Content uploaded by Magdalena Rzewuska
Author content
All content in this area was uploaded by Magdalena Rzewuska on Oct 18, 2017
Content may be subject to copyright.
Research Paper
Trajectories of acute low back pain: a latent class
growth analysis
Aron S. Downie
a,b,
*, Mark J. Hancock
c
, Magdalena Rzewuska
a
, Christopher M. Williams
a
,
Chung-Wei Christine Lin
a
, Christopher G. Maher
a
Abstract
Characterising the clinical course of back pain by mean pain scores over time may not adequately reflect the complexity of the clinical
course of acute low back pain. We analysed pain scores over 12 weeks for 1585 patients with acute low back pain presenting to
primary care to identify distinct pain trajectory groups and baseline patient characteristics associated with membership ofeach cluster.
This was a secondary analysis of the PACE trial that evaluated paracetamol for acute low back pain. Latent class growth analysis
determined a 5 cluster model, which comprised 567 (35.8%) patients who recovered by week 2 (cluster 1, rapid pain recovery); 543
(34.3%) patients who recovered by week 12 (cluster 2, pain recovery by week 12);222 (14.0%) patients whosepain reduced but did not
recover (cluster 3, incomplete pain recovery); 167 (10.5%) patients whose pain initially decreased but then increased by week 12
(cluster 4, fluctuating pain); and 86 (5.4%) patients who experienced high-level pain for the whole 12 weeks (cluster 5, persistent high
pain). Patients with longer pain duration were more likely to experience delayed recovery or nonrecovery. Belief in greater risk of
persistence was associated with nonrecovery, but not delayed recovery. Higher pain intensity, longer duration, and workers’
compensation were associated with persistent high pain, whereas older age and increased number of episodes were associated with
fluctuating pain. Identification of discrete pain trajectory groups offers the potential to better manage acute low back pain.
Keywords: Latent class growth analysis, Low back pain, Primary care, Trajectory, Clinical course
1. Introduction
Low back pain (LBP) is a major cause of disability
36
and is an
extremely common condition presenting to primary care.
3,28,45
Most guidelines for acute LBP advise a minimal treatment
approach on the premise that the clinical course is typically
favorable.
28
When patients are considered as a group, this
optimistic view seems consistent with the research evidence.
35,41
For example, a systematic review of 15 cohort studies reported
pooled mean pain scores, on a 0 to 100 scale of 52, 23, 12, and 6
at baseline, 6, 12, and 52 weeks follow-up, respectively.
41
However, the review also noted a moderate degree of person-to-
person variability with the SD of pain and disability outcomes
typically approximately 20 points on a 0 to 100 scale at all follow-up
time points.
41
The large variability in outcomes at each time point
suggests that it may be overly simplistic to describe the clinical
course of LBP by only considering the group mean over time.
There has been recent interest in studying the clinical course of
individual patients with LBP.
4,8,14
For example, Dunn et al.
14
studied 342 patients from primary care with back pain over 1 year
using latent class analysis. The method identified 4 clusters that
differed across pain status and impact, disability, and psycholog-
ical characteristics. However, the study provided little detail about
the early pain course and enrolled a mixed cohort of patients who
may not adequately describe the acute pain experience. Recently,
Kongsted et al.
29
studied 1082 patients from primary care by text
messagingusing latent class analysis to reveal clusters of LBP. The
study also enrolled a mixed cohort of patients with more than half
reporting leg pain and a low recovery rate at 12 weeks (;54%)
when compared with other studies.
9,17,18
The authors comment
that more appropriate clustering techniques may exist for
longitudinal data, such as latent class growth analysis (LCGA).
44
Given that the group level course of acute and chronic LBP is quite
different, it is likely that pain trajectory groupings identified in
chronic and mixed duration samples are not the same as those in
acute LBP. To date, no research has studied individual pain
trajectories exclusively in acute LBP using data from a large
inception cohort. Understanding the patient traits that are
associated with each pain trajectory group could advance
understanding of the mechanism(s) underlying the trajectories
and also assist in developing targeted therapies for each trajectory.
The aim of this study was to identify pain trajectory groups for
patients with acute LBP presenting to primary care and then to
explore individual characteristics that influenced group member-
ship. We hypothesized that for each trajectory group identified,
membership would be associated with clinical and demographic
characteristics measured at baseline.
2. Methods
Our interest was in the trajectories of patients’ pain during the
course of an episode of acute LBP. We used statistical procedures
Sponsorships or competing interests that may be relevant to content are disclosed
at the end of this article.
a
George Institute for Global Health, University of Sydney, Sydney, Australia,
b
Department of Chiropractic, Faculty of Science and Engineering, Macquarie
University, Sydney, Australia,
c
Department of Health Professions, Faculty of
Medicine and Health Sciences, Macquarie University, Sydney, Australia
*Corresponding author. Address: George Institute for Global Health, GPO Box
5389, Sydney NSW 2001, Australia. Tel.: 161 2 9657 0382; fax: 161 2 9657 0301.
E-mail address: adownie@georgeinstitute.org.au (A. S. Downie).
PAIN 157 (2016) 225–234
©2015 International Association for the Study of Pain
http://dx.doi.org/10.1097/j.pain.0000000000000351
January 2016·Volume 157 ·Number 1 www.painjournalonline.com 225
Copyright
!
2015 by the International Association for the Study of Pain. Unauthorized reproduction of this article is prohibited.
to (1) identify these trajectories and (2) characterise membership of
each trajectory using baseline patient characteristics. The data
were taken from a previous trial.
52
2.1. Data source
This is a secondary analysis of the PACE trial,
52
a placebo-
controlled trial of paracetamol for acute LBP. In brief, 235 primary
care clinicians screened 4606 consecutive patients who sought
care for an episode of acute LBP (,6 weeks duration and of at
least moderate-intensity pain as defined by item 7 of the Short
Form Health Survey, SF-36). A total of 1652 subjects were
recruited from across Sydney, Australia, during 2009 to 2012,
and pain outcomes were measured regularly over the following 12
weeks. There was no difference found in pain (or other) outcomes
for those receiving paracetamol vs placebo. Baseline information
was recorded by research staff through telephone interview and
occurred within 48 hours after the initial primary care consultation,
before study entry and randomization. Follow-up data at 1, 2, 4,
and 12 weeks were contemporaneously recorded into a booklet
by the subject and then transcribed through a telephone interview
with research staff or directly into an online database by the
subject.
52
A detailed description of the trial methods can be found
in the PACE study protocol
50
and statistical analysis plan.
