SPINE Volume 27, Number 24, pp 2835–2843

©2002, Lippincott Williams & Wilkins, Inc.

A Clinical Prediction Rule for Classifying Patients with

Low Back Pain Who Demonstrate Short-Term
Improvement With Spinal Manipulation

Timothy Flynn, PT, PhD,*‡ Julie Fritz, PT, PhD,† Julie Whitman, PT, DSc,‡
Robert Wainner, PT, PhD,*‡ Jake Magel, PT, DSc,‡ Daniel Rendeiro, PT, DSc,‡
Barbara Butler, PT,‡ Matthew Garber, PT, DSc,‡ and Stephen Allison, PT, PhD*

Attempts to identify effective interventions for patients

Study Design. A prospective, cohort study of patients with low back pain (LBP) have been largely unsuccess-
with nonradicular low back pain referred to physical ful.65,66 One explanation offered for the lack of evidence
therapy.
Objective. Develop a clinical prediction rule for identi-
is the inability to define subgroups of patients most likely
fying patients with low back pain who improve with spi- to respond to a particular intervention.8,39 Without the
nal manipulation. ability to match patients to specific interventions, clini-
Summary of Background Data. Development of clini- cians are left without evidence or guidance for their de-
cal prediction rules for classifying patients with low back
cision-making. Because of the difficulty in subgrouping
pain who are likely to respond to a particular intervention,
such as manipulation, would improve clinical decision- patients with LBP based on pathoanatomy,67 attempts
making and research. have been made to classify patients based on findings
Methods. Patients with nonradicular low back pain from the history and physical examination.17,49,60,64 De-
underwent a standardized examination and then under- veloping effective, clinically applicable methods for clas-
went a standardized spinal manipulation treatment pro-
gram. Success with treatment was determined using per- sifying patients with LBP could improve decision-
cent change in disability scores over three sessions and making and outcomes by matching interventions to the
served as the reference standard for determining the ac- patients they are most likely to benefit. Classification
curacy of examination variables. Examination variables methods would also enhance the power of clinical re-
were first analyzed for univariate accuracy in predicting
success and then combined into a multivariate clinical
search by permitting researchers to study more homoge-
prediction rule. neous groups of patients.56 Identifying methods for clas-
Results. Seventy-one patients participated. Thirty-two sifying patients with LBP has been identified as an
had success with the manipulation intervention. A clinical important research priority.1,6,66
prediction rule with five variables (symptom duration,
Manipulation is an intervention commonly used in
fear–avoidance beliefs, lumbar hypomobility, hip internal
rotation range of motion, and no symptoms distal to the the treatment of individuals with LBP. Several random-
knee) was identified. The presence of four of five of these ized trials have found manipulation to be more effective
variables (positive likelihood ratio ⫽ 24.38) increased the than placebo52,54,71 or other interventions.16,22,36,63
probability of success with manipulation from 45% to 95%. However, other studies have not shown any benefits for
Conclusion. It appears that patients with low back pain
likely to respond to manipulation can be accurately iden-
manipulation versus other interventions.10,26 –28 The dis-
tified before treatment. [Key words: manipulation, classi- parity may be partly attributable to the admission of a
fication, low back pain, likelihood ratio, clinical prediction heterogeneous group of patients with LBP without an
rule] Spine 2002;27:2835–2843 attempt to identify a priori those likely to benefit from
the intervention. Other randomized trials have found
From the *U.S. Army-Baylor University Graduate Program in Physical manipulation to be more beneficial for a subgroup of
Therapy, San Antonio, Texas, the †Department of Physical Therapy, patients with more acute symptoms30,42 or more limited
University of Pittsburgh, Pittsburgh, Pennsylvania, and the ‡U.S. Army- straight-leg raise range of motion.47 No previous studies
Baylor University, Postprofessional Doctoral Program in Orthopaedic
& Manual Physical Therapy, Fort Sam Houston, Texas. have sought to develop a multifactorial classification rule
Supported in part by a grant from the Foundation for Physical that would maximize the prediction of success with ma-
Therapy. nipulation before the intervention.
Acknowledgment date: October 31, 2001. First revision date: March 7,
2002. Acceptance date: May 6, 2002. One subgroup of patients for whom manipulation is
The manuscript submitted does not contain information about medical proposed to be effective comprises individuals with sac-
device(s)/drug(s).
Foundation funds were received in support of this work. No benefits in
roiliac (SI) region dysfunction.16,19,22,29,62,71 Numerous
any form have been or will be received from a commercial party related clinical findings have been promoted as capable of iden-
directly or indirectly to the subject of this manuscript. tifying patients with SI dysfunction; however, previous
The opinions and assertions contained herein are the private views of
the author and are not to be construed as official or as reflecting the
studies have questioned the reliability and validity of
views of the Department of the Army or the Department of Defense. many of these tests.4,9,20,21,37,40,41,44,45,50,53,61 Previous
Address correspondence to Timothy W. Flynn, PT, PhD, U.S. Army- studies have generally examined individual tests for SI
Baylor University Graduate Program in Physical Therapy, 3151 Scott
Road, Room 1303, Fort Sam Houston, TX 78234-6138, USA; E-mail: dysfunction in isolation and typically have not included
timothy.flynn@cen.amedd.army.mil other potentially important factors from the history or

