Comparing surgical interventions for intertrochanteric hip fracture by blood loss and operation time: a network meta-analysis

EMBASE, MEDLINE, CNKI, and the Cochrane Library were searched. All searches were limited to RTCs involving humans. Searches were not limited by language, publication date, or publication status. We also reviewed the reference lists of all selected studies to identify any additional relevant articles. The searches incorporated the following keywords: “intertrochanteric fractures,” “proximal femoral nail anti-rotation,” “dynamic hip screw,” “gamma nail,” “percutaneous compression plate,” and “artificial femoral head replacement.”

Studies that satisfied the following inclusion criteria were eligible for the analysis: (1) RCTs involving human participants treated for intertrochanteric hip fractures; (2) evaluations including either surgical time or blood loss, with complete data; (3) patients aged over 18 years; (4) the highest-quality study or the most recent publication in cases where data were duplicated; (5) comparison of at least two of the five treatments studied (DHS, PFNA, GN, PCCP, FHR).

The data were reviewed independently by two extractors. Extractors resolved any disagreements by consensus. The following details were extracted from each study: name of the first author, country of origin, sample size, year of publication, treatments compared, design, population characteristics, outcome measures, and fracture classification. The outcome measures of interest were surgical time and blood loss, as measured by change from baseline. The reviewers evaluated the methodological quality of all individual studies using the JADAD scale [5]. Particularly, the reviewers assessed the procedures of randomization sequence generation, appropriate blinding, allocation concealment, and reporting of withdrawal.

Eligible trials comprising direct and indirect comparisons were summarized qualitatively. In addition, essential clinical outcomes and methodological variables were recorded. We performed a traditional “head-to-head” meta-analysis to calculate the pooled estimates of the standard mean difference (SMD) and 95% confidence intervals (CIs). We conducted the statistical analysis using Stata version 11.0 (Stata Corporation, College Station, TX, USA) and developed a DerSimonian-Laird random effects model for comparisons between two strategies [6]. The Cochrane Q test and the I² statistic were assessed to determine the effect of between-study heterogeneity [7, 8]. Begg’s test was also conducted to investigate publication bias [9].

We also performed a random effects Bayesian network meta-analysis to enhance the head-to-head meta-analysis. Using this method, all five treatments could be simultaneously considered. In addition, this method increases statistical power by merging both direct and indirect comparison evidence across all five interventions. Therefore, conclusions could be educed on the relative effectiveness of interventions, such as GN fixation and PCCP use, which have never before been directly compared.

The network meta-analyses were undertaken using Bayesian Markov chain Monte Carlo (MCMC) simulation methods with minimally informative prior distributions. Thousands of simulated iterations were run based on the data and description of the proposed distributions of relevant parameters in WinBUGS software (version 1.4.3. MRC Biostatistics Unite, Cambridge, UK) [10]. The first 10,000 iterations (burn-in) were discarded, and 40,000 further iterations were run. Pooled effect sizes are reported as posterior medians of SMDs, and the corresponding 95% credible intervals (CrIs) were applied using the 2.5th and 97.5th percentiles of the posterior distribution, which can be interpreted similar to conventional 95% CIs [11]. We also calculated the percentage contribution of each effect size in the evidence base, i.e., the probability that each treatment was the most effective therapy or the second most effective therapy, etc. The distribution of ranking probabilities is presented graphically.

Meta-regression analyses were conducted to evaluate whether the publication year and type of fractures could explain the latent heterogeneity. The inconsistency between direct and indirect comparisons was evaluated using the loop-specific method [12]. This approach evaluates the difference (inconsistency factor) and 95% CI, between direct and indirect estimations for a specific comparison. Inconsistency was defined as disagreement between direct and indirect evidence with a 95% CI excluding 0.

After reviewing all selected studies, 24 trials involving five treatments met the selection criteria [13‐36]. A total of 3097 individuals with sufficient data were included in the quantitative synthesis (meta-analysis). The study selection process, including reasons for exclusion, is presented by a flowchart (Fig. 1). The characteristics of all involved studies are summarized in Table 1. Figure 2 shows all comparisons analyzed within the network. The study sample size varied from 60 to 426 participants, with a median sample size of 129. The mean age of patients varied from 62.2 to 83.6 years. According to the Evans or AO/OTA classification, all fractures were categorized as stable (Evans type I/II and AO/ATO type 31-A1) or unstable (Evans type III–V and AO/ATO type 31-A2/A3). In 7 trials, all selected patients had unstable fractures; in 15 trials, all selected patients had compound fractures, either stable or unstable. We conducted a quality assessment of all included studies (Table 2) and found that the studies had a broad range of overall reporting quality. The proportions of studies with definite descriptions of randomization, allocation concealment, blinding, dropout, and intention-to-treat analysis were 83.33% (20/24), 70.83% (17/24), 62.5% (15/24), 87.5% (21/24), and 91.67% (22/24), respectively. According to meta-regression analyses, blood loss and operation time were not affected by differences across trials in terms of the publication year or type of fractures (Table 3). No evidence of inconsistency between the direct and indirect estimates was observed in this meta-analysis (Table 4). No publication bias was observed among the included studies (Table 5).

The results of our network meta-analysis comparing the five treatments regarding the outcome of blood loss are reported in Table 6. The outcome of five treatments ranked by the blood loss is also presented. Except for FHR, all interventions had greater effects on blood loss than DHS treatment. A relatively greater increase in blood loss was seen in DHS treatment than for PFNA (SMD 1.96; 95% CI, 1.014–1.963) or PCCP (SMD 1.26; 95% CI, 0.31–2.20) interventions (Table 6). This finding indicated that the PFNA intervention leads to the least blood loss among the five treatments studied. PFNA intervention consistently showed the highest probability of having the superior ranking position among all approaches, and PCCP use showed the second highest probability (Fig. 3).

Table 6 shows the results of the network meta-analysis on the outcome of operation time. Except for FHR, all interventions had greater effects on operation time than DHS treatment. A relatively longer operation time was observed for DHS than for PFNA (SMD 0.75; 95% CI, − 0.02–0.75) or PCCP (SMD 0.61; 95% CI, − 0.20–1.44) interventions. A significantly shorter duration was observed for the PFNA approach than for GN (SMD − 0.49; 95% CI, − 1.3–0.33), or FHR use (SMD − 0.86; 95% CI, − 2.15–0.42). Therefore, PFNA, PCCP, and GN interventions resulted in significantly greater decreases in operation time. Figure 4 shows that among the five treatments, PFNA treatment exhibited the highest probability of reducing the operation time, and PCCP treatment had the second highest probability.