The participating centers were chosen because of their reputation for excellence in orthopaedic surgery across the United States and Canada and because of their performance of a high volume of primary and revision arthroplasties. A list of participating surgeons and their affiliated centers is provided in the note at the end of this article.
After institutional review board approval was obtained at each participating center, consecutive patients with a failed total knee arthroplasty presenting with indications for revision were approached about participation in the study. Table I details the inclusion and exclusion criteria. Patients who agreed to participate and gave informed consent were then included in the study. Data were collected at each site by dedicated study coordinators and forwarded to the central study coordinator for entry into the database.
Preoperatively, demographic information including age, sex, and body mass index as well as laterality, flexion contracture and extension contracture angles, and comorbidities was gathered. Using a validated self-reported comorbidities questionnaire5, patients reported whether they had heart disease, high blood pressure, lung disease, diabetes, ulcer or stomach disease, kidney disease, liver disease, anemia or other blood disease, cancer, depression, osteoarthritis or degenerative arthritis, back pain, or rheumatoid arthritis. The surgeon described the modes of failure of the primary total knee arthroplasty, which included polyethylene wear or breakage, instability, extensor mechanism, infection, femoral implant loosening or migration, tibial implant loosening or migration, patellar implant loosening or migration, femoral bone lysis, tibial bone lysis, patellar bone lysis, breakage of the implant, metal wear, malalignment, and other reasons. Modes of failure were not mutually exclusive since a failure of a total knee arthroplasty can be due to multiple reasons. The specific techniques of the revision surgery were also reported and included the use of cement for the tibial and femoral fixation, the use of stems, and the use of augments. The specifics of the revision surgery also included whether the femoral and/or tibial components were revised. We did not include information about patellar revision for this study.
At baseline and every six months up to two years, the functional status of each patient was assessed with use of the two Short Form-36 (SF-36) summary scales (the normalized physical component score [PCS] and mental component score [MCS])6-8, the Western Ontario and McMaster Universities Osteoarthritis Index (WOMAC) stiffness, pain, and difficulty-of-function components9,10, and the Lower-Extremity Activity Scale (LEAS)11. An increase in the scores for the SF-36 PCS (range, 0 to 100 points), SF-36 MCS (range, 0 to 100 points), and LEAS (range, 1 to 18 points) indicates improvement, whereas an increase in the WOMAC stiffness score (range, 0 to 8 points), pain score (range, 0 to 20 points), or difficulty-of-function score (range, 0 to 68 points) indicates deterioration. (The WOMAC scores were not normalized to a 0 to 100-point scale.)
We determined that the minimum sample size needed to conduct our analysis, with assumption of a significance level (alpha) equaling 0.05, a power (beta) equaling 0.80, and an estimated WOMAC effect size of 30% of the standard deviation9,12-15, to be 250 participants. Also, as is appropriate for any multicenter study of this nature, we performed a pilot study before undertaking the present, final investigation to ensure uniformity of indications, data management, and follow-up among centers16. A standard set of questionnaires, for both patients and surgeons, was used for all documentation to avoid any deviation in the data collection process. In addition, strict inclusion and exclusion criteria were applied from the beginning of the study, and all coordinators responsible for collecting the data were blinded to the study design and hypotheses. Also, since this was a multicenter study, a mixed-model approach was utilized in a separate analysis to test the center effect on the outcomes17. An individual center was modeled as a random variable with each of the outcome measures (SF-36 PCS, SF-36 MCS, three WOMAC component scores, and LEAS score) collected at the first two times of follow-up (six and twelve months) treated as a dependent variable, adjusting for baseline measures of the respective outcome measure. The p value for the center effect was consistently not significant across measures and follow-up time points.
Analysis of variance was conducted to examine differences in average changes in functional status over time. A (multivariate) piecewise general linear mixed model was then used to model functional improvement following revision total knee arthroplasty and to identify independent predictors of such improvement18. This method of analysis takes into account the correlated nature of repeated measures of the same subject as well as the nonlinear patterns in functional improvement by allowing a change in slope for different time periods. We hypothesized that the patients who had undergone revision, like those who have undergone primary total knee arthroplasty, would continue to have improvement during the first year following surgery, with a peak functional status at one year and a plateau thereafter4,12,13. We therefore estimated two slopes (baseline to twelve months [slope 1] and twelve to twenty-four months [slope 2]) for each of the six functional measures (the SF-36 PCS and MCS; the WOMAC stiffness, pain, and difficulty-of-function scores; and the LEAS) using piecewise general linear mixed models that adjusted for age, sex, number of comorbidities, body mass index, side operated on, preoperative flexion contracture and extension contracture angles, reasons for failure of the primary total knee arthroplasty, and surgical technique (use of cement for fixation, use of femoral and/or tibial stems, use of augments for the femur and/or tibia, and whether one or two components were revised). A full description of the piecewise general linear mixed model analytic method is provided in the Appendix.
