Research Article

*Correspondence to Leena Sharma, Feinberg School of Medicine, Northwestern University, 300 E. Superior, 3-715, Chicago IL 60611

Funded by:
  National Institute of Arthritis and Musculoskeletal and Skin Diseases; Grant Number: P60-AR-48098, R01-AR-48748
  National Center for Research Resources; Grant Number: RR-00048

Received: 10 March 2003; Accepted: 4 September 2003

10.1002/art.11420  About DOI

Knee osteoarthritis (OA) is a leading cause of chronic disability in older persons ([1]). As defined by Jette et al, disability is the impaired performance of expected socially defined life tasks, in a typical sociocultural and physical environment ([2]). Physical function limitation, or difficulty with physical tasks and activities, is fundamental to the development of disability in OA. Pain is likely to be a central factor in the physical function limitation of knee OA, both due to its direct effects on function and as a route through which other factors operate. However, pain is not the only source of functional limitation in knee OA ([3]). In the Framingham study, limited function was more likely to occur in conjunction with moderately severe OA and infrequent pain than with milder OA and frequent pain ([4]). Identification of factors, in addition to pain, that contribute to poor physical function provides an opportunity to broaden the strategies to prevent disability.

The literature dealing with physical function limitation in knee OA includes a wealth of cross-sectional studies but few longitudinal studies. These published OA studies and/or longitudinal reports not limited to subjects with arthritis have introduced several factors as potential determinants of physical function outcome in knee OA, including obesity, comorbidity, depressive symptoms, low social support, and low levels of physical activity. Self-efficacy, defined by Bandura ([5]) as the belief in one's capacities to mobilize the internal resources and course of action needed to meet given situational demands, predicted physical function outcome in a longitudinal study of older adults with knee pain ([6][7]).

Local impairments of knee function in knee OA may also have some bearing on difficulty with physical tasks and activities. Several local factors influence knee function, including varus-valgus knee stability, by contributing to tibiofemoral congruence and load distribution across the articular surface; hip-knee-ankle alignment, by proportionately dividing load between the medial and lateral compartments; muscle activity, by adding stability, controlling the stop and start of joint motion, and compensating for gravity; and joint proprioception, or the perception of joint position, by providing input to the nervous system to guide periarticular muscle activity. Although these factors may be impaired in some individuals prior to disease development, they also may be made worse by OA ([8-10]). In the examination of the relationship between the disease itself and limitation of function in knee OA, the disease has been considered globally, using radiographic severity as a surrogate, whereas the impact of specific disease-associated impairments on limitation of function has received little attention.

There is no established approach to define and analyze physical function over time in studies of knee OA. The examination of change in a function measure between baseline and followup is limited by its failure to capture the possibility of a sustained high level of function as a good outcome or a sustained low level of function as a bad outcome. Also, the meaning of small increments of change in a function measure, treated as a continuous variable, is unclear. Ideally, the approach taken to assess outcome should capture an individual's experience over time, and should achieve this in an interpretable way.

Advancing knowledge of factors contributing to a poor physical function outcome will aid in the development of strategies to prevent function limitation and consequential disability. The goal of this prospective study was to identify the factors that predict a poor physical function outcome assessed over 3 years in individuals with knee OA.

PATIENTS AND METHODS

Sample.

Mechanical Factors in Arthritis of the Knee (MAK) is a prospective, observational, longitudinal study of function decline and OA disease progression in a cohort of individuals with established knee OA that began in 1997, and is described in previous reports (for example, refs. [9-14]). Participants were recruited from the community through advertising in periodicals targeting elderly persons, announcements at 67 neighborhood organizations, letters to members of the registry of the Buehler Center on Aging at Northwestern University, and local medical center referrals.

