Available online at www.sciencedirect.com

www.EuropeanJournalPain.com

European Journal of Pain 12 (2008) 314–320

General practitioner sickness absence certification for lowback pain is not directly associated with beliefs about back pain

Paul J. Watson a,*, Julia Bowey a,b, Gari Purcell-Jones b, Tom Gales c

a Department of Health Sciences, University of Leicester, Gwendolen Road, Leicester LE5 4PW, UKb Department of Anaesthesia and Pain Management, St. Helier Hospital, Jersey, UK

c States of Jersey Department of Employment and Social Security, St. Helier, Jersey, UK

Received 16 March 2007; received in revised form 23 May 2007; accepted 14 June 2007Available online 30 July 2007

Abstract

Recent research has demonstrated a relationship between healthcare practitioner beliefs about low back pain and recommenda-tions about activity, work restrictions and work absence. None of the research to date has looked at the relationship between prac-titioner beliefs and actual behaviour. This study investigated the internal consistency of the pain attitudes and beliefs scale (PABS)and if general practitioner (GP) beliefs about back pain were more predictive of sickness certification for non-specific low back pain(NSLBP) than a general predisposition to sick certify patients with other non-specific conditions (common mental illness and non-specific upper respiratory disorders). Ninty-four eligible general practitioners were invited to participate in the study and data from83 (88.3%) were included in the full analysis. Evaluation of the internal consistency of the PABS found the biomedical subscale wasgood (a = 0.781) but the psychosocial subscale was poor (a = 0.396) after item elimination both subscales improved; biomedicala = 0.790, psychosocial a = 0.602. GP sickness certification behaviour for 1 year was gathered from the Department of Employmentand Social Security database. Multiple regression analysis demonstrated that neither the biomedical nor the psychosocial subscale ofthe PABS predicted the number of sickness certificates issued even after controlling for the time employed as a GP, number of hoursworked per week and the number of NSLBP patients seen. Certification for other conditions was predictive of NSLBP certificatesissued. These results demonstrate that sickness absences certification for NSLBP is predicted by sickness certification behaviour ingeneral and not by scores on the PABS.� 2007 European Federation of Chapters of the International Association for the Study of Pain. Published by Elsevier Ltd. Allrights reserved.

Keywords: Low back pain; Sickness certification; Work; Primary care

1. Introduction

The management of non-specific low back pain(NSLBP) remains a major consumer of health careresources in developed countries. However the costsof healthcare are dwarfed by the costs associated with

1090-3801/$32 � 2007 European Federation of Chapters of the International

reserved.

doi:10.1016/j.ejpain.2007.06.002

* Corresponding author. Tel.: +44 116 2584613; fax: +44 1162588317.

E-mail address: pjw25@le.ac.uk (P.J. Watson).

work absence, wage replacement and other social ben-efits associated with inability to work (Maniadakis andGray, 2000). The evidence for the psychosocial natureof predictors of work absence from this condition isconsiderable (Main et al., 2005) and recent attentionhas focused on the influence of fear avoidance beliefs(Vlaeyen and Linton, 2006). The fear avoidance modelof non-specific low back pain disability suggests thatpeople who perceive that back pain is harmful andthat physical activity (including work) is to be avoided

Association for the Study of Pain. Published by Elsevier Ltd. All rights

P.J. Watson et al. / European Journal of Pain 12 (2008) 314–320 315

during episodes of back pain are more likely tobecome disabled or absent from work than thosewho do not (Vlaeyen and Linton, 2000). Once a backpain episode starts patients have expectations aboutthe type of treatment which is appropriate and thedevelop beliefs regarding the possible outcomes andprognosis from back pain (Main et al., 2005). Further-more it has been suggested that patients learn thesebeliefs and cognitions through a wide variety of influ-ences including society, family and information fromor attitudes of healthcare practitioners (Vlaeyen andLinton, 2006).

