The Knee 17 (2010) 15–18

Contents lists available at ScienceDirect

The Knee

Predicting the longer term outcomes of total knee arthroplasty

Rajiv Gandhi a,⁎, Herman Dhotar a, Fahad Razak a,b, Peggy Tso a, J. Roderick Davey a, Nizar N. Mahomed a

a Division of Orthopedic Surgery, University of Toronto, 399 Bathurst St, Toronto ON, Canada M5T 2S8b Population Health Research Institute, McMaster University, Canada

⁎ Corresponding author. TorontoWestern Hospital, EaToronto ON, Canada M5T 2S8. Tel.: +1 416 603 5642; fa

E-mail address: Rajiv.gandhi@uhn.on.ca (R. Gandhi)

0968-0160/$ – see front matter © 2009 Elsevier B.V. Aldoi:10.1016/j.knee.2009.06.003

a b s t r a c t

a r t i c l e i n f o

Article history:Received 29 January 2009Received in revised form 3 June 2009Accepted 9 June 2009

Keywords:Knee arthroplastyComorbidityMental healthOutcomes

Weasked thequestion ofwhat are thepatient level predictors (age, gender, bodymass index, education, ethnicity,mental health, and comorbidity) for a sustained functional benefit at a minimum of 1 year follow-up after totalknee arthroplasty(TKA). Five hundred fifty-one consecutive patients were reviewed from our joint registrybetween the years of 1998 and 2005. Baseline demographic data and the outcome scores of theWestern OntarioMcMaster University Osteoarthritis Index (WOMAC) and Medical Outcomes Short-Form 36 (SF36) scores wereextracted from thedatabase. Longitudinal regressionmodelingwas performed to identify the predictive factors ofinterest.Wehad27%of data pointsmissing. Themean follow-up inour cohortwas3.0 years (range 1–8years) andtherewere no revisions performed during this time. Clinical outcome scoreswere found to be relatively constantfor 3–4 years after surgery and then demonstrated a gradual decline after that. Older age, year of follow-up,greater comorbidity, and a poorer mental health state at time of surgery were identified as negative prognosticfactors for a sustained functional outcome following TKR (Pb0.05). Knowledge of these factors that predictoutcomes should be used in setting appropriate patient expectations of surgery.

© 2009 Elsevier B.V. All rights reserved.

1. Introduction 1 year follow-up. With this analysis, the null hypothesis is that age,

It has been estimated that 7.5% of the population over the age of55 years have some degree of knee pain and functional limitationassociated with radiographic evidence of osteoarthritis(OA) [1,2].Total knee arthroplasty(TKA) has been shown to provide reliable painrelief and quality of life improvement in the short and long term [3–6].

Few authors however have evaluated the patient level predictors ofan improved longer term outcome in TKA. Knowledge of these factorsallows the surgeon to appropriately counsel patients on realisticexpectations and potentially determine optimal timing of surgery.Factors such as gender [7], age [7] and mental health status [8,9] havebeen suggested to negatively impact long term function. Many ofthese studies present results that have not been adjusted for potentialconfounders of the relationship between the predictive factor and theoutcome. Moreover, no studies have used longitudinal regression, orrepeated measures analysis, which is the most powerful analysis as itaccounts for all data points from baseline to last follow-up andevaluates the effect of time on an outcome. It is essential to account forthe effect of time (and aging) on an outcome as cross-sectionalpopulation studies have demonstrated an inverse relationshipbetween SF-36 scores and older age groups [10,11].

The primary objective of our study was to use longitudinalregression modeling to identify the patient level predictors for asustained functional outcome following TKA for OA at a minimum of

st Wing 1-439, 399 Bathurst St,x: +1 416 603 3437..

l rights reserved.

gender, body mass index(BMI), education, ethnicity, mental health,and comorbidity do not affect knee arthroplasty outcomes.

2. Materials and methods

This study was designed to use longitudinal regression techniquesto identify the predictive factors for a sustained functional outcomefollowing TKA.

2.1. Study sample

As part of our prospective total joint arthroplasty database,patients are recruited from a single Canadian academic institution,the TorontoWestern Hospital, while on a waiting list for primary kneereplacement surgery. All patients are consented to participate by aresearch coordinator not involved in the medical care of the patients.All data are collected by patient self report. Patients not returning forfollow-up then had the questionnaires mailed to their home. Phonecalls were also subsequently made to encourage full participation.