Inclusion/exclusion criteria were based on recommendations from the multidisciplinary workshop sponsored by the National Institute of Arthritis and Musculoskeletal and Skin Diseases and National Institute on Aging ([15]). Inclusion criteria were presence of definite tibiofemoral osteophytes (Kellgren/Lawrence [K/L] radiographic severity grade  2) in 1 or both knees, confirmed on study protocol radiographs, and a Likert category rating of at least  a little  difficulty with 2 or more of the Western Ontario and McMaster Universities Osteoarthritis Index (WOMAC) physical function scale items. Exclusion criteria were having received a corticosteroid injection within the previous 3 months or a history of avascular necrosis, rheumatoid or other inflammatory arthritis, periarticular fracture, Paget's disease, villonodular synovitis, joint infection, ochronosis, neuropathic arthropathy, acromegaly, hemochromatosis, Wilson disease, osteochondromatosis, gout, pseudogout, or osteopetrosis. Approval was obtained from the Office for the Protection of Research Subjects Institutional Review Board of Northwestern University.

Measurement of candidate risk factors.

Candidate risk factors were measured at baseline and at 18 months of followup. Varus-valgus laxity was measured using a previously described device and protocol that provide thigh and ankle immobilization, a stable knee angle, and fixed varus and valgus load ([9][11]). Laxity was measured as the angular deviation after varus and valgus load (expressed in degrees). All laxity measurements were performed by the same examiner and assistant (LS and SC). Our reliability with this device in persons with knee OA and varying body habitus was very good (within-session intraclass correlation coefficient [ICC] 0.85-0.96, between-session ICC 0.84-0.90) ([9][11]). Total rotation, which is the sum of varus and valgus rotation for each knee, was analyzed as a continuous variable.

Quadriceps strength and hamstring strength were measured isokinetically at 120°/second to assess maximal torque during movement (expressed in ft-lbs), using a computer-driven isokinetic dynamometer (Cybex Medical, Hauppauge, NY). A single tester assessed all patients, in accordance with a previously described protocol ([11]). To determine measurement reliability, actual test repetitions for each type of test, each muscle group, and each leg were analyzed. All ICCs exceeded 0.98.

Alignment was measured from a single anteroposterior radiograph of the full lower extremity, which was acquired using a graduated grid cassette. Alignment of the knee joint was assessed as the angle formed by the intersection of the mechanical axes of the femur (the line from femoral head center to femoral intercondylar notch center) and the tibia (the line from ankle talus center to the center of the tibial spine tips), following previously described acquisition and measurement protocols ([12]). One experienced reader (LS) made all measurements. Reliability of the alignment assessments for both varus and valgus knees was high (ICCs 0.98-0.99). The absolute value of severity of malalignment was analyzed as a continuous variable and was expressed as degrees from neutral (0°) in either the varus or valgus direction.

Proprioceptive inaccuracy was measured as the difference between actual knee-motion onset and the participant's detection of motion, using a previously described apparatus and protocol that provide computer-controlled, passive, angular knee motion at 0.3° per second and a precise measurement of angular displacement ([10][13]). Visual, auditory, vibration, cutaneous tension, and pressure cues to limb motion were eliminated. The participant was seated with the back supported, hips flexed to 70°, and knees flexed to 90°. In each trial, both legs were moved to a starting position of 45° knee flexion. ICCs for the assessments of participants with and without OA by a single experienced tester were 0.94 for the left leg and 0.95 for the right leg. Proprioceptive inaccuracy was analyzed as a continuous variable.

Body mass index (BMI) was measured as the weight (in kg) divided by the height (in m²). Knee pain intensity was measured using separate 100-mm visual analog scales (VAS) for pain in the right and left knees in the last week. Radiographs were obtained with individuals in a standing, semiflexed position, following the Buckland-Wright protocol (including knee position, beam alignment, markers to account for magnification, and fluoroscopic confirmation of knee position). The specific MAK protocol has been described previously ([9][11][12][14]).

Disease severity was assessed using the K/L global radiographic grading system (0 = normal, 1 = possible osteophytes, 2 = definite osteophytes and possible joint space narrowing, 3 = moderate/multiple osteophytes, definite narrowing, some sclerosis, and possible attrition, 4 = large osteophytes, marked narrowing, severe sclerosis, and definite attrition). Bilateral knee OA was defined as a K/L grade of  2 in both knees. Comorbidity was assessed using a questionnaire adaptation of the Charlson Index (higher score = greater comorbidity) ([16]). High test-retest reliability and strong correlation with the medical record-based Charlson Index has been demonstrated ([16]).

