Using meta-analysis we compared the survival and clinical outcomes of cemented and uncemented techniques in primary total knee replacement. We reviewed randomised controlled trials and observational studies comparing cemented and uncemented fixation. Our primary outcome was survival of the implant free of aseptic loosening. Our secondary outcome was joint function as measured by the Knee Society score. We identified 15 studies that met our final eligibility criteria. The combined odds ratio for failure of the implant due to aseptic loosening for the uncemented group was 4.2 (95% confidence interval (CI) 2.7 to 6.5) (p < 0.0001). Subgroup analysis of data only from randomised controlled trials showed no differences between the groups for odds of aseptic loosening (odds ratio 1.9, 95% CI 0.55 to 6.40, p = 0.314). The weighted mean difference for the Knee Society score was 0.005 (95% CI −0.26 to 0.26) (p = 0.972).
There was improved survival of the cemented compared to uncemented implants, with no statistically significant difference in the mean Knee Society score between groups for all pooled data.
In total knee replacement (TKR) the components can be secured to the host bone through either cemented or uncemented fixation. Cementing has been shown to have low rates of aseptic failure in long-term follow-up, with good clinical outcomes.1–4 However, observed signs of osteolytic activity at the cement-bone interface has raised questions about the long-term durability of cemented fixation in younger patients.5,6 Cement has also been shown to deform and degrade over years, and has weak resistance to tension and shear forces.7,8
Uncemented fixation for TKR has been developed in an attempt to improve the longevity of implants, particularly in younger patients.9 It is thought that a more physiological bond between the bone and the implant would result in improved survival from aseptic loosening. Nevertheless, the signs of osteolysis have also been observed with uncemented prostheses.10,11 Furthermore, early migration seen in uncemented fixation by radiostereometric analysis (RSA) studies has raised concerns about whether the longevity of cemented components could be matched.12 Limited by sample size and power, studies comparing cemented and uncemented fixation have failed to demonstrate a clear superiority of one technique over the other.1,13–17
We conducted a meta-analysis of randomised controlled trials (RCTs) and observational studies comparing cemented and uncemented fixation in TKR. The primary outcome of interest was survival of the implant free of aseptic loosening at a minimum follow-up of two years. The secondary outcome was measurement of joint function by the Knee Society score (KSS).18,19
Materials and Methods
We included articles relevant to patients undergoing primary TKR, comparison of fully cemented and fully uncemented TKRs, reported outcome measures of either the KSS or survival from revision for aseptic loosening, and a study that had been published or remained unpublished if it was an RCT or an observational study.
Two of the authors (RG, DT) independently completed a computerised search of the electronic databases PubMed MEDLINE,20 OVID MEDLINE21 (1950 to May 2008), and EMBASE22 (1980 to 2008) using the following search terms (‘Arthroplasty, Replacement, Knee’ [Mesh] OR Knee Prosthesis) AND (cemented fixation OR cementless fixation OR hybrid fixation OR cemented prosthesis OR cementless prosthesis OR hybrid prosthesis OR hydroxyapatite). We also searched the Cochrane Database of Systematic Reviews,23 the Cochrane Central Register of Controlled Trials24 and Clinicaltrials.gov.25 After reviewing the title of the study we retrieved the abstract if we felt it was appropriate. We reviewed these abstracts independently and chose those studies that were potentially relevant. The bibliographies of each article that met our inclusion criteria were reviewed for any further relevant studies. We also searched the archives of orthopaedic meetings for potential abstracts to limit any publication bias in our study. We reviewed the American Academy of Orthopaedic Surgeons26 (2001 to 2008), Knee Society27 (2001 to 2008), Canadian Orthopedic Association28 (2003 to 2008) and British Orthopaedic Association29 (2002 to 2008) meetings.
Assessment of study quality.
