Introduction

Musculoskeletal pain is a major cause of personal suffering

and economic loss following work-related injury (Schulte
2005). In the state of New South Wales, over 36 000
workplace injuries occur annually (WorkCover NSW
2004/5). Musculoskeletal sprains and strains account for
62% of these injuries at a cost of AU $434 million. The
percentage of injured workers who develop persistent pain
and activity limitation has risen by 16% over the past decade
(WorkCover NSW 2004/5). These fndings are consistent
with other westernised countries. Globally it is estimated
that 100 million workplace injures occur annually (Leigh
et al 1999) and the subsequent development of persistent
pain and activity limitation has been recognised as a major
public health problem (Feuerstein 2005).
The emerging paradigm in occupational injury management
is early identifcation of injured workers at risk of chronic
pain and activity limitation (Gatchel 2005) particularly
for those with back pain (Fritz and George 2002, Hogg-
Johnson and Cole 2003, Linton et al 2005, Pransky et al
2006, Schulz et al 2005, Turner et al 2006). While there is a
clear imperative to recognise workers with poor prognosis
in the early stage after injury, the current reality is that the
population of workers with long-term musculoskeletal pain
continues to rise (Waddell 1996). To date, there has been
limited research into prognosis for this population (Dionne
et al 2005, Proctor et al 2005, Sullivan et al 2005, van
der Giezen et al 2000) and available research has focused
predominantly on identifying prognostic factors. It is not
immediately clear how prognostic factors alone can be
used to facilitate clinical management. However, predictors
can be combined to develop

clinical prediction rules. This
approach can enable calculation of the likelihood of an
outcome for individual patients (Childs and Cleland 2006,
Randolph et al 1998) and can minimise bias in judgment, in
accordance with current principles of evidence-based health
care. This process is becoming increasingly recognised as a
useful adjunct to the evaluative process in clinical decision-
making, particularly for patients with musculoskeletal
injuries (Beattie and Nelson 2006, Childs and Cleland 2006).
To date, there are few clinical prediction rules for patients
with work-related injuries, and there are no published rules
for those injured in an Australian workplace setting. In a
landmark study in Canada, Dionne and colleagues (2005)
developed a clinical prediction rule to identify workers with
persistent back pain at risk of poor occupational outcome.
The predictive accuracy of return to work was not much
better than chance. Further research is clearly needed to
provide clinical prediction rules with higher predictive
accuracy for this population.
Clinical prediction rules can be derived and validated for
injured Australian workers with persistent musculoskeletal
pain: an observational study
Jennifer A Hewitt
1
, Julia M Hush
2
, Melody H Martin
1
, Robert D Herbert
2
and Jane Latimer
2
1
Rehab One Physiotherapy
2
University of Sydney
Australia
Questions: Can clinical prediction rules be derived for injured Australian workers with persistent musculoskeletal pain? Are
they valid? Design: Longitudinal observational study. Participants: 847 injured workers with persistent musculoskeletal pain
undergoing rehabilitation. Outcome measures: At baseline, 12 putative predictors were measured. At 9 weeks, short-term
outcomes such as pain (visual analogue scale), activity limitation (Functional Rating Index) and work upgrade (increase in work
hours or duties) were measured. At 6 months, long-term work status (working or not working) was measured. Results: Data
were obtained from 85% of the participants who were followed up at both 9 weeks (720 of 847) and 6 months (247 of 290).
