w w w . e l s e v i e r . c o m / l o c a t e / p a i n

PAIN�

151 (2010) 162–173

Work and neck pain: A prospective study of psychological, social,and mechanical risk factors

Jan Olav Christensen a,b,*,1, Stein Knardahl a,1

a The National Institute of Occupational Health, Norwayb Department of Psychology, University of Oslo, Norway

a r t i c l e i n f o

Article history:Received 19 November 2009Received in revised form 20 May 2010Accepted 1 July 2010

Keywords:Neck painOccupationalPsychosocialProspectiveMechanicalExposure

0304-3959/$36.00 � 2010 International Associationdoi:10.1016/j.pain.2010.07.001

* Corresponding author. Address: The National InstP.O. Box 8149 Dep, N-0033 Oslo, Norway. Tel.: +47 23 1

E-mail address: joc@stami.no (J.O. Christensen).1 Tel.: +47 23 19 51 00; fax:+47 23 19 52 00.

a b s t r a c t

To determine the impact of occupational psychological/social and mechanical factors on neck pain, a pro-spective cohort study with a follow-up period of 2 years was conducted with a sample of Norwegianemployees. The following designs were tested: (i) cross-sectional analyses at baseline (n = 4569) and fol-low-up (n = 4122), (ii) prospective analyses with baseline predictors, (iii) prospective analyses with aver-age exposure over time [(T1 + T2)/2] as predictor, and (iv) prospective analyses with measures of changein exposure from T1 to T2 as predictors. A total of 2419 employees responded to both the baseline andfollow-up questionnaire. Data were analyzed using ordinal logistic regression. After adjustment forage, sex, neck pain at T1, and other exposure factors that had been estimated to be confounders, the mostconsistent risk factors were role conflict (highest OR 2.97, 99% CI: 1.29–6.74) and working with arms raisedto or above shoulder level (highest OR 1.37, 99% CI: 1.05–1.78). The most consistent protective factors wereempowering leadership (lowest OR 0.53, 99% CI: 0.35–0.81) and decision control (lowest OR 0.60, 99% CI:0.36–1.00). Hence, psychological and social factors are important precursors of neck pain, along withmechanical factors. Although traditional factors such as quantitative demands and decision control playa part in the etiology of neck pain at work, in this study several new factors emerged as more important.

� 2010 International Association for the Study of Pain. Published by Elsevier B.V. All rights reserved.

1. Introduction known, and several studies have reported very low levels of trape-

Neck pain is a widespread health problem in the working pop-ulation. The annual prevalence has been reported to be above 40%[1,3,24]. Neck pain may severely disrupt the well being and pro-ductivity of the individual. Hence, it is a major public health issue,even if the medical consequences for each individual are notsevere.

Traditionally, biomechanical loads have been assigned the mainrole in explaining relationships between work and musculoskeletaldisorders (MSDs) [6]. However, MSDs are common in office work-ers and remain as prevalent conditions even if biomechanical loadshave been reduced in most occupations. Hence, psychological fac-tors are acknowledged as contributors to the pathogenesis of MSD.Nevertheless, the pathogenetic mechanisms of neck pain remainobscure. Several hypotheses have maintained that sustained mus-cle contraction with metabolic demands in muscle fibers exceedingthe energy supply leads to energy crisis which could produce pain[19,25]. What actually activates the nociceptive neurons is still un-

for the Study of Pain. Published by

itute of Occupational Health,9 52 58; fax: +47 23 19 52 00.

zius muscle activity during office work (e.g. [45]). Duringpsychological challenge, active coping behavior includes a cardio-vascular response pattern in which blood flow to muscles increases[45]. It has been hypothesized that factors involved in the regula-tion of muscle blood vessels may activate nociceptors [30,31].However, it is also possible that the pain may originate from ten-dons or periostea or other components outside the muscle tissue.

Systematic reviews have concluded that several occupationalpsychological and social factors have been found to be related toneck and upper limb pain, such as lack of supervisory support[2,38,49], low social support [2,7,47], low job control [6,7,47,49],high job strain, conflicts at work, low job security [2], and, mostconsistently, job demands [2,7,38,47,49]. However, consistency offindings across studies is low, and several methodological short-comings have been pointed out [6,7,16,22,32]. Perhaps most severeare ill-defined conceptualizations of psychological and social fac-tors [37] and a paucity of studies with longitudinal prospectivedesign. In their critical systematic review, Hartvigsen et al. [21]found that evidence of associations between psychosocial factorsat work and low back pain was lacking. Variation between studies,they claimed, indicated spurious findings due to widespread use ofnon-standardized, non-validated instruments for measuring psy-chosocial factors. Furthermore, they suggested that this tendencywas reinforced by multiple testing within and across studies. Davis

Elsevier B.V. All rights reserved.

J.O. Christensen, S. Knardahl / PAIN�

151 (2010) 162–173 163

and Heaney [11] reported that only one third of the studies in theirreview employed well-validated multiple-item measures. Also,psychological demands may be confounded by physical demands,yet there has been few studies that take both of these factors intoaccount [3,37]. Almost all previous studies have measured only asmall selection of psychological and social work factors [32].

The objective of the present study was to determine whichoccupational psychological and social factors contribute to risk ofneck pain, taking into account several factors not previouslyreported as well as aspects of biomechanical exposure. With a pro-spective study design we aimed to take into account changes inexposure over time as well as temporal associations between expo-sures and subsequent neck pain.

2. Methods

2.1. Design

The study design was longitudinal and prospective, with allvariables measured at baseline and at follow-up 2 years later. Sincethe question of the mechanisms of work-related musculoskeletalsymptoms is unresolved, the required time period before exposureproduces a health effect (neck pain) is unknown [15]. Therefore,several statistical designs testing associations over time were per-formed. In order to evaluate variance in patterns of association be-tween psychological and social factors and neck pain with differenttime periods, data were analyzed cross-sectionally at both timepoints as well as prospectively.

The time period was considered the same for all respondentswithin each company, since the electronic forms were closed forall respondents in each company simultaneously, with the oppor-tunity of responding until that point in time. The average time per-iod from the end of baseline to the end of follow-up forrespondents was 23 months (range 17–36).

2.2. Subjects

Subjects were recruited from organizations in Norway that werecontacted and offered to participate in the study. Baseline data weregathered from November 2004 until November 2006, and follow-updata were gathered from September 2006 until February 2009.

In return for participating and making the data available for re-search, companies received written reports and oral presentationsof results as a tool for organizational development and an aid formonitoring the organizational work environment. The surveywas web-based, although the participants with limited access tocomputers at work were given the option of filling out a paperversion.

Employees and management in the companies were informedat the organizational level first. Subsequently, all employeesexcluding those on sick leave were mailed a letter with informa-tion about the survey. This letter contained either a personalizedcode for logging into the web-questionnaire or a paper version ofthe questionnaire with a pre-stamped return envelope, in additionto information about the survey. The written information ex-plained the aims of the study and assured that responses wouldbe treated confidentially, in strict accordance with the generalguidelines and specific license from the Norwegian Data Inspector-ate. Employees were given the opportunity of filling out the ques-tionnaire at work, but it was also possible to fill it out from home orany other location. Each subject had the opportunity to log into theweb-questionnaire an unlimited number of times to change orcomplete their answers during the survey period.

The questionnaire gathered data about background informa-tion, work organization, psychological and social factors at work,

organizational change, attitudes towards work, personality, physi-cal activity, smoking, alcohol use, mental health, mastery of work,working ability, and health complaints. This study is based on partsof this information. In addition, all companies except one suppliedlists containing information about address, age, sex, department,and classification of occupation for all employees prior to the sur-vey. Occupation was classified according to the standard classifica-tion of occupations (STYRK), developed by Statistics Norway(www.ssb.no) based on the International Standard Classificationof Occupation (ISCO-88).

By the time of data analysis, 8928 employees had been invitedto participate at baseline, of which 4569 responded (51.2%), form-ing cross-sectional sample 1. A responder was defined as a personwho had completed the neck pain item and at least one predictor.At follow-up, a total of 8836 employees were invited, and 4122responded (46.7%), forming cross-sectional sample 2. Paper-and-pencil versions were returned by 1082 participants (12.1% of pos-sible respondents, 23.7% of actual respondents) at baseline, and388 at follow-up (4.4% of possible, and 9.4% of actual respondentsat follow-up, including those not participating at baseline).

Respondents were recruited from 20 organizations, 10 private(2703 invited and 1060 responded at baseline, 2434 invited and791 responded at follow-up) and 10 public (6225 invited and3509 responded at baseline, 6402 invited and 3331 responded atfollow-up), with a range of 13–2316 invited employees at baselineand 14–2036 at follow-up. The participating organizations repre-sented a wide variety of job types, with four municipalities(1315, 1060, 1113, and 434 invited at baseline), one insurancecompany (2317 invited at baseline), and two public organizations(297 and 1754 invited at baseline) comprising 8290 of 8928(92.9%) invited employees at baseline. Eight companies were un-able to participate twice, and thus were not included in this study.