Each RCT was independently assessed by two of the authors (RG, DT) to grade the quality of the study design using a 21-point scale.30 The observational studies were graded on an 11-point scale which assigned one point for each of the following: well-defined eligibility criteria that would limit the potential for confounding by indication; a comparison of demographic data between groups; and a description of a sample size calculation. Two points were assigned for well-reported outcome measures, including clear reporting of frequencies for survival data and/or pre- and post-operative reported means and sd for the KSS; blinded outcome assessment; limited loss to follow-up (< 80%); and appropriate statistical analysis. Both reviewers were blinded to each paper’s title, authors and institution of origin. Conflicts were resolved through consensus. Intra-class correlations were performed to assess the level of agreement between reviewers.
For each eligible study, two of us (RG, DT) extracted relevant data for both the intervention and the control groups. These included demographic data, details of the surgical technique and implants used, relevant outcomes data, loss to follow-up, and sources of funding. Survival was defined as freedom from surgical revision of either the tibial or femoral component due to aseptic loosening at a minimum follow-up of two years.
Heterogeneity describes between-study variability, which can be related to clinical and methodological differences between studies.31 We used two strategies to assess statistical heterogeneity between studies for both of our outcome measures. First, we used the methods of Hedges and Olkin32 to test for significance and homogeneity. A p-value < 0.1 was considered suggestive of statistical heterogeneity, as these tests are traditionally underpowered. Second, we calculated the I2 statistic, which describes the percentage of total variation across studies that is attributable to heterogeneity rather than chance.33 A value of < 25% is considered to reflect low heterogeneity, 50% moderate heterogeneity, and 75% high heterogeneity.33 We also developed hypotheses regarding potential sources of heterogeneity which included variations in the implants used, and the quality of the study (RCT versus comparison observational study). We planned sensitivity analysis by type of study and hypothesised that observational studies would overestimate the effect estimates in comparison with RCTs.
Publication bias was assessed with funnel plots, which demonstrate the relationship between the sample size of the studies and the precision in estimating the treatment effect. The x-axis represented the logarithm of the odds ratio (OR) as a measure of treatment effect whereas the y-axis represented a measure of study size. Bias is revealed if the plots are asymmetrical about the pooled log OR, whereas a plot resembling an inverted funnel shows that no bias is present.34 Bias can be seen particularly when there is a gap in the lower quadrants, where studies of small sizes may have biased results.
For the categorical outcomes of survival we used the OR as the summary statistic. This ratio represents the odds of the failure in the uncemented group over the odds of failures in the cemented group. An OR > 1 indicates greater failure in the uncemented group, and the point estimate of the OR is considered statistically significant only if the 95% confidence interval (CI) does not include the value 1. We used the Mantel-Haenszel method35 to combine the ORs for the outcome of interest. For the continuous outcome of KSS, we pooled data across studies by using weighted mean differences and 95% CIs. All data were combined with the random effects model, as there was moderate evidence of heterogeneity between studies for the outcome of KSS.36
The analysis was carried out by using Comprehensive Meta-analysis version 2.0 (Biostat, Englewood, New Jersey).
We identified 1292 studies from our search, and after applying our eligibility criteria 15 manuscripts qualified for systematic review and data synthesis.4,9,13,15–17,37–45 Of these, six did not provide the KSS, leaving only nine studies for that outcome.13,16,37,43–45 We excluded 1260 studies from the title or abstract, four because they were duplicate publications of the same patient cohort, ten because they did not clearly describe the method of fixation used for all implants, and two for not providing separate outcomes for each group. We excluded one study, a 15-year follow-up, in favour of a previously reported ten-year review of the same patients. We believe that the 15-year follow-up would present great clinical heterogeneity amongst the studies included in our analysis for odds of survival.
All but one paper was published in English. One was in French, with the translation provided by the publisher.4 Five studies were prospective randomised trials,4,13,16,40,41 and ten were observational studies.9,15,17,37–39,42–45 In the two RCTs, patients were quasi-randomised according either to their year16 or their day41 of birth. One RCT began randomising patients according to the year of birth, but later switched to using computer-generated random numbers.13 Another study used random number tables as a method of sequence generation, but allocation concealment was not discussed.4 The fifth RCT did not state the method of sequence generation or allocation concealment.40 In three of the studies13,16,42 the follow-up clinical assessments was performed by an independent observer. For our primary outcome, follow-up ranged from two to 11 years. The details of the studies are shown in Table I⇓.