Predictors of outcome included high baseline pain and activity limitation, long duration of previous intervention, not working,
non-English speaking background, and the area of pain. Accuracy was highest for clinical prediction rules predicting pain and
level of activity limitation at 9 weeks (R
2
= 0.67 and 0.69 respectively) and work status at 6 months (LR = 0.24). Conclusion:
Accurate clinical prediction rules have been derived and validated for injured workers with persistent musculoskeletal pain,
predicting activity limitation, pain, and work outcomes following exercise-based rehabilitation. Further research to validate these
prediction rules in other populations and to assess the effectiveness of tailoring intervention based on the estimated prognosis
would be valuable. [Hewitt JA, Hush JM, Martin MH, Herbert RD, Latimer J (2007) Clinical prediction rules can be derived
and validated for injured Australian workers with persistent musculoskeletal pain: an observational study. Australian
Journal of Physiotherapy 53: 269276]
Key words: Prognosis, Physical Therapy (Specialty), Work
Australian Journal of Physiotherapy 2007 Vol. 53 Australian Physiotherapy Association 2007 269
Hewitt et al: Clinical prediction rules for injured Australian workers
for the frst 6 to 9 weeks. Sessions were carried out in a
gymnasium for one hour, three times per week, and included
prescription of graded exercises (strength, endurance, ftness,
stretch) and activities (simulated work) based on injury
and work goals. Participants also performed a daily home
program of activities and exercises and completed a diary
that was checked at each physiotherapy session to monitor
compliance. All physiotherapists had received training in
cognitive behavioural techniques (education, goal setting,
and pacing). Education included reassurance about their
condition, explanation of scan fndings, and discussion of
the benefts of activity and the consequences of extended
down-time. After the supervised component of the program,
participants were instructed to exercise independently, and
long-term self-management was encouraged. Professional
interpreters were used as required. Short term outcomes
(pain, activity limitation, upgrade in hours or duties at work)
Australian Journal of Physiotherapy 2007 Vol. 53 Australian Physiotherapy Association 2007 270
Research
The research questions investigated in this study were:
1. Can clinical prediction rules be derived for injured
Australian workers with persistent musculoskeletal
pain?
2. Are they valid?
Method
Design
People with sub-acute or chronic musculoskeletal pain
in NSW, Australia, referred to a private physiotherapy
practice for exercise following a work injury were observed
longitudinally. Typically, their previous intervention
consisted of passive techniques that focused on symptom
relief. First, putative predictors were measured. Participants
then took part in an exercise-based physiotherapy program
Week
Exercise-based
physiotherapy program
0
9
26
Injured workers referred for exercise-based physiotherapy program
Measured 12 putative predictors
(n = 847)
Measured pain, activity limitation, work upgrade
(n = 720)
Measured work status
(n = 247)
Figure 1. Design and fow of participants through the study.
Participants lost to follow-up (n = 127)
Doctors decision (n = 53)
Participants decision (n= 74)
Randomly split
Randomly split
Predictors identifed
(n = 360)
Clinical Prediction Rule validated
(n = 360)
Predictors identifed
(n = 124)
Clinical Prediction Rule validated
(n = 123)
Australian Journal of Physiotherapy 2007 Vol. 53 Australian Physiotherapy Association 2007 271
Hewitt et al: Clinical prediction rules for injured Australian workers
were measured 9 weeks after commencing the program.
Long-term measurement of work status (working or not
working) was carried out at 6 months. Measurements were
collected by physiotherapists not involved in the intervention
program. The data from 9 weeks and 6 months were each
randomly divided into two halves. The frst half was used
to identify predictors from which a clinical prediction rule
for each outcome was developed. The second half was used
to test the predictive accuracy of each model. The design of
the study is illustrated in Figure 1. The study was approved
by the University of Sydney Human Research Ethics
Committee.
Participants
Patients were considered eligible for inclusion in the
study if they presented with work-related musculoskeletal
pain of 6 weeks duration, had been given a medical
clearance by their doctor to participate in the exercise-based
physiotherapy program, and had been given approval by the
third party payer (insurance company) to attend the program.
They were excluded from the program if they demonstrated
signs or symptoms suggestive of a red fag condition (eg,
tumour, systemic illness, infammatory disease or infection,
screened using the Modifed Core Network Screening
Questionnaire, Maher et al 2001), were in the acute stage of
their injury (0-6 weeks), or demonstrated signs of entrenched
psychological or psychosocial factors warranting referral to a
multidisciplinary pain clinic (Guzman et al 2003, Haldorsen
et al 2002). Physiotherapists performed comprehensive
assessments incorporating subjective, physical, functional,
and psychosocial measures to assist in determining each
participants eligibility for inclusion in the program.
Outcome measures
At baseline, 12 putative predictors were measured. In
order to increase the clinical utility of the fndings for
physiotherapists, putative predictors were selected that
were clinical features, easily and routinely recorded during
a standard physiotherapy assessment (Box 1).
At 9 weeks, outcomes were selected that addressed
issues relevant to the individual (such as pain and activity
limitation) as well as economic issues (upgrade in work
hours and duties) which are of particular interest to third
party payers, and workers compensation authorities. Pain
was measured using a 10-cm visual analogue scale where 0
= no pain, and 10 = the worst pain imaginable, and activity
limitation was measured using the Functional Rating Index
scored as a percentage. The Functional Rating Index is a
hybrid instrument of the Oswestry Disability Questionnaire
and the Neck Disability Index and has been shown to be
psychometrically sound with regard to reliability, validity,
and responsiveness in patients with spinal pain (Fiese et al
2001, Hush 2006). In the present study, a small modifcation
of the questionnaire was made to enable use with all
participants. The modifcation was made to the statement
Box 1: The twelve putative predictors measured at
baseline.
