ORIGINAL ARTICLE

Avoidance of Activities in Early Symptomatic KneeOsteoarthritis: Results from the CHECK Cohort

Jasmijn F. M. Holla, MSc &

Marike van der Leeden, PT, PhD & Dirk L. Knol, PhD &

Wilfred F. H. Peter, PT & Leo D. Roorda, MD, PhD &

Willem F. Lems, MD, PhD & Janet Wesseling, PT, MSc &

Martijn P. M. Steultjens, PhD & Joost Dekker, PhD

Published online: 13 March 2012# The Society of Behavioral Medicine 2012

AbstractBackground Pain-related avoidance of activities is hypoth-esized to lead to lower muscle strength and thereby activitylimitations. Negative affect (e.g., low vitality, depression) isthought to strengthen the tendency to avoid activities.Purpose The aim of this study was to assess the validity ofthis “avoidance model” in patients with early symptomaticknee osteoarthritis (OA).Methods Cross-sectional data (n0151) were used. Theassociations between pain, negative affect, avoidance, mus-cle strength, and activity limitations were modeled usingstructural equation modeling.Results Pain and negative affect were associated with lowermuscle strength via avoidance (mediation by avoidance).Avoidance was associated with activity limitations vialower muscle strength (mediation by muscle strength).

There were also direct associations between pain, negativeaffect, avoidance, muscle strength, and activity limitations.Conclusions The results support the validity of theavoidance model, which explains the associations betweenpain, negative affect, avoidance, muscle strength, and activitylimitations in patients with early symptomatic knee OA.

Keywords Knee osteoarthritis . Avoidance model . Activitylimitations .Muscle strength . Pain . Depression

Introduction

Activity limitation (e.g., difficulties in walking, stairclimbing) is one of the primary health complaints inosteoarthritis (OA) patients [1]. One of the factors that

J. F. M. Holla (*) :M. van der Leeden :W. F. H. Peter :L. D. Roorda : J. DekkerAmsterdam Rehabilitation Research Center | Reade,P.O. Box 58271,1040 HG Amsterdam, The Netherlandse-mail: j.holla@reade.nl

M. van der Leeden : J. DekkerDepartment of Rehabilitation Medicine,VU University Medical Center,Amsterdam, The Netherlands

M. van der Leeden : J. DekkerEMGO Institute for Health and Care Research,VU University Medical Center,Amsterdam, The Netherlands

D. L. KnolDepartment of Epidemiology and Biostatistics,VU University Medical Center,Amsterdam, The Netherlands

W. F. LemsJan van Breemen Research Institute | Reade,Amsterdam, The Netherlands

W. F. LemsDepartment of Rheumatology, VU University Medical Center,Amsterdam, The Netherlands

J. WesselingDepartment of Rheumatology and Clinical Immunology,University Medical Center Utrecht,Utrecht, The Netherlands

M. P. M. SteultjensInstitute for Applied Health Research and School of Health andLife Sciences, Glasgow Caledonian University,Glasgow, Scotland, UK

J. DekkerDepartment of Psychiatry, VU University Medical Center,Amsterdam, The Netherlands

ann. behav. med. (2012) 44:33–42DOI 10.1007/s12160-012-9353-x

may influence the degree of activity limitations is theway in which patients cope with their symptoms [2].Avoidance of pain-related activities is a way to copewith pain. Several studies in knee OA patients havedemonstrated that the pain coping strategy “avoidanceof activities” or “resting” is related to activity limitations[3–5].

During the past decades, several theoretical models havebeen developed to explain activity limitations in chronicpain patients [6]. Whereas the first models were based onclassical and operant conditioning, the latest models arebased on a cognitive approach owing to the prevailingassumption that individuals actively process informationregarding internal stimuli and external events [6]. In 1993,based upon previous work in chronic pain (mainly low backpain) [7–10], Dekker et al. proposed a cognitive behavioralmodel of activity limitations in OA: the avoidance model[11].

The avoidance model [11, 12] offers an explanation ofhow avoidance leads to activity limitations. According tothis model (Fig. 1), a patient initially experiences painduring activity. This leads to the expectation that renewedactivity results in more pain [13] and consequently to theavoidance of activities. In the short term, avoidance mayhave the desired effect of less pain due to the decreased loadon the affected joint. However, in the longer term, inactivityresults in physical deconditioning, most notably lower mus-cle strength. Lower muscle strength leads to an increase inactivity limitations [12].

As shown in Fig. 1, the model hypothesizes an indirectrelationship between pain during activities and lower musclestrength via avoidance [11]. Avoidance of activities ishypothesized to mediate the relationship between pain

and lower muscle strength: a mediator variable is con-ceptualized as the mechanism through which one variableinfluences another variable [14, 15]. Likewise, the rela-tionship between avoidance and activity limitations ishypothesized to be mediated by lower muscle strength[11, 12].

In addition, it is hypothesized that negative affect plays amoderating role in the avoidance model (Fig. 1) [11]. Amoderator is a variable that strengthens (or weakens) therelation between a predictor and an outcome [14, 15]. Neg-ative affect, defined as a broad range of aversive mood statesincluding fatigue, low vitality, nervousness, and depression,has been shown to be associated with pain and activitylimitations in OA patients [2, 11]. Depressive symptomsare a risk factor for physical inactivity [16] and are associ-ated with decreased pain tolerance [17]. Therefore, negativeaffect is thought to strengthen (or moderate) the tendency toavoid activities.

