Full text of DOI:10.1207/S15327558IJBM0704_05

INTERNATIONAL JOURNAL OF BEHAVIORAL MEDICINE, 7(4), 340–352
Copyright © 2000, Lawrence Erlbaum Associates, Inc.
Low Back Pain Patients in Primary Care:
Subgroups Based on the
Multidimensional Pain Inventory
Eva Johansson and Per Lindberg
Classification of patients into homogeneous subgroups is an important objective in
primary care management of low back pain patients. The purpose of this study is to (a)
identify and describe cluster profiles basedonself-reportedmultidimensional pain in-
ventory Scale (MPI) scores among subacute and chronic nonspecific low back pain
patients; (b) describe characteristics of the clusters in relation to disability, life satis-
faction, functional self-efficacy, and exercise self-efficacy; and (c) compare grouping
by clusters based on self-reported MPI scores with grouping by symptom duration.
Eighty-eightindividualsparticipated. Thesehadamedianlowerbackpaindurationof
7
months (range 1 to 144 months). Three clusters were identified; these were labeled
interpersonally distressed, adaptive copers, and dysfunctional. The clusters dif-
fered significantly in disability and functional self-efficacy scores, but not in life sat-
isfaction and exercise self-efficacy scores. The results of this study in a primary care
setting are discussed in relation to previous results in pain clinic settings.
Key words: classification, low back pain, primary care, multidimensional pain inventory,
disability, self-efficacy
Low back pain (LBP) affects about 80% of the general population at some point in
life, and 25% of adults report back pain in a given year (Skovron, 1992). Most LBP
patients are managed in primary care settings (Deyo & Phillips, 1996). Although
LBP typically improves considerably within 1 week after onset, less severe pain of-
Eva Johansson and Per Lindberg, Department of Public Health and Caring Sciences, Section of Car-
ing Sciences, Uppsala Science Park, Uppsala, Sweden.
ThisstudywassupportedbytheSwedishFoundationforHealthCareSciencesandAllergyResearch.
Correspondence concerning this article should be addressed to Eva Johansson, Department of Public
Health and Caring Sciences, Section of Caring Sciences, Uppsala Science Park, S–751 83 Uppsala,
Sweden. E-mail: eva.johansson@ccs.uu.se

MPI SUBGROUPS IN PRIMARY CARE
341
ten continues for 1 to 3 months. Long-term follow-ups (1 year or more) show a re-
current course (van den Hoogen, Koes, van Eijk, Bouter, & Deville, 1998; Von
Korff & Saunders, 1996), and the proportion of primary care patients with LBP
who experience poor recovery may be larger than is generally recognised (Cherkin,
Deyo, Street, & Barlow, 1996; Croft, Macfarlane, Papageorgiou, Thomas, &
Silman, 1998; Deyo & Phillips, 1996).
Although reliable classification systems based on physical factors are needed
for the diagnosis and assessment of individual patients, physical impairment is
only moderately related to disability and even less to pain intensity (Waddell,
1
987). Therefore, it may be argued that classification systems for LBP patients
should include assessment of psychological and behavioural factors as well as
physical factors. This may be especially relevant in primary care settings, where
patients are not as highly selected as in specialised clinics in secondary and tertiary
settings.
Turk and Rudy (1987) developed an empirically derived multivariate classifi-
cation system, the Multiaxial Assessment of Pain (MAP), which aims at inte-
grating physical, psychosocial, and behavioural data for chronic pain patients.
The psychosocial and behavioural components were assessed by the West Ha-
ven–Yale Multidimensional Pain Inventory (MPI; Kerns, Turk, & Rudy, 1985).