Duration of injury (mth)
Duration of previous intervention (mth)
Surgery to compensable area (y/n)
Area injured (non-spinal/spine)
Pain intensity (10-cm VAS)
Activity limitation (FRI) (%)
Gender (M/F)
Age (yr)
Time off work (mth)
Work status (working/not working)
Non-English speaking background (y/n)
Interpreter required (y/n)
Table 1. Characteristics of participants at baseline.
Characteristic All participants
(n = 847)
Participants
measured
at 9 weeks
(n = 290)
Duration of injury (mth),
mean (SD)
11 (14.6) 9.91 (9.6)
Duration of previous
intervention (mth), mean
(SD)
5.23 (5.6) 4.91 (5.4)
Surgery n (%)
Spine
Not spine
Compensable
126 (15)
720 (85)
153 (18)
59 (20)
232 (80)
58 (20)
Area injured (primary), n (%)
Lumbar spine
Shoulder
Cervical spine
Lower limb
Upper limb
Thoracic spine
Other
474 (56)
110 (13)
101 (12)
76 (9)
51 (6)
17 (2)
18 (2)
154 (53)
32 (11)
6 (2)
44 (15)
23 (8)
29 (10)
6 (2)
Pain intensity (10-cm VAS),
mean (SD)
5.8 (2.9) 5.6 (2.1)
Activity limitation FRI (0 to
100), mean (SD)
52 (17) 51 (17)
Gender, n male (%) 559 (66) 197 (68)
Age (yr), mean (SD) 39.7 (10.7) 39.7 (10.8)
Time off work (mth), mean
(SD)
2.4 (4.5) 2.2 (3.4)
Work status, n (%)
Working
Normal duties
Suitable duties
Full time
Part time
Reduced hours
609 (72)
119 (14)
727 (86)
441 (52)
101 (12)
305 (36)
214 (74)
32 (11)
258 (89)
160 (55)
26 (9)
104 (36)
Non-English speaking
background, n (%)
295 (35) 78 (27)
Interpreter required, n (%) 102 (12) 32 (11)
Fear avoidance beliefs
(FABQ)
Physical activity (0 to 24),
mean (SD)
Work (0 to 42), mean (SD)

16.4 (6.0)

28.7 (9.8)

16.9 (5.7)

27.6 (9.9)
at the top of the questionnaire: We wish to understand how
much your pain [instead of your back/neck pain] has affected
your ability to manage your everyday activities. None of the
items was altered and no items refer to the area of pain. High
reliability and responsiveness of this modifed version of the
Functional Rating Index has been reported (Chansirinukor
et al 2004). This process has also been performed on other
back pain-specifc scales for application to the wider
chronic pain population (eg, Roland Morris was adapted in
Stroud et al 2004, and the Oswestry Disability Index was
adapted in Whittink et al 2004, and Fish and Chang 2007).
Work upgrade (whether the participant returned to work or
upgraded work hours or duties) was measured categorically
as Yes/No.
At 6 months, long-term measurement of work status
(measured as Yes/No) was achieved by postal questionnaire
and phone call to the participants, or by contacting the
insurance company covering the workers compensation
claim. Long-term follow up was limited to the frst
consecutive 290 participants to complete the program due
to limited resources.
Data analysis
Separate models were developed to predict each outcome
of interest, using multiple regression analyses. A split-half
approach was used whereby the data set was randomly
divided into two halves. The frst half was used to identify
predictors by stepwise regression. A stepwise approach (p
to enter = 0.05, p to remove = 0.10) was used to identify
signifcant predictors from the 12 putative predictors. The
second half was used to test the predictive accuracy of each
model. For each outcome, a clinical prediction rule was
developed from the regression coeffcients of each predictor.
Australian Journal of Physiotherapy 2007 Vol. 53 Australian Physiotherapy Association 2007 272
Research
Table 2. Mean (95% CI) regression coeffcients of predictors, clinical prediction rule and accuracy of prediction for activity
limitation at 9 weeks.