In a previous study, evidence was obtained for the medi-ating role of lower muscle strength in the relationship be-tween avoidance of activities and activity limitations inpatients with established knee OA [12]. In that study, themediating role of avoidance and the moderating role ofnegative affect were not examined. Furthermore, althoughit is expected that the avoidance model applies to early OA,this has not been empirically tested. In early OA, in whichsymptoms commence and patients experience pain duringactivity for the first time, the processes of mental and phys-ical adaptation depicted in the model are thought to beinitiated.

The aim of the present study was to assess the validity ofthe avoidance model in patients with early symptomaticknee OA. It was hypothesized that (1) higher pain is asso-ciated with lower muscle strength via avoidance of activities(mediation of avoidance), (2) avoidance is associated with ahigher degree of activity limitations via lower musclestrength (mediation of muscle strength), and (3) negativeaffect (i.e., feelings of fatigue, low vitality, depression, andnervousness) strengthens the relationship between higherpain and avoidance of activities (moderation of negativeaffect).

Methods

Participants

A cross-sectional study was conducted in a sample of 151participants with early symptomatic knee OA from the Co-hort Hip and Cohort Knee (CHECK) study [18]. Two-yearfollow-up data were used.

CHECK is a prospective cohort study of 1,002 individu-als with early symptomatic OA of the knee or hip. On entry,

pain during activities

avoidance of activities

lower muscle strength

activity limitations

negative affect

Fig. 1 Hypothesized model of the associations between pain, negativeaffect, avoidance of activities, lower muscle strength, and activitylimitations in knee osteoarthritis patients

34 ann. behav. med. (2012) 44:33–42

all participants had pain or stiffness of the knee or hip,and were aged 45–65 years. They had not yet consultedtheir physician for these symptoms, or the first consulta-tion was within 6 months before entry. Participants withany other pathological condition that could explain thesymptoms were excluded (e.g., other rheumatic disease,previous hip or knee joint replacement, congenital dys-plasia, osteochondritis dissecans, intra-articular fractures,septic arthritis, Perthes’ disease, ligament or meniscusdamage, plica syndrome, Baker’s cyst). Additional exclu-sion criteria were comorbidity that did not allow physicalevaluation and/or follow-up of at least 10 years, malig-nancy in the last 5 years, and inability to understand theDutch language.

The CHECK cohort was formed from October 2002till September 2005. Nationwide, 10 general and academ-ic hospitals in the Netherlands are participating, locatedin urbanized and semi-urbanized regions. General practi-tioners in the surroundings of the participating centerswere invited to refer eligible persons. All patients thatvisited the general practitioner on their own initiative,potentially fulfilling the inclusion criteria, were referredto one of the 10 participating centers. In addition, par-ticipants were recruited through advertisements andarticles in local newspapers and on the Dutch ArthritisAssociation website. The physicians in the participatingcenters checked whether referred patients as well aspatients from their outpatient clinics fulfilled the inclu-sion criteria.

After 2 years of follow-up, all participants with kneesymptoms recruited through Reade, Center for Rehabilita-tion and Rheumatology, in Amsterdam (n0151), were invit-ed to additionally participate in the present study. Theadditional measurements needed (muscle strength and aperformance-based measure of activity limitations) wereintegrated in the existing measurement schedule forCHECK.

The CHECK study was approved by the medical ethicscommittees of all participating centers. The additional meas-urements necessary for the present study were approved bythe Medical Ethical Committee of the Slotervaart Hospitaland Reade. All participants gave their written informedconsent before entering the study.

Measures

Pain

Pain during activities was assessed with the pain subscaleof the Western Ontario and McMaster Universities Oste-oarthritis Index (WOMAC) [19]. This subscale consistsof five items, which assess the amount of knee painexperienced during activities the past 48 h. Items are

answered on a five-point scale. The subscale scores rangefrom 0 to 20, with higher scores indicating more pain.The WOMAC is widely used in clinical research and hasbeen shown to be reliable, valid, and responsive for usein OA patients [WOMAC pain subscale—Cronbach’s α00.88, intraclass correlation coefficient (ICC)00.77][19–21].

Negative Affect

Negative affect was assessed using the vitality and men-tal health subscale of the Short-Form 36 Health Survey(SF-36) [22]. The vitality subscale consists of four items,which assess feelings of energy and fatigue. Items areanswered on a six-point scale. The subscale scores arelinearly converted to a 0 to 100 scale, with higher scoresindicating more feelings of energy and less feelings offatigue. Background information on the SF-36 as well asstandard scoring algorithms are available elsewhere [23].The mental health subscale consists of five items, whichassess feelings of depression and nervousness. Items areanswered on a six-point scale. The subscale scores arelinearly converted to a 0 to 100 scale, with higher scoresindicating less feelings of depression and nervousness.The SF-36 has been shown to be a reliable and validmeasure of health-related quality of life in different patientgroups including OA (mean Cronbach’s α across sub-scales and samples with different chronic conditions00.84)[22, 24].

Avoidance of Activities

Avoidance of activities was assessed with the resting sub-scale of the Pain Coping Inventory (PCI) [25]. This subscaleconsists of five items, which assess the level to whichpatients avoid activities when experiencing pain. Items areanswered on a four-point scale, ranging from 1 (hardly ever)to 4 (very often) in terms of frequency with which thestrategy (e.g., stopping activities or confining oneself tosimple activities) is applied when dealing with pain. Thesubscale scores range from 5 to 20, with higher scoresindicating a more frequent use of avoidance of activitiesas a coping style. The PCI has been shown to be areliable and valid measure of pain coping in OA patients(Cronbach’s α>0.68) [4, 25, 26].