Reliable questionnaires such as the pain attitudesand beliefs scale (PABS) have been developed to assessthe beliefs of healthcare practitioners and these havebeen used in studies of Physical and Manual Thera-pists and general practitioners (Houben et al.,2005a,b; Jellema et al., 2005). The results suggest thathealth care providers’ (HCPs) attitudes and beliefsabout back pain influence their recommendationsabout activity and work. These previous studies havedemonstrated that those HCPs with a more biomedicalfocus on back pain (a high score in the biomedicalsubscale of the PABS) are more likely to recommendwork restrictions or work absence than those who havea psychosocial focus to management (a high score onthe psychosocial subscale) who may see working withlow back pain as not harmful. All of this work hasbeen conducted using questionnaires and case exam-ples. The relative orientation of participants has beenassessed (biomedical or psychosocial) and the scoresfrom these are matched to the recommendation madeon a sample patient in a vignette, photograph, per-sonal appraisal of adherence to guidelines, or videorating (Coudeyre et al., 2006; Houben et al.,2005a,b). Although these studies have consistentlydemonstrated a correlation between beliefs (question-naire scores) and explicit behaviour (recommendationson the patient example) it has not been tested whetherthis is consistent with actual behaviour in the clinicalsetting or if sickness certification behaviour is specificto beliefs about a specific condition. Alternatively, itmay be a more generalised phenomenon, in that someHCP are more likely to sanction work absence regard-less of the condition. Furthermore, large variations inthe issuing of sickness certificates is already recognised(Tellnes et al., 1990).

If HCP certification is clearly explained by beliefsabout back pain, then if a reduction in certification forback pain is desirable, it would require a campaigndirected at changing these beliefs. If certification forback pain is better explained by the general practi-tioner’s overall propensity for sanctioning work absencethen a campaign directed only at back pain beliefs mightbe less effective than targeting the factors which drivesickness certification in general.

This study investigated:

1. The reliability of the PABS in a GP population.2. The validity of the PABS identifying GP sickness

absence recommendations.3. If certification for NSLBP was more closely associ-

ated with the GPs beliefs about back pain or to theirgeneral propensity to issue sickness certification forother common non-specific conditions.

2. Method

2.1. Participants

All 99 general practitioners registered, practicing andresident in the island of Jersey during 2006 were invitedto participate in the study, five were ineligible for thestudy; two were no longer resident, one practiced onlyalternative medicine, one was a dermatology specialistonly, and one had failed to re-register in 2006. Of the94 eligible 85 (90.4%) agreed to participate.

All participants were sent an adapted version of thePain Attitudes and Beliefs Scale in April 2006. Thiswas adapted from the versions by Ostelo et al. (2003)and developed by Houben et al. (2005b) and includedall items previously validated. The PABS consists oftwo subscales: biomedical orientation and psychosocialorientation. The biomedical subscale consists of 14items, we excluded one item which was specific to phys-iotherapy (it is the task of the physiotherapist to remove

the cause of back pain). The GPs were invited to com-plete the questionnaire and provide demographic detailsof gender, years in general practice, number of hoursworked weekly and number of patients seen per month.Failure to return the questionnaire within 2 weeks elic-ited a telephone reminder followed by a final postaland telephone reminder two weeks later if required.

In the United Kingdom, including Jersey, sicknesscertification is provided through the general practi-tioner. In Jersey, all absences of greater than 2 daysrequire certification from a GP in order for paymentof sickness benefit. All sickness certificates are receivedby the States of Jersey Employment and Social SecurityDepartment and entered on a database. Each certificatecontains the diagnosis, name of the GP and the patient’sunique social security number. Previous estimates oferror for data entry place this below 2% (Watsonet al., 1998). Data was gathered on all short term inca-pacity allowance (STIA) sickness certification for eachGP in the study for the period January 1st 2005 toDecember 31st 2005. Sickness certification in this con-text referred to an inability to attend or be availablefor work for the period defined by the generalpractitioner.

316 P.J. Watson et al. / European Journal of Pain 12 (2008) 314–320

2.2. Diagnostic groups

Three groups of conditions were investigated andwere defined as follows:

2.2.1. Non-specific low back painThis group included all those diagnoses labelled with

non-specific low back pain attributions including, backpain, acute back pain, lumbago, back strain. Diagnosesof sciatica, intervertebral disc prolapse, back surgeryand inflammatory disorders were excluded.

2.2.2. Common mental illness

Conditions grouped together as common mental illi-ness were depression, anxiety, panic disorders and stress.

Excluded from this category were bi-polar disorders,substance addiction, eating disorders and hospitalisa-tion for any mental illness.