Our inclusion criteria for this study were being age 18 and above, adiagnosis of primary osteoarthritis, and a minimum of 1 year follow-up. All surgeries were unilateral surgeries and were performed by oneof two fellowship trained arthroplasty surgeons between the years of1998 and 2005. Both surgeons used the same implant (Genesis II,Smith and Nephew, Memphis, Tennessee) and used a cementeddesign. All patients were treated with immediate weight bearing aftersurgery and an identical postoperative protocol of antibiotics and

Table 1Demographic data and baseline functional scores for the knee replacement cohort.

Demographic covariate

Mean Age (SD) 67.4 (9.8)% Males 202/551(36.6%)Mean BMI kg/m2 (SD) 30.1 (6.3)% White ethnicity 416/551(75.5%)% Higher education 233/551(42.3%)Mean comorbidity (SD) 2.8 (1.7)Mean baseline WOMAC score (SD) 51.6 (17.3)Mean baseline SF-36 PF score(SD) 25.0 (20.8)Mean baseline SF-36 RP score(SD) 22.6(35.2)

16 R. Gandhi et al. / The Knee 17 (2010) 15–18

thromboembolism prophylaxis. The study protocol was approved bythe Human Subject Review Committee.

2.2. Collection of data

We extracted baseline demographic data of age, gender, ethnicityand BMI from the database. Highest level of educationwas recorded aseither higher education level (university or above) or low educationlevel (high school or below). Comorbidity was defined by the 14categories of chronic illness adapted from the Cumulative IllnessRating Scale (CIRS) [12,13]. The CIRS covers the domains of 1) cardiac,2) vascular, 3) hematological, 4) respiratory, 5) otorhinolaryngologicaland ophthalmological, 6) upper gastrointestinal, 7) lower gastro-intestinal, 8) hepatic and pancreatic, 9) renal, 10) genitourinary, 11)musculoskeletal and tegumental, 12) neurological, 13) endocrine,metabolic and breast, and 14) psychiatric systems.

Ethnicity was collected also by patient self report under thecategories of White, Black, European, Asian, or Aboriginal. Patientscould choose as many as were appropriate. We had no patients underthe category of Aboriginal. Those patients selectingWhite or Europeanwere collapsed into a White category. Asian refers to individuals whoclassified themselves as South Asian (India, Pakistan, Bangladesh andSri Lanka) or East Asian (China, Japan, Taiwan, and Korea).

Patient functional status was assessed preoperatively and thenannually with the Western Ontario McMaster University OsteoarthritisIndex (WOMAC) scale [14]. A greater score on the WOMAC scalerepresents poorer function or greater pain and stiffness [14]. Thepsychometric properties of theWOMAC scorewith respect to reliability,validity, and responsiveness have beenwell established in the literature[15]. Patient health related quality of life was assessed by the MedicalOutcomes Study Short-Form 36 (SF-36) preoperatively and at yearlyfollow-ups [16–18]. The SF-36 has 8 subscales that generically measurehealth status using a 0–100 scoring scale [19]. Contrary to theWOMAC, ahigher SF-36 score represents better quality of life.

Fig. 1. Unadjusted WOMAC scores plotted over time for TKR (number of data points ateach year of follow-up is given in parenthesis).

2.3. Statistical analysis

In our dataset of 551 consecutive patients, each patient contributeda minimum of two functional scores (baseline and at least one follow-up). Missing data in a longitudinal analysis indicates that not all Npatients have data on all T repeated measurements [20]. We had 143patients missing their 1 year WOMAC follow-up data point howeverall of these patients contributed their 2 year follow-up and thus wereincluded in the analysis. The options for managing missing datainclude Last Value Carried Forward (LVCF), multiple imputationmethod, or simply to leave the data as missing [20]. Multipleimputation assumes normally distributed data and generally givespoint estimates that would be closest to those that would be derivedfrom a complete dataset [20]. The Generalized Estimating Equations(GEE) method allows for efficient analysis of longitudinal data byaccounting for the within subject correlation between repeatedmeasures and also by including all provided follow-up data fromeach subject, even if it is not complete [21]. The analysis performedwith multiple imputation and leaving the data missing provided thesame results and therefore we present the data without imputation.