Muscle Strength

Muscle strength in Newton meter (Nm) of the quadricepsand hamstrings was assessed using an isokinetic dynamom-eter (EnKnee; Enraf-Nonius, Rotterdam, the Netherlands).Measurements were made in sitting position at an angularvelocity of 60° per second, which reflects a speed of

ann. behav. med. (2012) 44:33–42 35

movement usually applied during daily tasks, e.g., walking.These measurements have been shown to be reproducibleand valid [27, 28]. A single physical therapist assessed allpatients according to a standardized protocol. Followinginstruction, patients performed one test measurement. Aftera 30-s rest, patients performed three maximal quadricepsstrength measurements during knee extension and threemaximal hamstrings strength measurements during kneeflexion.

For analysis, the maximum voluntary contractionobtained from three measurements of the quadriceps andthree measurements of the hamstring muscles of the indexknee (most affected knee) were summed and divided by twoto obtain a measure of total muscle strength around theindex knee. The measure was corrected for weight bydividing it by the patient’s weight [29].

Most participants identified their most affected (index)knee in the clinical interview. For participants with bilateralsymptoms, we defined an index knee based on the followingdecision tree: (1) highest Kellgren and Lawrence score [30],(2) lowest degree of active knee flexion, (3) highest painduring active knee flexion, and (4) crepitus during kneeflexion. In participants for whom we could not define anindex knee based on these signs, we randomly assigned anindex knee.

Activity Limitations

Activity limitations were assessed both with a self-reportand a performance-based measure. Self-reported activitylimitations were assessed with the physical functioning sub-scale of the WOMAC: the WOMAC-PF (Cronbach’s α00.96, ICC00.92) [19, 20]. This subscale consists of 17 itemsthat are answered on a five-point scale, ranging from 0(none) to 4 (extreme) in terms of the degree of difficultiesone has in executing activities (e.g., walking or stair climb-ing). The subscale scores ranges from 0 to 68, with higherscores indicating more activity limitations.

Performance-based activity limitations were assessedwith a timed stair-climbing test. The stairs had 12 steps witha rise of 16 cm and a run of 30 cm. Participants began thetask by standing on a line that was 58 cm from the first step.Participants were instructed to climb the stairs step by stepas quickly as they felt safe and comfortable. They wereencouraged not to use the handrail, but were not prohibitedfrom doing so for safety. The task ended when participantsstood with both feet upstairs. The task was scored as thetotal time needed to climb the stairs. A longer time tocomplete the task indicates more activity limitations. Excel-lent test–retest reliability was reported for a comparablestair-climbing task in patients with knee OA [Pearson’scorrelation coefficient (r) of 3-month test–retest reliabilitydata00.87] [31].

Statistical Analysis

The data were explored using SPSS version 15.0 (SPSSInc., 2006). Prior to the analyses, we checked if the assump-tions for linear regression (e.g., no strong multicollinearity,homoscedasticity, linearity) were met [32].

The relationships as proposed by the avoidance model(Fig. 1) were modeled using Structural Equation Modeling(SEM) in Mplus version 6.12 [33]. SEM allows the simul-taneous modeling of several related regression relationships.Because all variables in the model were continuous, linearregression models were used. Regression coefficients wereestimated using robust maximum likelihood estimation(MLR).

In a structural equation model, a variable can be a depen-dent variable in one relationship and an independent vari-able in another. These variables are referred to as mediatingvariables [33]. In assessing mediation, it is important tomake a distinction between various associations and theircorresponding weights. The total association between anindependent variable and a dependent variable is composedof an indirect association between the independent variableand the dependent variable via the mediator and a directassociation between the independent variable and the de-pendent variable. For example, in Fig. 2 path a representsthe association between the independent variable pain andthe mediator variable avoidance, and path b represents theassociation between the mediator variable avoidance and thedependent variable muscle strength (indirect association).Path d represents the direct association between pain andmuscle strength. Avoidance is statistically shown to be amediator of the association between pain and musclestrength if the indirect association between pain and musclestrength (path a × path b) differs significantly from zero[34]. The rationale behind this method is that mediationdepends on the extent to which the independent variable isassociated with the mediator (path a) and the extent towhich the mediator is associated with the outcome variable(path b) [35]. If one of these associations (path a or path b)does not exist, the regression coefficient of the mediator(path a × path b) is not significantly different from zero:there is no mediation.

To obtain one outcome measure for activity limitations,the variables WOMAC-PF and timed stair-climbing testwere combined into one latent variable for activity limita-tions. Because of the high correlation between the subscalesfor vitality and mental health of the SF-36 (r00.74), thesevariables were combined into one observed variable fornegative affect by averaging the scores. To facilitate theinterpretation of the results, prior to analysis the negativeaffect score was multiplied by −1 so that a higher scoreindicated more feelings of fatigue, low vitality, depression,and nervousness. In this way, for all variables in the model a

36 ann. behav. med. (2012) 44:33–42

higher score indicated a higher degree of the variable ofinterest (i.e., a higher degree of pain, negative affect, avoid-ance, muscle strength, and activity limitations).