Physical activity was assessed using the Physical Activity Scale for the Elderly (PASE), a self-report measure of global activity that assesses recreational, occupational, and household activities (higher score = greater activity) ([17]). PASE was designed to assess activities commonly engaged in by older persons, and its construct validity and test-retest reliability have been demonstrated in community-dwelling older adults ([17]). Aerobic exercise was scored according to the amount of walking (for exercise), swimming, water aerobics, bicycling (including stationary), and exercise on aerobic equipment (stairmaster, rowing, or skiing machine) per week.

Psychological status was assessed using the mental health and role-functioning emotional subscales of the 36-item Short-Form (SF-36) (higher value = better). Criterion validity has been demonstrated for these individual scales, with significant and consistent associations with a range of validation criteria ([18]). Reliability for the individual scales has been demonstrated ([19]). Self-efficacy for lower extremity function was assessed using the lower extremity questions of the Arthritis Self-Efficacy Scale function subscale (self-rating of degree of certainty in one's ability to walk 100 feet on flat ground in 20 seconds, walk 10 steps downstairs in 7 seconds, and get out of an armless chair without using hands for support) (higher value = better) ([20]). Construct and concurrent validity and test-retest reliability for the subscales have been demonstrated in patients with arthritis ([20]). Social support was assessed using the Medical Outcomes Study Social Support Survey (higher value = better) ([21]). This survey represents several dimensions of support: emotional/informational, tangible, affectionate, and positive social interaction. High convergent and discriminant validity of the items, construct validity, and reliability have been demonstrated in patients with chronic conditions ([21]).

Physical function outcome from baseline to 3 years.

Physical function was assessed at baseline and at 3 years by using the Likert version of the WOMAC physical function scale and the rate of chair-stand performance. The WOMAC is a disease-specific (for knee or hip OA) self-report instrument with 17 questions comprising the physical function scale. It is extensively validated and widely recommended and used in studies of individuals with knee OA ([22][23]). The possible range of the physical function scale is 0-68, and a higher score indicates worse function. The WOMAC scores were categorized using quintiles of the MAK cohort at baseline, ranging from individuals with the best WOMAC function scores to those with the worst function, as follows: first quintile (WOMAC 0-7), second quintile (WOMAC 8-14), third quintile (WOMAC 15-22), fourth quintile (WOMAC 23-33), and fifth quintile (WOMAC >33).

As a task closely signifying knee status, chair-stand performance was assessed, i.e., time required for 5 repetitions of rising from a chair and sitting down, using the protocol described by Guralnik et al ([24]). Among community-dwelling elderly individuals, those with better performance in this chair-stand test had significantly less mobility-related disability and less disability in activities of daily living 4 years later ([24]). The rate of performance on the chair-stand test was used in our analysis; this value was expressed as the number of stands per minute calculated from the time required to complete 5 chair stands. The use of rate allows the inclusion of individuals who cannot complete the test (i.e., those with a rate of 0). The chair-stand rate was categorized using MAK cohort quintiles at baseline, ranging from those with the worst chair-stand performance to those with the best, as follows: first quintile ( 15.0), second quintile (>15.0 and  20.4), third quintile (>20.4 and  23.8), fourth quintile (>23.8 and  27.5), and fifth quintile (>27.5).

To characterize the baseline to 3-year function experience of each participant, quintile grids were used. The quintile grid captures practical, meaningful profiles of good and poor function over time. The WOMAC outcome grid is shown in Figure 1. The shaded squares represent a good baseline to 3-year WOMAC function outcome, i.e., moving into a higher function group or remaining within the 2 highest function groups. Squares that are not shaded represent a poor baseline to 3-year WOMAC function outcome, i.e., moving into a lower function group or remaining within the 3 lowest function groups. The chair-stand outcome grid is shown in Figure 2. Good and poor baseline to 3-year chair-stand outcomes are represented by shaded squares (moving into a faster chair-stand group or remaining within the 2 fastest groups) and nonshaded squares (moving into a slower group or remaining within the 3 slowest groups), respectively.

Statistical analysis.