Regression coeffcients of predictors
Baseline activity limitation = 0.72 (0.63 to 0.82)
Interpreter required = 8.93 (13.79 to 4.08)
Duration of previous intervention = 2.31 (0.89 to 3.72)
Baseline work status = 4.15 (7.74 to 0.56)
Constant = 13.66 (5.59 to 21.74)
Clinical Prediction Rule
Activity limitation = 0.72 (baseline activity limitation)
8.93 (interpreter required)
+ 2.31 (duration of previous intervention)
4.15 (baseline work status)
+ 13.66
Accuracy of prediction
R
2
= 0.69
Activity limitation 0100 FRI scale; interpreter requirement (0 = yes, 1 = no); duration previous symptom focused treatment =
natural log (months of previous treatment + 0.125); work status (0 = not working, 1 = working).
Table 3. Mean (95% CI) regression coeffcients of predictors, clinical prediction rule and accuracy of prediction for pain
intensity at 9 weeks.
Regression coeffcients of predictors
Baseline pain = 0.41 (0.28 to 0.53)
Baseline activity limitation = 0.04 (0.02 to 0.05)
Non-English speaking background = 0.94 (1.38 to 0.51)
Duration of previous intervention = 0.27 (0.06 to 0.49)
Constant = 0.41 (0.42 to 1.23)
Clinical Prediction Rule
Pain = 0.41 (baseline pain)
+ 0.04 (baseline activity limitation)
0.94 (Non-English speaking background)
+ 0.27 (duration of previous intervention)
+ 0.41
Accuracy of prediction
R
2
= 0.67
Pain intensity scale 010; activity limitation 0100; FRI scale (0 = minimum activity limitation, 100 = maximum activity
limitation); Non-English speaking background (0 = other, 1 = English); duration of previous intervention = natural log (months
of previous intervention + 0.125).
The predictive accuracy of models with continuous
outcomes was expressed as R
2
values. For dichotomous
outcomes, positive and negative likelihood ratios (LR+,
LR) were calculated because clinicians can use these ratios
to determine the probability of outcomes and they have been
reported previously for occupational prognosis (Fritz and
George 2002). Additionally, odds ratios were calculated for
dichotomous predictors to indicate the strength of individual
prognostic factors.
Results
Participants
Eight hundred and forty-seven injured workers commenced
the exercise-based physiotherapy program. Baseline
characteristics of these participants are summarised in
Table 1. All participants were compensable, being referred
following a work injury under the Workers Compensation
system in New South Wales, Australia. Participants had
sustained injuries, on average, 11 months previously from
manual handling (55%), falls, slips and trips (15%), motor
vehicle accident (12%), repetitive factory work (7%), being
hit by an object (4%), repetitive offce work (1%), and other
mechanisms (6%). Of the 847 participants, 720 (85%)
completed the program and were measured at 9 weeks.
Participants were lost to follow-up because they chose to
withdraw from the program (n = 74), or were advised by
their treating doctor to cease exercise because of medical
concerns unrelated to their injury (n = 41) or related to their
work injury (n = 12). Long-term follow-up of work status
was conducted at 6 months on the frst 290 participants for
whom data had been obtained at 9 weeks. The baseline
characteristics of the 290 participants followed up at 6
months were similar to those of the whole cohort (Table
1). Of the 290 participants, 247 (85%) were measured at 6
months.
Clinical prediction rules
The four variables predictive of a higher level of activity
limitation at 9 weeks were high baseline activity limitation,
requirement for an interpreter, long duration of previous
intervention, and being off work at baseline (Table 2). The R
2

value of the clinical prediction rule for this outcome is 0.69
(Table 2). The variables predictive of higher pain intensity
at 9 weeks were high baseline pain intensity, high baseline
activity limitation, a non-English speaking background and
Australian Journal of Physiotherapy 2007 Vol. 53 Australian Physiotherapy Association 2007 273
Hewitt et al: Clinical prediction rules for injured Australian workers
Table 4: Mean (95% CI) regression coeffcients of predictors, clinical prediction rule and accuracy of prediction for work
upgrade at 9 weeks.
Regression coeffcients of predictors
Baseline work status = 1.63 (1.03 to 2.23)
Duration of previous intervention = 0.41 (0.69 to 0.14)
Area injured = 0.72 (1.27 to 0.16)
Interpreter required = 0.97 (0.16 to 1.77)
Constant = 1.00 (1.97 to 0.03)
Clinical Prediction Rule
Odds (work upgrade) = e
1.00
x e
1.63 (baseline work status)
x e
0.41 (duration previous intervention)
x e
0.72 (area injured)
x e
0.97 (interpreter)
Probability of a work upgrade = Odds (work upgrade) / Odds (work upgrade) + 1
Accuracy of prediction
LR+ = 1.53
LR = 0.52
Work status (0 = not working, 1 = working); duration previous intervention = natural log (months of previous intervention +
0.125); area injured (0 = not spine, 1 = spine); interpreter requirement (0 = yes, 1 = no).