Of the 8928 employees that were invited to participate at base-line, 6361 (71.2%) were also invited at follow-up. Of these, 3749(58.9%) responded at baseline, 3093 (48.6%) responded at follow-up, and 2419 (38.0%) responded at both times. Responding at bothtime points was defined as completing neck pain items at bothbaseline and follow-up and at least one predictor at baseline.

2.3. Outcome measure

The outcome measure was the reported intensity of neck painduring 4 weeks prior to answering the questionnaire (see [44]).This item was part of a questionnaire measuring 21 differenthealth complaints. Health complaints were measured by asking‘‘have you been bothered by .... (i.e. neck pain) the last 4 weeks”,with optional answers ‘‘not bothered” (1), ‘‘a little bothered” (2),‘‘rather intensely bothered” (3), and ‘‘very intensely bothered”(4). If the subject answered ‘‘a little bothered”, ‘‘rather intenselybothered”, or ‘‘very intensely bothered”, the question was followedby items reflecting the duration of the health complaint, whether ithad been experienced at work, and whether it was believed to becaused by work.

2.4. Psychological and social work factors

The psychological and social factors were assessed by the Gen-eral Nordic Questionnaire for Psychological and Social Factors atWork (QPSNordic) [10]. The following scales were studied (a coeffi-cients are calculated from the total baseline and follow-up sam-ples, respectively): quantitative demands (four items, a = 0.76and 0.76), decision demands (three items, a = 0.63 and 0.64), deci-sion control (five items, a = 0.69 and 0.69), control over workintensity (four items, a = 0.80 and 0.82), role conflict (three items,a = 0.68 and 0.68), role clarity (three items, a = 0.82 and 0.82), sup-port from immediate superior (three items, a = 0.85 and 0.86),

164 J.O. Christensen, S. Knardahl / PAIN�

151 (2010) 162–173

empowering leadership (three items, a = 0.88 and 0.88), fair lead-ership (three items, a = 0.81 and 0.81), predictability during thenext month (three items, a = 0.65 and 0.64), commitment to orga-nization (three items, a = 0.87 and 0.86), social climate (threeitems, a = 0.72 and 0.71), and positive challenge at work (threeitems, a = 0.75 and 0.75). The scales were constructed on the basisof the following frequency scoring: ‘‘1 = very seldom or never”,‘‘2 = somewhat seldom”, ‘‘3 = sometimes”, ‘‘4 = somewhat often”,and ‘‘5 = very often or always”.

All relationships were not necessarily expected to be linear ormonotonic (see [13,14]). Therefore, scale scores were categorizedso that the effect of different levels of exposure could be investi-gated separately. Scores between 1.00 and 1.80 were assignedthe value ‘‘1”, scores between 1.81 and 2.60 the value ‘‘2”, scoresbetween 2.61 and 3.40 the value ‘‘3”, scores between 3.41 and4.20 the value ‘‘4”, and scores between 4.21 and 5.00 the value ‘‘5”.

In order to estimate the effect of exposure over time on subse-quent neck pain, the baseline and follow-up measures of exposurewere averaged. The resulting variables measure exposure on ascale of 1–5. For regression analyses, the scale was categorized intofive categories according to the procedure described in the previ-ous paragraph.

Due to a low number of responses in some categories, ‘‘1” and‘‘2” were combined for role clarity, social climate and positive chal-lenge in analyses of cross-sectional samples, and for decision de-mands, role clarity, predictability during the next month, socialclimate, and positive challenge in analyses of the prospectivesample.

To investigate the effect of increasing and decreasing exposureor different levels of unchanged exposure across time, analyseswere also performed with change in exposure from baseline to fol-low-up as predictor. Hence, the categorized measures were tricho-tomized to reflect ‘‘low” (1 and 2), ‘‘middle” (3), and ‘‘high” (4 and5) levels. Any change across these categories from baseline to fol-low-up was considered a change. Possibly, this operationalizationleaves too little contrast between categories, i.e. not enoughchange to detect effects. Also, changes in exposure may interactwith baseline levels of exposure. For instance, increases from highlevels could be quite different from increases from low levels.However, more specific classifications were not studied becauseof statistical power issues: for the different psychological/socialexposures an average of only 41 employees changed from ‘‘low”to ‘‘high” and 40 employees changed from ‘‘high” to ‘‘low”. It isimportant to keep in mind that classifying ‘‘change” according todirection alone may have affected effect estimates.

The resulting categories of change were ‘‘constant low”,‘‘decrease”, ‘‘constant middle”, ‘‘increase”, and ‘‘constant high”. Dueto few responses in the ‘‘constant low” category when multivariateregressions were run, ‘‘constant low” and ‘‘constant middle” werecombined for ‘‘role clarity”, ‘‘predictability during the next month”,‘‘social climate”, and ‘‘positive challenge”.

2.5. Mechanical exposure factors

Perceived physical workload was measured with three itemsunder the heading ‘‘to what degree do your work tasks consistof. . .”. The items were ‘‘lifting or handling objects that weighapproximately 1–5 kg with your own muscular strength”, ‘‘liftingor handling objects that weigh approximately 6–15 kg with yourown muscular strength”, ‘‘lifting or handling objects that weighmore than approximately 15 kg with your own muscular strength”.The response categories for these questions were ‘‘1 = seldom ornever”, ‘‘2 = sometimes”, ‘‘3 = daily”, and ‘‘4 = many times pr. day”.These items were combined into one scale (a = 0.90 at both totalbaseline sample and total follow-up sample). This scale wasweakly negatively correlated with the scale quantitative demands

(Spearman’s q = �0.13, p = 0.00 and q = �0.11, p = 0.00, in thebaseline and follow-up samples, respectively), thereby displayingdiscriminant validity. For the regression analyses, the variablewas recoded: ‘‘1.00–1.50” set to ‘‘1”, ‘‘1.51–2.50” set to ‘‘2”,‘‘2.51–3.50” set to ‘‘3”, and ‘‘3.51–4.00” set to ‘‘4”.

Working with arms raised to or above shoulder level was mea-sured with the item ‘‘work positions where you have to raise yourarms to or above shoulder level”, using the same response catego-ries as the physical work load items.

The average exposure across time points was also estimated forphysical workload and working with arms raised to or aboveshoulder level. For regression analyses, the variables were recoded:‘‘1.00–1.50” was set to ‘‘1”, ‘‘1.51–2.50” was set to ‘‘2”, ‘‘2.51–3.50”was set to ‘‘3”, and ‘‘3.51–4.00” was set to ‘‘4”.

The change in physical workload was estimated by dichotomizingthe categorized factor into ‘‘low” (1) and ‘‘high” (2–4), and then com-bining baseline and follow-up measures to form the categories ‘‘con-stant low”, ‘‘decrease”, ‘‘increase”, and ‘‘constant high”. The changein working with arms raised to or above shoulder level was estimatedthe same way with the single item at baseline and follow-up.

2.6. Statistical analyses

All statistical analyses were conducted with PASW Statistics(formerly SPSS), version 17.0.

2.6.1. Non-response analysisIn order to estimate the potential confounding effects of non-re-

sponse, a multivariate binary logistic regression model was com-puted to calculate the odds of being a responder at baselinebased on sex and age group. All invited employees at baseline wereincluded in this analysis.

2.6.2. Attrition analysisTo assess the potential presence of attrition bias, univariate bin-

ary logistic regression analyses were conducted to identify psycho-logical, social, and mechanical factors associated with respondingat both time points as opposed to responding only at baseline.All subjects that had been invited at both baseline and follow-upand had responded at baseline were included in these analyses.Factors found to predict response at follow-up were then enteredinto a multivariate binary logistic regression model to computethe odds of being a responder at follow-up controlling for the otherrelated factors as well as age, sex, and neck pain intensity atbaseline.

2.6.3. Main analysesBecause of the ordinal nature and non-normality of the out-

come scores, ordinal logistic regressions, by Generalized LinearModels (GZLMs) in PASW, were conducted to predict the outcomefrom psychological, social, and mechanical factors.

Cumulative odds ratios were calculated for each category ofeach predictor. Cumulative odds ratios compare the odds of out-come being above vs. below each of the possible cutpoints (i.e. painintensity of 2–4 vs. 1, 3–4 vs. 1–2, and 4 vs. 1–3) for all levels of thepredictor variable with the odds for the reference category [29]. Anassumption of proportional odds across all categories of the out-come is necessary for this model to be valid and this assumptionwas tested.

Age, sex, and baseline level of neck pain were selected a priorifor inclusion in all multivariate analyses. Educational level was alsomeasured, but was omitted from multivariate analyses due to ahigh rate of missing data (39% among baseline responders). How-ever, the confounding influence of educational level was estimatedaccording to the procedure described below, and such confoundingwas indicated only for the change measure of physical workload.

J.O. Christensen, S. Knardahl / PAIN�

151 (2010) 162–173 165

Categorizing predictors helps detect nonlinear or threshold ef-fects, but may decrease statistical power [51]. Therefore, analyseswere also conducted with the original scale scores of the psycho-logical and social factors entered as continuous variables. Odds ra-tios for continuous predictors are interpreted as the change in oddsper unit change of the predictor.