2.2.3. Respiratory disorders

All minor upper respiratory tract disorders, e.g.,influenza, common cold.

Where the classification of the condition was notclear the researcher (JB) cross referenced with the GPdescription entered from the original certificate to clarifythe actual diagnosis. If following cross referencing thereason for certification was still unclear this recordwas removed from the analysis.

Ethical approval to conduct the study was given bythe States of Jersey Health and Social Services Depart-ment Local Research Ethical Committee and the per-mission of the Employment and Social SecurityDepartment.

2.3. Statistical analysis

SPSS version 14 was used to conduct the analysis.Data were examined for normal distribution using aKolmolgorov–Smirnov test. Zero-order correlationswere performed to investigate the relationship betweenthe two subscales on the PABS and the number of cer-tificates issued.

2.3.1. Item means (response trends)

The mean score obtained on an item should be closeto the centre of the possible range of scores (DeVellis,1991). Items with a mean score near to the extreme ofa possible response range will have low variances (DeV-ellis, 1991). Therefore, standardised values of skewnesswere computed for each PABS item. Items with a skew-ness value greater than 1.96 have a response trend thatdeviates from a pattern of normal distribution (Field,2000) and should be excluded from an instrument.

2.3.2. Internal consistency of the PABS

A measure that assesses a particular constructshould comprise items that are homogeneous. That

is, all the items should assess different facets of thesame construct. Each item of a scale should be moder-ately correlated with each other and each item shouldcorrelate with the total score. Any two items whichcorrelate above 0.70 are likely to be assessing the sameaspect of the construct and one should be consideredredundant.

Cronbach’s a (Cronbach, 1951) was calculated toassess internal consistency. It has been suggested thatalpha levels should be above 0.60 and preferably above0.70 but probably not higher than 0.90 (Nunnally andBernstein, 1994). Item–total correlations were calcu-lated by correlating the score of each item with thetotal score omitting the score of each individual itemin turn. Items with item–total correlations less than0.20 are likely to be assessing a different construct fromthe other items on that measure (Streiner and Norman,1995).

2.3.3. Multiple regression analysis

A stepwise hierarchical multiple regression analysiswas used with the number of sickness certificates issuesas the dependent variable. To control for the generalpractitioner variables, the number of patients seen, yearsas a GP and hours worked were all entered on the firststep. The scores on the PABS subscales were enteredon the second step, the number of certificates issuedfor the other two categories of conditions was enteredon the third step.

The independent variables used in each of theregression analyses had Variance Inflation Factors thatwere considerably less than 10 (Myers, 1990) and toler-ance levels that were all notably higher than 0.2(Menard, 1995), indicating that the data were notaffected by multicollinearity. One case had a Cooks dis-tance of greater than 2 and was excluded from theanalysis. Following this none of the cases in the analy-ses exerted an undue influence on the final models,based on Mahalanobis’ distances (P < 0.001) (Tabach-nick and Fidell, 2001), Cook’s distances (less than 1)and leverage values (less than three times the averagevalue) (Stevens, 1992).

3. Results

3.1. General practitioners

Of the 85 GPs who agreed to participate 1 wasexcluded from the study as they had only recentlybecome a GP and had no certification history for2005, 1 was excluded later following the results of themultiple regression (described above).

The analysis of the PABS is based on responses from83 GPs. The characteristics of the GPs are given inTable 1.

Table 1GP characteristics

Mean (SD)

� Gender 74% males� Hours worked 39.5 (15.5)� Years as GP 17.8 (9.2)� LBP patients seen/month 15.3 (12.3)� Biomedical subscale 36.4 (7.5)� Psychosocial subscale 20.8 (2.9)

P.J. Watson et al. / European Journal of Pain 12 (2008) 314–320 317

3.2. The pain and attitudes beliefs scale

The PABS as developed by Houben et al. had an ini-tial Cronbach’s alpha of 0.781 for the biomedical sub-scale and 0.396 for the psychosocial subscale.

Following item reduction one item on the biomedicalsubscale was excluded as not contributing to the totalscore (if therapy does not result in a reduction in backpain, there is a high risk of severe restrictions). Thisincreased the alpha to 0.790. Inter-item correlation didnot reveal any redundant items (high inter-itemcorrelation).