We fitmultivariable longitudinal regressionmodels to identify thosefactors that predict an improved functional status at a minimum 1 yearfollow-up. With this technique, yearly follow-up scores on the samepatient are not considered independent. Separate models were createdfor each of the three dependent variables, the total WOMAC score, theSF-36 role physical (RP) score, and SF-36 physical function (PF) score. Inlongitudinal regression, the dependent variables are the correspondingchange in the outcome score fromyear to year, and thus naturally adjustfor baseline level of function. The covariates entered into the modelswere age, gender, ethnicity, BMI, comorbidity, and level of education.Ethnicity was collapsed into the categories of white and non-white forthe models. All covariates were retained in the models whethersignificant or not to maintain face validity of the models.

All statistical analysis was done with SPSS version 13.0 (Chicago,Illinois). Parameter estimates for regression modeling and their 95%confidence intervals (CI) are reported. All reported P values are 2tailed with an alpha of 0.05.

3. Results

The mean follow-up for our cohort of 551 knee replacement patients was 3.0 years(range 1–8 years). The demographic data and baseline functional scores for the cohortare given in Table 1. No patients required revision surgery during the study. No patientsdied during the follow-up period of this study.We had 27% of data points missing in thisstudy. Responders were not significantly different from non-responders in age, BMI,gender, ethnicity, or comorbidity.

Figs. 1–3 graphically show the mean WOMAC, mean SF-36 RP, and mean SF-36 PFscores for our cohort across the years of follow-up.

Longitudinal regression showed that an older age, female gender, year of follow-upand poorer mental health state were predictive of a less sustained functional outcomeon the WOMAC scale(Pb0.05, Table 2). For the outcome of SF-36 PF, a greater age,

Fig. 2. SF-36 physical function scores plotted over time for TKR (number of data pointsat each year of follow-up is given in parenthesis).

Fig. 3. SF-36 role physical scores plotted over time for TKR (number of data points ateach year of follow-up is given in parenthesis).

Table 3Longitudinal regression model predicting SF-36 physical function scores for TKR at amean 3 year follow-up.

Beta coefficient (95% C.I.) for predictingSF-36 physical function scores

P value

Year of follow-up 0.39 (−0.45, 1.23) 0.37Age −0.44 (−0.72, −0.17) 0.002Gender (ref male) 1.99 (−2.62, 6.61) 0.40BMI 0.06 (−0.28, 0.40) 0.73Comorbidity −2.10 (−3.76, −0.45) 0.013Higher education 8.30 (3.53, 13.10) b0.001Ethnicity (White) −0.93 (−6.86, 5.00) 0.76SF-36 MH 0.11 (0.01, 0.21) 0.031

Note: in this model, a lower outcome score indicates a poorer outcome.

17R. Gandhi et al. / The Knee 17 (2010) 15–18

greater comorbidity, lesser education, and a poorer mental health state predicted a lesssustained functional outcome (Pb0.05, Table 3). For the outcome of SF-36 RP, a greaterage, greater comorbidity, a poorer mental health state, and year of follow-up wereindependent predictors of a poorer outcome (Pb0.05, Table 4).

4. Discussion

To our knowledge, this study is the first to use longitudinalregression to identify the patient level predictors (age, gender, BMI,education, ethnicity, mental health, and comorbidity) for a sustainedfunctional improvement following TKA. We found that a greaterpatient age at time of surgery, greater comorbidity, and a poorermental health state at baseline independently predicted a lesssustained functional outcome at a mean 3 year follow-up.

Year of follow-up was also a significant predictor of outcomeindependent of all patient level demographic factors indicating thattime itself is an important variable for understanding the decline infunction following TKA. Fig. 1 shows thatWOMAC scores are relativelyconstant for the first 4 years after surgery and then demonstrate agradual decline after that. Figs. 2 and 3 indicate a similar patternwhereby health related quality of life showed a gradual decline after3 years. Simple linear regression performed with the baseline and lastyear of follow-up outcome scores would assume a straight linerelationship between these two points, however this data indicatesthat is not the true relationship. This further lends strength to theargument for using repeated measures analysis as the statisticalmethod for studying longer term outcomes in joint arthroplasty.

Similar to our findings, Brander et al. showed that preoperativedepression negatively affects functional outcomes of TKA 5 years aftersurgery [8]. These authors however measured outcomes with the KneeSociety Score (KSS) which is yet unvalidated. Bourne et al. performedonly unadjusted analyses and found a similar finding to us in thatadvancing age predicted a less sustained functional outcome at aminimum of 5 year follow-up [7]. Moreover these authors reported thatthe female gender predicted a lesser functional outcome in only one of

Table 2Longitudinal regression model predicting WOMAC scores for TKR at a mean 3 yearfollow–up.