Table 5. Mean (95% CI) regression coeffcients of predictors, clinical prediction rule and accuracy of prediction for work status
at 6 months.
Regression coeffcients of predictors
Baseline work status = 1.33 (0.66 to 1.99)
Area injured = 0.77 (0.04 to 1.51)
Constant = 0.12 (0.80 to 0.55)
Clinical Prediction Rule
Odds (work status) = e
0.12
x e
1.33 (baseline work status)
x e
0.77 (area injured)
Probability of being at work = Odds (work status) / Odds (work status) + 1
Accuracy of prediction
LR+ = 1.21
LR = 0.24
Work status (0 = not working, 1 = working); area injured (0 = not spine, 1 = spine).
long duration of previous intervention (Table 3). The R
2

value for this clinical prediction rule is 0.67 (Table 3).
Variables predictive of not upgrading at work by 9 weeks
included being off work at baseline (OR 5.10, 95% CI
2.81 to 9.26), longer duration of previous intervention, an
injury to the spine (OR 0.49, 95% CI 0.28 to 0.85), and
requirement for an interpreter (OR 2.63, 95% CI 1.18 to
5.88) (Table 4). The likelihood ratios (LR+ 1.53, LR 0.52)
suggest the predictive accuracy of this clinical prediction
rule is limited (Table 4).
Work status at 6 months was strongly predicted by two
factors: baseline work status (OR 3.77, 95% CI 1.93 to 7.34)
and area of injury (OR 2.17, 95% CI 1.12 to 4.20), such that
participants had lower odds of working at 6 months if they
had been off work at baseline and had a non-spinal injury
(Table 5). The likelihood ratios for the clinical prediction
rule for work status at 6 months (LR+ 1.21 and LR 0.24),
indicate that this prediction rule should be most useful for
identifying injured workers who are at higher risk of being
off work 6 months after commencing the rehabilitation
program (Table 5). Table 6 illustrates how two of the clinical
prediction rules could be used to estimate patient outcomes,
at 9 weeks (Cases 1 and 2) and 6 months (Cases 3 and 4).
Discussion
This longitudinal study has investigated prognosis in a
large cohort of people with persisting musculoskeletal
pain following a work injury. Baseline predictors of poor
outcome included being off work, longer duration of
previous intervention, high activity limitation and pain, and
being of a non-English speaking background or requiring an
interpreter. Clinical prediction rules to estimate prognosis
were developed and validated. The strengths of this study
are the prospective design, large sample size of consecutive
participants, adequate follow-up, and validation of the
algorithms.
The clinical prediction rules with the highest accuracy are
those that predict activity limitation and pain at 9 weeks
and poor work status at 6 months. In a previous study that
attempted to develop a clinical prediction rule to predict
outcome for workers with persistent back pain (Dionne et
al 2005), the predictive accuracy was not much more than
chance. Other published algorithms also report limited
accuracy. For instance, the models developed by Pransky
and colleagues (2006) explained only 12% of the variance
when predicting activity limitation. Therefore, the clinical
prediction rules developed in this study have predictive
accuracy that exceeds that of previous algorithms.
The high predictive accuracy of the models in the present
study is particularly interesting given that psychosocial
factors were not considered. There is some evidence that
such factors are predictive of occupational outcome in
people with persistent musculoskeletal pain (van der Giezen
et al 2000, Linton et al 2005) as well as for those with acute
occupational injuries (Burton et al 1995, Gatchel 2005,
Turner et al 2006). However, the intention of this study
was to evaluate the predictive capacity of measures taken
during a routine physiotherapy assessment, without the
need for specialised screening instruments or expertise. For
that reason, psychosocial factors were not evaluated in this
study. A possible direction for future research is to evaluate
whether inclusion of psychosocial factors in these prediction
rules further improves the predictive accuracy. In the clinical
setting, combining fndings from the clinical prediction rules
from this study with fndings from psychosocial screening
tools (eg, the Orebro Musculoskeletal Pain Questionnaire,
Linton and Boersma 2003) may assist in further estimating
prognosis using a biopsychosocial model.