Analyses were restricted to individuals who participated in both the baseline and 3-year assessments and who did not undergo total knee replacement during the study. Univariate screening was used to identify risk factors related to poor baseline to 3-year function outcome, using logistic regression. Odds ratios (ORs) from continuous variables are expressed per a specific clinically relevant increment, rather than per measured unit, to aid interpretation. For knee-specific variables, the maximal value of the 2 knees per participant was used for sample description and analyses. Adjusted ORs, controlling for age, BMI, knee pain intensity, disease severity, and other potential confounders, were estimated using multiple logistic regression.

The approach taken to consider additional potential confounders was based on scientific rationale, rather than derived by statistical criteria. There were several potential confounders in the age-physical function relationship. Strength and proprioceptive acuity are known to decline with age ([13][25][26]). Alternatively, an age effect might be due to less aerobic exercise, reduced self-efficacy, or greater comorbidity. Laxity is correlated, albeit modestly, with reduced strength ([11]) and greater malalignment ([12]). In theory, other psychosocial factors may confound the relationship between social support and function outcome. An effect of aerobic exercise may be mediated by strength, proprioceptive accuracy, mental health, or self-efficacy. Self-efficacy may be a confounder in the strength-function relationship.

To explore the robustness of the study results, sensitivity analyses included a subgroup analysis among the women, summing the values for the right and left knees (in contrast to using the maximal value), and altering the definition of a good outcome defined from the quintile grid. Finally, the outcome approach taken afforded us the opportunity to describe individuals with sustained high function and those with sustained low function.

RESULTS

Sample characteristics.

Among 285 participants, 257 (90%) completed the 3-year study. Reasons for not completing included incapacitation from another medical condition (8 participants), death (7 participants), moving out of the region (6 participants), and anxiety over coming into downtown Chicago due to the events of September 11, 2001 (7 participants). Among the 257 subjects who completed the study, 21 underwent total knee replacement during the study period and were excluded from the analyses. The 236 participants who completed the study without total knee replacement represent the sample in this report. The participants had a mean ± SD age of 68.6 ± 10.8 years, and 172 (73%) were women. The sample was predominantly white (80% versus 16% African American, 3% Hispanic, and 1% other), and 50% were married, 17% were single, 15% were divorced, and 18% were widowed.

The mean BMI at baseline was 30.6 kg/m² (±SD 6.1), the mean knee pain intensity on a VAS was 38.9 mm (±SD 26.8), the mean varus-valgus laxity was 5.9° (±SD 2.1), the mean quadriceps strength was 55.9 ft-lbs (±SD 29.2), the mean malalignment was 5.1° (±SD 3.4), and the mean proprioceptive inaccuracy was 2.4° (±SD 1.8). The higher K/L disease severity grade (of the 2 knees) was 2 in 43% of participants, 3 in 35%, and 4 in 22%. The 28 participants who did not complete the study were similar at baseline to those who completed the study without total knee replacement, except that the nonparticipants reported greater pain intensity (49.5 mm versus 38.9 mm on a VAS).

Of the 236 participants, 113 had a good baseline to 3-year WOMAC function outcome, and 123 had a poor outcome. Participants with a good and poor baseline to 3-year chair-stand outcome numbered 80 and 156, respectively. Univariate screening was used to identify the risk factors associated with a poor baseline to 3-year function outcome. Risk factors screened included baseline age, sex, marital status, comorbidity, radiographic severity of knee OA, prevalence of bilateral versus unilateral knee OA, BMI, knee pain intensity, mechanical factors (varus-valgus laxity, malalignment), neuromuscular factors (proprioceptive inaccuracy, quadriceps strength, hamstring strength), activity level (overall physical activity, amount of aerobic exercise), and psychosocial factors (mental health, role-functioning emotional subscale, self-efficacy, social support), as well as the baseline to 18-month change in each factor.

Relationship between candidate risk factors and WOMAC function outcome over 3 years.

Table 1 presents the mean value of each factor at baseline in both the good and the poor WOMAC outcome groups. The BMI, knee pain intensity, laxity, scores on mental health, self-efficacy, and social support, and amount of aerobic exercise per week were significantly different between the 2 outcome groups. Categorical factors that did not differ between participants with good outcome and those with poor outcome included sex (69% versus 79% female, respectively), marital status (50% versus 49% married, respectively), disease severity (higher K/L score of 2 in 47%, 3 in 35%, and 4 in 18% versus higher K/L score of 2 in 39%, 3 in 34%, and 4 in 27%, respectively), and prevalence of bilateral knee OA (89% in both groups).