Multiple testing was necessary to investigate a comprehensiveset of factors. Since this inflates the risk of type I error, a strict ap-proach was adopted by using 99% confidence intervals in mainanalyses. In addition a Bonferroni-corrected significance levelwas estimated based on the number of factors tested. This wasdone by dividing the overall significance level by the number offactors tested (i.e. 0.01/15 = 0.0007).

Exposures were modeled separately, not adjusted for all otherindependent variables. While entering all factors into one modelwould account for the maximum amount of variance and adjustfor confounding, it would partial out all explained variance that isshared between predictors. Confounding is not the only possiblereason for shared variance, and it is statistically indistinguishablefrom mediation [4]. Unless the aim is to identify the strongest pre-dictor or establish evidence of mediation, entering a mediator in amultivariate model constitutes overadjustment since it obscuresthe impact of the preceding factor [9,28]. Hence, it may motivatethe incorrect conclusion that the antecedent factor is theoreticallyirrelevant. One contributing reason for conflicting results or lack ofresults in epidemiologic research is the adjustment for mediators[9,12]. The aim of the present study was to investigate a compre-hensive range of factors in order to identify predictors of neck pain.These factors correlate to some degree and may be causally related[10].

Thus, in order to limit potential overadjustment each predictorwas adjusted only for other predictors that were first estimated toexert an influence above a certain threshold level, according to thestrategy recommended by Rothman [43]. These assessments werebased on the baseline sample data. Univariate regressions werefirst conducted for each exposure variable. Then each variablewas included in bivariate regressions with all other exposure vari-ables. Subsequently, the crude odds ratios from the univariateregressions were compared with the adjusted odds ratios fromthe bivariate regressions. When the inclusion of a potential con-founder resulted in a change in odds ratio of 10% or more that var-iable was treated as a real confounder. These confounders wereincluded in all subsequent multivariate analyses with the relevantexposure.

For the exposure change variables the confounder estimationprocedures were repeated with change measures as potentialconfounders.

2.6.4.1. Potentially etiologic factors – onset of pain. Factors that influ-ence the onset of neck pain (etiologic factors) may differ from fac-tors that influence the course of pre-existing pain (prognosticfactors). Musculoskeletal pain exhibits substantial variation overtime [34], hence a single report of no pain does not establish a sta-tus of no previous pain. However, in order to elucidate factors thatmay determine onset of neck pain, a separate analysis was per-formed on the 1915 subjects who reported of no pain (1 = ‘‘notbothered”) at baseline in the prospective sample. Multivariate bin-ary logistic regressions (with continuous baseline predictors) wererun to estimate the risk of neck pain at follow-up.

2.6.4.2. Potentially prognostic factors in subjects with pain. Multivar-iate binary logistic regressions (with continuous baseline predic-tors) were performed on the 1838 subjects who reported (2 = ‘‘alittle bothered”/‘‘rather intensely bothered”/‘‘very intensely both-ered”) at baseline to estimate the odds of recovery, i.e. reportingof no neck pain at follow-up.

3. Results

3.1. Subject characteristics

According to the STYRK classification of occupation, the threelargest groups among all employees that responded at baselinewere technicians and associated professionals (N = 2075, 45.4%), ser-vice workers and shop and market sales workers (N = 844, 18.5%), andprofessionals (N = 750, 16.4%) (Table 1). The rest of the categories onthe classification were legislators, senior officials and managers,skilled agricultural and fishery workers, craft and related trades work-ers, plant and machine operators and assemblers, elementary occupa-tions, and armed forces and unspecified (817, 17.9%). Baselinedistributions of employees who were invited twice but respondedat baseline only and employees who responded twice were similar(see Table 1).

The proportion of responders that reported that they were ‘‘alittle bothered” or more severely bothered by neck pain at baselinewas 49.0% (2239 employees) (Table 1). The prevalence of severeneck pain (‘‘rather intensely bothered” or ‘‘very intensely both-ered”) was 19.5%. For those who responded at both time points,these prevalence rates were 47.9 and 18.4, respectively.

3.2. Non-response and attrition analyses

Binary logistic regression with age and sex as predictorsrevealed that being in the three middle age groups (30–39,40–49, and 50–59) significantly increased the odds of respondingto the neck pain item and at least one predictor at baseline com-pared to being in the lowest age group (<30) (p < 0.05, analysesnot shown). Sex did not predict responding at baseline.

Multivariate binary logistic regression analysis showed thatquantitative demands, decision control, role conflict, social climate,age, and pain intensity at baseline had independent and signifi-cant relationships with responding at follow-up for those whohad already responded at baseline (p < 0.05, analyses not shown).Increased odds of responding at follow-up were observed for cat-egory 3 of quantitative demands, category 5 of decision control,categories 3 and 4 of role conflict, category 2 of social climate,and all three age groups between <30 and >59. Scoring 3 on painintensity at baseline decreased the odds of responding at follow-up.

3.3. Cross-sectional analyses

3.3.1. Baseline sampleUnivariate regressions showed that associations for all psy-

chological/social and mechanical variables were statistically sig-nificant except decision demands (p > 0.05) (analyses notshown). All factors except role clarity and physical workload re-mained statistically significant after Bonferroni-correction. ORswith p < 0.01 ranged from 0.34 (decision control) to 2.06 (workingwith arms raised). Neck pain was also univariately regressed onsex and age for the baseline sample (analyses not shown). Beinga woman was associated with higher neck pain intensity (OR1.81, 99% CI 1.55–2.12), and being in the age groups 40–49(OR 1.62, 99% CI 1.20–2.17) and 50–59 (OR 1.76, 99% CI 1.30–2.38) were associated with higher neck pain intensity whencompared with being <30.

Multivariate regressions showed statistically significant associ-ations for quantitative demands, decision control, control overwork intensity, role conflict, support from immediate superior,empowering leadership, fair leadership, predictability during thenext month, commitment to organization, social climate, andworking with arms raised to or above shoulder level (Tables 2a

Table 1Baseline characteristics of respondersa, non-responders, and responders to both time pointsb.

All workers invited at baseline (N = 8928) Workers who were invited at both baseline andfollow-up (N = 6361)

Baseline responders(n = 4569)

Baseline non-responders(n = 4359)

Baseline responders(n = 3749)

Responded at both times(n = 2419)

Age<30 364 (8.0%) 389 (8.9%) 243 (6.5%) 126 (5.2%)30–39 1223 (26.8%) 924 (21.2%) 945 (25.2%) 609 (25.2%)40–49 1468 (32.1%) 1041 (23.9%) 1242 (33.1%) 821 (33.9%)50–59 1202 (26.3%) 935 (21.4%) 1090 (29.1%) 722 (29.8%)>59 305 (6.7%) 379 (8.7%) 229 (6.1%) 141 (5.8%)Missing data 7 (0.2%) 691 (15.9%) 0 (0.0%) 0 (0.0%)

SexFemale 2984 (65.3%) 2390 (54.8%) 2449 (65.3%) 1542 (63.7%)Male 1575 (34.5%) 1278 (29.3%) 1300 (34.7%) 877 (36.3%)Missing data 10 (0.2%) 691 (15.9%) 0 (0.0%) 0 (0.0%)

Marital statusNot married 362 (7.9%) – – 263 (7.0%) 167 (6.9%)Married 2117 (46.3%) – – 1776 (47.4%) 1138 (47.0%)Cohabiting 501 (11.0%) – – 381 (10.2%) 229 (9.5%)Widowed 42 (0.9%) – – 34 (0.9%) 17 (0.7%)Divorced 205 (4.5%) – – 184 (4.9%) 114 (4.7%)Separated 57 (1.2%) – – 42 (1.1%) 29 (1.2%)Missing data 1285 (28.1%) – – 1069 (28.5%) 725 (30.0%)

Classification of occupationLegislators, senior officials and managers 348 (7.6%) – – 292 (7.8%) 225 (9.3%)Professionals 750 (16.4%) – – 568 (15.2%) 383 (15.8%)Technicians and associate professionals 2075 (45.4%) – – 1711 (45.6%) 1100 (45.5%)Clerks 344 (7.5%) – – 284 (7.6%) 172 (7.1%)Service workers and shop and market sales workers 844 (18.5%) – – 713 (19.0%) 429 (17.7%)Skilled agricultural and fishery workers 5 (0.1%) – – 2 (0.1%) 1 (0.0%)Craft and related trades workers 39 (0.9%) – – 32 (0.9%) 26 (1.1%)Plant and machine operators and assemblers 9 (0.2%) – – 7 (0.2%) 3 (0.1%)Elementary occupations 72 (1.6%) – – 64 (1.7%) 40 (1.7%)Armed forces and unspecified 0 (0.0%) – – 0 (0,0%) 0 (0.0%)Missing data 83 (1.8%) – – 76 (2.0%) 40 (1.7%)

Neck pain intensityNot bothered 2330 (51.0%) – – 1911 (51.0%) 1260 (52.1%)A little bothered 1348 (29.5%) – – 1101 (29.4%) 715 (29.6%)Rather intensely bothered 754 (16.5%) – – 631 (16.8%) 378 (15.6%)Very intensely bothered 137 (3.0%) – – 106 (2.8%) 66 (2.7%)

a Response at baseline is defined as having completed the neck pain item and at least one predictor.b Response to both time points is defined as having completed the neck pain item at both baseline and follow-up, and at least one predictor at baseline.