2835
2836 Spine • Volume 27 • Number 24 • 2002

examination. An exception is the work published by Cib-

ulka et al, who described a cluster of movement and
provocative tests purported to identify SI joint dysfunc-
tion.12–14 However, this test cluster was chosen based on
the authors’ clinical experience, and the predictive valid-
ity is unknown.12 Furthermore, most studies have used
short-term pain relief with an injection of an anesthetic
agent into the SI joint as the reference standard against
which validity has been judged.9,21,32,44 The validity of
using anesthetic joint blocks as a reference standard for
diagnostic studies has not been demonstrated.3,32,59 The
reference standard should represent the condition of in-
terest that the diagnostic test is attempting to identify.55
If the goal of a test is to identify patients likely to respond
to an intervention directed toward the SI region (e.g.,
manipulation), then the use of a criterion standard of
short-term pain relief with an injection is questionable. A Figure 1. Manipulation technique used in this study.
more viable alternative may be a longer duration re-
sponse to the intervention in question. A study using a categorize symptoms as low back, buttock/thigh, or distal to
more “therapeutic” reference standard, which considers the knee based on the distal-most extent of symptoms.70 The
a broader variety of clinical findings and attempts to Modified Oswestry Disability Questionnaire (OSW) assessed
combine clinical findings, will be more useful in identi- disability related to LBP.25 The Fear-Avoidance Beliefs Ques-
fying a subgroup of patients with SI dysfunction who are tionnaire (FABQ) was used to assess the patient’s beliefs about
likely to respond to spinal manipulation. The purpose of the influence of activity on LBP.69 The FABQ contains two
this study was to develop a clinical prediction rule for subscales: one is related to general physical activity and the
identifying patients with LBP likely to respond favorably other to work.69
to a specific manipulation technique. A variety of find- Patients underwent a standardized history and physical ex-
ings were considered against a reference standard of amination. History included mechanism of injury, nature of
current symptoms, and prior episodes of LBP. Patients were
change in disability.
asked to rank sitting, standing, and walking as to which was
Methods best and worst with respect to symptoms. The examination
included Waddell’s nonorganic signs.68 Range of motion and
A prospective cohort study of patients with LBP was conducted status change49 in symptoms with single lumbar movements
at two outpatient facilities: Brooke Army Medical Center and were recorded. Supine straight-leg raise and prone hip rotation
Wilford Hall Air Force Medical Center. Patients were between range of motion were measured. Posteroanterior spring test-
the ages of 18 and 60 years, referred to physical therapy with a ing43 was performed for pain provocation and mobility at each
diagnosis related to the lumbosacral spine, and had a chief lumbar level. Mobility was judged as normal, hypomobile, or
complaint of pain and/or numbness in the lumbar spine, but- hypermobile. Numerous special tests proposed to be diagnostic
tock, and/or lower extremity. The baseline Oswestry disability of SI dysfunction were performed (see Appendix). The tests
score had to be at least 30%. Exclusion criteria were current preg- were divided into three categories: position (tests assessing
nancy, signs consistent with nerve root compression (positive symmetry of bony landmarks), provocation (tests to reproduce
straight-leg raise at ⬍45°, or diminished lower extremity strength, symptoms), and mobility (tests assessing symmetry of pelvic
sensation, or reflexes), prior lumbar spine surgery, or a history of motion).37 The examination was repeated on the first 55 sub-
osteoporosis or spinal fracture. All patients were briefed on the jects by a second therapist blinded to the results of the first
purpose of the study and signed an informed consent approved by examiner to determine the reliability of the physical examina-
the Brooke Army Medical Center and Wilford Hall Air Force tion variables.
Medical Center Institutional Review Boards.
Treatment. Because response to treatment served as the refer-
Therapists. Eight licensed physical therapists participated. ence standard, all patients were treated with the same protocol
Four were residents in the U.S. Army-Baylor Post-Professional for two sessions. At the first session, the therapist performed a
Doctoral Program in Orthopedic and Manual Physical Ther- manipulation technique with the patient supine. The therapist
apy, and four were instructors in the program. This program is stood opposite the side to be manipulated. The patient was
designed to provide physical therapists serving in the U.S. Mil- passively side-bent away from the therapist. The therapist pas-
itary with advanced training in orthopedic and manual physi- sively rotated the patient and then delivered a quick posterior
cal therapy. A 1-day training session was conducted for partic- and inferior thrust through the anterior superior iliac spine
ipating therapists to standardize examination and treatment (Figure 1). The side to be manipulated was determined with the
techniques. following algorithm: first, the side of the positive standing flex-
ion test; if this test was negative, the side of tenderness during
Examination Procedures. Patients completed a baseline ex- sacral sulcus palpation was manipulated. If neither side was
amination including demographic information and a pain rat- tender, the side reported by the patient to be more symptomatic
ing using an 11-point scale.34 A pain diagram46 was used to was manipulated. If the patient was unable to identify a more
 Clinical Prediction Rule for Spinal Manipulation • Flynn et al 2837