Table 2 presents the unadjusted OR for a poor WOMAC outcome from baseline to 3 years, and the OR after adjustment for age, BMI, knee pain intensity, and disease severity. Baseline factors that significantly increased the likelihood of a poor WOMAC function outcome over 3 years were a higher BMI, greater knee pain intensity, and increased laxity, whereas better mental health, self-efficacy, and social support and more aerobic exercise were protective factors. The only factor in which the baseline to 18-month change predicted the WOMAC outcome over 3 years was knee pain intensity. After adjustment, the OR for the baseline to 18-month increase in pain intensity notably increased to 1.48, from an unadjusted OR of 1.32. For the other factors, the ORs either were not affected by adjustment or moved slightly toward unity.

Each of the following relationships persisted after further adjustment for additional potential confounders: laxity-WOMAC outcome after adjustment for alignment or strength (either as continuous or categorical variables), social support-WOMAC outcome after adjustment for mental health or self-efficacy, and aerobic exercise-WOMAC outcome after adjustment for strength, mental health, or self-efficacy.

Relationship between candidate risk factors and chair-stand performance outcome over 3 years.

Table 3 presents the mean value of each factor in the good outcome and poor outcome groups according to chair-stand performance rating. Age, proprioceptive inaccuracy, quadriceps strength, hamstring strength, scores on the role-functioning emotional and self-efficacy subscales, and amount of aerobic exercise per week differed significantly between the good and poor outcome groups. Categorical factors that did not differ between participants with good outcome and those with poor outcome included sex (75% versus 72% female, respectively), marital status (51% versus 49% married, respectively), disease severity (higher K/L grade of 2 in 41%, 3 in 36%, and 4 in 23% versus higher K/L grade of 2 in 45%, 3 in 34%, and 4 in 21%, respectively), and prevalence of bilateral knee OA (90% in both groups).

Table 4 presents the unadjusted OR for a poor chair-stand outcome, and the OR after adjustment for age, BMI, knee pain intensity, and disease severity. The risk of a poor chair-stand performance outcome over 3 years was increased by greater baseline age and proprioceptive inaccuracy (the increase in risk approached significance), whereas greater baseline quadriceps strength and hamstring strength, better self-efficacy, and more aerobic exercise were protective factors. None of the baseline to 18-month change variables were significant risk factors. For most factors, the OR was not affected by adjustment or was slightly attenuated. For the strength variables, the attenuation was somewhat greater.

Each of the following relationships persisted after further adjustment for additional potential confounders: age-chair-stand outcome after adjustment for strength, proprioceptive acuity, aerobic exercise, reduced self-efficacy, or greater comorbidity, and aerobic exercise-chair-stand outcome after adjustment for strength, proprioceptive acuity, or mental health. The strength-function relationship did not persist after adjustment for self-efficacy (i.e., for quadriceps strength, the OR adjusted for age, BMI, pain, disease severity, and self-efficacy was 0.95/20 ft-lbs, 95% CI 0.73-1.23)

Sex was not linked to outcome, and when the analyses were restricted to women, the results were similar. The number of men in the cohort was not sufficient to examine them as a separate group. Results were minimally altered by adjusting for baseline quintile, by considering the sum of values for the right and left knee as the knee-specific independent variables (as opposed to the maximal value), or by changing the definition of good outcome to include those who remained in the 3 highest function quintiles (as opposed to the 2 highest function quintiles).

Risk factor levels by subgroups of function outcome.

The outcome approach used in this study afforded us the opportunity to provide some characterization of the individuals in the outcome categories according to their baseline attributes. The outcome subgroups were divided among individuals with knee OA who, over 3 years, sustained a high level of function, improved in function, declined in function, or sustained a low level of function. The mean values of each factor are presented for each subgroup in Tables 5 (for WOMAC) and 6 (for chair-stand).