166 J.O. Christensen, S. Knardahl / PAIN�

151 (2010) 162–173

and b). Associations for all factors remained statistically significantafter Bonferroni-correction and ORs ranged from 0.36 (commit-ment to organization) to 2.55 (role conflict).

3.3.2. Follow-up sampleUnivariate regressions showed statistically significant associa-

tions for all psychological/social and mechanical variables exceptrole clarity with neck pain intensity (analyses not shown). All fac-tors except decision demands remained statistically significant afterBonferroni-correction. ORs ranged from 0.47 (commitment to orga-nization) to 2.24 (role conflict).

Multivariate analyses showed statistically significant associa-tions for quantitative demands, decision demands, decision con-trol, control over work intensity, role conflict, empoweringleadership, fair leadership, predictability during the next month,commitment to organization, positive challenge, and working witharms raised to or above shoulder level (Tables 2a and b). All factorsremained statistically significant after Bonferroni-correction, ex-cept decision demands, control over work intensity, and positivechallenge. ORs with p < 0.01 ranged from 0.40 (commitment toorganization) to 3.67 (role conflict).

3.4. Prospective analyses

3.4.1. Baseline exposure as predictorUnivariate regressions showed statistically significant associa-

tions for all baseline predictors except predictability during the nextmonth, decision demands, and role clarity (analyses not shown) withneck pain intensity at follow-up. All associations remained statisti-cally significant after Bonferroni-correction, except for quantitativedemands, control over work intensity, positive challenge, and physicalworkload. ORs with p < 0.01 ranged from 0.37 (decision control) to3.02 (role conflict).

In the multivariate analyses decision control, role conflict,empowering leadership, and positive challenge remained statisticallysignificant (Tables 3a and b). Statistical significance after Bonfer-roni-correction was maintained for role conflict and empoweringleadership. ORs with p < 0.01 ranged from 0.48 (positive challenge)to 2.97 (role conflict).

3.4.2. Average exposure as predictorUnivariate regressions showed statistically significant associa-

tions for all predictors except decision demands and role clarity(analyses not shown). All associations remained statistically

Table 2aCross-sectional analyses: multivariate ordinal logistic regressions with psychological, social, and mechanical factors as independent variables and neck pain intensity as outcome.Separate regressions were run with independent variables entered as categorical and continuous.

Psychological, social, and mechanical factors Confounders included inmultivariate analysesa

Baseline sample Follow-up sample

N OR 99% CI N OR 99% CI

Quantitative demands Category 1 Fair leadership, social climate 332 1.00 [ref] 228 1.00 [ref]2 1113 1.08 [0.79–1.48] 720 1.18 [0.80–1.75]3 1635 1.19 [0.88–1.62] 1102 1.34 [0.93–1.95]4 858 1.41 [1.02–1.96]* 615 1.49 [1.01–2.22]*

5 261 1.99 [1.32–3.01]** 173 2.58 [1.56–4.28]**

Continuous 4199 1.25 [1.12–1.38]** 2838 1.25 [1.10–1.42]**

Decision demands Category 1 85 1.00 [ref] 55 1.00 [ref]2 235 1.08 [0.59–2.01] 153 0.69 [0.32–1.50]3 1682 0.97 [0.57–1.68] 1165 0.84 [0.43–1.66]4 1560 1.03 [0.60–1.78] 1032 0.98 [0.50–1.94]5 747 1.15 [0.66–2.03] 510 1.07 [0.54–2.16]

Continuous 4309 1.09 [0.98–1.21] 2915 1.14 [1.00–1.30]*

Decision control Category 1 298 1.00 [ref] 176 1.00 [ref]2 1180 0.65 [0.47–0.89]** 835 0.80 [0.54–1.20]3 1665 0.62 [0.46–0.84]** 1148 0.59 [0.40–0.87]**

4 857 0.50 [0.36–0.69]** 614 0.49 [0.32–0.75]**

5 184 0.37 [0.23–0.60]** 125 0.41 [0.22–0.74]**

Continuous 4184 0.78 [0.71–0.87]** 2898 0.73 [0.64–0.83]**

Control over work intensity Category 1 638 1.00 [ref] 505 1.00 [ref]2 832 0.78 [0.61–1.01] 541 0.90 [0.67–1.22]3 922 0.96 [0.75–1.24] 609 1.02 [0.76–1.36]4 1184 0.75 [0.58–0.95]** 797 0.83 [0.62–1.10]5 878 0.74 [0.57–0.96]** 577 0.75 [0.55–1.03]

Continuous 4454 0.92 [0.86–0.99]* 3029 0.91 [0.84–0.99]*

Role conflict Category 1 931 1.00 [ref] 653 1.00 [ref]2 1334 1.20 [0.97–1.49] 958 1.39 [1.07–1.80]*

3 1795 1.57 [1.28–1.92]** 1192 1.83 [1.43–2.35]**

4 304 2.15 [1.55–2.97]** 201 2.82 [1.89–4.21]**

5 99 2.55 [1.52–4.27]** 43 3.67 [1.68–7.95]**

Continuous 4463 1.37 [1.25–1.51]** 3047 1.51 [1.34–1.71]**

Role clarity Category 1 and 2 Support from immediate superior,fair leadership, commitmentto organization

102 1.00 [ref] 69 1.00 [ref]3 370 1.31 [0.86–2.01] 228 1.12 [0.57–2.23]4 1094 1.16 [0.78–1.73] 817 1.01 [0.54–1.92]5 2540 1.12 [0.75–1.66] 1773 1.07 [0.57–2.02]

Continuous 4106 0.98 [0.87–1.11] 2887 1.06 [0.91–1.23]

Support from immediate superior Category 1 Fair leadership 145 1.00 [ref] 85 1.00 [ref]2 228 0.61 [0.36–1.05] 157 1.30 [0.65–2.60]3 928 0.58 [0.35–0.94]* 631 1.03 [0.54–1.95]4 1225 0.54 [0.33–0.89]** 853 0.90 [0.47–1.75]5 1830 0.50 [0.30–0.83]** 1218 0.82 [0.42–1.60]

Continuous 4356 0.87 [0.78–0.97]* 2944 0.92 [0.80–1.05]

a Age and sex were included as confounders in all multivariate models.* p < 0.01.

** p < 0.0007, which is the Bonferroni-corrected threshold based on the number of factors tested (0.01/15).

J.O. Christensen, S. Knardahl / PAIN�

151 (2010) 162–173 167

significant after Bonferroni-correction. ORs with p < 0.01 rangedfrom 0.32 (decision control) to 3.90 (role conflict).

The multivariate regressions showed statistically significantassociations for decision control, role conflict, role clarity, empow-ering leadership, fair leadership, predictability during the nextmonth, commitment to organization, and working with armsraised to or above shoulder level (Tables 3a and b). Role conflictand empowering leadership remained statistically significant afterBonferroni-correction. ORs with p < 0.01 ranged from 0.53(empowering leadership) to 3.01 (role clarity).

3.4.3. Change in exposure as predictorUnivariate regressions showed statistically significant associa-

tions for all predictors except decision demands, role clarity, predict-ability during the next month, and positive challenge (analyses notshown). Statistical significance was maintained for all factors afterBonferroni-correction except for control over work intensity. ORswith p < 0.01 ranged from 0.40 (constant high fair leadership) to2.69 (constant high role conflict).

The confounder estimation procedure revealed that none of thepsychological/social and mechanical factors acted as confounders.Hence, all multivariate regressions were conducted only withage, sex, and baseline pain as covariates.

Multivariate regressions showed statistically significant associ-ations for decision control, role conflict, empowering leadership, socialclimate, physical workload, and working with arms raised (Tables 4aand b). The association remained statistically significant after Bon-ferroni-correction for constant high social climate. ORs ranged from0.58 (constant high social climate) to 2.16 (constant high roleconflict).

3.4.4.3.4.4.1. Potentially etiologic factors – onset of pain. Of the employeesthat reported of no pain at baseline, 283 subjects reported of neckpain at follow-up (incidence rate 22.4%). Among those employeesthat were pain free at baseline, empowering leadership was a statis-tically significant predictor of neck pain at follow-up (OR 0.83,p < 0.01, analyses not shown).

Table 2bCross-sectional analyses: multivariate ordinal logistic regressions with psychological, social, and mechanical factors as independent variables and neck pain intensity as outcome.Separate regressions were run with independent variables entered as categorical and continuous.