symptomatic side, the therapist flipped a coin to determine the

side. Although the manipulation was performed on one side
only, Cibulka et al12 found changes in innominate tilt on both
sides of the pelvis after applying this technique. Therefore,
whereas the algorithm provided a consistent approach, it is
likely the manipulation affected bilateral SI regions.
After the manipulation, the therapist noted whether a cavi-
tation was heard or felt by the therapist or patient. If a cavita-
tion was experienced, the therapist proceeded to the other
treatment components. If no cavitation was produced, the pa-
tient was repositioned, and the manipulation was attempted
again. If no cavitation was experienced again, the therapist
attempted to manipulate the opposite side. A maximum of two
attempts per side was permitted. If no cavitation was produced
after the fourth attempt, the therapist proceeded to the other
treatment components. Two additional treatment components
were included: 1) instruction in a supine pelvic tilt range of
motion exercise (the patient was instructed to perform 10 rep- Figure 2. Initial and final Oswestry scores for the success and
nonsuccess groups. The mean percent change in the success
etitions, 3– 4 times daily; and 2) instruction to maintain usual
group was 73.2 ⫾ 15.8%. For the nonsuccess group, the mean
activity level within the limits of pain.5 percent change was 14.6 ⫾ 18.2%.
The second treatment session occurred 2– 4 days after the
first. Before the second session, each patient completed an
OSW questionnaire. Percentage improvement in OSW was cal-
culated (initial score ⫺ final score)/initial score ⫻ 100). If im- moderate shifts, and values ⬎10.0 generate large and often
provement was ⬎50%, the patient was categorized as a suc- conclusive shifts in probability. We chose to focus on the PLR
cess, and study participation ended. If the patient showed as opposed to the negative LR because we were attempting to
⬍50% improvement, the therapist repeated the examination predict success with manipulation based on positive test re-
and the manipulation procedures. The third session occurred sults. Potential prediction variables were entered into a step-
2– 4 days after the second. The patient again completed the wise logistic regression equation to determine the most parsi-
OSW, and the percentage improvement from the initial score monious set of predictors for success using a multivariate
was calculated. If ⬎50% improvement was noted, the patient model. A significance of 0.05 was required to enter a variable
was categorized as a success. If improvement was ⱕ50%, the into the model and a significance of 0.10 was required to re-
patient was categorized as a nonsuccess. At this point, the pa- move it. Variables retained in the regression model were used to
tient’s study participation ended, and further treatment was develop a multivariate clinical prediction rule for classifying
administered as needed. patients as likely responders to manipulation.