In a classification study that used self-reported MPI scores from 217 chronic
pain patients (Turk & Rudy, 1988), the authors identified and replicated three
different clusters. The first cluster, dysfunctional, was characterised by higher
than average levels of pain severity, life interference, and affective distress. The
second cluster, interpersonally distressed, was characterised by low levels of
perceived support from others in their environment. The third cluster, adaptive
copers, was characterised by lower levels of pain severity, life interference, and
affective distress, and greater perceptions of life control. The generality of the
classification was investigated in three different samples from a university out-
patient pain clinic (Turk & Rudy, 1990b): LBP (n = 200), headache (n = 200),
and temporomandibular disorders (TMD; n = 100). The authors reported that
9
5% of the LBP patients, 92% of headache patients, and 94% of TMD patients
could be classified accurately into one of the three patient profile types. Later,
the results were replicated in patients with fibromyalgia (Turk, Okifuji, Sinclair,
&
Starz, 1996).
Clinical utility of the classification system was presented in studies of pa-
tients with TMD (Rudy, Turk, Kubinski, & Zaki, 1995) and fibromyalgia (Turk,
Okifuji, Sinclair, & Starz, 1998) where patients were shown to respond differently
to treatment.
Recently, psychometric data from Dutch (Lousberg et al., 1999) and Swedish
(Bergström, Jensen, Linton, & Nygren, 1999) versions of the MPI have been re-
ported. The results support the reliability and validity of the MPI.

342
JOHANSSON AND LINDBERG
The Turk and Rudy approach offers a comprehensive biopsychosocial classifi-
cation system for LBP patients. Because the system was developed in a tertiary
care setting, the validity and clinical utility in primary care settings need to be in-
vestigated. One issue of interest would be whether classification by self-reported
MPI scores discriminates individuals in important LBP outcome variables such as
disability and life satisfaction and in common predictor variables such as functional
self-efficacy and self-efficacy related to exercise.
Thus, the purpose of this study is to (a) identify and describe cluster profiles
based on self-reported MPI scale scores among subacute and chronic nonspecific
LBP patients in a primary care setting; (b) describe characteristics of the clusters in
relation to disability, life satisfaction, functional self-efficacy, and exercise
self-efficacy; and (c) compare grouping by clusters based on self-reported MPI
scale scores with grouping by symptom duration, in this study subacute (4 weeks
to 3 months) and chronic (more than 3 months) on the same variables.
METHOD
Setting, Participants, and Procedure
ThestudywasconductedinaSwedishuniversitytownwithapopulationof180,000,
and three surrounding rural communities with a total population of 50,000. In the
city,publicservice,administration,andwhite-collarbranchesofbusinessdominate.
In the rural area, various small industries and farming dominate.
Participants in the study were recruited among persons seeking care at the physi-
caltherapydepartmentswithinthecountycouncilprimaryhealthcareorganisation.
About 30% of persons seeking care were self-referred.
The inclusion criteria were age 18 to 65; no signs of trauma; no malignant, in-
fectious, or systemic disease; ability to understand written and spoken Swedish;
and a duration of LBP for at least 4 weeks. Thus, both subacute (4 weeks to 3
months) and chronic (more than 3 months) LBP patients were included.
Participants were recruited consecutively from April 1995 through December
1
996 (n = 40; Sample 1), and from April 1997 through March 1999 (n = 48;
Sample 2). There were no differences in demographic or background data between
the two samples. By using two samples, a larger sample size for the classification
procedure was obtained. The individuals in Sample 2 participated in a randomised
study in which a cognitive–behavioral approach to exercise treatment for LBP
patients was compared to exercise only (Johansson, 1999).
The participants had a mean age of 42 years (SD 13, range 19–65). The me-
dian duration of low back pain was 7 months (25th percentile = 2 months, 75th
percentile = 24 months; range 1 to 144 months). About 27% of the participants
reported having had LBP for less than 3 months and 67% for more than 3
months. Six individuals (6%) did not report data on duration.

MPI SUBGROUPS IN PRIMARY CARE
343
A majority of the participants were women (71%), married or cohabiting
65%), and a large proportion (34%) had university-level education.
