A consecutive series of patients in need of a revision procedure for a failed total knee arthroplasty were prospectively followed in a multi-center cohort study involving 14 centers in the United States and one in Canada. Patients were spread relatively equally between sites and a total of 6 patients were lost to follow-up. These 6 all came from separate units. All patients had to meet specific inclusion and exclusion criteria prior to enrollment in the study. The inclusion criteria were that at the least, the tibial and/or the femoral component required reconstruction, signed informed consent was obtained from the subject, the patient was over 18 years of age, the patient was cognitively intact, fluent in English, and capable of completing the self-administered questionnaires and adhering to the study protocol, the patient had a primary TKA that had failed, not a re-revision. The exclusion criteria were patients having a TKA re-revision, revision for failed unicondylar prosthesis, patients with metastatic or primary tumour of the knee, reflex sympathetic dystrophy of the leg, subject medically unfit to undergo TKAR, progressive muscular condition (with quadriceps weakness), neurologic deficit of affected limb, knee pain associated with spinal pathology, patient declined participation.
After obtaining IRB approval from each site, patients with failed total knee arthroplasties were approached about study participation. Once the patient agreed to participate, the investigator obtained subject consent, and the subject was then included in the study. Subjects and investigators then completed respective baseline forms.
As is necessary with any multi-center study of this nature, great care was taken a priori in the design of the study to ensure uniformity of indications, data management and follow-up between centers . All documentation was performed using a standard set of proforma questionnaires, for both surgeons and patients, structured so as to not permit of any deviation in data collection. Strict inclusion and exclusion criteria were applied from the outset and the coordinators that helped collect the data were blinded to study design and hypotheses.
The specific information gathered from the patients and investigators were Short Form-36 (SF-36) both mental (MCS) and physical components (PCS), the Western Ontario and McMaster Universities Osteoarthritis (WOMAC) Index (pain, stiffness, and difficulty of function), the Knee Society Score (KSS) both functional and clinical components, the Lower Extremity Activity Scale (LEAS), and a physician derived severity score, which is a visual analogue scale. Nine scales in all were thus involved in subsequent calculations. These instruments include the most commonly used in arthroplasty studies, for both primary and revision procedures. Although it might be argued that a TKAR population is potentially more heterogeneous than a primary population, these instruments are used in identical fashion for both populations and any new approach based on these instruments has to be robust enough to measure improvement in any arthroplasty cohort.
Among the less familiar scales used here, the physician derived severity score has been previously utilized and validated by the authors as an investigative tool in assessing the subjective physician judgment of the severity of the patient's condition and likelihood of good outcome, specifically as this relates to the failed or failing knee implant . The LEAS is a simple, patient administered instrument developed and comprehensively validated by the current authors that assesses the actual activity level of lower limb arthritis and arthroplasty patients .
Baseline forms were completed prior to the revision procedure and a further set of follow-up forms were subsequently completed at six months postoperatively. As we were testing here only a new methodological approach to analyzing postoperative improvement, we did not pursue longer clinical follow up of this cohort for the purpose of this study. Each of the constituent scales used results in a single 'outcome score' for that scale. Although these scores often present difficulties in clinical interpretation for individual patients, particularly those with 'mixed' outcomes (such as good in one measurement in the scale but poor in another) they are very useful in analyzing cohort populations, and represent the best means we currently have for outcome analyses. The changes in these scores from baseline to follow up were then converted into measures of improvement by assigning a positive sign to improvement in each patient's condition. This modification was necessary as, for example, a decrease in one system might signal improvement versus another system where increasing scores indicate improvement and so on. The resultant scores for the patients were then combined in order to determine the improvement or otherwise that occurred for each system (WOMAC, KSS, LEAS, physician derived score and SF-36). Improvement for each of the scales was then normalized by its respective standard deviations so that it was possible to compare the magnitude of improvement of individual scales.
Exploratory orthogonal factor analysis with varimax rotation was applied to the change in scores between the two time points for all instruments. The essential purpose of the factor analysis was to determine if the measures of change could be grouped into factors in order to more parsimoniously describe the concept of improvement. This orthogonality or 'independence' of the factors was an assumption we made a priori, but it should be noted, however, that the dimensions of improvement are not necessarily independent. This assumption was nevertheless necessary in practice, as the use of non-orthogonal factor analysis at this point of the study would provide too much uncertainty in the factorial model. For example, changes between scores in individual patients have smaller systematic variations than single scores and because they are based on the difference between 2 scores have a higher random error than any one of the basic measures.
As the result of this orthogonal rotation, then, we obtained several factors. Each of them was represented as a combination of all nine measures of improvement. We then applied non-linear transformation of the formulas (V1 = (D1+D5+D6+D7)/4, V2 = (D3+D4+D8)/3, V3 = D2, V4 = D9) – all coefficients that were greater than 0.6 were assumed to be equal and all coefficients that were smaller than 0.6 were replaced with zero. In this equation, D1-D9 values refer to the changes of scale scores from baseline to follow-up for the 9 outcomes scales used in this study. The resulting "new" factors became a subject of mean value analysis, correlation analysis and interpretation.
Because various scales had different rates of data completeness, factor analysis was initially performed using only the subset of the TKAR cohort where values on all 9 scales were complete. Thereafter, various sensitivity analysis tests were done to investigate generalizability of the results to the entire TKAR cohort. In this regard, we investigated stability of the factorial structure, as well as mean values and correlation coefficients.