Of the 13 items included in the psychosocial eightwere excluded, three because of skewed response distri-bution and five did not correlate to the total score. Onlyfive items remained with a Cronbach’s alpha of 0.602which is at the lower limit of acceptability. All remainingitems correlated moderately by the aforementioned cri-teria and there were no redundant items. These finalitems are given in Table 2.

The two variables were negatively correlated but themoderate correlation r = �0.474 (P < 0.0001) demon-

Table 2Final items in the PABS psychosocial scale

�Mental stress can cause back pain even in the absence of tissuedamage�The cause of back pain is unknown�Functional limitations associated with back pain are the result of

psychosocial factors�There is no effective treatment to eliminate back pain�Learning to cope with stress promotes recovery from back pain

Table 3Zero-order correlations (n = 83)

Years as a GP Hours workedper week

Nupat

Hours worked 0.326**

Number of LBP patients per month 0.034 0.321**

Respiratory certificates 0.223* 0.424** .CMI certificates 0.244* 0.495** 0.NSLBP certificates 0.273* 0.303** 0.PABS psychosocial subscale �0.224* �0.013 0.PABS biomedical subscale 0.278* �0.078 �0.

NSLBP, non-specific low back pain; CMI, common mental illness.* P < 0.01.

** P < 0.001.

strated that these two subscales are not entirelyindependent.

3.3. Number of sick certificates issued

During 2005 the GPs in the study issued a combinednumber of 4280 certificates for non-specific low backpain, 7025 for common mental illness and 7891 fornon-specific respiratory conditions.

3.4. Relationship of variables of interest

Zero-order correlations between the GP variablesand the certification behaviour demonstrated signifi-cant correlations between the hours worked, years asa GP, and the number of certificates issued, the longera person had been a GP the more certificates forNSLBP they had issued, likewise the more hours theGP worked the greater number of certificates issued.There were significant correlations between certificationrates for all three groups of conditions. There was norelationship between the number of NSLBP patientsseen per month and the number of certificates issuedfor NSLBP (see Table 3).

The results of the zero-order correlations includingthe PABS subscales demonstrated that GPs who havebeen working longer were more likely to score highlyon the biomedical subscale. The opposite relationshipwas demonstrated for the psychosocial subscale, viz.,short duration as a GP gave a higher score. There wasno relationship between the biomedical or the psychoso-cial score and the number of certificates issued forNSLBP.

3.4.1. Multiple regression analysis

Stepwise multiple regression analysis demonstratedGP variables accounted for 19% of the variables butonly the number of hours worked by the GP was signif-icant (b = 0.321; t = 2.83; P < 0.01). It also confirmedthat the PABS subscales added no significant explana-tion of the number of certificates for NSLBP by GPs

mber of LBPients per month

Respiratorycertificates

CMI certificates NSLBP certificates

192179 0.469**

112 0.434** 0.699**

054 0.076 0.060 �0.048150 0.146 �0.061 0.067

Table 4Hierarchical multiple regression predicting number of certificates issued by GPs for NSLBP

Independent variables Total R2 R2 change F change ba t-Value

Step 1 0.19 0.19 6.10**

Hours worked/week 0.321 2.83*

Years as GP 0.134 1.24NSLBP patients/month 0.117 1.09

Step 2 0.23 0.04 2.26Psychosocial subscale 0.204 1.78Biomedical subscale 0.220 1.83

Step 3 0.54 0.31 24.79**

Respiratory certificates 0.366 3.75**

CMI certificates 0.426 4.23**

NSLBP, non-specific low back pain; CMI, common mental illness.a Standardised regression coefficient.* P 6 0.01.

** P 6 0.001.

318 P.J. Watson et al. / European Journal of Pain 12 (2008) 314–320

after entering the GP descriptors on the first step. Thenumber of certificates for CMI and respiratory condi-tions in the third step both added significantly to theexplanation of variance and the final model explainednearly 50% of the variance for the number of certificatesissued (see Table 4).