Australian Journal of Physiotherapy 2007 Vol. 53 Australian Physiotherapy Association 2007 274
Research
Table 6. Predicted outcome using clinical prediction rules for four cases.
Case Baseline score Predicted outcome
Case 1
Activity limitation [0100%] 50
Interpreter required [0 = yes, 1 = no] 1
Duration of previous intervention [months] 1.5
Work status [0 = no, 1 = yes] 1
Predicted activity limitation at 9 weeks = 38%
Case 2
Activity limitation score [0100%] 73
Interpreter required [0 = yes, 1 = no] 0
Duration of previous intervention [months] 6.5
Work status [0 = no, 1 = yes] 0
Predicted activity limitation at 9 weeks = 71%
Case 3
Work status [0 = no, 1 = yes] 1
Area injured [0 = not spine, 1 = spine] 1
Predicted probability of working at 6 months = 88%
Case 4
Work status [0 = no, 1 = yes] 0
Area injured [0 = not spine, 1 = spine] 0
Predicted probability of working at 6 months = 47%
Baseline work status was found to be a strong predictor of
work upgrade at 9 weeks and work status at 6 months. Proctor
and colleagues (2005) investigated prognosis in a large
study of injured workers with occupational musculoskeletal
disorders with an average duration of 19 months since
injury, representing a population similar to the present study.
These authors found that return to work reliably predicted
completion of a functional restoration program. Previous
research has reported a relationship between length of time
off work and the likelihood of returning to work (Fordyce
1995). However in our analyses, time off work did not
appear in the fnal multivariate models.
Health care utilisation (> 30 visits) was also found to be
highly predictive of outcome in the Proctor (2005) study.
This concurs with our results that a long duration of previous
intervention is predictive of high levels of activity limitation,
pain, and reduced likelihood of upgrading at work. Notably,
this effect is independent of chronicity, which was also
entered into the regression model. This fnding also aligns
with current evidence regarding effcacy of interventions
for persisting occupational conditions. Evidence-based
occupational health guidelines (Waddell and Burton 2001)
recommend explicitly that back pain participants having
diffculty returning to normal activities at 412 weeks
should cease symptomatic intervention, including manual
therapy and electrotherapy, and commence active exercise-
based intervention.
Self-reported pain and activity limitation were identifed as
prognostic factors in this study. This fnding is supported
by research that has consistently found these variables to be
positively associated with the duration of activity limitation,
in occupational and back injuries (Hogg-Johnson and Cole
2003, Sullivan et al 2005, Turner et al 2000).
A notable feature of the present study is the evaluation of
non-English speaking background (or requirement for an
interpreter) as predictive of outcome, as these individuals
have frequently been excluded from previous studies. These
variables were found to predict pain and activity limitation,
and likelihood of a work upgrade. Interpretation of these
results is complex. One possible explanation is that cultural
issues may impact on the perception or reporting of pain.
Translators may also infuence delivery of the program
particularly when providing reassurance and education.
Further research is required to better understand this
fnding.
A surprising feature of the results was that spinal pain was
predictive of failing to upgrade at work in the short term,
however having non-spinal pain was predictive of being
off work in the long term. One explanation for this may
be the high proportion of participants with upper or lower
limb injuries requiring surgery, indicating that the injuries
were severe. Conficting fndings regarding area of injury
as a prognostic factor have been reported previously (Hogg-
Johnson 1998, Turner et al 2000).
Valid prognostic evidence is recognised as useful in providing
participants with information about what the future is likely
to hold for them and their injury (Straus et al 2005). The
models developed in this study may be clinically useful
for this purpose, particularly following validation in other
populations. A further application of these fndings would
be to use the clinical prediction rules to assist in planning
intervention. The rationale is that participants identifed at
risk of poor prognosis may achieve better outcomes if they
receive more comprehensive or tailored interventions. The
current study does not provide data to support this assertion,
as clinical trials are required to test whether selection of
intervention based on estimated patient prognosis at baseline
is effective. However, one study by Haldorsen and colleagues
(2002) does provide data that confrm this approach. The
authors found that participants classifed as having a high
risk for poor outcome achieved better outcomes with an
extensive multidisciplinary program, whereas those who
were estimated to have a good outcome did equally well
with the brief program. These results support the proposal
that clinical prediction rules may be helpful in optimising
choice of intervention. Nonetheless, clinical prediction
rules should not be considered a replacement for clinical
judgment: expert opinion is still the cornerstone of clinical
decision making (Beattie and Nelson 2006). Where poor
prognosis is indicated, maximising resources rather than
withdrawal of intervention is more likely to result in better
outcomes. Evaluation of the effcacy of this approach is
required.