In terms of the baseline to 3-year WOMAC outcome, individuals with sustained high function had the best values for several factors, and had a strikingly lower BMI, less knee pain, and greater self-efficacy compared with those with sustained low function (Table 5). Persons who persisted at a low level of function (compared with those who improved) appeared to have greater laxity, less social support, and minimal change in the intensity of pain between baseline and 18 months (versus a decrease in pain in those whose function improved), and they were performing less aerobic exercise on average.

With regard to the baseline to 3-year chair-stand outcome, individuals with sustained high function had superior values for several factors, and had substantially better self-efficacy and quadriceps strength (Table 6). Persons who persisted at a low level of function (compared with those who improved) appeared to be older and weaker, and were performing less aerobic exercise.

DISCUSSION

The likelihood of a poor function outcome as assessed by the WOMAC over 3 years in persons with knee OA was increased by the presence of greater varus-valgus laxity, BMI, and knee pain intensity at baseline and a larger baseline to 18-month increase in pain intensity, and was decreased by better baseline mental health, self-efficacy, and social support and more aerobic exercise. The likelihood of a poor outcome as assessed by chair-stand performance was increased by an older age and greater proprioceptive inaccuracy (approaching significance), and was reduced by greater strength and self-efficacy and more aerobic exercise. The effect of strength was somewhat reduced by adjusting for age, pain, and BMI, and was lost after adjusting for self-efficacy. The WOMAC and chair-stand performance rating identified different factors, supporting the concept that self-report and performance measures each capture unique aspects of functioning and should not be substituted for each other. The change in most risk factors from baseline to 18 months did not predict the outcome over 3 years, possibly reflecting the generally small amount of change between baseline and 18 months.

The level of self-efficacy predicted the outcomes by both the self-report and the performance function measures. Differences in self-efficacy between persons who sustained high function and those who declined in function or sustained low function appeared larger than for any other factor. Our results are consistent with those reported by Rejeski et al, who found, in the Observational Arthritis Study in Seniors (OASIS), that in older persons with knee pain, self-efficacy predicted both the self-reported and stair-climb performance outcomes, after adjusting for pain and strength ([7]). In the current study, there was similarly no evidence of confounding by either the intensity of pain or strength. In addition, we found no evidence of confounding by mental health status or aerobic exercise.

This study is the first to show evidence of the relationship between greater baseline varus-valgus laxity and increased odds of a poor baseline to 3-year WOMAC function outcome. Both persons who sustained high function and those who improved had lower levels of laxity. Cross-sectional relationships between laxity and function (see ref. [11], which also involves the cohort of the current study), and between symptoms of instability and function ([27]) have been reported in knee OA. Varus-valgus laxity increased the likelihood of OA after ligament injury ([28][29]), was worse in persons with OA (even in their nonarthritic knees) than in elderly controls, and was made worse by aspects of the disease ([9]). The mechanism of the laxity-function relationship may, in theory, be related to symptoms of pain or instability, the effects of laxity on knee function, or both. We found no evidence of mediation by pain, or confounding by age, strength, or alignment.

The link between aerobic exercise and function outcome is consistent with the long-term benefit of aerobic exercise on disability in activities of daily living, as demonstrated in the Fitness, Arthritis, and Seniors Trial ([30]), and is consistent with the cross-sectional relationship between low aerobic capacity and poor function ([6]). Those subjects who sustained high function were performing the greatest amount of aerobic exercise. Adjustment for each of several factors, BMI, knee pain intensity, mental health, proprioceptive inaccuracy, self-efficacy, or strength, did not attenuate the relationship between aerobic exercise and function outcome, suggesting that aerobic exercise did not act through these factors, at least as measured here. The effect of aerobic exercise may be mediated by improvement in aerobic capacity or aspects of neuromuscular function not captured by strength and the accuracy of knee-motion detection.

Specific, coordinated quadriceps and hamstring activity is required to achieve the knee and hip movements that occur during the chair-stand task. Given the muscle specificity of this task, and the abundant evidence of a cross-sectional relationship between strength and function ([6][11][31-35]), it is surprising that the protective effect of strength was not greater and that there was attenuation after adjustment. However, these results are consistent with the relatively modest effect of strengthening exercise on physical function in the long term ([36]), and with the finding, in both of the studies that have considered this question, that strength does not protect against knee OA disease progression ([14][37]). As a muscle parameter, strength may not be the best surrogate for the vast contribution from muscle to joint function and task performance; other measures of muscle activity may be more closely related to the risk of poor physical function.