Psychological, social, and mechanical factors Confounders includedin multivariate analysesa

Baseline sample Follow-up sample

N OR 99% CI N OR 99% CI

Empowering leadership Category 1 495 1.00 [ref] 314 1.00 [ref]2 684 0.76 [0.57–1.02] 440 0.85 [0.60–1.21]3 1603 0.64 [0.50–0.83]** 1101 0.62 [0.45–0.84]**

4 999 0.54 [0.41–0.71]** 699 0.52 [0.38–0.73]**

5 700 0.47 [0.35–0.63]** 501 0.56 [0.40–0.80]**

Continuous 4481 0.81 [0.75–0.87]** 3055 0.83 [0.76–0.91]**

Fair leadership Category 1 Commitment to organization 98 1.00 [ref] 59 1.00 [ref]2 192 1.15 [0.63–2.11] 149 1.73 [0.83–3.65]3 831 0.77 [0.45–1.33] 573 1.01 [0.52–2.00]4 1131 0.76 [0.45–1.30] 797 0.84 [0.43–1.66]5 1968 0.61 [0.36–1.04] 1396 0.73 [0.37–1.44]

Continuous 4220 0.85 [0.77–0.94]** 2974 0.79 [0.70–0.90]**

Predictability during thenext month

Category 1 62 1.00 [ref] 28 1.00 [ref]2 107 0.80 [0.37–1.73] 63 1.78 [0.58–5.59]3 408 0.82 [0.42–1.59] 266 1.31 [0.50–3.58]4 1161 0.69 [0.37–1.31] 756 0.98 [0.39–2.59]5 2749 0.54 [0.29–1.02] 1926 0.80 [0.32–2.10]

Continuous 4487 0.81 0.73–0.90** 3039 0.76 [0.67–0.87]**

Commitment toorganization

Category 1 192 1.00 [ref] 114 1.00 [ref]2 337 0.87 [0.56–1.35] 203 0.93 [0.53–1.63]3 1146 0.62 [0.42–0.90]* 834 0.62 [0.39–1.01]4 1277 0.47 [0.32–0.69]** 894 0.53 [0.33–0.86]**

5 1355 0.36 [0.25–0.53]** 1009 0.40 [0.25–0.64]**

Continuous 4307 0.72 [0.66–0.78]** 3054 0.74 [0.67–0.82]**

Social climate Category 1 and 2 Fair leadership, commitmentto organization

203 1.00 [ref] 109 1.00 [ref]3 1054 0.65 [0.44–0.96]* 774 1.37 [0.81–2.35]4 1452 0.59 [0.39–0.88]** 1001 0.91 [0.53–1.58]5 1414 0.54 [0.35–0.82]** 1023 0.88 [0.50–1.56]

Continuous 4123 0.86 [0.75–0.98]* 2907 0.86 [0.73–1.01]

Positive challenge Category 1 and 2 Commitment to organization 140 1.00 [ref] 76 1.00 [ref]3 630 1.08 [0.68–1.72] 403 0.83 [0.45–1.55]4 1396 0.98 [0.63–1.55] 990 0.68 [0.37–1.25]5 1885 0.92 [0.58–1.44] 1334 0.63 [0.35–1.16]

Continuous 4051 0.96 [0.85–1.08] 2803 0.85 [0.73–0.99]*

Physical workload Category 1 Working with arms raisedto or above shoulder level

2668 1.00 [ref] 1756 1.00 [ref]2 1061 1.08 [0.89–1.31] 733 1.22 [0.96–1.54]3 438 1.08 [0.81–1.43] 326 1.16 [0.83–1.63]4 218 1.03 [0.70–1.52] 142 1.14 [0.71–1.83]

Continuous 4385 1.03 [0.93–1.14] 2957 1.09 [0.95–1.25]

Working with arms raisedto or above shoulderlevel

Category 1 2961 1.00 [ref] 1980 1.00 [ref]2 990 1.21 [1.01–1.44]* 693 1.22 [0.98–1.52]3 374 1.28 [0.98–1.67] 263 1.42 [1.03–1.96]*

4 177 1.73 [1.18–2.54]** 112 2.08 [1.31–3.28]**

Continuous 4502 1.17 [1.07–1.28]** 3048 1.23 [1.10–1.38]**

a Age and sex were included as confounders in all multivariate models.* p < 0.01.

** p < 0.0007, which is the Bonferroni-corrected threshold based on the number of factors tested (0.01/15).

168 J.O. Christensen, S. Knardahl / PAIN�

151 (2010) 162–173

3.4.4.2. Potentially prognostic factors in subjects with pain. Of theemployees that reported of pain at baseline, 284 subjects reportedof no pain at follow-up (incidence rate 15.5%). Role conflict (OR0.71, p < 0.0007) and commitment to organization (OR 1.31,p < 0.01) were statistically significant predictors of being pain freeat follow-up (analyses not shown).

4. Discussion

This study clearly demonstrated the role of both psychological/social and mechanical factors for the experience of neck pain. Astrength was the comprehensive set of exposures measured. Themost robust predictors were role conflict and empowering leader-ship. These factors were statistically significant in all regressions,with p < 0.0007 in 4 of 5 regressions. Decision control was statisti-cally significant in all analyses. Mechanical factors were notconfounders.

We have not found studies relating role conflict or empoweringleadership to neck pain. Role conflict has been linked to strain [8],well-being [42], and sickness absence [36]. Empowering leadershiphas been reported not to predict sick leave due to MSD [5]. Hence,the current findings represent important new information thatwarrants further investigations.

Working with arms raised is an established risk factor for neckand shoulder pain [18,20,33,48], and was both cross-sectionallyand prospectively related to outcome in this study. Physical work-load was only statistically significant across time, for increasedand constant high levels.

Systematic reviews have concluded that job demands is the psy-chological factor most consistently associated with neck/shoulderpain [2,6,7,37,38,47,49]. The current results partly supported therole of quantitative demands. Decision demand was not a risk fac-tor. This seems to be at odds with the previous research. One com-mon measure of job demands in Karasek’s Job Content

Table 3aProspective analyses: multivariate ordinal logistic regressions with psychological, social, and mechanical factors measured at baseline (T1) and averaged across time [(T1 + T2)/2]as independent variables, and neck pain intensity at follow-up as outcome. Separate regressions were run with predictors entered as categorical and continuous.

Psychological, social, and mechanical factors Confounders included inmultivariate analysesa

Baseline exposure as predictor Average exposure as predictor

N OR 99% CI N OR 99% CI

Quantitative demands Category 1 Fair leadership, social climate 150 1.00 [ref] 98 1.00 [ref]2 589 0.84 [0.51–1.38] 490 0.76 [0.42–1.41]3 932 0.94 [0.59–1.51] 1041 0.90 [0.51–1.61]4 455 0.86 [0.52–1.44] 408 1.00 [0.55–1.86]5 125 1.19 [0.62–2.28] 65 1.40 [0.60–3.28]

Continuous 2251 0.99 [0.85–1.17] 2102 1.10 [0.92–1.32]

Decision demands Category 1 and 2 160 1.00 [ref] 148 1.00 [ref]3 898 0.91 [0.58–1.44] 758 0.90 [0.56–1.47]4 846 0.82 [0.52–1.29] 1012 0.95 [0.60–1.53]5 385 0.93 [0.57–1.54] 249 0.99 [0.56–1.73]

Continuous 2289 0.97 [0.82–1.14] 2167 1.05 [0.87–1.26]

Decision control Category 1 137 1.00 [ref] 74 1.00 [ref]2 605 0.69 [0.42–1.13] 546 0.86 [0.45–1.66]3 886 0.70 [0.43–1.12] 913 0.70 [0.37–1.33]4 485 0.60 [0.36–1.00]* 469 0.64 [0.33–1.26]5 122 0.66 [0.33–1.30] 93 0.61 [0.26–1.43]

Continuous 2235 0.89 [0.76–1.04] 2095 0.82 [0.69–0.98]*

Control over work intensity Category 1 340 1.00 [ref] 282 1.00 [ref]2 431 1.09 [0.74–1.62] 436 1.16 [0.77–1.75]3 479 1.26 [0.86–1.85] 543 1.31 [0.89–1.95]4 620 0.97 [0.67–1.42] 629 1.01 [0.68–1.50]5 511 1.01 [0.69–1.49] 446 0.97 [0.64–1.48]

Continuous 2381 0.98 [0.88–1.09] 2336 0.96 [0.85–1.07]

Role conflict Category 1 477 1.00 [ref] 358 1.00 [ref]2 717 1.01 [0.73–1.40] 1006 1.32 [0.94–1.88]3 980 1.24 [0.91–1.69] 816 1.58 [1.11–2.27]*

4 164 1.48 [0.91–2.39] 156 2.89 [1.72–4.87]**

5 40 2.97 [1.29–6.74]** 15 2.10 [0.48–8.32]Continuous 2378 1.25 [1.08–1.45]** 2351 1.41 [1.19–1.67]**

Role clarity Category 1 and 2 Support from immediatesuperior, fair leadership,commitment

52 1.00 [ref] 38 1.00 [ref]3 188 1.28 [0.54–3.14] 134 3.01 [1.08–8.90]*

4 601 1.56 [0.69–3.67] 656 2.29 [0.88–6.39]5 1332 1.42 [0.63–3.32] 1228 2.33 [0.89–6.49]

Continuous 2173 1.00 [0.83–1.20] 2056 1.04 [0.84–1.29]

Support from immediatesuperior

Category 1 Fair leadership 71 1.00 [ref] 28 1.00 [ref]2 114 1.16 [0.50–2.74] 104 0.96 [0.26–3.68]3 494 1.13 [0.53–2.45] 450 0.98 [0.28–3.66]4 642 0.84 [0.39–1.87] 799 0.73 [0.20–2.77]5 999 1.05 [0.48–2.33] 836 0.82 [0.22–3.13]

Continuous 2320 1.00 [0.84–1.17] 2217 0.99 [0.81–1.22]

a Baseline measures of age, sex, and neck pain (T1) were included as confounders in all multivariate models.* p < 0.01.