51
2.2. Analysed participant characteristics
Our primary outcome was pain intensity numerical rating scale
(NRS) values, which were collected at 1, 2, 4, and 12 weeks. In
addition, baseline demographic (age, sex, and compensation
status) and clinical data (pain level, duration and referral beyond
knee, number of previous episodes, days of reduced activity,
perceived risk of persistence, sleep quality, quality of life, and
disability) were extracted from the PACE data set.
2.3. Statistical analysis
The analysis assumed that there exist a number of distinct and
unobserved trajectory groups for acute LBP that are revealed by
the patterns of recorded pain level data across 5 time points. The
data analysis comprised 3 steps. First, a LCGA model for ordinal
data was established to identify the minimum number of clusters
that explained variation in the data (step 1). Second, each subject
was assigned to one of the identified clusters (step 2).
25,38
Finally,
multinomial logistic regression was applied to assess the
relationship between cluster membership and baseline charac-
teristics (step 3).
37
Internal (bootstrap) validation of the regression
model was undertaken.
42
2.3.1. Latent class growth analysis
Pain course was modeled over time using pain scores at 5 time
points (baseline, week 1, 2, 4, and 12). A requirement of LCGA is
normal distribution of the dependent variable.
6
Pain score
distribution was non-normal at each time point. The solution was
to trichotomize pain scores analogous to the method described by
others,
13,14,43
with cut points used previously.
19,26,48,50
This
resulted in “pain recovery” (NRS #1); “low–moderate pain” (NRS
2-4); and “high pain” (NRS $5) strata. Latent class growth analysis
uses maximum likelihoodto assign subjects to a cluster, so missing
values are handled without need for imputation. We excluded
subjects missing greater than 2 of the 5 time points. Up to an
8-cluster solution was considered, with the optimum number of
clusters determined using goodness of fit criteria, namely, Akaike’s
information criterion (AIC); Bayesian information criterion (BIC); and
the bootstrap likelihood ratio test.
25,39
Pragmatic consideration
included high entropy
7
and high average posterior probability of
belonging to each cluster
24
; a minimum cluster size of 5%; and
a distinctive pain course for each trajectory.
39
2.3.2. Multinomial logistic regression analysis
After pain trajectories and subject membership were determined,
multivariate multinomial logistic regression analysis was used to
determine the characteristics of each cluster with the strength of the
relationship determined using proportional classification for all
subjects.
2,47
The risk of belonging to each cluster for a given
characteristic was compared to the base cluster (defined by the
cluster with most rapid recovery) and expressed as a relative risk
ratio (RRR).
16
Baseline factors considered important to include in
the multivariate regression were those hypothesised to influence
cluster membership and were selected a priori (Table 1). Contin-
uous baseline variables were expressed as risk per SD increase.
Non-normally distributed count and continuous variables were
dichotomized. The variables’ duration of current episode,number of
previous episodes,andquality of sleep were dichotomized using
amediansplitasothermethodsusingoptimalcutpointsare
recommended against.
1
Testing for collinearity was performed with
tolerance of ,0.1 and variance inflation factor (VIF) #0.2 or $5.0
indicating possible collinearity.
40
Full and reduced unconditional
model predictive efficiency was determined by goodness of fit
indices, mainly AIC,
34
with the lowest score (AIC
min
)indicatingthe
better fitting model. Reduced models excluded all combinations of
the 5 least significant variables found on univariate analysis.
The robustness of the predictors in the multinomial logistic
regression model was tested using Monte Carlo modelling (non-
naive bootstrap).
30,42
The study population parameters (estimates
of relative risk per cluster pair) were compared to 1000 bootstrap
replications of the study data.
21
The threshold for nonignorable
bias was set at 0.25 3SE of the bootstrap sample.
15,21
STATA
version 13.1 for Mac OS (Stata Corp, College Station, TX) was used
to analyse model collinearity, and Mplus version 7.31 (Muth´en &
Muth ´en, Los Angeles, CA) was used to generate the LCGA,
multivariate regression, and bootstrap validation models.
3. Results
3.1. Sample characteristics
Overall, of the 1652 individuals enrolled in PACE, 4.1% (n 567)
were excluded from this study on the basis of missing greater than
2 of 5 pain scores. Pain scores were then available for the
remaining 95.9% (n 51585) of subjects at baseline. Of these,
98.9% (n 51568), 99.1% (n 51570), 97.4% (n 51543), and
96.0% (n 51522) were recorded for weeks 1, 2, 4, and 12,
respectively. Table 2 displays baseline characteristics for included
subjects stratified by cluster for the final 5-cluster LCGA model.
3.2. Latent class growth analysis (selection of ideal number
of clusters)
The 3-cluster to 8-cluster models were tested using goodness of
fit indices to determine the best model and are displayed in
Table 3. The rate of change in log likelihood diminished rapidly
after the 5-cluster model (22LL difference 5498.9, 87.7, 121.5,
44.5, 25.6, and 23.6 for 3-8 cluster models, respectively),
whereas the bootstrap likelihood ratio test indicated up to an
8-cluster model was possible (Pvalue approached nonsignifi-
cance, P50.033). Fit statistics reached minima for the 8-cluster
226 A.S. Downie et al.·157 (2016) 225–234 PAIN
®
Copyright
!
2015 by the International Association for the Study of Pain. Unauthorized reproduction of this article is prohibited.
(AIC
min
511,488) and 6-cluster (BIC
min
511,651) models,
respectively. Entropy was adequate for all solutions (0.79, 0.79,
0.78, 0.75, 0.77, and 0.77 for 3-cluster to 8-cluster models,
respectively). Given these results, the 4-cluster to 6-cluster
models warranted further (pragmatic) evaluation.
25
For each of the 4-cluster to 6-cluster models, the lowest
average posterior probabilities were 0.86, 0.82, and 0.74, with
minimum cluster membership of 11.6%, 5.4%, and 4.2%,
respectively. Table 4 details descriptors for the 4-cluster to 6-
cluster models to aid in assessment of model distinctiveness. All
models displayed 1 stable rapid recovery cluster. There was
variation in the proportion of delayed, incomplete, persistent high,
and fluctuating pain recovery clusters with decreasing member-
ship at higher levels of pain persistence as successive clusters
were added. For example, the 5-cluster model was the first to
identify persistent high pain with 86 subjects, a potentially
important subgroup. Beyond the 5-cluster model, no increase
in clinically distinct subgroups was identified. For example, the
6-cluster model divided the number of incomplete recovery
clusters into 2, which added unnecessary complexity. Therefore,
the 5-cluster model was chosen and satisfied efficiency,
parsimony, and its ability to identify 5 discrete pain courses.