(
Measures
Psychosocial and behavioural aspects of chronic pain. Psychosocial
and behavioural aspects of chronic pain were measured by the Swedish version
Carlsson, Bergström, & Jensen, 1994) of the West Haven–Yale MPI (Kerns et
(
al., 1985). The MPI comprises 52 items in nine subscales: (a) reports of pain se-
verity, (b) perceptions of pain interference with life, (c) perceived life control,
(d) affective distress, (e) appraisals of the amount of support from significant
others, (f) perceived punishing responses from significant others to patient dis-
play of pain, (g) perceived solicitous responses from significant others to patient
display of pain, (h) perceived distracting responses from significant others to pa-
tient display of pain, and (i) the frequency of performing common activities, la-
beled a general activities scale. The answering format is a seven-grade (0 to 6)
scale with various anchors depending on the question. A mean score is computed
for each scale. The Swedish version of the MPI has shown satisfactory internal
consistency and test–retest reliability (Bergström et al., 1998).
Disability. Disability was measured by the Roland and Morris Disability
Questionnaire, which consists of 24 items relevant for patients with back problems
(Roland & Morris, 1983). The psychometric properties of the Swedish Roland and
Morris Disability Questionnaire were found to be satisfactory, with a test–retest re-
liability of ICC = .88 (Johansson & Lindberg, 1998).
Perceived satisfaction with key psychosocial life areas. Perceived sat-
isfaction with key psychosocial life areas was measured with the Quality of Life
Scale, which is a seven-item inventory that assesses participants’ degree of satis-
faction with their current level of functioning (Chibnall & Tait, 1990). Test–retest
reliability over an average interval of 75 days was acceptable with a reported r of
0
.73 (Chibnall & Tait, 1990). The Quality of Life Scale was translated into Swedish
by us and tested for internal consistency (α = 0.87) and test–retest reliability (r =
0
.79) in Sample 2 of this study.
Functional self-efficacy. Functional self-efficacy was measured with the
Self-Efficacy Scale, which is an eight-item inventory constructed to assess self-ef-
ficacy expectations specifically related to various basic physical activities
(Estlander, Vanharanta, Moneta, & Kaivanto, 1994; Kaivanto, Estlander, Moneta,

344
JOHANSSON AND LINDBERG
&
Vaharanta, 1995). The Self-Efficacy Scale was translated into Swedish by us and
tested for internal consistency (α = 0.91) and test–retest reliability (r = 0.77) in
Sample 2 of this study.
Exercise self-efficacy. Exercise self-efficacy was measured with the Ex-
ercise Self-Efficacy Scale, which is a six-item inventory constructed to assess
self-efficacy beliefs specifically related to confidence to exercise in the face of
potential barriers: work schedule, physical fatigue, boredom related to exercise,
minor injuries, other time demands, and family and home responsibilities
(Dzewaltowski, 1989; Yordy & Lent, 1993). The Exercise Self-Efficacy Scale
was translated into Swedish by us and tested for internal consistency (α = 0.85)
and test–retest reliability (r = 0.64) in Sample 2 of this study.
Data Analysis
To obtain complete data sets, missing values in occasional items were substituted
by the scale or subscale mean of the individual. This was done for 2 participants in
Sample 1 and for 6 participants in Sample 2.
In the Swedish version of the MPI, participants who live alone are instructed
not to respond to the items making up the subscales punishing responses, solici-
tous responses, and distracting responses. This was the case for 18 individuals in
this sample. These missing scores were substituted by the corresponding group
means for each subscale.
Exploratory data analysis was performed to identify outliers or skewed distri-
butions in the MPI subscales, disability, life satisfaction, functional self-efficacy,
and self-efficacy for exercise measures (Tabachnic & Fidell, 1996). The punishing
responses subscale was significantly and positively skewed, and a potential outlier
was identified in this subscale.