The strength-function outcome relationship was lost after additional adjustment for self-efficacy. Our results support a close relationship between strength, knee pain intensity, and self-efficacy in their effect on physical function in knee OA. Pain may acutely reduce the maximal voluntary contraction and lead to chronic activity revision or avoidance. A downward spiral of pain, weakness, and reduced self-efficacy may lead to substantial reduction in activity. In support of this paradigm, subjects with knee OA and pain were weaker than those without pain ([34]), and in OASIS, there was evidence of an interaction between strength and pain ([38]) as well as strength and self-efficacy ([7]), in evaluating the effects on physical function.

Proprioceptive inaccuracy, measured as a greater delay before detecting slow, passive knee motion, was associated with a poor chair-stand performance outcome (approaching significance). Various approaches to assess joint-position sense have been developed and applied in clinical studies, which reflect the status of different pathways that contribute to proprioceptive awareness under a variety of circumstances. A relationship between proprioceptive inaccuracy assessed at study baseline and physical function over the ensuing years has not previously been reported. Proprioceptive accuracy declines with age ([13]), was worse in persons with OA even in their nonarthritic knee than in healthy elderly subjects ([10]), and was correlated with physical function in cross-sectional studies of knee OA ([13][39-41]). The results of the current study are noteworthy, especially since this was a passive test; dynamic joint-position sense inaccuracy may play a greater role in physical tasks and activities and more strongly predict physical function outcome.

Knee alignment strongly influences the risk of tibiofemoral OA disease progression ([12]). In the current study, malalignment as a continuous variable was not significantly linked to function outcome. However, as we have previously reported, the malalignment-function relationship in knee OA appears to be nonlinear; participants with 2 knees that had alignment of  5° (either varus or valgus) were more likely to experience a poor function outcome than were those with both knees more neutrally aligned ([12]).

Both baseline knee pain intensity and the baseline to 18-month change in pain predicted function outcome. There is a paucity of information concerning the longitudinal relationship between pain intensity and physical function in knee OA. In the OASIS study, baseline knee pain intensity during ambulation and transfer predicted a decline in performance during stair-climb and car tasks, respectively, but not self-reported function ([38]). The relationship between knee pain intensity and function decline was not significant after accounting for self-efficacy and the self-efficacy-strength interaction ([7]). In the National Health and Nutrition Examination Survey I, presence of knee pain predicted difficulty in mobility, transfer, and instrumental activities of daily living 10 years later ([42]). Other reports describe a cross-sectional relationship between knee pain and function in those with knee OA ([6][43]) and in studies not limited to persons with OA ([35][44]).

Age predicted the chair-stand performance outcome, which is consistent with the finding that age was associated with an increased risk of losing mobility over the next 4 years in the Established Populations for Epidemiologic Studies of the Elderly ([45]). The current study is notable in 2 ways. First, a longitudinal relationship between age and function outcome, adjusting for comorbidity, was demonstrated in a cohort of individuals with established knee OA. Second, the examination of neuromuscular factors afforded an opportunity to explore mechanisms of the age effect. Although greater age is associated with reductions in strength and proprioceptive acuity, the age-function relationship did not appear to be mediated by these factors. It remains possible that age exerts its effect via other neuromuscular factors, i.e., sensory and/or motor impairments not captured by strength or the accuracy of detection of passive motion.

We found a relationship between BMI at baseline and the WOMAC function outcome over 3 years. A longitudinal relationship between BMI and physical function in individuals with knee OA has not been previously reported. In cross-sectional studies not limited to individuals with OA, obesity was associated with self-reported difficulty with activities (by Health Assessment Questionnaire, in the lower limb) even after adjusting for pain ([43]), and higher body weight was associated with slower task performance ([35]).