The 5-cluster model comprised cluster 1, rapid pain recovery
(n 5567; 35.8%); cluster 2, pain recovery by week 12 (n 5543;
34.3%); cluster 3, incomplete pain recovery (n 5222; 14.0%);
cluster 4, fluctuating pain (n 5167; 10.5%); and cluster 5, persistent
high pain (n 586; 5.4%). Figure 1 displays the proportion of
subjects reporting none, low–moderate, or high pain at each time
point for each cluster in the 5-cluster LCGA model. Figure 2
displays the mean pain course for the 5-cluster LCGA model.
Cluster 1 (rapid pain recovery) was characterized by an early
rapid decrease in the proportion of subjects who reported any
pain over the first 4 weeks (39%, 0%, and 0% at weeks 1, 2, and
4). At week 12, all subjects reported pain recovery. Subjects in
cluster 1 reported a large decrease in their mean pain score over
the first 2 weeks (25.7; SD 51.5).
Cluster 2 (pain recovery by week 12) was characterized by
a decrease in the proportion of subjects who reported any pain
over the first 4 weeks (93%, 83%, and 33% at weeks 1, 2, and 4).
By week 12, all subjects had pain recovery. Subjects in cluster 2
Table 2
Baseline characteristics for each cluster in the 5-cluster model.
All: clusters
combined
(n 51585)
Cluster 1: rapid
recovery
(n 5567)
Cluster 2: recovery
by week 12
(n 5543)
Cluster 3:
incomplete recovery
(n 5222)
Cluster 4:
fluctuating pain
(n 5167)
Cluster 5: persistent
high pain
(n 586)
Taking paracetamol (%) 1056 (66.6) 366/567 (64.6) 365/543 (67.2) 155/222 (69.8) 111/167 (66.5) 59/86 (68.6)
Age (y) 45.15 (15.76) 43.48 (14.94) 44.94 (16.20) 45.21 (16.21) 50.07 (15.49) 47.67 (15.47)
Female (%) 741 (46.7) 260/563 (46.2) 243/542 (44.8) 113/222 (50.1) 83/167 (49.7) 42/85 (49.4)
Compensable (%) 111 (7.0) 33/566 (5.8) 43/541 (7.9) 13/222 (5.9) 11/167 (6.6) 11/85 (12.9)
Baseline pain 6.25 (1.89) 5.93 (1.99) 6.28 (1.86) 7.11 (1.39) 5.72 (1.93) 6.90 (1.51)
Back pain days 9.84 (9.92) 7.06 (8.15) 9.55 (9.69) 12.87 (10.35) 13.15 (11.5) 15.73 (11.2)
Pain beyond knee (%) 309 (19.5) 74/567 (13.1) 104/541 (19.2) 69/221 (31.1) 36/165 (21.8) 26/86 (30.2)
Previous episodes 6.85 (15.43) 5.52 (13.13) 6.17 (13.44) 7.36 (16.85) 10.07 (15.86) 11.83 (25.89)
Days of reduced activity 3.60 (5.9) 2.36 (3.62) 3.67 (5.55) 5.5 (8.03) 4.51 (7.74) 4.97 (7.93)
Risk of persistence 4.50 (2.7) 3.83 (2.77) 4.34 (2.71) 5.74 (2.53) 5.30 (2.61) 6.05 (2.47)
Sleep quality 1.50 (0.8) 1.43 (0.77) 1.49 (0.78) 1.71 (0.79) 1.56 (0.82) 1.76 (0.84)
Quality of life—physical 42.56 (10.14) 45.37 (9.74) 42.41 (9.7) 38.77 (10.38) 39.5 (9.27) 36.73 (10.01)
Quality of life—mental 44.56 (7.95) 46.06 (7.45) 45.09 (7.85) 42.03 (8.45) 43.33 (7.42) 39.47 (8.01)
Disability 13.03 (5.48) 12.43 (5.64) 13.14 (5.36) 14.25 (5.3) 12.63 (5.48) 14.08 (5.12)
Data are mean (SD) or number (%).
Table 1
Baseline predictor variables considered for multinomial logistic regression.
Baseline measure Mean (SD) or number (%) Recoded for regression analysis
Taking paracetamol 1056/1585 (66.6%)
Age 45.15 (15.75)
Female 741/1585 (46.80%)
Currently on workers’ compensation 111/1585 (7.00%)
Baseline pain (NRS) 6.25 (1.89)
Duration of low back pain at baseline (d) 9.84 (9.92) 0 if score #5; 1 if score .5 (median split)
Pain referral beyond knee 309 (19.5%)
Number of previous episodes 6.85 (15.14) 0 if score #1; 1 if score .1 (median split)
Risk that pain may become persistent 4.55 (2.79)
Quality of sleep 1.52 (0.79) 0 if score 0 and 1; 1 if score 2 and 3 (median split)
SF12v2 (PCS) 42.43 (10.15) 100-score
SF12v2 (MCS) 44.57 (7.95) 100-score
Duration of reduced activity (d) 3.62 (5.93)
RMDQ 13.04 (5.48)
Taking paracetamol—as required dosing or time contingent group. Baseline pain—numerical rating scale (NRS), scored from 0 (no pain) to 10 (worst possible pain). Risk of persistence “In your view, how large is the risk that
your current pain may become persistent?” scored from 0 (no risk) to 10 (very large risk). Sleep quality “During the past week, how would you rate your sleep quality overall?”, item 6 Pittsburgh sleep quality index, scored 0 (very
good), 1 (fairly good), 2 (fairly bad), or 3 (very bad). Quality of Life—physical (PCS) and mental (MCS) components from SF12v2; population mean 550 and SD 510. Duration (days) of reduced activity “In relation to the current
episode, on how many days did back or leg pain force you to cut down on the things you usually do, for more than half a day?” Disability—Roland Morris Disability Questionnaire (RMDQ), scored from 0 (no disability) to 24 (high
disability).
January 2016·Volume 157 ·Number 1 www.painjournalonline.com 227
Copyright
!
2015 by the International Association for the Study of Pain. Unauthorized reproduction of this article is prohibited.
reported a moderate decrease in mean pain over the first 2 weeks
(23.3; SD 50.1).
Cluster 3 (incomplete pain recovery)wascharacterizedby
adecreaseintheproportionofsubjectswhoreportedhighpain
over 12 weeks (95%, 82%, 38%, and 4% at weeks 1, 2, 4, and 12)
with a reciprocal increase in the proportion of low–moderate pain
over the same period (5%, 18%, 61%, and 77%). At week 12, the
LCGA model classified 19% (n 541) of subjects who reported no
pain at week 12 into this cluster. This was due to classification based
on subjects’ trajectory regardless ofrecoverystatus(unconditional
LCGA model). Subjects in cluster 3 reported only a small decrease in
mean pain over the first 2 weeks (21.5; SD 50.3).