Moderation was tested using an interaction term be-tween pain and negative affect [15, 32]. Bootstrap meth-ods were used to estimate standard errors (SE) and 95%confidence intervals (95% CI) for both indirect (media-tion) and direct associations [34–36]. The bootstrap wasused because this method does not assume normality ofthe sampling distribution of the indirect (mediation) ef-fect [34, 36]. Bootstrapped standard errors were comput-ed using 1,000 draws.

Goodness of fit of the structural equation models wasassessed by the chi-square test of model fit (χ2, criterionp>0.05), the root mean square error of approximation(RMSEA, criterion<0.06), the comparative fit index(CFI, criterion>0.95), and the standardized root meansquare residual (SRMR, criterion<0.08) [37]. When thehypothesized model did not fit the data adequately, themodel was revised. The goodness of fit of the revisedmodel was tested and compared with the previous modelusing the Satorra–Bentler scaled chi-square differencetest (TRd) [38, 39]. Modification indices [33] were usedto guide the revisions which were primarily based onknowledge of direct associations and plausibility. Thecorrelation coefficient squared (R2) was calculated toexpress the amount of variability in avoidance, musclestrength, and activity limitations that was explained bythe other variables in the model [32].

Results

Study Population

Characteristics of the study population and mean scores onpain, negative affect (i.e., vitality and mental health), avoid-ance of activities, muscle strength, and activity limitationsare presented in Table 1.

Goodness of Fit of the Structural Equation ModelsUsed to Assess the Validity of the Avoidance Model

The model fit indices of the structural equation modelsused to test the hypotheses described in the “Introduc-tion” are given in Table 2. The hypothesized model(Fig. 1) did not adequately fit the data. Therefore,paths were added or removed stepwise to improve themodel fit. The steps made to revise the model arepresented in Table 2. From model 1a (Fig. 1), theinteraction between pain and negative affect was re-moved (model 1b), and paths were added between painand activity limitations (model 1c), avoidance and ac-tivity limitations (model 1d), pain and muscle strength(model 1e), and negative affect and activity limitations(model 1f) (Table 2). The final revised model in whichboth indirect and direct associations between the vari-ables were included fit the data well according to thecriteria CFI >0.95 and SRMR <0.08 but not RMSEA.This final model is shown in Fig. 2.

i: 0.37 (<0.001)

a: b: c:

0.17 -0.27 -0.16

(0.03) (<0.001) (<0.001)

Painduring

activities

Avoidance of

activities

Muscle strength

Activity limitations

e: 0.20 (<0.001)

d: -0.25 (<0.001)

f: 0.71 (<0.001)

WOMAC-PF Timed stair-

climbing test

0.95 (<0.001) 0.58 (<0.001)

j: 0.14 (0.01)

Negative affect

Fig. 2 Final structural equation model describing the associationsbetween pain during activities, negative affect, avoidance of activities,muscle strength, and activity limitations (model 1f). WOMAC-PFPhysical Functioning scale of the Western Ontario and McMasterUniversities Osteoarthritis Index. Pain during activities was assessedwith the pain subscale of the WOMAC. Negative affect was computedby averaging the subscale scores for vitality and mental health of the

Short-Form 36 Health Survey (SF-36). The negative affect score wasmultiplied by −1 so that higher scores indicated a higher degree ofnegative affect. Avoidance of activities was assessed with the restingsubscale of the Pain Coping Inventory (PCI). Muscle strength wasassessed using an isokinetic dynamometer. The associations arepresented as standardized regression coefficients (p value)

ann. behav. med. (2012) 44:33–42 37

The Mediating Role of Avoidance of Activities: Hypothesis 1

The final structural equation model used to test the validityof the avoidance model (model 1f) is shown in Fig. 2.The indirect association between pain and musclestrength via avoidance (Fig. 2—path a×path b) wasmarginally statistically significant [standardized regressioncoefficient (β)0−0.05, p00.09]. Besides this trend towards

an indirect association, pain was directly associated withmuscle strength (β0−0.25, p0<0.001). Of the total associa-tion between pain and muscle strength, 15.3% was mediatedby avoidance.When negative affect was excluded frommodel1f, the indirect association between pain and muscle strengthvia avoidance was statistically significant (β0−0.08, p00.01).In the model without negative affect, 25.2% of the totalassociation between pain and muscle strength was mediatedby avoidance. The hypothesis that avoidance mediates therelationship between higher pain and lower muscle strengthwas confirmed in the analysis without negative affect.

The Mediating Role of Lower Muscle Strength: Hypothesis2

The hypothesis that avoidance of activities is associatedwith a higher degree of activity limitations via lower musclestrength (Fig. 2—path b×path c) was confirmed. The indi-rect association between avoidance and activity limitationsvia muscle strength was significant (β00.04, p00.01). Be-sides this indirect association, avoidance was directly asso-ciated with activity limitations (β00.20, p0<0.001). Of thetotal association between avoidance and activity limitations,18.1% was mediated by muscle strength.

The Moderating Role of Negative Affect: Hypothesis 3

Negative affect did not strengthen the association betweenpain and avoidance of activities. The interaction term betweenpain and negative affect was not significantly (p00.24)associated with avoidance of activities.