Cluster analysis (Aldenderfer & Blashfield, 1984) was conducted on the MPI
subscale scores for the total sample. The k-means clustering method was used be-
cause it addresses how subsets of participants fit together across measures. Briefly,
this method seeks to create nonoverlapping clusters by minimising within-group
variance and maximising between-group variance (Aldenderfer & Blashfield,
1
984). To determine the optimum number of clusters to retain, the C index method
(Hubert & Levin, 1976) was used. This method examines the within-cluster dis-
tances of different cluster solutions and indicates the solution with the smallest
within-cluster distances.
Discriminant function analysis was used to investigate the degree to which
cases were correctly classified into clusters. Standard multiple regression analysis
was used to predict Roland and Morris disability scores from the MPI subscales

MPI SUBGROUPS IN PRIMARY CARE
345
where clusters differed the most (excluding pain severity and interference to avoid
spurious results due to high correlations between these scale scores and the Roland
and Morris disability score).
Differences between clusters in disability scores (total sample), life satisfac-
tion, functional self-efficacy, and self-efficacy for exercise scores (Sample 2) were
analysed with analysis of variance.
Differences in disability scores (total sample), life satisfaction, functional
self-efficacy, and self-efficacy for exercise scores (Sample 2) between subacute
and chronic individuals, and differences in all variables with regard to gender were
analysed with Student’s t test. The level of significance was set at p < .05.
RESULTS
The mean MPI subscale scores and standard deviations for the total sample are
given in Table 1. The possible range of each subscale score is 0 to 6 points. There
weregenderdifferencesinthepainseverityscaleandinthegeneralactivityscale.
The C index method indicated that a three-cluster solution, among k-means so-
lutions from two to four clusters, represented optimal explanations for the profile
structure of the patients’ reported MPI scores.
The cluster profiles are shown in Figures 1a to 1c. Cluster 1 (34% of the sam-
ple, n = 30) was characterised by low levels of pain and interference along with
low levels of perceived support from others, solicitous responses, and distracting
TABLE 1
Mean Scores and Standard Deviations in the MPI Subscales for the Total Samplea
Menb
Womenc
M
SD
M
SD
MPI Subscale
Pain Severity
Interference
Life Control
Affective Distress
Support
Punishing Responses
Solicitous Responses
Distracting Responses
General Activity
2.8
2.8
3.4
2.3
3.5
.9
2.7
2.1
2.6
a
a
a
a
a
a
a
a
a
.9
1.1
1.0
1.0
1.5
1.0
1.1
1.0
.9
3.4
b
1.1
1.3
1.1
1.5
2.0
.9
1.4
1.0
.8
2.5
3.5
2.5
3.3
.8
2.6
2.0
3.1
a
a
a
a
a
a
a
b
Note. Means in the same row that do not share subscripts differ at p < .05 in Student’s independent
t test.
a
b
c
N = 88. n = 26. n = 62.

346
JOHANSSON AND LINDBERG
FIGURE 1 Mean MPI scale scores in the three clusters. (a) = “Interpersonally Distressed”
cluster (n = 30), (b) = “Adaptive Copers” cluster (n = 39), (c) = “Dysfunctional” cluster (n = 19).
PS = pain severity, I = interference, LC = life control, AD = affective distress, S = support, PR =
punishing responses, SR = solicitous responses, DR = distracting responses, and GA = general
activity level.
responses. This cluster was labeled interpersonally distressed. Cluster 2 (44% of
the sample, n = 39) was characterised by low levels of pain severity and interfer-
ence along with high levels of perceived support and solicitous responses. This
cluster was labeled adaptive copers. Cluster 3 (22% of the sample, n = 19) was
distinguished by high levels of pain severity and interference along with high
levels of perceived support and solicitous responses. This cluster was labeled
dysfunctional. Cluster 1 and 2 both displayed high levels of life control and low

MPI SUBGROUPS IN PRIMARY CARE
347
levels of affective distress, whereas Cluster 3 showed the opposite pattern. All
three clusters displayed low levels of punishing responses and midscale levels of
general activity.