Beta coefficients (95% C.I.) for predicting WOMAC scores P value

Year of follow up 0.66 (0.005, 1.32) 0.048Age 0.37 (0.19, 0.54) b0.001Gender (ref male) 4.55 (1.32, 7.77) 0.006BMI −0.06 (−0.31,0.19) 0.64Comorbidity 0.80 (−0.15, 1.76) 0.100Higher education 1.26 (−1.88, 4 .40) 0.43Ethnicity (white) 3.71 (−0.36, 7.78) 0.074SF-36 MH −0.11 (−0.17, −0.05) b0.001

Note: in this model, a greater outcome score indicates a poorer outcome.

their four functional outcome scores while we found that female genderpredictedapoorer functional outcomeinoneof the threeoutcomescoresin our study. Nunez et al. used linear regressionmodeling to examine therelationship between patient factors and each of the domains ofWOMAC: pain, stiffness and function at 3 years following TKA. Theirmost significant finding was that severe obesity (BMI 35.0–39.9 kg/m2)predicted greater reported pain after surgery [22]. Our study, similar toothers, showed no significant association between BMI and long termoutcomes following TKA [7,8,23]. Fewauthors have examined the impactofmedical comorbidity on long term functional outcomes following TKA.Lingard et al. showed that comorbidity was a significant predictor ofoutcome at 2 years follow-up [9].

The strength of this analysis is that it utilizes all data pointsbetween baseline and the latest follow-up to examine the impact oftime on change in function. Other studies in the orthopedic literatureon this topic only use the latest follow-up score in the analysis and assuch, a patient with a gradual decline in functionwill appear the sameas a patient with a stable outcome and a later more abrupt decline infunction and does not consider the effect of time on function.Moreover, we used validated outcome measures in our study.

There are potential limitations of our study. First, although wereported no revisions in our cohort, we did not examine radiographsfor radiolucent lines and potential implant loosening. Second, our datarepresent the experience of a high volume academic hospital and assuch our findings are only directly generalizable to similar hospitals.Third, we had a 27% incidence of missing data points in our analysishowever we found no difference in demographic covariates betweenresponders and non-responders which would lessen the opportunityfor selection bias. Fourth, there is the potential for unmeasuredresidual confounders of the outcome in our analysis.

In summary, we identified older age, year of follow-up, greatercomorbidity, and a poorer mental health state as negative prognosticfactors for a sustained functional outcome following TKA. Functionalresults are relatively constant for the first 3 to 4 years followingsurgery, and then show a gradual decline following that. Patientshould be appropriately counseled prior to surgery to ensure realisticexpectations following surgery on longer term outcomes.

Table 4Longitudinal regression model predicting SF-36 role physical scores for TKR at a mean 3year follow-up.

Beta coefficient (95% C.I.) for predictingSF-36 role physical scores

P value

Years of follow-up −2.22 (−3.63, −0.80) 0.002Age −0.60 (−0.96, −0.24) 0.001Gender (ref male) −2.11 (−9.75, 5.52) 0.59BMI −0.02 (−0.59, 0.55) 0.95Comorbidity −3.39 (−5.74, −1.04) 0.005Higher education 2.07 (−5.26, 9.41) 0.58Ethnicity (White) −6.86 (−16.34, 2.63) 0.16SF-36 MH 0.21 (0.06, 0.36) 0.007

Note: in this model, a lower outcome score indicates a poorer outcome.

18 R. Gandhi et al. / The Knee 17 (2010) 15–18

Acknowledgement

The authors thank Woojin Yoon for his assistance with the statis-tical analysis.

References

[1] Dieppe P, Basler HD, Chard J, Croft P, Dixon J, Hurley M, et al. Knee replacementsurgery for osteoarthritis: effectiveness, practice variations, indications andpossible determinants of utilization. Rheumatology (Oxford) 1999;38:73–83.

[2] Felson DT, Naimark A, Anderson J, Kazis L, Castelli W,Meenan RF. The prevalence ofknee osteoarthritis in the elderly. The Framingham Osteoarthritis study. ArthritisRheum 1987;30:914–8.

[3] Heck DA, Robinson RI, Partridge CM, Lubitz RM, Freund DA. Patient outcomes afterknee replacement. Clin Orthop Relat Res 1998;356:93–110.