Therefore, the interaction between pain and negativeaffect was removed from the model (Table 2). In the revisedmodel, negative affect was hypothesized to lead to lowermuscle strength via the mediator avoidance of activities (i.e.,avoidance as a mediator between negative affect and lower

Table 2 Model fit indices for the structural equation models used to validate the avoidance model

Model χ2 dƒ p value TRd RMSEA CFI SRMR

Criteria >0.05 <0.06 >0.95 <0.08

1a (Fig. 1) 158.02 11 <0.001 0.298 0.460 0.224

1b: 1a− interaction pain×negative affect 202.79 8 <0.001 2.02 0.402 0.382 0.215

1c: 1b+pain→activity limitations 55.15 7 <0.001 182.33** 0.213 0.847 0.101

1d: 1c+avoidance→activity limitations 29.67 6 <0.001 24.61** 0.162 0.925 0.079

1e: 1d+pain→muscle strength 19.15 5 0.002 14.01** 0.137 0.955 0.043

1f: 1e+negative affect→activity limitations (final model; Fig. 2) 12.24 4 0.02 5.82* 0.117 0.974 0.040

Model 1a–1f refer to the steps taken to improve the model fit as described in the “Results” section

χ2 chi-square test of model fit, df degrees of freedom, TRd Satorra–Bentler scaled chi-square difference test, RMSEA root mean square error ofapproximation, CFI comparative fit index, SRMR standardized root mean square residual

*p<0.05, **p<0.01

Table 1 Characteristics of the study population (n0151)

Characteristic Value

Age (years), mean±SD 58.5±5.0

Female, n (%) 120 (79.5)

Body mass index (kg/m2), mean±SD 25.6±3.8

Knee symptoms, n (%)

Unilateral 47 (31.1)

Bilateral with index knee 93 (61.6)

Bilateral with equal symptoms 11 (7.3)

Duration of knee symptoms (years), median (IQR) 3.7 (2.8–5.0)

Clinical knee OA (ACR criteria), n (%) 118 (78.1)

KL grade ≥2, n (%) 29 (19.2)

WOMAC pain score (range 0–20), mean±SD 5.0±3.8

SF-36 vitality score (range 0–100), mean±SD 63.9±19.7

SF-36 mental health score (range 0–100), mean±SD 75.6±16.4

PCI resting score (range 5–20), mean±SD 9.3±2.6

Muscle strength (N/kg), mean±SD 1.0±0.3

WOMAC physical function score (range 0–68),mean±SD

15.5±12.1

Timed stair climbing test (s), mean±SD 5.2±2.0

IQR interquartile range, SD standard deviation, OA osteoarthritis, ACRAmerican College of Rheumatology, KL grade Kellgren and Lawrencegrade, WOMAC Western Ontario and McMaster Universities Osteoar-thritis Index, SF-36 Short-Form 36 Health Survey, PCI Pain CopingInventory

38 ann. behav. med. (2012) 44:33–42

muscle strength). The results showed that negative affectwas significantly associated with muscle strength via avoid-ance (Fig. 2—path i×path b, β0−0.10, p00.004). Negativeaffect was not directly associated with muscle strength: thedirect path between negative affect and muscle strength wasnot added to the final model. Thus, the association betweennegative affect and muscle strength was fully mediated byavoidance. The results confirm the hypothesis that avoid-ance mediates the relationship between negative affect andlower muscle strength.

Pain and negative affect explained 21.8% of variance inavoidance of activities. Pain, negative affect, and avoidanceexplained 17.8% of variance in muscle strength. Pain, neg-ative affect, avoidance, and muscle strength explained87.4% of variance in activity limitations (Fig. 2). The hy-pothesized indirect association between pain and activitylimitations via avoidance and muscle strength accountedfor 9.9% of the total association between pain and activitylimitations. The indirect association between negative affectand activity limitations via avoidance and muscle strengthaccounted for 38.2% of the total association between nega-tive affect and activity limitations.

Direct Associations Between Pain, Negative Affect,Avoidance, Muscle Strength, and Activity Limitations

To improve the fit of the hypothesized model (Fig. 1), directpaths between the variables were added (Fig. 2). The resultsshowed that besides the hypothesized indirect associations,most variables in the model were directly associated witheach other: (1) pain was positively associated with avoid-ance (path a), negatively associated with muscle strength(path d), and positively associated with activity limitations(path f); (2) negative affect was positively associated withavoidance (path i) and activity limitations (path j); (3)avoidance was negatively associated with muscle strength(path b), and positively associated with activity limitations(path e); and (4) muscle strength was negatively associatedwith activity limitations (path c). The Pearson correlationbetween pain and negative affect was 0.39 (p<0.001).

Discussion

In the present study, the validity of the avoidance model wasassessed in patients with early symptomatic knee OA. Weinvestigated the associations between pain, negative affect,avoidance of activities, muscle strength, and activity limi-tations using SEM. We hypothesized that (1) avoidance ofactivities plays a mediating role in the relationship betweenhigher pain and lower muscle strength: this hypothesis wasconfirmed. The indirect relationship between pain and mus-cle strength via avoidance was marginally significant; when

negative affect was excluded from the model the indirectassociation was significant. We hypothesized that (2) lowermuscle strength plays a mediating role in the relationshipbetween avoidance and activity limitations: this hypothesiswas confirmed, as the indirect relationship between avoid-ance and activity limitations via muscle strength was signif-icant. We hypothesized that (3) negative affect would play amoderating role in the avoidance model, by strengtheningthe tendency to avoid activities. This hypothesis was notconfirmed. Instead, our results suggest that negative affectleads to lower muscle strength, via avoidance of activities.In other words, avoidance is a mediator between negativeaffect and lower muscle strength. This—slightly revised—avoidance model (Fig. 3) seems to offer a valid explanationof the associations between pain, negative affect, avoidanceof activities, lower muscle strength, and activity limitations.