The discrimination of the model was good (Wilks’s λ = 0.11, approximately
F[18, 154] = 17.8, p < .001). Ninety-eight percent of the cases were correctly clas-
sified into one of the three clusters.
There were no significant differences between the clusters regarding the dis-
tribution of participants in demographic variables (gender, age, civil status, and
education).
For the total sample, there were significant mean differences between clusters
in the Roland and Morris disability scores (F[2, 85] = 10.42, p < .001). To further
validate the cluster solution in relation to disability, prediction of Roland and Mor-
ris disability scores by the MPI subscales (excluding pain severity and interfer-
ence) in which the cluster profiles differed the most (i.e., affective distress,
support, and solicitous responses) was performed. Entered as independent vari-
ables in a multiple regression analysis, the scores for the total sample on these
subscales explained 26% of the variance in the Roland and Morris disability scores
(R = 0.51, F[3, 84] = 9.83, p < .001). Only affective distress contributed signifi-
cantly to the prediction of disability scores (β = 1.58, t[84] = 3.94, p < .01). Based
on the regression equation, Roland and Morris disability scores were predicted
from the affective distress, support, and solicitous responses subscale scores of
each cluster. The actual scores in each cluster showed the same pattern as the
scores predicted from the affective distress, support, and solicitous responses
subscale scores (Table 2).
For Sample 2, analyses of variance indicated significant differences between
the clusters in the functional self-efficacy scores (F[2, 45] = 5.28, p < .01), but not
in the life satisfaction scores or the exercise self-efficacy scores (Table 3).
TABLE 2
Individual Subscale Scores for Each Cluster in the Affective Distress,
Support, and Solicitous Responses Subscales, the Predicted Diability Scores,
a
and the Actual Disability Scores in Each Cluster
Interpersonally Distressedb
Adaptive Copersc
Dysfunctionald
MPI subscale
Affective Distress
Support
Solicitous Responses
2.0
1.2
1.5
2.2
4.1
2.9
3.7
5.1
3.9
Predicted disability score
Actual disability score
8.6
8.1
10.2
10.0
13.4
14.5
a
b
c
d
N = 88. n = 30. n = 39. n = 19.

348
JOHANSSON AND LINDBERG
TABLE 3
Mean Scores in Life Satisfaction, Functional Self-Efficacy, and Exercise Self-Efficacy,
a
and Differences Between Clusters in Sample 2
Interpersonally
Distressed
b
Adaptive Copersc
Dysfunctionald
Quality of Life Scale
Self-Efficacy Scale
Exercise Self-Efficacy Scale
35.1
45.0
34.5
a
a
a
34.4
41.6
30.5
a
a
a
30.8
30.5
30.6
a
b
a
Note. Means in the same row that do not share subscripts differ at p < .05 in the Tukey honestly
significance difference comparison.
a
b
c
d
N = 48. n = 15. n = 19. n = 14.
There were no significant differences between subacute and chronic patients
in disability scores (total sample) or in life satisfaction scores or any of the
self-efficacy scores (Sample 2).
DISCUSSION
The results of this study support the presence of subgroups differing in
psychosocial and behavioural aspects among LBP patients in a primary care set-
ting. The three clusters identified in this study show clear similarities with the clus-
ters among chronic LBP patients in a pain clinic previously reported by Turk and
Rudy (Turk & Rudy, 1988; Turk & Rudy, 1990b) but are also dissimilar in some re-
spects. The main difference seems to be the lower and more differentiated levels of
pain severity and life interference in our sample. This, however, seems logical con-
sidering the different settings. It is also consistent with earlier results showing that
pain levels differ significantly between chronic LBP patients seen in pain clinics
and primary care or general practice (Crook, Weir, & Tunks, 1989; Deyo, Bass,
Walsh, Schoenfeld, & Ramamurthy, 1988). Those patients who eventually reach
pain clinics through a referral filtering process are those who show greater disabil-
ityandpsychosocialproblems(Turk&Rudy, 1990a). Anotherdifferenceisthatthe
general activity mean scale score was 3 in all three clusters, whereas the data of
Turk and Rudy (Turk & Rudy, 1990b) indicate mean scale scores ranging from 1.5
to 2 in the general activity scale. This, too, may be explained by the different set-
tings. Following the previous argument (Turk & Rudy, 1990a), those who remain
in primary care management should be less disabled and thus more active. Finally,
all three clusters displayed low levels of perceived punishing responses. In the orig-
inal clustering (Turk & Rudy, 1990b), the interpersonally distressed profile is
clearly higher in this respect than the other two.