The reviewers achieved good agreement on evaluation of the quality of the study for both the RCTs and the observational studies (intra-class correlation, 0.87, 95% CI 0.12 to 0.99; and 0.76, 95% CI 0.24 to 0.99, respectively).
Figure 1⇓ shows a funnel plot for these studies reporting the log of the ORs of survival as a measure of treatment effect. The plot demonstrates some mild asymmetry, but all studies fall within the 95% CI axis for a given standard error. There may be a few studies missing from the lower right-hand corner of the plot, and so there is minimal evidence of publication bias.
The combined OR for failure due to aseptic loosening for the uncemented group was 4.2 (95% CI: 2.7 to 6.5) (p < 0.001). There was no evidence of statistical heterogeneity between studies, p = 0.545, I2 = 0.00. Figure 2⇓ shows a forrest plot of the cumulative data from the 15 studies revealing a more favourable outcome for cemented fixation.
The weighted mean difference for the KSS was 0.005 (95% CI −0.26 to 0.26), p = 0.972. There was evidence of statistical heterogeneity between studies (p = 0.001). The I2 statistic was 77.2. Figure 3⇓ shows a forrest plot of the cumulative data from the nine studies.
Sensitivity analysis of study type revealed no evidence of statistical heterogeneity among RCTs (p = 0.937, I2 = 0) or the observational studies (p = 0.356, I2 = 9.5) for our primary outcome of odds of survival. Under the sensitivity analysis, the cumulative effect estimates for the RCTs in isolation demonstrated no significant differences between groups for our primary outcome (OR 1.9, 95% CI 0.55 to 6.40, p = 0.314) (Fig. 4⇓). There were only three CTs that reported the KSS, and so we did not perform a sensitivity analysis by study type for our secondary outcome.
In addition to aseptic loosening of the tibial and/or femoral components, the studies reported other complications. In the cemented group, the most common complications were patellar subluxation and dislocation (4.3% to 11.8%),4,17,41,42 infections and sepsis (0.9% to 2.9%),4,13,17,37,39,42 the need for subsequent patellar resurfacing (1.1% to 1.2%),13,16 and loosening or failure of the patellar component (5.6% to 39.1%).39,40 Several patients also suffered from idiopathic chronic pain (2.6%),17 developed deep-vein thrombosis (8%),38 or had ruptured patellar ligaments (0.9%).42
In the non-cemented group, several patients had developed infections (0.8% to 4.1%),4,13,16,17,37 patellar subluxations and dislocations (2.9% to 6%),4,42 and failure of the patellar components (4% to 35.8%).15,18,39,42 Several cases also required subsequent patellar resurfacing (1.8% to 3.4%),13,16 manipulations for restricted range of movement (7.7% to 8%),38,42 and revisions for idiopathic pain (0.8%).17 Collins et al38 reported a 7.7% incidence of deep-vein thrombosis.
The results of this meta-analysis, with inclusion of RCTs and observational studies, demonstrate a statistically significant benefit towards improved survival of the cemented compared to uncemented components, with follow-up ranging from two to 11 years. However, subgroup analysis of the RCT data in isolation showed no differences between groups for odds of implant survival. There were no clinical differences in outcomes between groups as measured by the KSS.
Many of the studies we reviewed showed no differences in implant survival or functional scores between the two groups of patients. However, only one gave details of a power calculation to ensure that the study had an adequate sample size with sufficient power to find a significant difference if one did exist.16 To detect a significant change of a 50% difference in failure rate between groups (e.g. 10% vs 5%) at a power of 80%, and a type I error of 5%, a study would need 474 patients per arm. Few studies of this size exist in the literature, which emphasises the importance of meta-analysis techniques for increasing the size a conclusion by pooling the best available evidence.