Cluster 4 (fluctuating pain) was characterized by a diminishing
proportion of subjects who reported any pain over the first 4 weeks
(93%, 80%, and 66% at weeks 1, 2, and 4), followed by all subjects
reporting increased pain at week 12. At week 12, no subject had
pain recovery. Mean pain course over the first 4 weeks for subjects
classified with fluctuating pain was similar to those classified with
pain recovery by week 12, but with an increase in mean pain score
between week 4 and week 12 (1.9; SD 50.5).
Cluster 5 (persistent high pain) was comprised entirely subjects
who reported high pain over the first 4 weeks (98%, 100%, and
100% at weeks 1, 2, and 4). At week 12, all subjects continued to
report high pain. Subjects in cluster 5 reported negligible change
in mean pain scores over the first 2 weeks (0.0; SD 50.2), with
only marginal decrease in mean pain over the whole study period
(20.4; SD 50.2).
3.3. Multinomial logistic regression
Multivariate multinomial logistic regression models compared
baseline characteristics of all subjects between clusters, with
rapid pain recovery (cluster 1) as a reference. There was low
chance of collinearity (tolerance range 50.62-0.96; mean VIF 5
1.28) when all variables were included. The AIC statistic that
evaluated model best fit achieved a minimum value after removal
of the variables “trial allocation,” “quality of sleep,” and “disability”
(AIC
min
52894) compared with the full model (AIC 52913).
Change in regression coefficients between the reduced and full
models ranged from 0% (age) to 5.5% (quality of life physical
subscale). Given the low chance of collinearity and marginal
difference in magnitude of regression coefficients between full
and reduced models, we were able to retain all a priori
hypothesized predictors. The final model included baseline
measures of age, sex, compensation status, pain (intensity,
duration, number of previous episodes, and pain beyond the
knee), duration of reduced activity, perceived risk of persistence,
quality of sleep, quality of life, disability, and trial group allocation.
Internal validation of the regression model, which used the
difference in parameter estimates (bias) between the study
sample and average of 1000 bootstrap samples for all covariates
in each cluster pair, was of lesser magnitude than the threshold
set for nonignorable bias (equal to 0.25 3SE of bootstrap
data).
15,21
Compared to the study sample, the risk ratio for all
variables in the bootstrap sample was in the same direction and of
similar magnitude. Bootstrapping had the effect of changing
some marginally nonsignificant variables in the original analyses
to significant in the bootstrapped analyses but did not result in
important changes in the magnitude of the relative risk. For
example, when measuring relative risk of pain recovery by week
12 compared to rapid pain recovery, the relative risk for the
variable “previous episodes” changed from 1.29 (95% confi-
dence interval [CI], 0.89-1.86) in the original analyses to 1.30
(95% CI, 1.02-1.67) in the bootstrapped analyses.
3.4. Risk of belonging to delayed pain recovery clusters
compared with rapid pain recovery
Figure 3 reveals the relative risk of belonging to each cluster
compared to rapid pain recovery.
Relative risk ratios for continuous variables are expressed per
SD increase in the predictor variable.
Table 3
Goodness of fit criteria for 3-cluster to 8-cluster models. Shaded region indicates models considered for pragmatic evaluation.
Goodness of fit criteria Model
3-cluster 4-cluster 5-cluster 6-cluster 7-cluster 8-cluster
Sequential model comparison (K11 cluster vs
K2cluster)
3 vs 2 4 vs 3 5 vs 4 6 vs 5 7 vs 6 8 vs 7
LL (K11) 25859.5 25815.6 25759.1 25736.9 25724.1 25712.3
22LL difference 498.9 87.7 121.5 44.5 25.6 23.6
BLRT 22LL diff (Pvalue) — ,0.001 ,0.001 ,0.001 ,0.001 0.033
Information criterion
AIC 11743.0 11663.3 11558.3 11521.8 11504.2 11488.7
BIC 11807.4 11749.2 11665.7 11650.7 11654.5 11660.5
Entropy 0.789 0.789 0.777 0.751 0.771 0.772
22LL difference, change in LL as each additional cluster is added; AIC, Akaike’s information criterion; BIC, Bayesian information criterion; BLRT 22LL diff (Pvalue), bootstrap likelihood ratio test for (k21) vs k-class model
(P.0.05 suggests no further improvement in model fit by adding another cluster); Entropy, the ability of the model to provide well-separated clusters (range: 0-1 with higher values superior); LL, log-likelihood.
Table 4
Trajectory descriptors for the 4-cluster to 6-cluster solutions.
4-cluster model* 5-cluster model 6-cluster model
Cluster 1 36% rapid pain recovery 36% rapid pain recovery 36% rapid pain recovery
Cluster 2 39% pain recovery by week 12 34% pain recovery by week 12 20% pain recovery by week 4
Cluster 3 14% persistent pain 14% incomplete recovery 18% incomplete recovery type I
Cluster 4 12% fluctuating pain 11% fluctuating pain 17% incomplete recovery type II
Cluster 5 5% persistent high pain 5% persistent high pain
Cluster 6 4% fluctuating pain
* Cluster (column) membership does not sum to 100% because of rounding.
228 A.S. Downie et al.·157 (2016) 225–234 PAIN
®
Copyright
!
2015 by the International Association for the Study of Pain. Unauthorized reproduction of this article is prohibited.
3.4.1. Comparison of pain recovery by week 12 (cluster 2)
with rapid pain recovery (cluster 1)
Subjects with greater pain duration were 1.81 times more likely
(95% CI, 1.17-2.78) to recover by week 12 than have rapid pain
recovery. Subjects on workers’ compensation were 2.06 times
more likely (95% CI, 1.03-4.14) to recover by week 12 than have
rapid pain recovery.
3.4.2. Comparison of incomplete recovery (cluster 3) with
rapid pain recovery (cluster 1)
Subjects with higher baseline pain (RRR 52.00; 95% CI, 1.45-
2.85), of greater pain duration (RRR 53.41; 95% CI, 1.94-5.97),
increase in perceived risk of persistence (RRR 51.53; 95% CI,
1.16-2.07) and decrease in quality of life (mental, MCS) (RRR 5
1.44; 95% CI, 1.09-1.94), were more likely to have incomplete
pain recovery than have rapid pain recovery.