4. Discussion

The PABS biomedical subscale was demonstrated tobe robust with good internal consistency. The removalof one item which did not correlate with the total scoreonly improved the alpha marginally. The psychosocialsubscale did not initially reach the minimum acceptablelevel for internal consistency but after item removal andimprovement it just reached this level (0.60) (Nunnallyand Bernstein, 1994). Previous development of thisquestionnaire was done with Physical Therapists in theNetherlands (Ostelo et al., 2003) and the results in thisstudy demonstrate the caution which must be exercisedwhen using questionnaires with different professionalgroups. The psychosocial subscale of the original ques-tionnaire cannot be recommended for use in generalpractitioners, and despite the improvements made in thisstudy it still requires further development. We were notable to directly compare the scores on each scale in thisstudy with previous work because of the difference in thenumber of items per scale.

One previous study (Jellema et al., 2005) used thepsychosocial subscale to detect changes in the orienta-tion of GPs and the success of training in a more psy-chosocial approach to NSLBP. Although training didlead to a difference in patient reported GP behaviour,one must be concerned about the accuracy of the psy-chosocial subscale as a measurement of change inattitude.

GPs who had been in practice for a long-time weremore likely to have a biomedical view of NSLBP; those

with a shorter duration as a GP were more likely toreport a psychosocial orientation. Whether this is theresults of a more psychosocial approach to GP trainingin recent years we cannot tell.

The lack of prediction for either the biomedical orpsychosocial subscales in the issuing of sickness certifi-cates even after controlling for time as a GP, NSLBPhistory and hours worked is surprising in view of previ-ous work. This calls into question the validity of thePABS where recommendations about work absenceare concerned. It also demonstrates how complex thedecision to issue a sickness certificate is. Usually, GPshave known their patients for many years and have builtup a close relationship with them. In previous studiesGPs were found to see the maintenance of their relation-ship with their patient as more important than challeng-ing cognitions about illness and work (Campbell andOgden, 2006; Chew-Graham and May, 1999, 2000; Hus-sey et al., 2004). Many were willing to relinquish the roleas gate-keeper to work absence as they felt it threatenedtheir relationship with the patient (Hussey et al., 2004).Their relationship with the patient is continuous, unlikeother health professionals who can put time limitationson treatment or indicate that their treatment is notappropriate.

In a series of studies of GP management of chroniclow back pain some GPs reported that they would issuea sickness certificate because the patient would only goand seek one from another GP if they did not. TheGPs recognised that they colluded with the patient insanctioning work absence, finding it difficult to chal-lenge the patient’s attributions or modify their (thepatients) explanatory models of pain (Chew-Grahamand May, 1999; Hussey et al., 2004). In this respectGPs with a psychosocial orientation might collude inthe patient’s biomedical view of their problem to avoidconfrontation and maintain a good relationship. TheGP therefore considers more information in coming toa decision than simply the signs and symptoms of the

P.J. Watson et al. / European Journal of Pain 12 (2008) 314–320 319

current condition or their own perspective. They usuallyhave a detailed knowledge of the patient’s social back-ground and previous work history. One qualitativestudy (Campbell and Ogden, 2006) demonstrated thatGPs are more likely to feel sympathy towards a patientwho has family problems and are less likely to see themas work shy irrespective of the presenting condition.They were also more likely to have sympathy and morelikely to issue a sick certificate if the person had psycho-logical problems as a primary or secondary condition. Inthis scenario biomedically orientated GPs might beswayed by the psychosocial presentation of the patientrather than the examination findings.

The lack of association between the PABS and certi-fication behaviour might be explained by the differencebetween the implicit and explicit behaviour of the GPs.Experimental studies have demonstrated a lack of rela-tionship between the reasoned explicit attitude, such asassessed by questionnaire, and implicit or ‘‘automatic’’attitudes (Houben et al., 2005a). Although in Houbensstudy (Houben et al., 2005a) there was some relationshipbetween implicit attitudes to back pain and recommen-dations about treatment via video these were notentirely consistent and the authors recognised that thismight not reflect what happens in practice. Although itis proposed that implicit attitudes are a better reflectionof spontaneous behaviour and explicit are more predic-tive of controllable and reasoned behaviour (Greenwaldand Banaji, 1995) this has not been tested in clinicalpractice during a brief (in Jersey typically 10 min)consultation.