Further, validation of these clinical prediction rules in other
clinical populations would enhance generalisability. It
should be noted that the sample from which these clinical
prediction rules were derived consisted of injured workers
referred for secondary intervention programs within the New
South Wales Workers Compensation system. The use of the
split-half approach in this study to evaluate the predictive
accuracy of each model enhances the ability to generalise
these prediction rules to similar populations.
This study would have been strengthened by collecting 6
month outcomes from all 720 participants who completed
the program rather than the frst 290 participants. Limited
resources prevented this. However, this sub-cohort appeared
to be representative of the total cohort. Data were sought via
postal questionnaire to participants. Attempts were made
to collect data from non-respondents by telephone and by
contacting the participants insurance company. In future,
collaborative research between clinicians and insurance
companies may provide improved systems for longer term
follow-up.
A further limitation of this study was that information about
the characteristics of the injured workers who dropped out
of the program was not obtained. However, the percentage
of participants who dropped out of the program (15%) is
unlikely to seriously threaten the validity of the study
(Straus et al 2005).
In conclusion, this study makes a unique contribution
to the body of literature regarding prognosis of injured
workers. This is the frst report of clinical prediction rules
with high predictive accuracy that have been developed for
injured workers with persisting musculoskeletal pain. These
algorithms may be a useful adjunct to assist in clinical
decision-making. Further validation of these models on
new samples would be useful as would evaluating the
effectiveness of choice of intervention based on patient
prognoses determined using these clinical prediction rules.
Acknowledgements: This study was supported by Rehab
One Physiotherapy Pty Ltd, Sydney, who provided baseline
and follow-up data for all participants in this study. The
authors would like to acknowledge the staff of Rehab
One Physiotherapy for their co-operation, support, and for
contributing data to this study.
Australian Journal of Physiotherapy 2007 Vol. 53 Australian Physiotherapy Association 2007 275
Hewitt et al: Clinical prediction rules for injured Australian workers
Correspondence: Dr Julia Hush, Faculty of Health Sciences,
University of Sydney, PO Box 170, Lidcombe, NSW 1825,
Australia. Email: j.hush@usyd.edu.au
References
Burton K, Tillotson K, Main C, Hollis S (1995) Psychosocial
predictors of outcome in acute and subchronic low back
trouble. Spine 20: 722728.
Beattie P, Nelson R (2006) Clinical prediction pules: what
are they and what do they tell us? Australian Journal of
Physiotherapy 52:157163.
Childs J, Cleland J (2006) Development and application of
Clinical Prediction Rules to improve decision-making in
physical therapist practice. Physical Therapy 86: 122131.
Chansirinukor W, Maher C, Latimer J, Hush J (2004) Comparison
of the Functional Rating Index and the 18-Item Roland Morris
Disability Questionnaire: responsiveness and reliability. Spine
30: 141145.
Dionne C, Bourbonnais R, Frmont P, Rossignol M, Stock S,
Larocque (2005) A clinical return-to-work rule for participants
with back pain. Canadian Medical Association Journal 172:
15591567.
Feise R, Michael Menke J (2001) A Functional Rating Index: a
new valid and reliable instrument to measure the magnitude
of clinical change in spinal conditions. Spine. 26: 7886.
Fish D E, Chang W (2007) Treatment of ilio-psoas tendinitis
after a total hip arthroplasty with botulinum toxin type A. Pain
Physician 10: 565571.
Fordyce W (1995) Back Pain in the Workplace. Seattle:
International Association fort he Study of Pain.
Fritz J, George S (2002) Identifying psychosocial variables
in participants with acute work-related low back pain: the
importance of fear-avoidance beliefs. Physical Therapy 82:
973983.
Feuerstein M (2005) Introduction: The World Challenge of
Work Disability. Journal of Occupational and Environmental
Medicine 15: 451452.
Gatchel R (2005) Preventing chronic disability in participants with
acute low back pain. Journal of Musculoskeletal Medicine 22:
586588.
Guzman A, Esmail R, Karjalainen K, Almivaara A, Irvin E,
Bombardier C (2003) Multi-Disciplinary Bio-Psychosocial
Rehabilitation for Chronic Low Back Pain. The Cochrane
Library, Issue 2. Oxford: Update Software.