Source of Funding
This study was funded by the Orthopaedic Research and Education Foundation, American Geriatrics Society, and The Knee Society. Dr. Ghomrawi was also supported in part by the Weill Cornell Medical College Center for Education and Research on Therapeutics (CERT) Program from the Agency for Healthcare Research and Quality, Grant Number U18 HS016075.
Three hundred and eight patients (318 knees) from seventeen sites consented to participate in the study and were enrolled. A description of the baseline demographic characteristics is provided elsewhere19,20. Fifty-five percent of the participants were women, 83% were white, 14% were black, and 3% were of other races. The mean age of the participants was 68.7 years (range, thirty-four to eighty-five years). The majority (86%) of the participants were retired (one-third because of disability and/or illness). The mode of failure of the arthroplasty was classified as infection for fifty-six knees (18%) and aseptic in 262 knees (82%). Surgeons reported that the tibial component was revised in 248 knees (78%) and the femoral component, in 226 knees (71%). Both the tibial component and the femoral component were revised in 210 knees (66%), and one component was revised in the remaining 108 knees. The main reasons for failure reported in this series included instability (n = 92, 28.9%), tibial bone lysis (n = 87, 27.4%), polyethylene wear (n = 78, 24.5%), femoral bone lysis (n = 72, 22.6%), and tibial loosening (n = 71, 22.3%).
Two hundred and twenty-one subjects (221 revised knees) were seen at the two-year follow-up time point (follow-up rate, 71.8%). Five subjects died because of reasons not related to the revision, seventy-eight were lost to follow-up despite repeated attempts to contact them, and four had a repeat revision before the two-year point. Details of the loss to follow-up are provided in Figure 1. The follow-up cohort (n = 221) did not differ significantly (p > 0.05) from the patients without complete follow-up (n = 87) in terms of the baseline SF-36 PCS, SF-36 MCS, WOMAC (pain, stiffness and difficulty-of-function), or LEAS scores; the number of comorbidities; or the body mass index. They were, however, older on average (mean age, 68.7 years compared with 64.4 years for the patients with incomplete follow-up, p = 0.03). The two cohorts were very similar in terms of modes of failure, including infection rate.
The mean baseline scores for the functional measures indicate that patients treated with revision total knee arthroplasty constitute a severely disabled population with averages below the age-matched normative values in the United States20. The average follow-up functional status scores are displayed in Table II. The greatest improvement in the mean values was observed from baseline to six months.
Patterns of Functional Improvement
The slopes for the period of zero to twelve months, shown in Table III, indicate significant improvements in the SF-36 PCS, SF-36 MCS, LEAS, and three WOMAC component scores. However, in the second year after the operation, the slopes were not significantly different from zero (that is, they indicated a plateau) except in the cases of the WOMAC pain score (slope = 0.67 ± 0.21 [standard deviation], p < 0.01) and WOMAC difficulty-of-function score (slope = 1.66 ± 0.63, p < 0.05). In those two cases, the slopes were significantly positive, indicating worsening—i.e., an increased perception of pain and difficulty of function in the second year after the surgery. The results are displayed graphically in Figure 2.
Predictors of Functional Improvement
There appeared to be few predictors of improvement in functional status, and they were outcome-instrument-specific (Table IV). Patients with a higher number of comorbidities reported significantly (p < 0.01) worse outcomes consistently across the instruments (SF-36 PCS coefficient = -1.83 ± 0.35, SF-36 MCS coefficient = -1.92 ± 0.47, WOMAC pain coefficient = 0.58 ± 0.17, WOMAC stiffness coefficient = 0.15 ± 0.07, WOMAC difficulty-of-function coefficient = 2.32 ± 0.62, and LEAS coefficient = -0.49 ± 0.1). In addition, the SF-36 PCS was higher for subjects who had polyethylene wear or breakage (coefficient = 3.46 ± 1.47, p < 0.05) and those who had malalignment (coefficient = 5.41 ± 2.35, p < 0.05) as the mode of failure, whereas tibial bone lysis was associated with lower SF-36 PCS scores (coefficient = -5.46 ± 1.91, p < 0.01). Higher pain levels, as measured with the WOMAC pain score, were associated with obesity (a body mass index of >30 kg/m2) (coefficient = 0.08 ± 0.04, p < 0.05). Finally, patients in whom the primary total knee arthroplasty had failed as a result of malalignment had higher LEAS scores (coefficient = 1.42 ± 0.69, p < 0.05).