When we excluded negative affect from model 1f, theindirect association between pain and muscle strength viaavoidance changed from marginally statistically significant(p00.09) to statistically significant (p00.01). Apparently theassociation between pain and avoidance is partially explainedby negative affect. Several studies have shown that pain andnegative affect are associated [40–42], and also in the presentstudy these variables were moderately to highly correlated[32]. These findings indicate that although it is important toseparate pain and negative affect conceptually, empiricallythese concepts are not completely separable.

In the final structural equation model, direct associationsbetween pain, negative affect, avoidance, muscle strength,and activity limitations were added to improve the model fit(see Fig. 2). Prior research has shown that pain, negative affect,and avoidance of activities are also directly associated withactivity limitations in knee OA patients [2, 4, 12, 16, 43, 44].Originally, we did not include these direct associations in ouranalyses, as the present study was aimed at assessing themechanisms described in the avoidance model. These direct

pain during activities

lower muscle strength

activity limitations

negative affect

avoidance of activities

Fig. 3 Results-based model of the associations between pain, negativeaffect, avoidance of activities, lower muscle strength, and activitylimitations in knee osteoarthritis patients

ann. behav. med. (2012) 44:33–42 39

associations indicate the existence of other pathways betweenpain and activity limitations than those hypothesized in theavoidance model. Alternative pathways may involve poorvoluntary effort [45] or low self-efficacy beliefs [12].

In the final model (Fig. 2), the indirect association be-tween pain and activity limitations via avoidance and mus-cle strength accounted for 9.9% of the total associationbetween pain and activity limitations. The indirect associa-tion between negative affect and activity limitationsaccounted for 38.2% of the total association between nega-tive affect and activity limitations. Again, these percentagesindicate that avoidance is not the only mechanism explain-ing activity limitations in knee OA: the present study showsthat avoidance is one mechanism, among other mechanisms.

All individual parameter estimates of the paths in themodels were statistically significant (Fig. 2), and the modelhad an adequate model fit according to the criteria CFI >0.95and SRMR <0.08 (Table 2). The final model did notmeet the fit criteria for the χ2 test and RMSEA. However,these fit indices are very sensitive to sample size: the χ2 tendsto over-reject true populationmodels in large sample sizes andthe RMSEA tends to over-reject true population models insample sizes ≤250 [37]. To minimize the type I and type IIerror rate, in studies with a sample size ≤250 Hu and Bentler[37] recommended to use concurrent values of CFI >0.95 andSRMR <0.08 as criteria for adequate fit. The final model metthese criteria.

Our confirmation of hypothesis 2 (i.e., mediation bymuscle strength) is in line with the results of a previousstudy in patients with established knee OA [12]. The addedvalue of the present study is that we also examined themediating role of avoidance of activities (hypothesis 1)and the role of negative affect in the avoidance model(hypothesis 3). In addition, the present study was directedat patients with early symptomatic knee OA: it is assumedthat the relationships described in the avoidance model areinitiated in early stage OA because in this stage patientsexperience activity-related pain for the first time, leading tothe described process of adaptation.

The relationships proposed by the avoidance model wereanalyzed using structural equation models in which the twomeasures of activity limitations were combined into a latentvariable, while pain, negative affect, avoidance, and musclestrength were analyzed as observed variables. These modelswere used instead of full latent variable models because thiswas more appropriate for the present study’s sample size. Astrength of the present study is that we analyzed the associ-ation between avoidance and activity limitations using alatent variable constructed of both a self-report and aperformance-based measure of activity limitations. There-fore, the associations found between avoidance and activitylimitations cannot be explained by a correlation betweenself-report measures.

Model 1f (Figs. 2 and 3), in which negative affect washypothesized to lead to lower muscle strength via the medi-ator avoidance of activities, was formulated based on theresults of the analyses instead of a stated hypothesis. Theresults of the final structural equation model are interesting:they seem to imply that negative affect strengthens pain-related avoidance and thereby leads to lower musclestrength. However, because the final model was not formu-lated based on a stated hypothesis, the findings are in needof replication. In addition, we used a cross-sectional designto study a model describing longitudinal relationships. Thisis a limitation of the present study. The next step is tovalidate the avoidance model in a longitudinal study, whichwill provide stronger evidence for causality.

Avoidance of activities was assessed with the restingsubscale of the Pain Coping Inventory, which is a reliableand valid [4, 25, 26], but self-report measure of the level towhich patients avoid activities when experiencing pain.Because self-report measures are susceptible to responseand recall bias, in further research it is recommended toadditionally use an accelerometer to measure avoidance ofactivities. Negative affect was assessed using the vitality andmental health subscales of the SF-36, which is a genericquestionnaire designed to measure health-related quality oflife. Negative affect might be better assessed with a morespecific questionnaire designed to measure aversive moodstates; unfortunately, such a questionnaire is not included inthe CHECK study.

In summary, the avoidance model seems to offer a validexplanation of the associations between pain, negative af-fect, avoidance of activities, muscle strength, and activitylimitations in patients with early symptomatic knee OA.Pain and negative affect were positively associated withavoidance, and avoidance was negatively associated withmuscle strength (mediation by avoidance). Avoidance ofactivities was negatively associated with muscle strength,and muscle strength was negatively associated with activitylimitations (mediation by lower muscle strength). Therewere also direct associations between pain, negative affect,avoidance, muscle strength, and activity limitations. Theseresults are summarized in Fig. 3. Further confirmation in alongitudinal study is required.

