Introduction

The alignment and component position are key factors for the success of total knee arthroplasty (TKA) [12, 20, 24, 31, 35, 37]. To accurately achieve the desired alignment and component position, increasing new techniques have been applied in TKA, including computer navigation (Navigation) [9], patient-specific instruments (PSI) [36, 40] and surgical robots (Robot) [11]. In the past 20 years, there have been many reports on the comparison between these new techniques and conventional TKA (CON) with variable outcomes [1, 2, 6, 14, 17, 30, 34].

Up to now, there has been no literature conducting a comprehensive comparison and analysis of the above four surgical techniques. Network meta-analysis (NMA) can help to fill this gap. NMA is an extension of conventional pairwise meta-analysis. It can perform direct and indirect comparisons at the same time, even when the two measures have never been compared via head-to-head evaluation [22]. Besides, the best intervention measures can be evaluated by the value of surface under the cumulative ranking area (SUCRA) [10].

This study only included randomised controlled trials (RCTs) with level of evidence I, and conducted a thorough and comprehensive evaluation on the radiological and clinical outcomes of the four surgical techniques, i.e. Navigation, PSI, Robot, and CON, and attempted to rank the above surgical techniques. The authors hypothesised that Navigation, PSI, and Robot could improve the accuracy of alignment, but have no significant improvement on the clinical outcomes.

Methods

This NMA strictly complied with “PRISMA Extension Statement” [7]. The complete PRISMA checklist could be found in Appendix A. This NMA has been registered on the INPLASY (INPLASY202060018).

Search strategy

A comprehensive search was conducted in three databases (PubMed, EMBASE and Cochrane Library) from their inception to June 1, 2020 using a combination of MeSH terms and free words. Please refer to Appendix B for more details on search strategies.

Inclusion and exclusion criteria

Types of studies

This study only included RCTs; non-English articles, animal studies, cadaver studies, case reports, comments, letters, editorials, protocols, guidelines, unpublished articles, and review papers were excluded.

Types of participants

Patients who underwent primary TKA and were ≥ 18 years old were included, regardless of gender and race. Only the most recently published articles among the multiple articles on the same research subjects by the same author or team were included. However, if the study subjects or outcome indicators were different, they would be separately included in this NMA.

Types of interventions

RCTs containing two or more interventions of Navigation, PSI, Robot and CON were included in this NMA; the included studies were not limited to two-arm RCTs.

Types of outcomes

The radiological outcomes included: (1) mechanical axis outliers; (2) coronal femoral component angle outliers; (3) coronal tibial component angle outliers; (4) sagittal femoral component angle outliers; and (5) sagittal tibial component angle outliers. Deviations of more than 3° from the target value were defined as outliers. The clinical outcomes included: (1) short-term Knee Society Score (KSS) knee scores (follow-up period < 5 years); (2) short-term KSS function scores; (3) medium-and-long-term KSS knee scores (follow-up period ≥ 5 years); and (4) medium- and long-term KSS function scores.

Data extraction and quality assessment

Two reviewers (KL and LML) conducted the screening process of the article by reading the title and abstract, and then further evaluated the article by reading the full text. Data were extracted from the included literature according to the pre-designed table, including study characteristics, patient demographics and the risk of bias. If the data that needed to be included in the meta-analysis were lost or were only shown in the form of pictures, the authors would try to contact the author for further information. If no responses were received, data would be extracted by digital ruler software or excluded. Two investigators (KL and LML) used the Cochrane Risk of Bias Tool for RCTs to independently evaluate the bias of the included literature. Different opinions in the process were resolved by discussion or passed to a third person (LG).

Data analysis

An NMA was conducted for outcomes of the four surgical techniques in a Bayesian approach. Data were combined with a random-effects model and Markov Chain Monte Carlo was implemented to the model. In the analysis process, the prior distribution was set as normal distribution and three chains were used for simulation. The number of iterations was set to 50,000, and the first 5000 were used for the annealing algorithm to eliminate the impact of the initial value. For binary and continuous variables, odds ratios (OR) and mean differences (MD) were selected, respectively. When the 95% confidence interval (95% CI) of OR contained 1 or the 95% CI of MD contained 0, the result was considered to have no statistical significance. The interventions were ranked by the SUCRA value, which showed the percentage of effectiveness of each treatment and ranged from 0 to 100%. Intervention with larger SUCRA values was generally considered to have a better effect [10]. A network graph was drawn to reflect the number and distribution of the included literature. Meanwhile, funnel plots were applied to reflect the publication bias of outcomes that included more than 10 RCTs. Inconsistency factor (IF) and node-splitting method were used to evaluate the consistency. The I2 statistic was used to statistically assess the presence of heterogeneity. Subgroup analyses or sensitivity analyses were planned if necessary. The calculation was performed by WinBUGS (Version 1.4.3, Biostatistics the Medical Research Council, Cambridge, United Kingdom), R software (Version 4.0.2, R foundation for statistical Computing, Vienna, Austria), and Stata software (Version 15.0, Stata Corp, Texas, USA).

Results

A total of 73 RCTs with 4209 TKAs were included in this NMA (Fig. 1). The characteristics and risk of bias assessment of the included literature are shown in Table 1 and Fig. 2, respectively; the network graph was shown in Fig. 3. For more details on the included literature and the risk assessment of bias, see Appendix C.