On the basis of the numbers available, the surgical technique (use of stems, use of cement for fixation of stems, use of augments, and revision of both components instead of one component) did not show any significant effect on outcome as assessed with any of the six functional measures. The scores were not sensitive to whether the tibial or femoral component had been fixed with or without cement, whether stems had been used, or whether augments had been used. The scores were also not sensitive to whether one or two components had been revised. We hypothesized that the effect of the surgical technique on outcome differed depending on the reason for failure and/or time, or what is known in statistical terms as an interaction effect. To test this hypothesis, we ran models with the addition of interaction terms to measure this differential effect. For example, an interaction term was created for tibial cement use with tibial bone lysis and with tibial implant loosening and migration. We also included interaction terms separately for the surgical technique with the follow-up time period and for the reason for failure with the follow-up time period to see if the reason for failure or the surgical techniques impacted the improvement either initially or after twelve months. Finally we included a three-way interaction (reason for failure × surgical technique × follow-up time period). None of these interaction terms showed any significant effect, and we therefore report only the main effects in Table IV.
In this study, improvement in function following revision total knee arthroplasty had a pattern similar to that following primary knee arthroplasty in that it peaked at one year postoperatively. Our cohort of patients showed significant improvement in several functional outcomes in their first year following revision surgery. However, the magnitude of the improvement was smaller than that after primary surgery. A meta-analysis of the literature on primary knee arthroplasty revealed that patients had average improvements in the SF-36 PCS score and the WOMAC cumulative score of 16 and 28.5 points, respectively, compared with 3.74 and 10.14 points in our group of patients4. Our findings are similar to those of Hozack et al., who found the improvements after primary surgical procedures to be greater than those after revisions in patients with similar baseline scores21.
The improvements in the scores on both the SF-36 and the LEAS, which are general health measures, were sustained (the slopes were not significantly different from zero) in the second year after the revisions. However, two of the three WOMAC components, which are knee-specific measures, indicated a slight but significant increase in pain and difficulty of function. These declines in function may signal that the onset of deterioration after revision arthroplasty may start as early as the second year following the surgery. This should caution surgeons to closely monitor patients for a substantial period of time after a revision in order to understand the changes in their functional status.
While our results clearly demonstrate a nonlinear pattern of improvement in patients who underwent revision total knee arthroplasty and while the magnitude of this improvement was further quantified with use of a rigorous statistical model, our findings should be interpreted with caution for two main reasons. Primarily, our analysis assumed one inflection point (the point at which a change in the slope is expected) at twelve months postoperatively to distinguish two periods: a period of improvement and a period of stability (a plateau). However, the nonlinear nature of improvement spanning the twenty-four-month period is not entirely explained by one inflection point. For example, it is likely that an inflection point is needed when examining functional improvement patterns during the period from baseline to twelve months to distinguish sharp improvement in the first few months from slower improvement in the later months. In addition, we assumed that the inflection point for all of the measures occurs at exactly twelve months. It is likely that some of the measures peak at different times (from six to twelve months or from twelve to eighteen months) than others. Among all of the measures examined, the LEAS may be the most appropriately modeled with an inflection point exactly at twelve months (Fig. 2). Because we collected data only at time points that are conventional in orthopaedic outcomes studies, we were not able to utilize the general linear mixed model fully to address these two limitations.
The discrepancy in findings between knee-specific and general health instruments has been reported in other studies12,13. It is noteworthy in the present study of revision surgery because it hints at the progressive sequence of this deterioration, which starts at the knee level and then may expand to affect general health. Moreover, this study highlights the potential usefulness of the WOMAC pain and function scores as tools to monitor early failures of revision total knee arthroplasty.