The final step in the analysis appears to demonstratethat some GPs have a lower threshold for issuing sick-ness certificates than others and this is reflected acrossthe board in the two other illness categories investigated.This demonstrates that it is potentially the GPs percep-tion of his/her role in sickness certification versus main-taining people at work per se rather than disease specificattributions about work that influences certification. IfGPs have a low threshold for issuing sickness certificatespatients who themselves are more likely to absent fromwork might gravitate to such GPs in a system whichallows them to choose their own GP. However, we mustbe careful in this interpretation, those GPs with high cer-tification rates may disproportionately serve patientsfrom certain social and cultural groups where workabsence is know to be higher (Main et al., 2005).

One of the difficulties presenting the general practi-tioner in sanctioning or not sanctioning low back painabsence is the lack of identifiable pathology in themajority of cases. Indeed psychosocial factors are muchstronger predictors of absence from low back pain and ahost of other conditions than the pathology or the inten-sity of the symptoms. In this respect low back painshares common predictors of work absence with othercommon conditions such as common mental illness

and seasonal respiratory disorders (influenza and thecommon cold). The GP can only make decisions basedon the patient’s report of how severely the symptomsaffect their ability to work.

4.1. Limitations

In interpreting the results it must be understood thatthe psychosocial subscale remained weak even after itemexclusion, this might have weakened the study. How-ever, the biomedical subscale was robust and still couldnot explain sickness certification. In looking purely atthe number of sick certificates issued we have notaddressed the important issue of the duration ofabsence. Sanctioning absence of longer duration orrepeated absence certification might be more closelyrelated to attitudes and beliefs than certification for ashort duration. Sickness certification is also one of themore extreme measures of activity restriction recom-mended by GPs and in using this as the main variablethe study has not addressed other, more subtle, recom-mendations on activity and restriction which may havebeen given, such as advice to rest or stay active.

Sickness certification and GP beliefs data were notgathered concurrently. There is no evidence that theGP beliefs might have changed between 2005 and early2006, there was no intervention to change GP beliefsand no guidelines on the management of back pain wereissued during this period. To have assessed beliefs at thesame time as recording sickness certification might havedrawn attention to the GPs beliefs and influencedbehaviour. However, timing may have influenced ourresults. The numbers in the study were relatively smallfor this type of analysis which may have inflated theR2 (Tabachnick and Fidell, 2001) but will not have sub-stantially affected the lack of predictive value of thePABS or the importance of certification for other condi-tions in predicting certification for NSLBP. Finally, theresults may only be representative of the health care sys-tem that exists in Jersey. Different healthcare systemswhere the GP does not normally sanction sicknessabsence so early or where different sickness absence peri-ods qualification periods exist might demonstrate differ-ent results.

Future studies into attitudes and beliefs about backpain and recommendations about work must also con-sider the influence of the patient’s attitudes on the out-come and the dynamic of the consultation. It is rathertoo simple to believe that the HCP’s attitude alone isthe main determinant of the outcome of theconsultation.

The PABS was specifically developed for use in pop-ulations of physical and manual therapists. Although ithas been used with GPs it still requires further valida-tion. We chose it over other instruments such as theFear Avoidance Beliefs Questionnaire (Waddell et al.,

320 P.J. Watson et al. / European Journal of Pain 12 (2008) 314–320

1993), Back Beliefs Questionnaire (Symonds et al., 1996)and Health Care Providers Pain and ImpairmentRelationship Scale (Rainville et al., 2000) because itwas derived from fear avoidance questionnaires but alsohad many other items about activity as well as work,both of which are the subject of an on going researchproject. The work scale of the FABQ might have beenmore appropriate to sickness certification for back pain.

5. Conclusions

The PABS biomedical subscale has good internalconsistency but the psychosocial subscale requires fur-ther development for use with general practitioners.General practitioner beliefs about low back pain werenot related to sickness certification for non-specific lowback pain but were closely associated with sickness cer-tification behaviour for other conditions. The influenceof the dynamics of the GP and patient relationshipand consultation requires further investigation.

References

Campbell A, Ogden J. Why do doctors issue sick notes? Anexperimental questionnaire study in primary care. Fam Pract2006;23(1):125–30.

Chew-Graham C, May C. Chronic low back pain in general practice:the challenge of the consultation. Fam Pract 1999;16(1):46–9.

Chew-Graham CA, May C. ‘Partners in pain’ – the game ofpainmanship revisited. Fam Pract 2000;17(4):285–7.

