total study population, and study population with type 31-A1 fracture. In addition, patient characteristics were collected including age, sex, implant type, and number of patients treated using extramedullary or intramedullary fixation.

Data were collected for the following outcome measures: functional outcomes: Harris hip score, pain, parker mobility score, and walking ability (cases of unassisted/good walking ability were compared with assisted or no walking ability). Complications: reoperation, non-union, cut-out, peri-implant fracture, conversion to prosthesis, implant/fixation failure, superficial and deep infection, malunion, limb-length discrepancy, heterotopic ossification, osteolysis in fixed implants, and mortality. Surgical outcomes/operation characteristics: operation time, blood loss, bone-healing time, quality of reduction, hospital stay, fluoroscopy time, and costs/cost-effectiveness. All outcomes were included as defined by individual presenting articles.

If case data were not described sufficiently in the full-text or supplementary materials, the authors were contacted by e-mail once. If this approach was unsuccessful, missing data were imputed based on the average standard deviation for the total included population across all studies with no missing data and adjusted for population size. This was done for the standard deviations of Harris hip score in Sevinc et al.[27], the standard deviations of 1-year pain scores by Matre et al. [28], and standard deviations for bone-healing time by Cho et al. [29].

Two reviewers (LHTN and ACP) independently assessed the methodological quality of all included studies using the risk of bias 2 (RoB 2) tool for assessing risk of bias in randomized trials and MINORS, a methodological index for non-randomized studies [30, 31]. Disagreement was resolved by consensus. The RoB 2 is structured into a fixed set of domains of bias, trial design, conduct, and reporting. Within each domain, a series of signaling questions indicate features that are relevant to risk of bias. A proposed judgement about the risk of bias for each domain is generated by an algorithm, based on the signaling questions. Judgement can be 'Low' or 'High' risk of bias, or can express 'Some concerns’. The MINORS ranges from 0 (poor quality) to 24 (high quality).

Data were analyzed using Review Manager (Revman, version 5.3.5). Input in the analysis was sample size, mean, and standard deviation (SD) for continuous outcomes and sample size and number of cases with the specific outcome for binary outcomes. Random effects models were used for measuring treatment effects because of the expected heterogeneity due the inclusion of both RCT and observational studies and comparison of multiple types of devices in different countries and clinical settings. No (age-)adjusted analyses were performed, all data were analyzed as described in the original studies. Treatment effects of the binary outcome measures were pooled using the Cochran–Mantel–Haenszel statistic and are presented as risk ratio (RR) with 95% confidence intervals (CI). Treatment effects of the continuous outcome measures were pooled using the inverse variance weighting method and are presented as mean difference (MD) or standardized mean difference (SMD) with 95% confidence intervals (CI). SMD was used when included studies used different measurement scales for the same variable. All analyses were stratified on study design (randomized controlled trials (RCTs) or observational studies) and were presented as Forest plots. Differences between pooled estimates for the two types of study design were compared using the χ²-test test. Heterogeneity between studies was assessed qualitatively using the Cochran’s Q-test and quantified using the I² statistic. Statistical significance was assumed using a p value threshold of 0.05. Publication bias was assessed by visual inspection of funnel plots for all studied variables (Online Resource 3).

The literature search and selection of included studies is shown in Fig. 1. The primary search resulted in 14,577 records. After deduplication, 7,213 records were screened on title and abstract. Out of the 473 articles assessed for eligibility, a total of 19 studies (five RCTs [17, 32‐35] and 14 observational studies [18, 19, 27‐29, 36‐44]) were included in the final analysis. All studies were published between 2005 and 2022.

The 19 included studies (Table 1) comprised a total of 21,793 patients. The RCTs included a total of 397 patients and the observational studies a total of 21,396 patients. Extramedullary fixation was used in 15,910 patients and 5,883 patients were treated using intramedullary fixation. The follow-up ranged from 6 to 60 months. A complete overview of reported variables per study is shown in Table 2.

The details and distribution of the quality assessment using the RoB 2 for RCTs and MINORS for observational studies are described in Tables 3 and 4. The overall bias assessment in the RoB 2 ranged from high overall risk of bias in Pajarinen et al. [32] to low overall risk of bias in Tao et al. [35] and Ovesen et al. [33]. The mean score for the MINORS was 18 (SD 2.5) and ranged from 15 (moderate quality) to 22 (high quality).

A summary of all study results and meta-analyses is provided in Table 5. The most (clinically) relevant and statistically significant outcome measures are described below. Forrest plots for additional outcome measures can be found in Online Resource 4 (OR4).