The advantages of cemented fixation is that it is less technically challenging, as bone cuts do not require a perfect fit with the prosthesis and cement can fill the defects.1 Cement may also potentially create an effective barrier to poly-ethylene debris generated from the articular surface, thereby preventing osteolysis and implant loosening.46 Cemented implants also display a different pattern of migration, initially achieving more reliable stable fixation.46 The RSA studies comparing cemented and unce-mented components have found that the uncemented sustained greater early migration, which may potentially lead to late clinical failures.12,47,48 Additionally, cement allows the opportunity to deliver antibiotics directly into the joint for those at high risk of infection.49,50 Cementless techniques on the other hand, offer potential advantages of shorter operating and tourniquet times,4 but cementless components are estimated to cost three times more than the equivalent cemented implants.51
We found that cemented prostheses offered better survival than the uncemented when all studies were combined, but when we combined the data from just the randomised trials, survival was equivocal. One potential explanation for the different finding is that the uncemented implants were generally used in younger patients in the observational studies, as seen in Table I⇑. Some authors have shown that the failure rate for knee arthroplasty is greater in younger patients, as they have a higher level of activity and higher demand.52,53 Although longer-term follow-up of these patients would be important, many authors have shown that between 3% and 50% of primary TKRs undergo revision within the first five years, and so we believe our study represents an important period of follow-up.52,54,55 The majority of revisions occurring in our review, for both cemented and uncemented components, were for loosening on the tibial side. This is consistent with other authors.52,55,56
In the studies included in this analysis a variety of implants were used. Two used the Miller-Galante knee system (Zimmer, Warsaw, Indiana) for both the cemented and the uncemented groups. This system was designed with a metal-backed patellar component and is known to have a high failure rate.17,41 No studies used hydroxyapatite (HA)-coated uncemented implants. The RSA studies have demonstrated a greater earlier stability of these implants compared to non-HA-coated uncemented components, which may translate into improved survival. Epinette and Manley57 have demonstrated 99.2% survival in primary TKR for HA-coated uncemented components at a mean of 11.2 years of follow-up. The RSA studies, however, have shown equivalent stability between HA-coated uncemented and cemented components and the latter have a significantly lower cost.48,58 The recent introduction of tantalum metal cementless components has shown promising early results, but the expense of the component may limit their routine use.59
There are several strengths of this meta-analysis. First, we conducted a thorough search of the literature, including non-English language publications and unpublished abstracts. Second, there was no statistical evidence of heterogeneity for our primary outcome. Third, although we found no statistical evidence of heterogeneity between studies, we performed sensitivity analysis based on a prior hypothesis for our primary outcome.
A potential limitation of any meta-analysis is that the strength of point estimates and conclusions is only as strong as the primary studies and the rigour with which they were conducted. Moreover, the ability to extract all relevant data on potential confounders between study groups is limited by the extent of reporting in the original studies.60 Although only five of 15 studies were randomised trials,4,13,16,40,41 the observational studies had groups that were well matched for gender and pre-operative diagnosis, which limits the opportunity for selection bias.15,37,39,42,43
The results of our study suggest that cemented fixation in TKR offers equivalent clinical outcomes and at least as good as, if not better survival than uncemented fixation at medium-term follow-up (2 to 11 years). Considering the higher cost of the uncemented components, the cemented components offer an economic advantage with comparable clinical outcomes. With the demographic transition of an increased incidence of TKR being performed in young patients,61–63 the question of the most durable fixation becomes even more important. The mean age of our patient cohort was approximately 63 years (16 to 97), and therefore our results are generalised only to this age group. Future randomised trials of sufficient power using validated outcome measures should be designed to examine the best fixation method for TKR in a young population.
No benefits in any form have been received or will be received from a commercial party related directly or indirectly to the subject of this article.
- Received August 28, 2008.
- Accepted March 16, 2009.
- © 2009 British Editorial Society of Bone and Joint Surgery