3.4.3. Comparison of fluctuating pain (cluster 4) with rapid
pain recovery (cluster 1)
Subjects of older age (RRR 51.56; 95% CI, 1.15-2.20), with
greater pain duration (RRR 52.68; 95% CI, 1.36-5.30), who had
multiple previous episodes (RRR 52.75; 95% CI, 1.43-5.31),
who believed they were at greater risk of persistence (RRR 5
1.74; 95% CI, 1.20-2.63), and who had lower physical (PCS)
(RRR 51.98; 95% CI, 1.31-3.13) and MCS (RRR 51.40; 95% CI,
1.02-1.99) quality of life scores were more likely to experience
fluctuating pain than have rapid pain recovery.
3.4.4. Comparison of persistent high pain (cluster 5) with
rapid pain recovery (cluster 1)
Subjects who were on workers’ compensation (RRR 53.84;
95% CI, 1.41-10.43), with higher baseline pain (RRR 51.88;
95% CI, 1.29-2.84), who had greater pain duration (RRR 5
4.89; 95% CI, 2.25-10.62), who believed that they were at
greater risk of persistence (RRR 51.66; 95% CI, 1.12-2.54),
and who had lower PCS (RRR 52.31; 95% CI, 1.46-3.83) and
MCS (RRR 52.06; 95% CI, 1.43-3.08) quality of life scores
were more likely to experience persistent high pain than have
rapid pain recovery.
4. Discussion
4.1. Summary of main findings
We identified 5 distinct pain trajectory patterns for patients with
acute nonspecific LBP presenting to primary care. Two of the
identified clusters included subjects who recovered from their
pain by 12 weeks but differed substantially in the rate at which
pain diminished and time of pain recovery. One cluster
comprised subjects whose pain diminished but did not recover
Figure 1. Proportion of subjects reporting pain recovery, low–moderate pain, or high pain at each time point for each of the 5 clusters.
January 2016·Volume 157 ·Number 1 www.painjournalonline.com 229
Copyright
!
2015 by the International Association for the Study of Pain. Unauthorized reproduction of this article is prohibited.
Figure 2. Mean pain course for each of the 5 trajectory groups in the 5-cluster latent class growth analysis model. Error bars are cluster SDs. NRS, numerical rating
scale.
230 A.S. Downie et al.·157 (2016) 225–234 PAIN
®
Copyright
!
2015 by the International Association for the Study of Pain. Unauthorized reproduction of this article is prohibited.
Figure 3. Risk profile for each cluster compared to cluster 1 (rapid pain recovery). Plot of average pain over time for investigated cluster vs reference cluster is
provided to assist with interpretation of risk. Regression coefficients at .95% significance are coloured blue. *RRR per SD increase. †Quality of life reverse scored
(100-score), ie, greater risk with lower quality of life. RRR, relative risk ratio; CI, confidence interval.
January 2016·Volume 157 ·Number 1 www.painjournalonline.com 231
Copyright
!
2015 by the International Association for the Study of Pain. Unauthorized reproduction of this article is prohibited.
by 12 weeks, another cluster comprised subjects with
fluctuating pain, and the last contained subjects who experi-
enced persistent high-level pain over the study period.
Characteristics for each cluster identified were reported and
compared to the rapid pain recovery cluster. Compared to
rapid pain recovery, we found subjects who had pain duration
of greater than 5 days experienced delayed recovery or
nonrecovery by 12 weeks. Similarly, belief in greater risk of
persistence was associated with a lack of pain recovery at 12
weeks compared to rapid pain recovery.Higherpainintensity,
longer duration, and workers’ compensation were strongly
associated with persistent high pain, whereas older age and
increased number of episodes were uniquely associated with
fluctuating pain when compared to rapid pain recovery.
4.2. Strengths and limitations of the study
The key strengths of our study are the large inception cohort (n 5
1585, drawn from original sample of 1652) recruited from primary
care for a new episode of LBP,
52
with substantially complete data
($96%) for 5 time points over 3 months. Robust methodology
was used to determine distinct pain trajectory groups with
membership. We a priori selected predictors that we believed
likely to be important to the pain course, and the large sample size
allowed for inclusion of multiple predictors in the multivariate
regression model without overfitting.
The low number of subjects on workers’ compensation (n 5
102, 7%) influenced the CI for this predictor, but the prevalence
was similar to other studies investigating patients with acute LBP
presenting to primary care.
10,12,22,46
Although the data originate
from a clinical trial, there was no effect of treatment on time to
recovery within the trial, and allocation to active or placebo did not
discriminate membership between trajectory groups. Monitoring
beyond 12 weeks may have helped further define pain trajectories
(eg, place more subjects into the fluctuating trajectory group if
monitored over a longer period), but we were primarily concerned
with exploring latent trajectories for people with acute LBP and
believe this information is critical to patients and clinicians.
Patterns of recovery identified in other studies that followed
patients for 1 year
5,14
were also identified in our study (rapid
recovery, delayed recovery, fluctuation pain, and persistent high
pain), suggesting that patterns of persistent pain behavior may be
apparent early in the recovery from acute LBP.
This study does not seek to provide prediction of recovery
given the study design, but does profile the patient who is more
likely to follow one of the identified 5 pain courses. For example,
rapid pain recovery membership was associated with episode
duration (patients with pain greater than 5 days were nearly twice
as likely to recover at 12 weeks vs 2 weeks). Similarly, fluctuating
pain membership was associated with older age when compared
to rapid pain recovery, but older age was not associated with
incomplete pain recovery or persistent high pain trajectory
patterns when compared to rapid pain recovery.
4.3. Strengths and weaknesses in relation to other studies,
discussing important differences in results
This is the first large study to explore pain trajectories for patients
with only acute LBP. It extends the previous work in mixed
cohorts of patients with acute and chronic LBP by showing that
distinct pain trajectory groups are apparent early in the course of
the condition. Information on individual pain course is infrequently
studied
35
but may be a key to understanding recovery when used
as an element within a prognostic framework.
11
Our study goes beyond previous studies,
4,8,14,27,31,43
which
have noted different trajectory groupings and differences in
membership, by identifying patient characteristics at baseline
that associate with pain course compared to ideal (rapid) pain
recovery. The large enrollment provided a base category of
more than 500 patients who recovered by week 2 to represent
rapid pain recovery. We have avoided the problems of previous
studies in this area by enrolling an inception cohort rather than
sampling prevalent cases
33
and by keeping loss to follow-up to
less than 5%, whereas in other studies, loss to follow-up has
been as high as 54%.
A number of psychosocial factors have been previously found
to be associated with pain trajectory groups (eg, anxiety,
14,31,43
depression,
14,43
and sick leave
14
). In our study, only a few
psychosocial factors were available, and so we can make only
limited inferences about the impact of psychological character-
istics on acute LBP trajectories.