Acknowledgments CHECK is funded by the Dutch Arthritis Asso-ciation on the lead of a steering committee comprising 16 memberswith expertise in different fields of OA chaired by Prof. J.W.J. Bijlsmaand coordinated by J. Wesseling, MSc. Involved are Erasmus MedicalCenter Rotterdam; Kennemer Gasthuis Haarlem; Leiden UniversityMedical Center; Maastricht University Medical Center; MartiniHospital Groningen/Allied Health Care Center for Rheumatologyand Rehabilitation Groningen; Medical Spectrum Twente Enschede/Ziekenhuisgroep Twente; Reade, Center for Rehabilitation andRheumatology (formerly Jan van Breemen Institute)/VU MedicalCenter Amsterdam; St. Maartenskliniek Nijmegen; University MedicalCenter Utrecht; and Wilhelmina Hospital Assen.

40 ann. behav. med. (2012) 44:33–42

Conflict of Interest Statement The authors have no conflict ofinterest to disclose.

References

1. Brooks PM. Impact of osteoarthritis on individuals and society:How much disability? Social consequences and health economicimplications. Curr Opin Rheumatol 2002; 14(5):573-577.

2. Dekker J, van Dijk GM, Veenhof C. Risk factors for functionaldecline in osteoarthritis of the hip or knee. Curr Opin Rheumatol2009; 21(5):520-524.

3. Hopman-Rock M, Kraaimaat FW, Odding E, Bijlsma JW. Copingwith pain in the hip or knee in relation to physical disability incommunity-living elderly people. Arthritis Care Res 1998; 11(4):243-252.

4. Steultjens MP, Dekker J, Bijlsma JW. Coping, pain, and disabilityin osteoarthritis: A longitudinal study. J Rheumatol 2001; 28(5):1068-1072.

5. van Baar ME, Dekker J, Lemmens JA, Oostendorp RA, BijlsmaJW. Pain and disability in patients with osteoarthritis of hip orknee: The relationship with articular, kinesiological, and psycho-logical characteristics. J Rheumatol 1998; 25(1):125-133.

6. Linton SJ. New avenues for the prevention of chronic musculo-skeletal pain and disability. Amsterdam: Elsevier Health Sciences,2002.

7. Turk DC, Flor H. Etiological theories and treatments for chronicback pain. II. Psychological models and interventions. Pain 1984;19(3):209-233.

8. Philips HC. Avoidance behaviour and its role in sustaining chronicpain. Behav Res Ther 1987; 25(4):273-279.

9. Keefe FJ, Gil KM. Behavioral concepts in the analysis of chronicpain syndromes. J Consult Clin Psychol 1986; 54(6):776-783.

10. Dolce JJ, Raczynski JM. Neuromuscular activity and electro-myography in painful backs: Psychological and biomechanicalmodels in assessment and treatment. Psychol Bull 1985; 97(3):502-520.

11. Dekker J, Tola P, Aufdemkampe G, Winckers M. Negative affect,pain and disability in osteoarthritis patients: The mediating role ofmuscle weakness. Behav Res Ther 1993; 31(2):203-206.

12. Steultjens MP, Dekker J, Bijlsma JW. Avoidance of activityand disability in patients with osteoarthritis of the knee: Themediating role of muscle strength. Arthritis Rheum 2002; 46(7):1784-1788.

13. Bandura A. Self-efficacy: Toward a unifying theory of behavioralchange. Psychol Rev 1977; 84(2):191-215.

14. Baron RM, Kenny DA. The moderator-mediator variable distinc-tion in social psychological research: Conceptual, strategic, andstatistical considerations. J Pers Soc Psychol 1986; 51(6):1173-1182.

15. Rose BM, Holmbeck GN, Coakley RM, Franks EA. Mediator andmoderator effects in developmental and behavioral pediatric re-search. J Dev Behav Pediatr 2004; 25(1):58-67.

16. Rosqvist E, Heikkinen E, Lyyra TM, Hirvensalo M, Kallinen M,Leinonen R et al. Factors affecting the increased risk of physicalinactivity among older people with depressive symptoms. Scand JMed Sci Sports 2009; 19(3):398-405.

17. Vlaeyen JW, Linton SJ. Fear-avoidance and its consequences inchronic musculoskeletal pain: A state of the art. Pain 2000; 85(3):317-332.

18. Wesseling J, Dekker J, van den Berg WB, Bierma-Zeinstra SM,Boers M, Cats HA et al. CHECK: Cohort Hip & Cohort Knee;similarities and differences with the Osteoarthritis Initiative. AnnRheum Dis 2009; 68:1413-1419.

19. Bellamy N, Buchanan WW, Goldsmith CH, Campbell J, Stitt LW.Validation study of WOMAC: A health status instrument formeasuring clinically important patient relevant outcomes to anti-rheumatic drug therapy in patients with osteoarthritis of the hip orknee. J Rheumatol 1988; 15(12):1833-1840.

20. Roorda LD, Jones CA, Waltz M, Lankhorst GJ, Bouter LM, vander Eijken JW et al. Satisfactory cross cultural equivalence of theDutch WOMAC in patients with hip osteoarthritis waiting forarthroplasty. Ann Rheum Dis 2004; 63(1):36-42.