MPI SUBGROUPS IN PRIMARY CARE
349
A more specific comparison of the cluster profiles in our sample with those
reported by Turk and Rudy (1990b) suggests that it is the interpersonally dis-
tressed cluster that deviates the most from the original profile even when differ-
ent levels of pain severity, interference, and general activity levels have been
taken into account. However, the characterising feature of the interpersonally
distressed cluster profile (i.e., low levels of perceived support from others, pun-
ishing responses, solicitous responses, and distracting responses) indicate a lack
of interaction, or communication, with significant others. This feature also dis-
tinguishes the interpersonally distressed cluster profile from the other two cluster
profiles in our sample.
Ninety-eight percent of the cases were correctly classified into one of the three
clusters. Because this was a post hoc prediction based on the discriminant func-
tions of the same group, this should only mean that there were few extreme cases
or areas where the classification function was inadequate (Tabachnic & Fidell,
1
996).
The clusters differed significantly in the Roland and Morris disability scores.
This may partly be due to high intercorrelations between the MPI interference
scale scores and the Roland and Morris disability scores. However, the Roland and
Morris disability scores predicted from the affective distress, support, and solici-
tous responses scales showed the same pattern, with the interpersonally distressed
cluster displaying the lowest scores and the dysfunctional cluster displaying the
highest scores. This may imply a general relation between disability and
psychosocial variables.
In Sample 2, the results show that the clusters differed significantly in func-
tional self-efficacy, with the dysfunctional cluster displaying the lowest scores.
These individuals may possibly be more vulnerable to worsening of their back
pain, considering the combination of low perceived life control, low functional
self-efficacy, and high affective distress. Individuals in the interpersonally dis-
tressed and adaptive copers clusters both showed lower levels of pain and disabil-
ity and may be better adjusted considering the higher levels of life control,
functional self-efficacy, and low affective distress.
There were no differences between subacute and chronic participants in any
of the outcome variables. This is in line with earlier results in primary care indi-
cating that pain intensity, perceived health, and daily functioning do not differ in
LBP patients classified as acute, subacute, or chronic (van den Hoogen, Koes,
van Eijk, Bouter, & Deville, 1997). Thus, classification based on behavioural
and psychosocial factors may be a better predictor of important LBP variables
than duration of symptoms.
There are some limitations in this study of importance for the conclusions that
can be drawn from the results. The sample was drawn from LBP patients seeking
care at county council primary care health organisations and is characterised by
being mainly female with a high education level. There are other settings that

350
JOHANSSON AND LINDBERG
could be regarded as primary, for example, private practitioners and occupational
health services, and where different patterns of duration and degree of problems
may be present. Thus, generalisation to other types of primary care settings, and to
other areas that are not characterised by university and other public administration,
cannot be taken for granted.
The sample also displayed a wide range in LBP duration. However, because
correlations with the MPI subscales and the other measures were low, duration of
complaints has probably not affected the results.
Means substitution was performed for 18 participants who had missing scores
the punishing responses, solicitous responses, and distracting responses subscales
due to the instructions regarding significant other in this section of the MPI. The
consequence of means substitution is reduced variance, which may make relations
between variables less distinct (Tabachnic & Fidell, 1996). Considering the clear
differences between the clusters in these subscales, it is unlikely that the results
have been seriously affected. Recently, a modification of the MPI, with a clarifica-
tion of the term significant other has been suggested if the MPI is going to be used
for classification purposes (Okifuji, Turk, & Eveleigh, 1999).