4.4. Meaning of the study: implications for clinicians
and policymakers
This study provides strong evidence that averaged pain
trajectories for all people with LBP, which are widely
reported,
35
do not adequately reflect the complexity of the
clinical course. The different trajectory patterns potentially
represent subgroups, which may logically require quite
different interventions. For example, engaging early intensive
therapy to address biopsychological factors may be effective
for patients likely to follow a persistent pain trajectory, whereas
those likely to recover by week 4 may require only watchful
waiting or minimal intervention. Patients likely to improve early
but then worsen may represent a group prone to recurrence
that would benefit from effective strategies to prevent re-
currence. Further research is required to develop simple
approaches for clinicians to identify patients likely to belong
to a particular cluster and also test if targeting treatments
based on likely cluster membership lead to improved out-
comes. This study provides further argument that defining
chronicity based on time may be overly simplistic,
49
given that
differentiation of our persistent high pain group occurred as
early as week 2 and had a distinctive baseline profile compared
to patients who had rapid pain recovery.
4.5. Unanswered questions and future research
A key research question is whether management focused on
the patient attributes associated with cluster membership can
effectively shift a patient from a less favorable course to a more
favorable course or whether the identified pain courses
represent patient phenotypes.
20,23
Another question is
whether communicating pain course heterogeneity in a graph-
ical way during the clinical encounter could assist patient
understanding of pain course
53
or support recovery through
modification of pain beliefs.
32
Further investigation is also
required to determine whether simple clinical tools,
appropriate for use in clinical practice, can be developed that
enable early identification of patients who will follow each
trajectory pattern. If this is achieved, it would allow clinicians to
provide more individualized information on course to their
patients.
Conflict of interest statement
The authors have no conflicts of interest to declare.
232 A.S. Downie et al.·157 (2016) 225–234 PAIN
®
Copyright
!
2015 by the International Association for the Study of Pain. Unauthorized reproduction of this article is prohibited.
This research received no specific grant from any funding
agency in the public, commercial, or not-for-profit sectors. A. S.
Downie and M. J. Hancock are employed by Macquarie
University; C. -W. C. Lin and C. G. Maher are employed by the
George Institute for Global Health; C. G. Maher is supported by
National Health and Medical Research Council Fellowship
(APP1002081); C. -W. C. Lin is supported by a Career De-
velopment Fellowship from the National Health and Medical
Research Council, Australia (APP1061400); and C. M. Williams is
employed by Hunter Medical Research Institute.
Article history:
Received 14 April 2015
Received in revised form 17 August 2015
Accepted 31 August 2015
Available online 7 September 2015
References
[1] Altman DG, Royston P. The cost of dichotomising continuous variables.
BMJ 2006;332:1080.
[2] Asparouhov T, Muthen B. Auxiliary variables in mixture modeling: 3-step
approaches using Mplus. Mplus Web Notes, Muth ´en & Muth ´en, 2014.
No. 15.
[3] Atlas SJ, Deyo RA. Evaluating and managing acute low back pain in the
primary care setting. J Gen Intern Med 2001;16:120–31.
[4] Ax ´en I, Bodin L, Bergstr ¨om G, Halasz L, Lange F, L ¨ovgren PW,
Rosenbaum A, Leboeuf-Yde C, Jensen I. Clustering patients on the
basis of their individual course of low back pain over a six month period.
BMC Musculoskelet Disord 2011;12:99.
[5] Axen I, Leboeuf-Yde C. Trajectories of low back pain. Best Pract Res Clin
Rheumatol 2013;27:601–12.
[6] Bauer DJ, Curran PJ. Distributional assumptions of growth mixture
models: implications for overextraction of latent trajectory classes.
Psychol Methods 2003;8:338–63.
[7] Celeux G, Soromenho G. An entropy criterion for assessing the number of
clusters in a mixture model. J Classification 1996;13:195–212.
[8] Chen C, Hogg-Johnson S, Smith P. The recovery patterns of back pain
among workers with compensated occupational back injuries. Occup
Environ Med 2007;64:534–40.
[9] Coste J, Delecoeuillerie G, Cohen de Lara A, LeParc JM, Paolaggi JB.
Clinical course and prognostic factors in acute low back pain: an
inception cohort study in primary care practice. BMJ 1994;308:577–80.
[10] Coste J, Lefranc¸ ois G, Guillemin F, Pouchot J. Prognosis and quality of life
in patients with acute low back pain: insights from a comprehensive
inception cohort study. Arthritis Rehum 2004;51:168–76.
[11] Croft P, Altman DG, Deeks JJ, Dunn KM, Hay AD, Hemingway H,
LeResche L, Peat G, Perel P, Petersen SE, Riley RD, Roberts I, Sharpe
M, Stevens RJ, Van Der Windt DA, Von Korff M, Timmis A. The science
of clinical practice: disease diagnosis or patient prognosis? Evidence
about “what is likely to happen” should shape clinical practice. BMC
Med 2015;13:20.
[12] Dionne CE, Koepsell TD, Von Korff M, Deyo RA, Barlow WE, Checkoway
H. Predicting long-term functional limitations among back pain patients in
primary care settings. J Clin Epidemiol 1997;50:31–43.
[13] Dunn KM, Campbell P, Jordan KP. Long-term trajectories of back pain:
cohort study with 7-year follow-up. BMJ Open 2013;3:e003838.
[14] Dunn KM, Jordan K, Croft PR. Characterizing the course of low back
pain: a latent class analysis. Am J Epidemiol 2006;163:754–61.
[15] Efron B, Tibshirani R. An introduction to the bootstrap. New York:
Chapman & Hall, 1993. p. 105–12.
[16] Gould W. Sg124: interpreting logistic regression in all its forms. STATA
Tech Bull 2000;53:18–29.
[17] Grotle M, Brox JI, Veierød MB, Glomsrød B, Lønn JH, Vøllestad NK.
Clinical course and prognostic factors in acute low back pain:
patients consulting primary care for the first time. Spine 2005;30:976–82.
[18] Gurcay E, Bal A, Eksioglu E, Hasturk AE, Gurcay AG, Cakci A. Acute low
back pain: clinical course and prognostic factors. Disabil Rehabil 2009;
31:840–5.
[19] Hancock MJ, Maher CG, Latimer J, McLachlan AJ, Cooper CW, Day RO,
Spindler MF, McAuley JH. Assessment of diclofenac or spinal
manipulative therapy, or both, in addition to recommended first-line
treatment for acute low back pain: a randomised controlled trial. Lancet
2007;370:1638–43.