21. Veenhof C, Bijlsma JW, van den Ende CH, van Dijk GM, PistersMF, Dekker J. Psychometric evaluation of osteoarthritis question-naires: A systematic review of the literature. Arthritis Rheum2006; 55(3):480-492.

22. Ware JE, Jr., Sherbourne CD. The MOS 36-item short-form healthsurvey (SF-36). I. Conceptual framework and item selection. MedCare 1992; 30(6):473-483.

23. Ware JE, Snow KK, Kosinski M, Gandek B. SF-36 Health SurveyManual and Interpretation Guide. Boston, MA: New EnglandMedical Center, The Health Institute; 1993.

24. Aaronson NK, Muller M, Cohen PD, Essink-Bot ML, Fekkes M,Sanderman R et al. Translation, validation, and norming of theDutch language version of the SF-36 Health Survey in communityand chronic disease populations. J Clin Epidemiol 1998; 51(11):1055-1068.

25. Kraaimaat FW, Evers AW. Pain-coping strategies in chronic painpatients: Psychometric characteristics of the pain-coping inventory(PCI). Int J Behav Med 2003; 10(4):343-363.

26. Perrot S, Poiraudeau S, Kabir M, Bertin P, Sichere P, Serrie A et al.Active or passive pain coping strategies in hip and knee osteoar-thritis? Results of a national survey of 4,719 patients in a primarycare setting. Arthritis Rheum 2008; 59(11):1555-1562.

27. Tan J, Balci N, Sepici V, Gener FA. Isokinetic and isometricstrength in osteoarthrosis of the knee. A comparative study withhealthy women. Am J Phys Med Rehabil 1995; 74(5):364-369.

28. Madsen OR, Bliddal H, Egsmose C, Sylvest J. Isometric andisokinetic quadriceps strength in gonarthrosis; inter-relationsbetween quadriceps strength, walking ability, radiology, sub-chondral bone density and pain. Clin Rheumatol 1995; 14(3):308-314.

29. Steultjens MP, Dekker J, van Baar ME, Oostendorp RA, BijlsmaJW. Muscle strength, pain and disability in patients with osteoar-thritis. Clin Rehabil 2001; 15(3):331-341.

30. Kellgren JH, Lawrence JS. Radiological assessment of osteo-arthrosis. Ann Rheum Dis 1957; 16(4):494-502.

31. Rejeski WJ, Ettinger WH, Jr., Schumaker S, James P, Burns R,Elam JT. Assessing performance-related disability in patients withknee osteoarthritis. Osteoarthritis Cartilage 1995; 3(3):157-167.

32. Field A. Discovering Statistics Using SPSS. 2nd ed. London:SAGE Publications Ltd, 2005.

33. Muthén LK, Muthén BO. Mplus User’s Guide. Sixth ed. LosAngeles, CA: Muthén & Muthén, 2010.

34. Shrout PE, Bolger N. Mediation in experimental and nonexperi-mental studies: New procedures and recommendations. PsycholMethods 2002; 7(4):422-445.

35. Mackinnon DP, Warsi G, Dwyer JH. A Simulation Study ofMediated Effect Measures. Multivariate Behav Res 1995; 30(1):41.

36. Preacher KJ, Hayes AF. Asymptotic and resampling strategies forassessing and comparing indirect effects in multiple mediatormodels. Behav Res Methods 2008; 40(3):879-891.

37. Hu L, Bentler PM. Cutoff criteria for fit indexes in covariancestructure analysis: Conventional criteria versus new alternatives.Struct Equ Modeling 1999; 6(1):1-55.

38. Satorra A, Bentler PM. A scaled difference chi-square test statisticfor moment structure analysis. Psychometrika 2001; 66(4):507-514.

ann. behav. med. (2012) 44:33–42 41

39. MPlus website. Chi-square difference test for MLM and MLR.Available at http://www.statmodel.com/chidiff.shtml. Accessibilityverified January 25, 2012.

40. Somers TJ, Keefe FJ, Godiwala N, Hoyler GH. Psychosocialfactors and the pain experience of osteoarthritis patients: Newfindings and new directions. Curr Opin Rheumatol 2009; 21(5):501-506.

41. Axford J, Heron C, Ross F, Victor CR. Management of kneeosteoarthritis in primary care: Pain and depression are the majorobstacles. J Psychosom Res 2008; 64(5):461-467.

42. Knoop J, van der Leeden M, Thorstensson CA, Roorda LD, LemsWF, Knol DL et al. Identification of phenotypes with different

clinical outcomes in knee osteoarthritis: Data from the OsteoarthritisInitiative. Arthritis Care Res (Hoboken) 2011; 63(11):1535-1542.

43. Creamer P, Lethbridge-Cejku M, Hochberg MC. Factors associatedwith functional impairment in symptomatic knee osteoarthritis.Rheumatology (Oxford) 2000; 39(5):490-496.

44. van der Esch M, Steultjens M, Harlaar J, Knol D, Lems W, DekkerJ. Joint proprioception, muscle strength, and functional ability inpatients with osteoarthritis of the knee. Arthritis Rheum 2007; 57(5):787-793.

45. O’Reilly SC, Jones A, Muir KR, Doherty M. Quadriceps weaknessin knee osteoarthritis: The effect on pain and disability. AnnRheum Dis 1998; 57(10):588-594.

42 ann. behav. med. (2012) 44:33–42