** p < 0.0007, which is the Bonferroni-corrected threshold based on the number of factors tested (0.01/15).

J.O. Christensen, S. Knardahl / PAIN�

151 (2010) 162–173 169

Questionnaire (JCQ) [26] incorporates ‘‘intense concentration” [26].This appears to correspond to decision demands of the QPSNordic

(e.g. ‘‘maximum attention”). Job demands of the JCQ also incorpo-rate ‘‘conflicting demands”. This appears to correspond to role con-flict (QPSNordic) (e.g. ‘‘receive opposing requests from two or morepersons”), which was the strongest predictor in our study. Hence,the ‘‘job demands” concept of previous studies may include factorsdiscriminately related to neck pain, which may explain inconsis-tent results.

Systematic reviews have designated decision control an impor-tant, but less consistent, predictor [2,6,7,38,47,49]. The presentstudy supported the role of decision control. Inconsistency is per-haps due to conceptual imprecision. ‘‘Decision latitude” (JCQ)encompasses ‘‘decision authority” and ‘‘skill discretion” [26]. Theformer is closest to decision control of the QPSNordic. The latter in-cludes the items ‘‘learn new things”, ‘‘requires creativity”, ‘‘highskill level”, and ‘‘develop own abilities”. This seems conceptuallyrelated to the less predictive positive challenge (QPSNordic) (e.g.‘‘are your special knowledge and skills useful in your work”, ‘‘isyour work challenging in a positive way”).

Supervisory support has received limited support as a predictorof neck pain [2,38,49]. In this study no statistically significant pro-spective ORs were detected for support from immediate superior.However, support was adjusted for fair leadership, which may beconceptually related [10] or a consequence of support from thesuperior. This is a general concern in multivariate analyses. Thenumber of predictors in each model was limited to avoid overad-justment. However, overadjustment cannot be ruled out. For in-stance, commitment to organization was included in severalanalyses. Commitment could be an outcome of other work factors(e.g. positive challenge) rather than a confounder.

Commitment to the organization was cross-sectionally related topain at both time points. Prospectively, only average commitmentas a continuous variable reached statistical significance. Organiza-tional commitment (QPSNordic) has recently been associated withsubjective health after 3.5 [35] and 4 years [23]. Hence, 2 yearsmay not be sufficient.

Prolonged exposure to demanding work situations with sus-tained high levels of physiological arousal and insufficient recoverymay induce poor health [16]. For decision control, role conflict,

Table 3bProspective analyses: multivariate ordinal logistic regressions with psychological, social, and mechanical factors measured at baseline (T1) and averaged across time [(T1 + T2)/2]as predictors, and neck pain intensity at follow-up as outcome. Separate regressions were run with predictors entered as categorical and continuous.

Psychological, social, and mechanical factors Confounders included inmultivariate analysesa

Baseline exposure as predictor Average exposure as predictor

N OR 99% CI N OR 99% CI

Empowering leadership Category 1 235 1.00 [ref] 168 1.00 [ref]2 363 0.74 [0.48–1.14] 447 0.79 [0.50–1.26]3 866 0.63 [0.43–0.93]* 776 0.71 [0.46–1.10]4 531 0.53 [0.35–0.81]** 665 0.54 [0.34–0.84]**

5 392 0.64 [0.41–0.99]* 310 0.73 [0.44–1.21]Continuous 2387 0.88 [0.79–0.99]* 2366 0.87 [0.77–0.98]*

Fair leadership Category 1 Commitment to organization 43 1.00 [ref] 22 1.00 [ref]2 93 1.26 [0.48–3.39] 78 0.72 [0.21–2.53]3 444 1.06 [0.44–2.57] 348 0.68 [0.22–2.23]4 600 0.91 [0.38–2.20] 872 0.59 [0.19–1.90]5 1052 0.82 [0.35–1.98] 850 0.47 [0.15–1.54]

Continuous 2232 0.89 [0.76–1.03] 2170 0.82 [0.68–0.99]*

Predictability during thenext month

Category 1 and 2 66 1.00 [ref] 36 1.00 [ref]3 191 0.95 [0.44–2.08] 173 0.95 [0.35–2.67]4 612 1.05 [0.53–2.15] 732 0.79 [0.31–2.07]5 1523 0.88 [0.45–1.77] 1419 0.65 [0.26–1.71]

Continuous 2392 0.90 [0.77–1.06] 2360 0.82 [0.68–1.00]*

Commitment toorganization

Category 1 88 1.00 [ref] 42 1.00 [ref]2 170 0.76 [0.38–1.51] 188 1.36 [0.56–3.42]3 577 0.68 [0.38–1.26] 530 0.98 [0.42–2.35]4 697 0.62 [0.34–1.14] 872 0.99 [0.43–2.35]5 741 0.67 [0.37–1.22] 621 0.85 [0.37–2.04]

Continuous 2273 0.91 [0.80–1.04] 2253 0.87 [0.75–1.00]*

Social climate Category 1 and 2 Fair leadership, commitment toorganization

102 1.00 [ref] 67 1.00 [ref]3 531 1.12 [0.61–2.07] 421 1.23 [0.60–2.58]4 783 0.94 [0.51–1.76] 1036 1.00 [0.48–2.12]5 769 0.92 [0.48–1.76] 567 0.92 [0.42–2.05]

Continuous 2185 0.89 [0.73–1.09] 2091 0.85 [0.66–1.10]

Positive challenge Category 1 and 2 Commitment to organization 57 1.00 [ref] 41 1.00 [ref]3 311 0.56 [0.27–1.19] 218 0.86 [0.34–2.18]4 712 0.51 [0.25–1.06] 852 0.64 [0.26–1.58]5 1055 0.48 [0.23–0.99]* 860 0.62 [0.25–1.56]

Continuous 2135 0.90 [0.75–1.07] 1971 0.85 [0.68–1.07]

Physical workload Category 1 Working with arms raised to orabove shoulder level

1390 1.00 [ref] 1416 1.00 [ref]2 611 1.25 [0.94–1.65] 544 1.27 [0.95–1.69]3 229 1.24 [0.81–1.90] 245 1.06 [0.68–1.64]4 120 1.04 [0.58–1.82] 60 1.15 [0.53–2.42]

Continuous 2350 1.12 [0.95–1.32] 2265 1.16 [0.96–1.41]

Working with arms raisedto or above shoulderlevel

Category 1 1539 1.00 [ref] 1680 1.00 [ref]2 563 1.19 [0.91–1.55] 496 1.23 [0.93–1.61]3 201 1.40 [0.95–2.07] 169 1.45 [0.95–2.18]4 91 1.17 [0.66–2.05] 20 1.49 [0.47–4.63]

Continuous 2394 1.12 [0.98–1.29] 2365 1.20 [1.03–1.41]*

a Baseline measures of age, sex, and neck pain (T1) were included as confounders in all multivariate models.* p < 0.01.

** p < 0.0007, which is the Bonferroni-corrected threshold based on the number of factors tested (0.01/15).

170 J.O. Christensen, S. Knardahl / PAIN�

151 (2010) 162–173

support from immediate superior, empowering leadership, fair leader-ship, commitment to organization, social climate and working witharms raised, this was indicated by constant high levels displayingthe strongest ORs among levels of the change/constancy factors.

For some factors the risk associated with constant high expo-sure was comparable to that of increasing exposure. Hence, pro-longed exposure cannot be concluded to have an effect beyondthat of novel exposure at follow-up.

For some factors, decreasing exposure seemed to pose a riskcomparable to that of increasing exposure. This could signifylong-term effects of high risk factors or low protective factors atbaseline. If pain processes were initiated at baseline, mechanismsof chronification could result in heightened pain levels in absenceof prolonged exposure to high risk factors or low protective factors[31]. Most previous studies assessed only baseline exposure. Thus,the current study adds information and strengthens assumptionsof causality by demonstrating that exposure at T1 predicts neckpain at T2 when exposure at T2 is lowered.

4.1. Methodological considerations

The baseline response rate was 51.2%. Possibly not all employ-ees received enough information to motivate them to answer thecomprehensive questionnaire. Self-selection is a threat to internalvalidity if it is related to the exposure and, independently of expo-sure, to the disease/outcome [40]. Prevalences of a range of healthcomplaints, including neck pain, have been shown not to differ be-tween responders and non-responders [17,46]. Prevalences in thepresent study were consistent with those commonly reported[17]. The survey involved a comprehensive range of work andhealth factors apart from those included in the current study. Ad-verse working conditions and illness was not a specific focus. Atbaseline 56.9% of the invited employees responded to one or moreitems and the response rate declined gradually throughout thequestionnaire. Hence, there is little reason to suspect self-selectionto be based on a combination of high exposure and high neck painintensity.

Table 4aProspective analyses: multivariate ordinal logistic regressions with the change in psychological, social, and mechanical factors as predictors and neck painintensity at follow-up as outcome.