There was a significantly longer operation time (9 studies, Fig. 10) for extramedullary fixation, with a mean of 53.5 min in 6704 patients versus 53.2 min in 1685 patients in the intramedullary group (MD 14.1, CI 6.98–21.29, I² = 96%, p < 0.001) [18, 19, 28, 29, 35, 39‐41, 43]. Operative blood loss (7 studies, Fig. 11) in the extramedullary group was significantly higher as well, with a mean of 351 mL in 305 patients versus a mean of 204 mL in 355 patients in the intramedullary group (MD 92.30, CI 13.49–171.12, I² = 98%, p = 0.02) [18, 19, 29, 35, 39, 40, 43].

No significant differences between the groups were found in the other surgical outcomes or operation characteristics: time to radiological union (MD − 0.04, CI − 1.10 to 1.02, I² = 84%, p = 0.94, OR4, Fig. 10) [19, 29, 35, 38, 39], good quality of reduction (RR 1.03, CI 0.95–1.11, I² = 0%, p = 0.46, OR4, Fig. 11) [19, 35, 38, 40, 41], hospital stay (MD 0.50, CI − 0.25 to 1.25, I² = 48%, p = 0.19, Online Resource 4, Fig. 12) [19, 35, 39, 41, 43], or fluoroscopy time (MD 13.12, CI − 3.28 to 29.52, I² = 73%, p = 0.12, Online Resource 4, Fig. 13) [19, 35, 41].

Only two studies reported data on hospital or surgery-related costs [41, 43]. Due to differences in included cost variables, no pooled effect could be calculated. None of the included studies reported data on cost-effectiveness.

This meta-analysis found that only the duration of surgery and intra-operative blood loss to be statistically significantly in favor of intramedullary fixation. These results should be interpreted with care. The level of heterogeneity of effects across the studies was over 90% with large differences in mean operation time or blood loss (e.g. a range of 25.6–136 min for extramedullary fixation [18, 19]). This large variety may partly be explained by differences in surgery protocols, device used, experience of surgeons, the small number of inclusions, and inclusion or exclusion of anesthesia times across the included studies. The largest study by Matre et al. [28], with 7643 patients, and arguably the best sample size for this comparison, found no significant difference in operation time. The only other review reporting duration of surgery, by Parker et al. [46], included studies all conducted before the year 2005 and concluded that no definitive conclusion could be drawn because of limited data.

As current literature shows no differences between intra- and extramedullary fixation in functional outcomes or complications, it is tempting to conclude that surgeons could use either type of device. However, the difference in device costs should also be considered in this decision. Extramedullary fixation or more specifically dynamic/sliding hip screw is most likely the most cost-effective implant [47, 48]. Intramedullary devices can cost well over $1000 more than extramedullary options with, as demonstrated by this review, the same surgical and clinical results for stable fractures. More expensive devices, such as PFNA, should be avoided when enough surgical expertise with extramedullary devices exists. This financial argument is also used in the Dutch Guideline for treatment of proximal femoral fractures and NICE guideline, as an argument for usage of extramedullary devices when clear evidence for superiority is absent [23, 24].

Some limitations should be considered when interpreting the results of this study: First, in every meta-analysis, there is a chance of publication bias, however, the funnel plots show no clear indication for one-sided publication for the included variables. The large majority of patients (> 98%) included in this study came from observational studies resulting in a larger risk of selection bias. Many of the studied variables have a moderate-to-high level of heterogeneity between studies. Part of this could be explained by random chance and can be corrected for by using random effect models. Another part could be explained by the (small) differences in devices used, different moments in time studies were conducted, differences in study populations (e.g. age, fracture type prevalence, and inclusion of patients with concomitant injuries), and international differences between included studies. Most of the included RCTs and to a lesser degree, observational studies appear underpowered or only adequately powered for the total group of combined stable and unstable trochanteric femoral fractures. As many of the studied outcomes and complications are rare, and occur at rates of 1% or lower, studies are often not adequately powered to study these outcomes for a restricted group of stable fractures. A combination of heterogeneity and low power could conceal treatment effects that now seem insignificant. Although the clinical relevance of these effects may be questionable. Larger and adequately powered RCTs or high-quality prospective observational studies comparing extramedullary and intramedullary fixation in, specifically, stable trochanteric femoral fractures are needed to provide a definitive answer to the question of superiority for treatment for stable trochanteric fractures. Considering the low frequency of complications and sparsity of data on functional outcome and patient reported outcomes, future research should focus on both functional and quality of life data, and provide a large scale cost-effectiveness analysis.