Coudeyre E, Rannou F, Tubach F, Baron G, Coriat F, Brin S, et al.General practitioners’ fear-avoidance beliefs influence their man-agement of patients with low back pain. Pain 2006;124(3): 330–7.

Cronbach L. Coefficient alpha and the internal structure of tests.Psychometrika 1951;16:297–334.

DeVellis R. Scale development: theory and application. Lon-don: Sage Publications; 1991.

Field A. Discovering statistics using SPSS for windows. Lon-don: Sage Publications; 2000.

Greenwald A, Banaji M. Implicit social cognitions: attitudes, selfesteem and stereotypes. Psychol Rev 1995;102:4–27.

Houben RM, Gijsen A, Peterson J, de Jong PJ, Vlaeyen JW. Do healthcare providers’ attitudes towards back pain predict their treatmentrecommendations? Differential predictive validity of implicit andexplicit attitude measures. Pain 2005a;114(3):491–8.

Houben RM, Ostelo RW, Vlaeyen JW, Wolters PM, Peters M, Stomp-van den Berg SG. Health care providers’ orientations towardscommon low back pain predict perceived harmfulness of physicalactivities and recommendations regarding return to normal activ-ity. Eur J Pain 2005b;9(2):173–83.

Hussey S, Hoddinott P, Wilson P, Dowell J, Barbour R. Sicknesscertification system in the United Kingdom: qualitative study ofviews of general practitioners in Scotland. BMJ 2004;328(7431):88.

Jellema P, van der Windt DA, van der Horst HE, Twisk JW,Stalman WA, Bouter LM. Should treatment of (sub)acute lowback pain be aimed at psychosocial prognostic factors? Clusterrandomised clinical trial in general practice. Br Med J2005;331(7508):84.

Main CJ, Phillips CJ, Watson PJ. Secondary prevention inhealth-care and occupational settings in musculoskeletal con-ditions. In: Schultz I, Gatchel G, editors. Handbook ofcomplex occupational disability claims: early risk identification,intervention and prevention. New York: Kluwer Academic;2005. p. 387–404.

Maniadakis N, Gray A. The economic burden of back pain in the UK.Pain 2000;84:95–103.

Menard S. Applied logistic regression. Sage university paperseries on quantitative application in social science. CA, 07-106; 1995.

Myers R. Classical and modern regression with applications. Boston,MA: Duxbury; 1990.

Nunnally J, Bernstein I. Psychometric theory. New York: McGraw-Hill; 1994.

Ostelo RW, Stomp-van den Berg SG, Vlaeyen JW, Wolters PM, de VetHC. Health care provider’s attitudes and beliefs towards chroniclow back pain: the development of a questionnaire. Manual Ther2003;8(4):214–22.

Rainville J, Carlson N, Polatin P, Gatchel RJ, Indahl A. Explorationof physicians’ recommendations for activities in chronic low backpain. Spine 2000;25(17):2210–20.

Stevens J. Applied multivariate statistics for the social sciences. Erl-baum; 1992.

Streiner D, Norman G. Health measurement scales: a practical guideto their development and use. New York: Oxford UniversityPress; 1995.

Symonds TL, Burton AK, Tillotson KM, Main CJ. Do attitudes andbeliefs influence work loss due to low back trouble? Occup Med(Oxf) 1996;46(1):25–32.

Tabachnick B, Fidell L. Using multivariate statistics. London: Allynand Bacon; 2001.

Tellnes G, Sandvik L, Moum T. Inter-doctor variation in sicknesscertification. Scand J Prim Health Care 1990;8:45–52.

Vlaeyen JW, Linton SJ. Fear-avoidance and its consequences inchronic musculoskeletal pain: a state of the art. Pain 2000;85(3):317–32.

Vlaeyen JW, Linton SJ. Are we ‘‘fear-avoidant? Pain 2006;124(3):240–1.

Waddell G, Newton M, Henderson I, Somerville D, Main CJ. A fear-avoidance beliefs questionnaire (FABQ) and the role of fear-avoidance beliefs in chronic low back pain and disability. Pain1993;52(2):157–68.

Watson PJ, Main CJ, Waddell G, Gales TF, Purcell-Jones G.Medically certified work loss, recurrence and costs of wagecompensation for back pain: a follow-up study of theworking population of jersey. Br J Rheumatol 1998;37(1):82–6.