Haldorsen E, Grasdal A, Skouen J, Risa A, Kronholm K, Ursin H
(2002) Is there a right treatment for a particular patient group?
Comparison of ordinary treatment, light multi-disciplinary
treatment and extensive multi-disciplinary treatment for long
term sick listed employees with musculoskeletal pain. Pain
95: 4963.
Hogg-Johnson S, Cole D (2003) Early prognostic factors for
duration on temporary total benefts in the frst year among
workers with compensated occupational soft-tissue injuries.
Occupational and Environmental Medicine 60: 244253.
Hogg-Johnson S, Cole D (1998) Early prognostic factors for
duration on benefts among workers with compensated
occupational soft tissue injuries. Toronto: Institute for Work
and Health, Working paper no 64R1.
Hush J (2006) Functional status questionnaires for spinal pain.
Australian Journal of Physiotherapy 52: 59.
Leigh J, Macaskill P, Kuosma E, Mandryk J (1999) Global
burden of disease and injury due to occupational factors.
Epidemiology 10: 626631.
Linton S, Boersma K (2003) Early identifcation of participants
at risk of developing a persistent back problem: the predictive
validity of the Orebro Musculoskeletal Pain Questionnaire.
The Clinical Journal of Pain 19: 8086.
Linton S, Gross D, Schultz I, Main C, Cote P, Pransky G, Johnson
W (2005) Prognosis and the identifcation of workers risking
disability: research issues and directions for future research.
Journal of Occupational Rehabilitation 15: 459474.
Maher C, Latimer J, Refshauge KM. (2001) Atlas of Clinical
Tests and Measures for Low Back Pain. Melbourne: Australian
Physiotherapy Association.
Pransky G, Verma S, Okurowski L, Webster B (2006) Length
of disability prognosis in acute occupational low back pain.
Spine 31: 690697.
Proctor T, Mayer T, Theodore B, Gatchel R (2005) Failure
to complete a functional restoration program for chronic
musculoskeletal disorders: a prospective 1-year outcome
study. Archives of Physical Medicine and Rehabilitation 86:
15091515.
Randolph A, Guyatt G, Calvin J, Doig G, Richardson W
(Evidence Based Medicine in Critical Care Group) (1998)
Understanding articles describing clinical prediction tools.
Critical Care Medicine 26: 16031612.
Schulte P (2005) Characterizing the burden of occupational injury
and disease. Journal of Occupational and Environmental
Medicine 47: 607622.
Schultz I, Crook J, Berkowitz J, Milner R, Meloche G (2005)
Predicting return to work after low back injury using the
psychosocial risk for occupational disability instrument: a
validation study. Journal of Occupational Rehabilitation 15:
365376.
Straus S, Richardson W, Glasziou P, Haynes R (2005) Evidence
Based Medicine. How to practice and teach EBM (3rd edn).
Edinburgh: Churchill Livingstone.
Stroud MW, McKnight PE, Jensen MP (2004) Assessment of
self reported physical activity in patients with chronic pain:
development of an abbreviated Roland-Morris disability scale.
Journal of Pain 5: 257263.
Sullivan M, Feuerstein M, Gatchel R; Linton S, Pransky G
(2005) Integrating psychosocial and behavioural interventions
to achieve optimal rehabilitation outcomes. Journal of
Occupational Rehabilitation 15: 475489.
Turner J, Franklin G, Fulton-Kehoe D, Sheppard L, Wickizer T,
Wu R, Gluck J, Egan K (2006) Worker recovery expectations
and fear-avoidance predict work disability in a population-
based workers compensation back pain sample. Spine 31:
682689.
Turner J, Franklin G, Turk D (2000) Predictors of chronic
disability in injured workers: a systematic literature synthesis.
American Journal of Industrial Medicine. 38:707722.
van der Giezen A, Bouter L, Nijhuis (2000) Prediction of return-
to-work of low back pain participants sicklisted for 34 months.
Pain 87: 285294.
Waddell G (1996) Low back pain: a twentieth century health
care enigma. Spine 21: 28202825
Waddell G, Burton, A (2001) Occupational health guidelines for
the management of low back pain at work: evidence review.
Occupational Medicine 51: 124135.
Whittink H, Turk DC, Carr DB, Sukiennik A, Rogers W (2004)
Clinical Journal of Pain 20: 133142.
WorkCover New South Wales (2004/5) Workers Compensation
Statistical Bulletin.
Australian Journal of Physiotherapy 2007 Vol. 53 Australian Physiotherapy Association 2007 276
Research