To conclude, the results of this study support the presence of subgroups that dif-
fer in psychosocial and behavioural aspects among nonspecific LBP patients in a
primary care setting. Taken together, these data provide support for the generality
of a classification approach developed by Turk and Rudy (Turk & Rudy, 1987,
1
988; 1990b, 1992) in pain clinic settings. Although some deviations from the
original cluster profiles were found in this primary care sample, these may well be
explained by the different settings. Replication in further primary care samples is
needed to ensure the validity of the cluster structure found in this study.
RERFERENCES
Aldenderfer, M., & Blashfield, R. (1984). Cluster analysis. (Vol. 44). Newbury Park: Sage.
Bergström, G., Jensen, I., Bodin, L., Linton, S., Nygren, L., & Carlsson, S. (1998). Reliability and factor
structure of the Multidimensional Pain Inventory—Swedish Language Version (MPI–S). Pain, 75,
1
01–110.
Bergström, K., Jensen, I., Linton, S., & Nygren, A. (1999). A psychometric evaluation of the Swedish
version of the Multidimensional Pain Inventory (MPI–S): A gender differentiated evaluation. Euro-
pean Journal of Pain, 3, 261–273.
Carlsson, S., Bergström, G., & Jensen, I. (1994). The Multidimensional Pain Inventory: A Swedish ver-
sion. Gothenburg, Sweden: Department of Psychology, University of Gothenburg.
Cherkin, D., Deyo, R., Street, J., & Barlow, W. (1996). Predicting poor outcomes for back pain seen in
primary care using patients’ own criteria. Spine, 21, 2900–2907.
Chibnall, J., & Tait, R. (1990). The Quality of Life Scale: A preliminary study with chronic pain pa-
tients. Psychology and Health, 4, 283–292.
Croft, P., Macfarlane, G., Papageorgiou, A., Thomas, E., & Silman, A. (1998). Outcome of low back
pain in general practice: A prospective study. British Medical Journal, 316, 1356–1359.

MPI SUBGROUPS IN PRIMARY CARE
351
Crook, J., Weir, R., & Tunks, E. (1989). An epidemiological follow-up survey of persistent pain suffer-
ers in a group family practice and specialty pain clinic. Pain, 36, 49–61.
Deyo, R., Bass, J., Walsh, N., Schoenfeld, L., & Ramamurthy, S. (1988). Prognostic variability among
chronic pain patients: implications for study design, interpretation, and reporting. Archives of Physi-
cal Medicine and Rehabilitation, 69, 174–178.
Deyo, R., & Phillips, W. (1996). Low back pain. A primary care challenge. Spine, 21, 2826–2832.
Dzewaltowski, D. (1989). Toward a model of exercise motivation. Journal of Sport & Exercise Psychol-
ogy, 11, 251–269.
Estlander, A.-M., Vanharanta, H., Moneta, G., & Kaivanto, K. (1994). Anthropometric variables,
self-efficacy beliefs, and pain and disability ratings on the isokinetic performance of low back pain
patients. Spine, 19, 941–947.
Hubert, L., & Levin, J. (1976). A general statistical framework for assessing categorical clustering in
free recall. Psychological Bulletin, 83, 1072–1080.
Johansson, E. (1999). Exercise-based physiotherapy management of patients with persistent, non-spe-
cific low back pain. A cognitive-behavioural approach to assessment and treatment in a primary
care setting. Uppsala, Sweden: Uppsala University.
Johansson, E., & Lindberg, P. (1998). Subacute and chronic low back pain. Reliability and validity of a
Swedish version of the Roland and Morris Disability Questionnaire. Scandinavian Journal of Reha-
bilitation Medicine, 30, 139–143.