Psychological, social, and mechanical factors Categories N ORa 99% CI

Quantitative demands Constant low 488 1.00 [ref]Decrease 463 1.00 [0.70–1.43]Constant middle 509 1.26 [0.89–1.78]Increase 456 1.37 [0.96–1.96]Constant high 359 1.36 [0.93–1.99]

Decision demands Constant low 56 1.00 [ref]Decrease 379 0.56 [0.26–1.20]Constant middle 526 0.74 [0.36–1.57]Increase 341 0.82 [0.39–1.77]Constant high 865 0.77 [0.38–1.60]

Decision control Constant low 459 1.00 [ref]Decrease 457 0.89 [0.63–1.26]Constant middle 454 0.81 [0.57–1.15]Increase 398 0.71 [0.49–1.03]Constant high 364 0.65 [0.44–0.95]*

Control over work intensity Constant low 591 1.00 [ref]Decrease 333 1.30 [0.90–1.89]Constant middle 192 1.29 [0.83–1.99]Increase 295 1.27 [0.86–1.87]Constant high 925 0.95 [0.70–1.28]

Role conflict Constant low 862 1.00 [ref]Decrease 494 1.28 [0.94–1.73]Constant middle 526 1.35 [1.00–1.83]*

Increase 400 1.49 [1.07–2.08]*

Constant high 69 2.16 [1.11–4.15]*

Role clarity Constant low/middle 89 1.00 [ref]Decrease 151 1.10 [0.55–2.21]Increase 160 0.53 [0.26–1.07]Constant high 1962 0.68 [0.39–1.20]

Support from immediate superior Constant low 68 1.00 [ref]Decrease 361 0.88 [0.44–1.76]Constant middle 204 0.76 [0.37–1.58]Increase 344 0.83 [0.42–1.67]Constant high 1324 0.59 [0.31–1.14]

a Baseline age, sex, and neck pain (T1) were included as confounders in all multivariate models.* p < 0.01.

** p < 0.0007, which is the Bonferroni-corrected threshold based on the number of factors tested (0.01/15).

J.O. Christensen, S. Knardahl / PAIN�

151 (2010) 162–173 171

A low response rate increases the possibility of self-selectionbias which may threaten external validity (generalizability) if thosewho do not respond differ from those who do [43]. Participantswere recruited through many organizations in several sectors rep-resenting a wide variety of occupations (see Table 1). Nevertheless,the results cannot be inferred to the general working population ora specific type of work. However, a higher response rate would notallow such generalization. Only random sampling from the wholepopulation of interest allows such general conclusions.

Repeated responders may be resistant to effects of adverseexposure. Attrition analyses revealed that baseline pain predictedfollow-up non-response. Hence, a healthy worker effect cannot beruled out. To investigate this, baseline cross-sectional multivari-ate regressions were rerun separately for baseline-only respond-ers and repeated responders. Six of 15 ORs were stronger forbaseline-only responders. Notably, decision demands and controlover work intensity were significant factors in this group, butnot for repeated responders. This suggests underestimation ofsome effects.

All data were collected by self-report. Reporting bias (e.g. due tonegative affectivity) influencing both exposure and outcome mea-sures may inflate associations (often referred to as common methodbias; CMB) [27,41]. Features of QPSNordic should attenuate reportingbias [10,41]: avoiding negative/positive connotations in responsescales (frequency of occurrence rather than ‘‘satisfied with”), re-versed items, and verbal labels for all response categories. Placing

exposure and outcome measures in different sections of the ques-tionnaire, rated on different scales, and carefully communicatingrespondent anonymity should also reduce CMB [41]. Baseline painadjustment eliminates CMB caused by biases that are stable acrosstime or mediated by stable outcome measures [50]. Nevertheless,CMB cannot be ruled out when interpreting the current results.

All prospective regressions included baseline pain. Since base-line exposure may have caused baseline pain, overadjustment isplausible. The strongest predictor of pain was previous pain (OR5.41). It was related to all baseline exposures except decision de-mands, hence limiting the variation left to uniquely predict fol-low-up neck pain. Therefore, ORs may be underestimated.

A conservative statistical significance level minimizes type I er-ror but may increase type II error. ORs should not be judged on thebasis of statistical significance alone [34]. For example, in Table 4aan OR of 0.59 for constant high support from immediate superiorwas not statistically significant, while an OR of 0.65 for constanthigh decision control was. Some factors displayed fairly consistentresults in one direction even if not always statistically significant(e.g. positive challenge). Furthermore, the lowest category was setas reference for all factors, many of which were negatively skewed.Few observations in the reference category will inflate confidenceintervals. This can be compensated by analyzing continuous pre-dictors (e.g. fair leadership and predictability in cross-sectional anal-yses). However, skewness may have influenced change predictorsin particular, since few workers responded ‘‘constant low” for

Table 4bProspective analyses: multivariate ordinal logistic regressions with the change in psychological, social, and mechanical factors as predictors and neck pain intensity at follow-upas outcome.

Psychological, social, and mechanical factors Categories N ORa 99% CI

Empowering leadership Constant low 330 1.00 [ref]Decrease 487 0.82 [0.56–1.19]Constant middle 423 0.77 [0.52–1.13]Increase 504 0.88 [0.61–1.27]Constant high 622 0.68 [0.47–0.98]*

Fair leadership Constant low 48 1.00 [ref]Decrease 338 1.14 [0.52–2.59]Constant middle 161 0.82 [0.34–1.97]Increase 338 1.05 [0.47–2.38]Constant high 1420 0.72 [0.34–1.57]

Predictability during the next month Constant low/middle 67 1.00 [ref]Decrease 180 1.71 [0.78–3.85]Increase 166 1.17 [0.53–2.66]Constant high 1947 1.07 [0.53–2.21]

Commitment to organization Constant low 112 1.00 [ref]Decrease 378 0.86 [0.49–1.50]Constant middle 273 0.71 [0.39–1.27]Increase 365 0.82 [0.47–1.44]Constant high 1125 0.68 [0.41–1.14]

Social climate Constant low/middle 284 1.00 [ref]Decrease 343 0.89 [0.59–1.34]Increase 339 0.70 [0.46–1.07]Constant high 1335 0.58 [0.41–0.81]**

Positive challenge Constant low/middle 149 1.00 [ref]Decrease 181 1.11 [0.62–1.99]Increase 183 1.20 [0.68–2.14]Constant high 1569 0.79 [0.50–1.25]

Physical workload Constant low 1159 1.00 [ref]Decrease 211 1.30 [0.86–1.94]Increase 205 1.62 [1.08–2.40]*

Constant high 708 1.40 [1.08–1.82]*

Working with arms raised to or above shoulder level Constant low 1317 1.00 [ref]Decrease 242 1.03 [0.70–1.51]Increase 209 1.20 [0.80–1.79]Constant high 597 1.37 [1.05–1.78]*

a Baseline age, sex, and neck pain (T1) were included as confounders in all multivariate models.* p < 0.01.

** p < 0.0007, which is the Bonferroni-corrected threshold based on the number of factors tested (0.01/15).

172 J.O. Christensen, S. Knardahl / PAIN�

151 (2010) 162–173

many factors. Change predictors were not analyzed as continuous.However, because many of these associations were unidirectionaland seemed to approximate linearity, post hoc regressions wereconducted with these variables entered as continuous. Resultsshowed that 10 of 15 factors were statistically significant(p < 0.01).

There was a sex imbalance in our sample. Effects of some psy-chological/social variables may be sex-specific [39]. Therefore, ad-justed prospective multivariate regressions with baseline exposureas predictors were rerun for each sex. These analyses gave no rea-son to suspect sex differences in magnitude of effects.

5. Conclusions

The relevance of previously investigated factors like workingwith arms raised, job demands, and decision control was con-firmed in this study. However, other factors, most notably role con-flict and empowering leadership were more consistent predictorsof neck pain. The range of psychological/social and mechanical fac-tors in this study represents an important potential for interven-tions aimed at reducing health problems among employees.

Conflict of interest

None declared.

Acknowledgements

This study was funded by the Norwegian Research Council. Theauthors have no conflict of interest related to the work in this man-uscript. The authors would like to thank the participating compa-nies for their willingness to participate, and Anne Lene Andersen,Shahrooz Elka, Margrethe Schøning, and Jan S. Emberland for theirassistance in the survey administration.

References

[1] Ahlberg-Hulten GK, Theorell T, Sigala F. Social support, job strain andmusculoskeletal pain among female health care personnel. Scand J WorkEnviron Health 1995;21:435–9.

[2] Ariëns GAM, van Mechelen W, Bongers PM, Bouter LM, van der Wal G.Psychosocial risk factors for neck pain: a systematic review. Am J Ind Med2001;39:180–93.

[3] Ariëns GAM, Bongers PM, Hoogendoorn WE, Houtman IL, van der Wal G, VanMW. High quantitative job demands and low coworker support as risk factorsfor neck pain: results of a prospective cohort study. Spine 2001;26:1896–901.

[4] Babyak MA. Understanding confounding and mediation. Evid Based MentalHealth 2009;12:68–71.

[5] Bergstrom G, Bodin L, Bertilsson H, Jensen IB. Risk factors for new episodes of sickleave due to neck or back pain in a working population. A prospective study withan 18-month and a three-year follow-up. Occup Environ Med 2007;64:279–87.