Fig. 1
figure 1

PRISMA flow diagram for selection of included RCTs

Full size image
Table 1 Characteristics of the included RCTs
Full size table
Fig. 2
figure 2

Risk of bias assessment of included RCTs

Full size image
Fig. 3
figure 3

Network graph of different outcomes. a Mechanical axis outliers; b coronal femoral component angle outliers; c coronal tibial component angle outliers; d sagittal femoral component angle outliers; e sagittal tibial component angle outliers; f short-term KSS knee scores; g short-term KSS function scores; h medium-and-long-term KSS knee scores; i medium-and-long-term KSS function scores

Full size image

Robot and Navigation were significantly better than PSI and CON in the control of lower limb alignment and component position (Table 2). Robot had the lowest probability of the outlier of lower limb alignment and component position, followed by Navigation (Table 3). Except for Navigation that had statistically significant difference compared to CON in medium-and-long-term knee scores, the four techniques showed no significant difference in KSS scores (Table 2). Navigation had the greatest probability of obtaining better KSS scores after surgery through ranking analysis (Table 3). The assessment revealed that heterogeneity and inconsistency were low for most outcomes (Appendix D). For more details on data analysis and publication bias, please refer to Appendix E and F, respectively.

Table 2 Odds ratio/mean difference (95% CI) of the various interventions
Full size table
Table 3 Ranking probabilities of the various interventions (%)
Full size table

Discussion

The most important finding of this study is that Navigation and Robot could significantly reduce the occurrence of malalignment and malposition compared with PSI and CON, but the above four techniques showed no clinical significance in postoperative outcomes.

Referring to the Knee Society roentgenographic evaluation system [5], this study evaluated alignment and component position from the following five aspects: mechanical axis outliers, coronal femoral component angle outliers, coronal tibial component angle outliers, sagittal femoral component angle outliers and sagittal tibial component angle outliers. The results showed that Robot and Navigation were significantly better than PSI and CON in the alignment and component position, and Robot had the lowest probability of outliers (Tables 2 and 3). Rhee et al. reported that Navigation could improve postoperative alignment compared with CON [34]. Van der List and Rebal also expressed the same view [33, 41]. The latest meta-analysis also agreed that Robot could significantly improve alignment and component position [1, 2, 29]. As of PSI, two latest meta-analyses also supported the views of this study [6, 17]. Besides, Pietsch et al. found a higher frequency of recuts with PSI [32]. Suggestions from PSI manufacturers for component sizes and positioning were often not accurate enough and intraoperative revision was required [4, 16, 38, 42]. Maybe these were the reasons why there was no significant difference between PSI and CON in alignment.

Knee Society clinical rating system [8] was adopted in this study to evaluate clinical outcomes; clinical scores were divided into short-term and medium-and-long-term. Except for significant difference in medium-and-long-term knee scores of Navigation compared to CON, these four techniques showed no significant difference in KSS scores. Navigation had the greatest probability of obtaining better KSS scores after surgery through ranking analysis (Tables 2 and 3). Navigation was better than CON in KSS and Western Ontario and McMaster University osteoarthritis index scores (WOMAC) during 5–8 years of follow-up [30]. Robot was also superior to CON in few postoperative clinical scores [29]. Kizaki and Mannan found that PSI did not significantly improve clinical scores compared with CON [13, 23]. The reason why so many studies showed no difference could be the commonly used scoring system nowadays, which is not sensitive enough [18]. Still, the statistical significance of patient reported outcome measures (PROM) does not necessarily represent clinical significance. Lee et al. believed that the minimal clinically important difference (MCID) of KSS knee score after TKA was 5.3–5.9 points [15], while Lizaur–Utrill’s research suggested it should be 9 points [19]. The statistical significances of Navigation with CON in medium-and-long-term knee scores (1.380 points) in this study were far from clinical significance. Navigation performed well in alignment and obtained higher scores in KSS scores, which validated the views that inaccurate alignment could result in poor clinical outcomes [12, 20, 24, 31, 35, 37]; after all, various factors could affect postoperative clinical outcomes, such as age, BMI, psychological status, soft tissue balance, component design, and rehabilitation.

This study has the following strengths compared with other meta-analyses. First, most of the previous articles were head-to-head two-arm studies. Second, the previous studies had either limited number of literatures [14, 34] or included non-RCTs, which enlarged the number [1, 2, 30]. Third, some previous articles focused only on medium-and-long-term clinical outcomes [14, 34], but RCTs with shorter follow-up are still meaningful for analysing the accuracy of alignment.

This study also has limitations. First, the prostheses in these included RCTs were different, and the navigation systems, robot systems, and PSI systems used were also diverse. Second, the biases of the included RCTs might also influence the results. Third, the indicators for evaluating clinical outcomes were various, so the authors could only the outliers and KSS scores to evaluate outcomes. Finally, there are few RCTs on Robot with long-term follow-up.

Navigation and Robot are much more expensive and require longer operation time than conventional TKA [18, 21, 39]. Besides, complications of these techniques are occasionally reported [3]. These disadvantages make the application of Navigation and Robot on normal primary TKA not so cost-efficient. However, for extremely challenging deformities [26], the advantages of Navigation and Robot may be more obvious [25,26,27,28].

Conclusion

Navigation and Robot did improve the accuracy of alignment compared with PSI and conventional instruments in TKA. Robot has the greatest advantage in achieving the desired alignment accurately, followed by Navigation; Navigation has the greatest advantage in postoperative clinical outcomes. However, the above four techniques showed no clinical significance in postoperative outcomes.