We found few significant predictors of functional improvement. The number of reported comorbidities was the most significant predictor of outcomes, forecasting less improvement of all six measures. Our findings were consistent despite the crude nature of our measures, which were self-reported and nonweighted. The association between an increased number of comorbidities and worse outcomes in our study of revisions was also observed in outcome studies of primary total knee arthroplasty22,23. Other significant predictors were inconsistent across measures and specific to one aspect of function. It is noteworthy that, of the few modes of failure that were significantly associated with outcomes, one (tibial bone lysis) was associated with worse outcomes and two (polyethylene wear and malalignment) were associated with better outcomes. These conclusions are important because they distinguish modes of failure that are rectifiable by revision surgery and the correction of which improves function from modes of failure that are not rectifiable and therefore become a threat to the outcomes of revision surgery. Further research is needed in this area.
The finding that surgical technique did not significantly affect improvement patterns, even after introduction of interaction terms, was surprising since revision surgery is difficult and technically demanding. This lack of effect may be attributable to the large variation in surgical techniques.
This study has a number of strengths. First, it is one of the few prospective cohort studies to focus particularly on the patterns of functional improvement following revision surgery. Second, to our knowledge, this is the first study of revision total knee arthroplasty to use a piecewise general linear mixed model analytic approach to model patterns of improvement. This approach is more appropriate than conventional regression analysis and is becoming popular in studies of functional outcomes. Not only does it model correlation between repeated measures of the same person, it also assesses fluctuations in function over time. Third, we used several instruments to simultaneously measure the functional status of the same patient at each time point. While the utility of this approach may be modest in studies of primary surgery, the use of multiple instruments has proved effective not only in identifying the early onset of deterioration following revision arthroplasty but also in identifying the domains in which the onset of deterioration is most likely to occur.
Despite its multicenter design, this study cohort may not be representative of the population of patients who undergo revision total knee arthroplasty. The seventeen centers are all high-volume joint replacement centers, the functional outcomes of which may be superior to those of the average community hospital. Moreover, the multicenter design of this study may have impeded our ability to observe the effect of modes of failure or surgical techniques on outcomes. In addition, despite a priori training, the surgeons may still have had slightly different definitions for the same mode of failure or may have carried out the surgical procedures with varying degrees of cementing or use of different prosthesis designs. Finally, this study did not account for the variation in rehabilitative efforts across patients, which may be critical to their functional recovery24.
Note: Central project team coordinators: L. Saleh, BSN, University of Toronto, Toronto, ON, Canada (1995-1999); A. Macaulay, PhD, Columbia University, New York, NY (1998-1999); J. Agel, MA, University of Minnesota, Minneapolis, MN (1999-2001); M. Celebresse, MA, MSc, University of Minnesota, Minneapolis, MN (2001-2003); and N. Miller, MSc, K. Schwartz, MSc, and H. Ghomrawi, MPH, University of Minnesota, Minneapolis, MN (2003-2006). NAKAR statisticians: B. Bershadsky, PhD, Cleveland Clinic, Cleveland, OH; L. Eberly, PhD, University of Minnesota, Minneapolis, MN; A. Gafni, PhD, McMaster University, Hamilton, ON, Canada; M. Kuskowski, PhD, University of Minnesota, Minneapolis, MN; G. Norman, PhD, University of Minnesota, Minneapolis, MN; R. Tweedie, PhD, University of Minnesota, Minneapolis, MN; and W. Novicoff, PhD, University of Virginia, Charlottesville, VA. Contributing investigators (contributed to the study by recruiting and inducting patients): M. Bostrom, MD, The Hospital for Special Surgery, New York, NY; R. Bourne, MD, London Health Sciences Centre, London, ON, Canada; H. Clark, MD, Insall Scott Kelly Institute, New York, NY; F. Cushner, MD, Insall Scott Kelly Institute, New York, NY; S. Haas, MD, The Hospital for Special Surgery, New York, NY; W. Healy, MD, Lahey Clinic, Burlington, MA; R. Laskin, MD, The Hospital for Special Surgery, New York, NY; J. Marsh, MD, University of Iowa, Iowa City, IA; J. McAuley, MD, Anderson Orthopaedic Research Institute, Alexandria, VA; R. McCalden, MD, London Health Sciences Centre, London, ON, Canada; C. Nelson, MD, University of Pennsylvania, Philadelphia, PA; C. Rorabeck, MD, London Health Sciences Centre, London, ON, Canada; J. Schmidt, MD, University of Minnesota, Minneapolis, MN; and N. Scott, MD, Insall Scott Kelly Institute, New York, NY.