[6] Bongers PM, IJmker S, van den Heuvel S, Blatter BM. Epidemiology of workrelated neck and upper limb problems: psychosocial and personal risk factors(part I) and effective interventions from a bio behavioural perspective (part II).J Occup Rehabil 2006;16:279–302.

J.O. Christensen, S. Knardahl / PAIN�

151 (2010) 162–173 173

[7] Bongers PM, Kremer AM, ter Laak J. Are psychosocial factors, risk factors forsymptoms and signs of the shoulder, elbow, or hand/wrist?: A review of theepidemiological literature. Am J Ind Med 2002;41:315–42.

[8] Bridger RS, Kilminster S, Slaven G. Occupational stress and strain in the navalservice: 1999 and 2004. Occup Med 2007;57:92–7.

[9] Christenfeld NJS, Sloan RP, Carroll D, Greenland S. Risk factors, confounding,and the illusion of statistical control. Psychosom Med 2004;66:868–75.

[10] Dallner M, Elo A-L, Gamberale F, Hottinen V, Knardahl S, Lindström K, SkogstadA, Ørhede E. Validation of the General Nordic Questionnaire (QPSNordic) forpsychological and social factors at work. Copenhagen: Nordic Council ofMinisters; 2000.

[11] Davis KG, Heaney CA. The relationship between psychosocial workcharacteristics and low back pain: underlying methodological issues. ClinBiomech 2000;15:389–406.

[12] de Croon EM. Studies in occupational epidemiology and the risk ofoveradjustment. Occup Environ Med 2006;63:787.

[13] de Jonge J. Linear and nonlinear relations between psychosocial jobcharacteristics, subjective outcomes, and sickness absence: baseline resultsfrom SMASH. J Occup Health Psychol 2000;5:256–68.

[14] de Jonge J, Schaufeli WB. Job characteristics and employee well-being: a test ofWarr’s vitamin model in health care workers using structural equationmodelling. J Organ Behav 1998;19:387–407.

[15] de Lange AH, Taris TW, Kompier MA, Houtman IL, Bongers PM. ‘‘The very bestof the millennium”: longitudinal research and the demand-control-(support)model. J Occup Health Psychol 2003;8:282–305.

[16] Faucett J. Integrating ‘‘psychosocial” factors into a theoretical model for work-related musculoskeletal disorders. Theor Issues Ergon Sci 2005;6:531–50.

[17] Fejer R, Kyvik KO, Hartvigsen J. The prevalence of neck pain in the worldpopulation: a systematic critical review of the literature. Eur Spine J2006;15:834–48.

[18] Gamperiene M. Work related risk factors for musculoskeletal complaints inthe spinning industry in Lithuania. Occup Environ Med 1999;56:411–6.

[19] Hägg GM. Static work load and occupational myalgia. In: Anderson PA, HobartDJ, Danoff JV, editors. Electromyographical kinesiology. Amsterdam: ElsevierScience Publishers B.V.; 1991. p. 141–4.

[20] Hales TR. Epidemiology of work-related musculoskeletal disorders. OrthopClin N Am 1996;27:679–709.

[21] Hartvigsen J, Lings S, Leboeuf-Yde C, Bakketeig L. Psychosocial factors at workin relation to low back pain and consequences of low back pain; a systematic,critical review of prospective cohort studies. Occup Environ Med 2004;61:e2.

[22] Hoogendoorn WE, van Poppel MN, Bongers PM, Koes BW, Bouter LM.Systematic review of psychosocial factors at work and private life as riskfactors for back pain. Spine 2000;25:2114–25.

[23] Hudek-Knezevic J, Krapic N, Maglica BK. Organizational stress and workattitudes as predictors of health outcomes: a prospective study. DrustvenaIstrazivanja 2009;18:129–49.

[24] Janwantanakul P, Pensri P, Jiamjarasrangsri V, Sinsongsook T. Prevalence ofself-reported musculoskeletal symptoms among office workers. Occup Med2008;58:436–8.

[25] Johansson H, Sojka P. Pathophysiological mechanisms involved in genesis andspread of muscular tension in occupational muscle pain and in chronicmusculoskeletal pain syndromes: a hypothesis. Med Hypotheses1991;35:196–203.

[26] Karasek R, Brisson C, Kawakami N, Houtman I, Bongers P, Amick B. The JobContent Questionnaire (JCQ): an instrument for internationally comparativeassessments of psychosocial job characteristics. J Occup Health Psychol1998;3:322–55.

[27] Kasl SV. Measuring job stressors and studying the health impact of the workenvironment: an epidemiologic commentary. J Occup Health Psychol1998;3:390–401.

[28] Katz MH. Multivariable analysis: a practical guide for clinicians. Cambridge:Cambridge University Press; 2006.

[29] Kleinbaum DG, Klein M. Logistic regression. New York: Springer; 2002.

[30] Knardahl S. Psychophysiological mechanisms of pain in computer work: the bloodvessel–nociceptor interaction hypothesis. Work and Stress 2002;16:179–89.

[31] Knardahl S. Psychological and social factors at work: contribution tomusculoskeletal disorders and disabilities. G Ital Med Lav Ergon2005;27:65–73.

[32] Kompier M, van der Beek AJ. Psychosocial factors at work and musculoskeletaldisorders. Scand J Work Environ Health 2008;34:323–5.

[33] Leclerc A. Incidence of shoulder pain in repetitive work. Occup Environ Med2004;61:39–44.

[34] Lempert R. The significance of statistical significance: two authors restate anincontrovertible caution. Why a book? Law Soc Inq 2009;34:225–49.

[35] Lohela M, Björklund C, Vingard E, Hagberg J, Jensen I. Does a change inpsychosocial work factors lead to a change in employee health? J OccupEnviron Med 2009;51:195–203.

[36] Lund T, Labriola M, Christensen KB, Bültmann U, Villadsen E, Burr H.Psychosocial work environment exposures as risk factors for long-termsickness absence among danish employees: results from DWECS/DREAM. JOccup Environ Med 2005;47:1141–7.

[37] Macfarlane GJ, Pallewatte N, Paudyal P, Blyth FM, Coggon D, Crombez G, LintonS, Leino-Arjas P, Silman AJ, Smeets RJ, van der Windt D. Evaluation of work-related psychosocial factors and regional musculoskeletal pain: results from aEULAR task force. Ann Rheum Dis 2009;68:885–91.

[38] Malchaire J, Cock N, Vergracht S. Review of the factors associated withmusculoskeletal problems in epidemiological studies. Int Arch Occup EnvironHealth 2001;74:79–90.

[39] Östergren PO, Hanson BS, Balogh I, Ektor-Andersen J, Isacsson A, Örbaek P,Winkel J, Isacsson SO. Incidence of shoulder and neck pain in a workingpopulation: effect modification between mechanical and psychosocialexposures at work? Results from a one year follow up of the Malmöshoulder and neck study cohort. J Epidemiol Commun 2005;59:721–8.

[40] Pearce N, Checkoway H, Kriebel D. Bias in occupational epidemiology studies.Occup Environ Med 2007;64:562–8.

[41] Podsakoff PM, MacKenzie SB, Lee JY, Podsakoff NP. Common method biases inbehavioral research: a critical review of the literature and recommendedremedies. J Appl Psychol 2003;88:879–903.

[42] Ron P. Relations between work stressors and well-being among nursingassistants in nursing homes. Aging Clin Exp Res 2008;20:359–67.

[43] Rothman KJ, Greenland S. Modern epidemiology. Philadelphia: LippincottWilliams and Wilkins; 1998.

[44] Steingrimsdottir OA, Vollestad NK, Roe C, Knardahl S. Variation in reporting ofpain and other subjective health complaints in a working population andlimitations of single sample measurements. Pain 2004;110:130–9.

[45] Strøm V, Røe C, Knardahl S. Work-induced pain, trapezius blood flux, andmuscle activity in workers with chronic shoulder and neck pain. Pain2009;144:147–55.

[46] van den Akker M, Buntinx F, Metsemakers JFM, Knottnerus JA. Morbidity inresponders and non-responders in a register-based population survey. FamPract 1998;15:261–3.

[47] van der Windt DAWM, Thomas E, Pope DP, De Winter AF, Macfarlane GJ,Bouter LM, Silman AJ. Occupational risk factors for shoulder pain: a systematicreview. Occup Environ Med 2000;57:433–42.

[48] Viikari-Juntura E, Martikainen R, Luukkonen R, Mutanen P, Takala EP,Riihimaki H. Longitudinal study on work related and individual risk factorsaffecting radiating neck pain. Occup Environ Med 2001;58:345–52.

[49] Walker-Bone KE, Palmer KT, Reading I, Cooper C. Soft-tissue rheumaticdisorders of the neck and upper limb: prevalence and risk factors. SeminArthritis Rheum 2003;33:185–203.

[50] Zapf D, Dormann C, Frese M. Longitudinal studies in organizational stressresearch: a review of the literature with reference to methodological issues. JOccup Health Psychol 1996;1:145–69.

[51] Zhao LP, Kolonel LN. Efficiency loss from categorizing quantitative exposuresinto qualitative exposures in case–control studies. Am J Epidemiol1992;136:464–74.