[4] Ethgen O, Bruyere O, Richy F, Dardennes C, Reginster JY. Health-related quality oflife in total hip and total knee arthroplasty. J Bone Joint Surg Am 2004;86:963–74.

[5] Jones CA, Voaklander DC, Johnston DW, Suarez-Almazor ME. Health-relatedquality of life outcomes after total hip and knee arthroplasties in a communitybased population. J Rheumatol 2000;27:1745–52.

[6] Diduch DR, Insall J, ScottWN, Scuderi G, Font-Rodriguez D. Total knee replacementin young, active patients. Long-term follow-up and functional outcome. J BoneJoint Surg Am 1997;79:575–82.

[7] Bourne RB, McCalden RW, MacDonald SJ, Mokete L, Guerin J. Influence of patientfactors on TKA outcomes at 5 to 11 years follow-up. Clin Orthop Relat Res2007;464:27–31.

[8] Brander V, Gondek S, Martin E, Stulberg SD. Pain and depression influenceoutcome 5 years after knee replacement surgery. Clin Orthop Relat Res2007;464:21–6.

[9] Lingard EA, Katz JN, Wright EA, Sledge CB. Predicting the outcome of total kneearthroplasty. J Bone Joint Surg Am 2004;86:2179–86.

[10] Bowling A, BondM, Jenkinson C, Lamping DL. Short-Form 36 (SF-36) health surveyquestionnaire: which normative data should be used? Comparisons between thenorms provided by the Omnibus Survey in Britain, the Health Survey for Englandand the Oxford Health Life Survey. J Public Health Med 1999;21:255–70.

[11] HopmanWM, Towheed T, Anastassiades T, Tenenhouse A, Poliquin S, Berger C, et al.Canadian normative data for the SF-36 health survey. CMAJ 2000;163(3):265–71.

[12] Linn BS, Linn MW, Gurel L. Cumulative illness rating scale. J Am Geriatr Soc1968;16:622–6.

[13] Miller MD, Paradis CF, Houck PR, Mazumdar S, Stack JA, Rifai AH, et al. Rating chronicmedical illness burden in geropsychiatric practice and research: application of theCumulative Illness Rating Scale. Psychiatry Res 1992;41:237–48.

[14] Bellamy N. WOMAC osteoarthritis index: a user's guide. London, ON; 1995.[15] Bellamy N, Buchanan WW, Goldsmith CH, Campbell J, Stitt LW. Validation study of

WOMAC: a health status instrument for measuring clinically important patientrelevant outcomes to antirheumatic drug therapy in patients with osteoarthritis ofthe hip or knee. J Rheumatol 1988;15:1833–40.

[16] Ware Jr JE, Sherbourne CD. The MOS 36-item short-form health survey (SF-36). I.Conceptual framework and item selection. Med Care 1992;30:473–83.

[17] McHorney CA, Ware Jr JE, Raczek AE. The MOS 36-Item Short-Form Health Survey(SF-36): II. Psychometric and clinical tests of validity in measuring physical andmental health constructs. Med Care 1993;31:247–63.

[18] McHorney CA, Ware Jr JE, Lu JF, Sherbourne CD. The MOS 36-item Short-FormHealth Survey (SF-36): III. Tests of data quality, scaling assumptions, and reliabilityacross diverse patient groups. Med Care 1994;32:40–66.

[19] Ware JE, Snow KK, Kosinski M, Gandek B. SF-36 health survey: manual andinterpretation guide. 2nd ed. Boston: The Health Institute, New England MedicalCenter; 1997. p. 10–26.

[20] Twisk J, de Vente W. Attrition in longitudinal studies: how to deal with missingdata. J Clin Epidemiol 2002;55:329–37.

[21] Liang KY, Zeger SL. Longitudinal data analysis using generalized linear models.Biometrika 1986;73:13–22.

[22] Nunez M, Nunez E, del Val JL, Ortega R, Segur JM, Hernandez MV, et al. Health-related quality of life in patients with osteoarthritis after total knee replacement:factors influencing outcomes at 36 months of follow-up. Osteoarthr Cartil2007;15(9):1001–7.

[23] Amin AK, Patton JT, Cook RE, Brenkel IJ. Does obesity influence the clinical outcomeat five years following total knee replacement for osteoarthritis? J Bone Joint SurgBr 2006;88:335–40.