[20] Hartvigsen J, Davidsen M, Hestbaek L, Sogaard K, Roos EM. Patterns of
musculoskeletal pain in the population: a latent class analysis using
a nationally representative interviewer-based survey of 4817 Danes. Eur J
Pain 2013;17:452–60.
[21] Henderson AR. The bootstrap: a technique for data-driven statistics.
Using computer-intensive analyses to explore experimental data. Clin
Chim Acta 2005;359:1–26.
[22] Henschke N, Maher CG, Refshauge KM, Herbert RD, Cumming RG,
Bleasel J, York J, Das A, McAuley JH. Prognosis in patients with recent
onset low back pain in Australian primary care: inception cohort study.
BMJ 2008;337:a171.
[23] Hooff MLV, Loon JV, Limbeek JV, Kleuver MD. The Nijmegen decision
tool for chronic low back pain. Development of a clinical decision tool for
secondary or tertiary spine care specialists. PLoS One 2014;9:1–12.
[24] Jones BL, Nagin DS. Advances in group-based trajectory modeling and an
SAS procedure for estimating them. Soc Method Res 2007;35:542–71.
[25] Jung T, Wickrama KAS. An introduction to latent class growth analysis and
growth mixture modeling. Soc Personal Psychol Compass2008;2:302–17.
[26] Kamper SJ, Stanton TR, Williams CM, Maher CG, Hush JM. How is
recovery from low back pain measured? A systematic review of the
literature. Eur Spine J 2011;20:9–18.
[27] Kent P, Kongsted A. Identifying clinical course patterns in SMS data using
cluster analysis. Chiropr Man Therap 2012;20:20.
[28] Koes BW, van Tulder M, Lin CWC, Macedo LG, McAuley J, Maher C. An
updated overview of clinical guidelines for the management of non-
specific low back pain in primary care. Eur Spine J 2010;19:2075–94.
[29] Kongsted A, Kent P, Hestbaek L, Vach W. Patients with low back pain
had distinct clinical course patterns that were typically neither complete
recovery nor constant pain. A latent class analysis of longitudinal data.
Spine J 2015;15:885–94.
[30] Langeheine R, Pannekoek J, Van De Pol F. Bootstrapping goodness-of-
fit measures in categorical data analysis. Soc Methods Res 1996;24:
492–516.
[31] Macedo LG, Maher CG, Latimer J, McAuley JH, Hodges PW, Rogers WT.
Nature and determinants of the course of chronic low back pain over
a 12-month period: a cluster analysis. Phys Ther 2014;94:210–21.
[32] Main CJ, Foster N, Buchbinder R. How important are back pain beliefs
and expectations for satisfactory recovery from back pain? Best Pract
Res Clin Rheumatol 2010;24:205–17.
[33] Mak K, Kum C. How to appraise a prognostic study. World J Surg 2005;
29:567–9.
[34] Malhotra MK, Sharma S, Nair SS. Decision making using multiple models.
Eur J Oper Res 1999;114:1–14.
[35] Menezes Costa L, Maher CG, Hancock MJ, McAuley JH, Herbert RD,
Costa LO. The prognosis of acute and persistent low-back pain: a meta-
analysis. CMAJ 2012;184:E613–24.
[36] Murray CJL, Vos T, Lozano R, Naghavi M, Flaxman AD, Michaud C, Ezzati
M, Shibuya K, Salomon JA, Abdalla S, Aboyans V, Abraham J, Ackerman
I, Aggarwal R, Ahn SY, Ali MK, AlMazroa MA, Alvarado M, Anderson HR,
Anderson LM, Andrews KG, Atkinson C, Baddour LM, Bahalim AN,
Barker-Collo S, Barrero LH, Bartels DH, Bas ´an
˜ez MG, Baxter A, Bell ML,
Benjamin EJ, Bennett D, Bernab ´eE,BhallaK,BhandariB,BikbovB,
Abdulhak AB, Birbeck G, Black JA, Blencowe H, Blore JD, Blyth F,
Bolliger I, Bonaventure A, Boufous S, Bourne R, Boussinesq M,
Braithwaite T, Brayne C, Bridgett L, Brooker S, Brooks P, Brugha TS,
Bryan-Hancock C, Bucello C, Buchbinder R, Buckle G, Budke CM,
Burch M, Burney P, Burstein R, Calabria B, Campbell B, Canter CE,
Carabin H, Carapetis J, Carmona L, Cella C, Charlson F, Chen H, Cheng
ATA, Chou D, Chugh SS, Coffeng LE, Colan SD, Colquhoun S, Colson
KE, Condon J, Connor MD, Cooper LT, Corriere M, Cortinovis M, de
Vaccaro KC, Couser W, Cowie BC, Criqui MH, Cross M, Dabhadkar KC,
Dahiya M, Dahodwala N, Damsere-Derry J, Danaei G, Davis A, Leo DD,
Degenhardt L, Dellavalle R, Delossantos A, Denenberg J, Derrett S, Des
Jarlais DC, Dharmaratne SD, Dherani M, Diaz-Torne C, Dolk H, Dorsey
ER, Driscoll T, Duber H, Ebel B, Edmond K, Elbaz A, Ali SE, Erskine H,
Erwin PJ, Espindola P, Ewoigbokhan SE, Farzadfar F, Feigin V, Felson
DT, Ferrari A, Ferri CP, F `evre EM, Finucane MM, Flaxman S, Flood L,
Foreman K, Forouzanfar MH, Fowkes FGR, Fransen M, Freeman MK,
Gabbe BJ, Gabriel SE, Gakidou E, Ganatra HA, Garcia B, Gaspari F,
Gillum RF, Gmel G, Gonzalez-Medina D, Gosselin R, Grainger R, Grant B,
Groeger J, Guillemin F, Gunnell D, Gupta R, Haagsma J, Hagan H, Halasa
YA, Hall W, Haring D, Haro JM, Harrison JE, Havmoeller R, Hay RJ,
Higashi H, Hill C, Hoen B, Hoffman H, Hotez PJ, Hoy D, Huang JJ,
Ibeanusi SE, Jacobsen KH, James SL, Jarvis D, Jasrasaria R, Jayaraman
S, Johns N, Jonas JB, Karthikeyan G, Kassebaum N, Kawakami N, Keren
A, Khoo JP, King CH, Knowlton LM, Kobusingye O, Koranteng A,
Krishnamurthi R, Laden F, Lalloo R, Laslett LL, Lathlean T, Leasher JL,
Lee YY, Leigh J, Levinson D, Lim SS, Limb E, Lin JK, Lipnick M, Lipshultz
January 2016·Volume 157 ·Number 1 www.painjournalonline.com 233
Copyright
!
