As the demand for primary total joint arthroplasties continues to increase, alternate means of longitudinal follow-up will be necessary. What information should be collected from our patients, and at what time interval remains a topic of discussion. In this retrospective review of 1,012 total knee arthroplasties we found that the probability of requiring revision surgery could be calculated using AKS data from 3 critical postoperative time intervals: 0, 3, and 15 months. This model can account for 67 % of the variance concerning the probability of the patients needing revision surgery. This model effectively discriminated between revision and non-revision patients. The age of the patient at the time of revision and the length of time between primary surgery and revision were not significant predictors for revision surgery. Our findings provide further support to previous findings that Oxford knee and hip scores at 6-months and Oxford hip scores at 5-years were predictive of subsequent early revision [15, 16]. A recent UK study showed that review of questionnaire and radiograph together (but not either alone) identified all patients in need of increased surveillance after TKA/THA [17]. Our findings of association of pain/function scores with revision risk are similar to these previous studies. An advance from our study is the development of a prediction model using scores from validated questionnaires.
The proposed prediction equation would require AKS to be measured preoperatively and at 3 and 15 months post-TKA (at least two office visits). Some providers may only be following their patients on an as needed basis after the initial wound check in the first 2–6 weeks post-TKA. Patient-reported outcome measures may be able to capture data that were captured in AKS scores (a physician-administered measure), and may be a more practical alternative to AKS. Our prediction equation needs to be replicated using patient-reported outcome measures.
Although consensus recommendations in the US have favored annual or biannual follow-up [4], this has not been globally accepted. Bankes and colleagues [18] reviewed 1000 questionnaires to the British Orthopaedic Association (BOA) concerning follow-up practices. They found that 50 % of surgeons followed their total hip patients less than 1 year. The majority of providers (78 %) followed them for under 5 years, and only 14 % had indefinite follow-up. The authors reported that cost was a major deterrent of annual follow-up for all arthroplasty patients [18]. In our population, 73.9 % of revisions occurred early and were captured within the first 4 years.
The utility of the physician visit for well functioning arthroplasties has been questioned. Bhatia and Obadare [19] performed a cost-benefit analysis of 100 consecutive patients and found that the 304 visits with radiographs for 100 patients over a 2 year period resulted in a cost of £23,397 ($38,970 in 2003 USD). There were 10 patients that had issues requiring interventions, of which three were found in clinic follow-ups and seven identified by General Practice referral or in the Emergency Department. They noted that most issues that required intervention were found at the first postoperative visit. Their recommendations were for routine follow-up for 6–12 weeks [19]. The BOA recommends AP/lateral radiographs at 5 years and every 5 years thereafter for long-term surveillance (Anon. Total Hip Replacement: A Guide to Best Practice 1999). The British National Health Service also notes the waiting list time for new arthroplasty patient evaluation as a significant consideration for resource utilization [18, 19].
Patient experience and satisfaction have been driving forces in US healthcare reform and optimization. Sethuraman surveyed 100 patients during routine arthroplasty follow-up about their preference for office visits [20]. A significant proportion (45 %) would have preferred not to have an office visit, citing wages lost and potential time spared as determining factors. None of these patients felt that the quality of their care would be jeopardized by not having an office visit. It was noted that the remaining 55 % noted that office visits and routinely seeing their surgeon was important for maintaining quality of care and satisfaction. The authors recommended asking patients their preference and following those not interested in office follow-up with radiographs and completion of an outcomes measure.
There are limitations to the conclusions that can be drawn from the current study. The ability to generalize results using our population’s logistical regression equation is limited to use in similar patient populations with unilateral total knee arthroplasty. Although women represent 66.2 % of the primary total knee arthroplasty patients receiving a joint in one year [21], they represent 1 % of this Veterans Administration arthroplasty patient population. Single-site study and a low number of revisions limit the generalizability. Another limitation is the outcome instrument utilized. This version of the Knee Society Score itself is not without limitations. It is non-specific in its reporting of clinical change and function for patients with bilateral arthroplasties. Lingard and colleagues evaluated the validity and reliability of the AKS compared to the WOMAC and SF-36. They found that although the AKS had adequate convergent construct validity, it had weaker responsiveness and poor inter-item correlation compared with either of the other measures [12]. Both the clinical and function scores reach their maximum improvement at 2-years, followed by subsequent decline as a function of age and as the patient’s number of comorbidities increase [22]. Konig and colleagues prospectively evaluated 276 TKA patients and found that after clinical scores and functional scores plateaued, they tended to decrease by 5 points/year after 2 years postoperative [22]. A notable weakness of this study is the need to normalize the time intervals for generating the regression models required to compare the revision patients’ AKS scores. We realize that standard clinic follow up visits likely occurs at 2 weeks, 6 weeks, 3 months, 6 months and 1 year after the arthroplasty. In order to optimize comparisons of the trends of each individual despite variability of follow-up intervals, we utilized rules to calculate intervening scores. The validity of our conclusions are clearly related to the assumptions made.
Our study predicts the risk of revision only, recognizing that interventions other than revision may be needed by some patients post-TKA. Future studies should examine non-revision interventions as well.
Although this study can act as a basis for future comparisons, validation of the prediction equation could occur by repeating this study with an increased number of visits at 3 month intervals for 48 months. This may be costly and impractical. Validation could also take the form of replicating the techniques used here to evaluate similar TKA registries that have collected AKS scores over time.
Further research is needed to evaluate other arthroplasty outcome instruments for their ability to predict revision at associated critical time intervals. It may also be important to determine if regression modeling can assist in risk stratification for patients following revision surgery, identifying risk for further additional surgical interventions. Interestingly, the preoperative AKS scores did vary significantly between revision and non-revision groups. Previous studies have shown that preoperative differences in depression, pain and anxiety, and gender are predictive of poor outcome following total knee arthroplasty [23–25]. Future research can be directed at quantifying the utility of using preoperative AKS scores to identify patients at risk for poor outcome prior to surgery. Latent and acute revision subgroups were both statistically different from each other, and from the non-revision group. Due to the low number of patients in each of these subgroups it was impossible to create logistical regression equations to predict membership in each of these groups. As we continue to follow our cohort and identify more members in these groups, further analysis can be performed to describe predictive implications.