Kaivanto, K., Estlander, A.-M., Moneta, G., & Vaharanta, H. (1995). Isokinetic performance in low
back pain patients: the predictive power of the self-efficacy scale. Journal of Occupational Rehabil-
itation, 5, 87–99.
Kerns, R., Turk, D., & Rudy, T. (1985). The West Haven–Yale Multidimensional Pain Inventory
(WHYMPI). Pain, 23, 345–356.
Lousberg, R., Van Breukelen, G., Groenman, N., Schmidt, A., Arntz, A., & Winter, F. (1999).
Psychometric properties of the multidimensional pain inventory, Dutch language version
(MPI–DLV). Behaviour Research and Therapy, 37, 167–182.
Okifuji, A., Turk, D., & Eveleigh, D. (1999). Improving the rate of classification of patients with the
multidimensional pain inventory (MPI): Clarifying the meaning of “significant other.” Clinical
Journal of Pain, 15, 290–296.
Roland, M., & Morris, R. (1983). A study of the natural history of back pain: Part I. Development of a re-
liable and sensitive measure of disability in low-back pain. Spine, 8, 141–144.
Rudy, T., Turk, D., Kubinski, J., & Zaki, H. (1995). Differential treatment responses of TMD patients as
a function of psychological characteristics. Pain, 61, 103–112.
Skovron, M. (1992). Epidemiologyoflowbackpain. Baillière’sClinicalRheumatology, 6, 559–573.
Tabachnic, B., &Fidell, L. (1996). Usingmultivariatestatistics. (3rded.). NewYork:HarperCollins.
Turk, D., Okifuji, A., Sinclair, J., & Starz, T. (1996). Pain, disability, and physical functioning in sub-
groups of patients with fibromyalgia. Journal of Rheumatology, 23, 1255–1262.
Turk, D., Okifuji, A., Sinclair, J., & Starz, T. (1998). Differential responses by psychosocial subgroups
of fibromyalgia syndrome patients to an interdisciplinary treatment. Arthritis Care and Research,
1
1, 397–404.
Turk, D., & Rudy, T. (1987). Toward a comprehensive assessment of chronic pain patients. Behaviour
Research and Therapy, 25, 237–249.
Turk, D., & Rudy, T. (1988). Toward an empirically derived taxonomy of chronic pain patients: Integra-
tionofpsychologicalassessmentdata. JournalofConsultingandClinicalPsychology, 56, 233–238.
Turk, D., & Rudy, T. (1990a). Neglected factors in chronic pain treatment outcome studies: Referral pat-
terns, failure to enter treatment and attrition. Pain, 43, 7–25.
Turk, D., & Rudy, T. (1990b). The robustness of an empirically derived taxonomy of chronic pain pa-
tients. Pain, 43, 27–35.

352
JOHANSSON AND LINDBERG
Turk, D., & Rudy, T. (1992). Classification logic and strategies in chronic pain. In D. Turk & R. Melzack
Eds.), Handbook of pain assessment (pp. 409–428). New York: Guilford.
(
van den Hoogen, H., Koes, B., van Eijk, J., Bouter, L., & Deville, W. (1997). Pain and health status of
primary care patients with low back pain. Journal of Family Practice, 44, 187–192.
van den Hoogen, H., Koes, B., van Eijk, J., Bouter, L., & Deville, W. (1998). On the course of low back
painingeneralpractice:aoneyearfollowupstudy. AnnalsoftheRheumaticDiseases, 57, 13–19.
VonKorff, M., &Saunders, K. (1996). Thecourseofbackpaininprimarycare. Spine, 21, 2833–2839.
Waddell, G. (1987). A new clinical model for the treatment of low back pain. Spine, 12, 632–644.
Yordy, G., & Lent, R. (1993). Predicting aerobic exercise participation: social cognitive, reasoned ac-
tion, and planned behavioral models. Journal of Sport & Exercise Psychology, 15, 363–374.