2015 by the International Association for the Study of Pain. Unauthorized reproduction of this article is prohibited.
SE, Liu W, Loane M, Ohno SL, Lyons R, Mabweijano J, MacIntyre MF,
Malekzadeh R, Mallinger L, Manivannan S, Marcenes W, March L,
Margolis DJ, Marks GB, Marks R, Matsumori A, Matzopoulos R, Mayosi
BM, McAnulty JH, McDermott MM, McGill N, McGrath J, Medina-Mora
ME, Meltzer M, Memish ZA, Mensah GA, Merriman TR, Meyer AC, Miglioli
V, Miller M, Miller TR, Mitchell PB, Mock C, Mocumbi AO, Moffitt TE,
Mokdad AA, Monasta L, Montico M, Moradi-Lakeh M, Moran A,
Morawska L, Mori R, Murdoch ME, Mwaniki MK, Naidoo K, Nair MN,
Naldi L, Narayan KMV, Nelson PK, Nelson RG, Nevitt MC, Newton CR,
Nolte S, Norman P, Norman R, O’Donnell M, O’Hanlon S, Olives C, Omer
SB, Ortblad K, Osborne R, Ozgediz D, Page A, Pahari B, Pandian JD,
Rivero AP, Patten SB, Pearce N, Padilla RP, Perez-Ruiz F, Perico N,
Pesudovs K, Phillips D, Phillips MR, Pierce K, Pion S, Polanczyk GV,
Polinder S, Pope CA III, Popova S, Porrini E, Pourmalek F, Prince M,
Pullan RL, Ramaiah KD, Ranganathan D, Razavi H, Regan M, Rehm JT,
Rein DB, Remuzzi G, Richardson K, Rivara FP, Roberts T, Robinson C,
De Le `on FR, Ronfani L, Room R, Rosenfeld LC, Rushton L, Sacco RL,
Saha S, Sampson U, Sanchez-Riera L, Sanman E, Schwebel DC, Scott
JG, Segui-Gomez M, Shahraz S, Shepard DS, Shin H, Shivakoti R, Singh
D, Singh GM, Singh JA, Singleton J, Sleet DA, Sliwa K, Smith E, Smith JL,
Stapelberg NJC, Steer A, Steiner T, Stolk WA, Stovner LJ, Sudfeld C,
Syed S, Tamburlini G, Tavakkoli M, Taylor HR, Taylor JA, Taylor WJ,
Thomas B, Thomson WM, Thurston GD, Tleyjeh IM, Tonelli M, Towbin JA,
Truelsen T, Tsilimbaris MK, Ubeda C, Undurraga EA, van der Werf MJ,
van Os J, Vavilala MS, Venketasubramanian N, Wang M, Wang W, Watt
K, Weatherall DJ, Weinstock MA, Weintraub R, Weisskopf MG,
Weissman MM, White RA, Whiteford H, Wiebe N, Wiersma ST,
Wilkinson JD, Williams HC, Williams SRM, Witt E, Wolfe F, Woolf AD,
Wulf S, Yeh PH, Zaidi AKM, Zheng ZJ, Zonies D, Lopez AD. Disability-
adjusted life years (DALYs) for 291 diseases and injuries in 21 regions,
1990–2010: a systematic analysis for the Global Burden of Disease Study
2010. Lancet 2012;380:2197–223.
[37] Nagin DS. Group-based modelling of development. Cambridge: Harvard
University Press, 2005. p. 114–5.
[38] Nagin DS, Tremblay RE. Developmental trajectory groups: fact or a useful
statistical fiction? Criminology 2005;43:873–904.
[39] Nylund KL, Asparouhov T, Muth ´en BO. Deciding on the number of
classes in latent class analysis and growth mixture modeling: a Monte
Carlo simulation study. Struct Equ Model 2007;14:535–69.
[40] O’brien RM. A caution regarding rules of thumb for variance inflation
factors. Qual Quant 2007;41:673–90.
[41] Pengel LHM, Herbert RD, Maher CG, Refshauge KM. Acute low back
pain: systematic review of its prognosis. BMJ 2003;327:323–7.
[42] Steyerberg EW, Harrell FE, Borsboom GJJM, Eijkemans MJC, Vergouwe
Y, Habbema JDF. Internal validation of predictive models. J Clin
Epidemiol 2001;54:774–81.
[43] Tamcan O, Mannion AF, Eisenring C, Horisberger B, Elfering A, Muller U.
The course of chronic and recurrent low back pain in the general
population. PAIN 2010;150:451–7.
[44] Twisk J, Hoekstra T. Classifying developmental trajectories over time
should be done with great caution: a comparison between methods.
J Clin Epidemiol 2012;65:1078–87.
[45] van Tulder M, Koes B, Bombardier C. Low back pain. Best Pract Res Clin
Rheumatol 2002;16:761–75.
[46] Vasseljen O, Woodhouse A, Bjørngaard JH, Leivseth L. Natural course of
acute neck and low back pain in the general population: the HUNT study.
PAIN 2013;154:1237–44.
[47] Vermunt JK. Latent class modeling with covariates: two improved three-
step approaches. Polit Anal 2010;18:450–69.
[48] Von Korff M, Deyo RA, Cherkin D, Barlow W. Back pain in primary care.
Outcomes at 1 year. Spine 1993;18:855–62.
[49] Von Korff M, Dunn KM. Chronic pain reconsidered. PAIN 2008;138:
267–76.
[50] Williams CM, Latimer J, Mahe r CG, McLachlan AJ, Cooper CW,
Hancock MJ, Day RO, McAuley JH, Lin CW. PACE—the first placebo
controlled trial of paracetamol for acute low back pain: design of
arandomisedcontrolledtrial.BMCMusculoskeletDisord2010;11:
169.
[51] Williams CM, Maher CG, Latimer J, McLachlan AJ, Hancock MJ, Day
RO, Billot L, Lin CW. PACE—the first placebo controlled trial of
paracetamol for acute low back pain: statistical analysis plan. Trials
2013;14:248.
[52] Williams CM, Maher CG, Latimer J, McLachlan AJ, Hancock MJ, Day RO,
Lin CW. Efficacy of paracetamol for acute low-back pain: a double-blind,
randomised controlled trial. Lancet 2014;384:1586–96.
[53] Wolter T, Szabo E, Becker R, Mohadjer M, Knoeller SM. Chronic low back
pain: course of disease from the patient’s perspective. Int Orthop 2011;
35:717–24.
234 A.S. Downie et al.·157 (2016) 225–234 PAIN
®
Copyright
!
2015 by the International Association for the Study of Pain. Unauthorized reproduction of this article is prohibited.