A relationship between depressive symptoms and physical function has been described in longitudinal studies not limited to individuals with arthritis, as summarized by Ormel et al ([46]), and in cross-sectional studies of knee OA ([31][34][47][48]). The demonstration in the current study that the status of mental health predicts subsequent physical function is important, given how function assessment might be biased by depressive symptoms.

The approach of the current study provided an opportunity to describe persons with knee OA in specific outcome subgroups. The values presented in Tables 5 and 6 are unadjusted, but in this format, they illustrate the attributes of individuals who sustain high function or low function as they might be encountered in clinical or screening settings. Those who sustained high function had the best values for most factors, but their high self-efficacy, low BMI, low pain intensity, and greater level of aerobic exercise were especially noteworthy.

The distinctive characteristics of the function subgroups would not have been evident if  change  in a function measure had been the defined outcome. The choice, in some observational studies, to focus on change may be intended as a way of inferring causation. However, in OA, change may require several years. The number of individuals who showed no change in function over 3 years was substantial: 98 (42%) remained in the same WOMAC group, and 105 subjects (44%) remained in the same chair-stand group. A focus on change ignores those with persistently high or low function, and thus effectively lumps these individuals together in the same group and reduces the ability to detect the effects of factors responsible for an individual's state of function. In a chronic disease that is slow to evolve, factors related to persistent low- or high-function states are of particular importance to the development of prevention or intervention programs.

This study has limitations. It would have been of value to separately analyze the effect of a risk factor on change in function status and on persistence in a given group. However, the modeling to achieve this required problematic post hoc exclusions of groups of subjects. Our approach gave weight to the need to identify factors linked to the persistence of low-function states or decline, both of which are problematic to the individual. We were able to confirm the findings separately in women, but not in men, due to the smaller number of men in the cohort. This confirmation in women was important, given that the impact of knee OA on physical function has been shown to be greater in older women than in other age and sex subgroups ([42]). The small amount of change in most factors between baseline and 18 months limited our ability to evaluate the impact of change in a given factor with outcome. Also, it is possible that a larger sample may have enabled us to detect an effect of the 18-month data. The absence of a link between comorbidity and function outcome may be related to the small number of comorbidities present, on average, in this cohort.

These results have implications for future studies and for strategies to optimize physical function in knee OA. Treatment of pain is believed to be the crux of disability prevention in knee OA, and the results of the current study provide additional, longitudinal support for this. However, several modifiable factors were identified that were independently linked to function outcome after adjusting for pain intensity, which supports a multifaceted approach. Results of short-term studies have suggested the greater benefit of combined approaches (e.g., acupuncture/diet/exercise versus pain therapy alone [49]). Moreover, the effect sizes for strengthening interventions appear to be enlarged by incorporating pain management, attention to psychosocial factors, self-management, or aerobic exercise into the intervention ([50][51]).

The current study supports a key place for self-efficacy enhancement in strategies to optimize physical function. Components of perceived self-efficacy include skill, experience, general self-worth, and motivation ([20]); each of these is addressed in self-management education, which has been shown to be beneficial in short-term studies in arthritis patients ([52]). Our results suggest that aerobic exercise may have greater impact than a general increase in physical activity. Although orthoses specifically for varus-valgus laxity may have value and should be developed, novel dynamic approaches are also emerging. Fitzgerald et al tailored agility and perturbation training techniques, which were used for anterior cruciate ligament deficiency, to address the functional instability of individuals with knee OA ([27]). Such techniques may also enhance periarticular muscle activity and dynamic proprioceptive acuity. The effect of exercise that is restricted to strengthening has been modest in long-term trials ([36]). Although strength maintenance is important, parameters of periarticular muscle function other than strength may better predict a good physical function outcome and may constitute a better target for disability-prevention strategies.

In conclusion, factors placing individuals with knee OA at greater risk of a poor function outcome by at least 1 of the 2 function measures included the local factors laxity and proprioceptive inaccuracy, as well as age, BMI, and knee pain intensity. Factors protecting against a poor function outcome included strength, the psychosocial factors mental health, self-efficacy, and social support, and the activity factor amount of aerobic exercise per week. The identification of these factors describes individuals with knee OA who are at greater risk of a poor physical function outcome and provides possible targets for rehabilitative and self-management strategies to prevent disability.