Results
Data Analysis. Kappa coefficients were calculated to deter-
mine the interrater reliability of the special tests for SI dysfunc- Seventy-five patients entered the study. Four subjects
tion. Patients were dichotomized based on success or nonsuc- (5%) did not return after the first session and were not
cess with respect to the treatment. Success or nonsuccess was included in the analysis. Two subjects left the study be-
then used as the reference standard. Individual variables from cause of personal or work-related circumstances. One
the self-reports, history, and physical examination were tested subject dropped out because of complications from an
for their univariate association with the reference standard us- ongoing episode of gastrointestinal distress, and one sub-
ing independent sample t tests for continuous variables and ␹2
ject failed to return for his final visit. Of the 71 patients
tests for categorical variables. Variables with a significance
level of P ⬍ 0.15 were retained as potential prediction vari-
completing the study, 29 (41%) were female and 59
ables; a more liberal significance level was chosen at this stage (83%) had a prior history of LBP. The mean age was
to avoid excluding potential predictive variables. For continu- 37.6 ⫾ 10.6 years (range 18 –59 years). The mean OSW
ous variables with a significant univariate association, sensitiv- score at baseline was 42.4 ⫾ 11.7, and at study conclu-
ity and specificity values were calculated for all possible cut-off sion 25.1 ⫾ 13.9. The mean percent improvement in
points and then plotted as a receiver operator characteristic OSW over the study period was 41.0 ⫾ 33.9% (range
(ROC) curve.31 The point on the curve nearest the upper left- ⫺29.0 –100%). Thirty-two patients (45%) were classi-
hand corner represents the value with the best diagnostic accu- fied as treatment successes, and 39 (55%) were nonsuc-
racy, and this point was selected as the cut-off defining a posi- cesses. Twenty patients were successes after one manipula-
tive test.18 Sensitivity, specificity, and positive likelihood ratios tion session, 12 after two sessions. The mean improvement
(PLR) were calculated for all potential prediction variables.57
in OSW in the success group over the study period was 32.5
The PLR is calculated as sensitivity/(1 ⫺ specificity) and indi-
cates the increase in the probability of success given a positive
⫾ 12.6 points, with a mean percent improvement of 73.2 ⫾
test result.58 A PLR of 1 indicates the test does nothing to alter 15.8%. In the nonsuccess group, the mean OSW improve-
the probability of success, whereas PLR values ⬎1 increase the ment was 6.2 ⫾ 7.8 points, with a mean percent improve-
probability of success given a positive test result. According to ment of 14.6 ⫾ 18.2% (Figure 2).
Jaeschke et al,33 PLR values between 2.0 and 5.0 generate small Among self-reported variables (Table 1), the FABQ
shifts in probability, values between 5.0 and 10.0 generate work subscale, the presence of symptoms in the back
2838 Spine • Volume 27 • Number 24 • 2002

Table 1. Self-Report Variables Used in the Study

Manipulation Success Manipulation
Self-Report Variable All Subjects (n ⫽ 71) (n ⫽ 32) Nonsuccess (n ⫽ 39) Significance

Pain rating 5.3 (2.0) 5.3 (2.0) 5.2 (1.7) 0.74

Oswestry score 42.4 (11.7) 44.4 (14.4) 40.8 (8.7) 0.21
Fear avoidance beliefs
Work subscale 12.9 (10.3) 9.1 (8.3) 16.0 (10.8) 0.003
Physical activity subscale 15.1 (5.3) 14.9 (5.0) 15.2 (5.6) 0.85
Pain diagram (%)
Low back symptoms only 31 41 23 0.11
Buttock/thigh symptoms present 76 69 82 0.19
Symptoms distal to knee 25 13 36 0.024
Values are means (SD) unless otherwise indicated.

only, and symptoms distal to the knee were retained as of their similarity, left and right hip internal rotations
potential prediction variables. From the history (Table were combined into a single variable. Cut-off scores and
2), duration of symptoms, increasing episode frequency, accuracy statistics for retained variables were calculated
and ranking standing as the worst position were re- (Table 5). Among historical variables, duration of symp-
tained. Five variables were retained from the clinical ex- toms ⬍16 days was most predictive of success (PLR ⫽
amination (Table 3): left and right hip internal rotation, 4.39).
hypomobility and pain with lumbar spring testing, and The 11 potential prediction variables were entered
peripheralization with single lumbar movement testing. into the logistic regression. Five were retained in the final
The special tests for SI dysfunction along with the reli- model: duration of symptoms ⬍16 days, at least one hip
ability coefficients are presented in Table 4. As a group, with ⬎35° of internal rotation, hypomobility with lum-
the provocation tests were more reliable with values bar spring testing, FABQ work subscale score ⬍19, and
ranging from fair to substantial agreement. The Gillet no symptoms distal to the knee (model ␹2 ⫽ 48.5, df ⫽ 5,
test was the only motion test to demonstrate at least P ⬍ 0.001, Nagelkerke R2 ⫽ 0.67). These five variables
moderate agreement. The anterior superior iliac spine were used to form the clinical prediction rule. Only six
and iliac crest in standing and the posterior superior iliac subjects (all in the success group) were positive for all five
spine symmetry in sitting were the only symmetry tests to retained prediction variables at baseline (Table 6). Four-
have at least fair agreement. Among the special tests for teen of 15 subjects with 4 of 5 variables present were in
SI dysfunction (Table 4), only the compression– the success group. Of subjects with two or fewer vari-
distraction test was retained in the predictive model, al- ables present, 25 of 27 were in the nonsuccess group.
though positive findings were more common in the non- Accuracy statistics were calculated for each level of the
success group. Cut-off values for retained continuous clinical prediction rule (Table 7). Based on the pretest
variables (duration of symptoms, FABQ work subscale, probability of success with manipulation found in this
left and right hip internal rotation) were obtained from study (45%), and the PLR values calculated, a subject
receiver operator characteristic curve analyses. Because with four or more variables present at baseline increases

Table 2. Variables From the Patient History Used in This Study

Manipulation Manipulation
Variable All Subjects Success Nonsuccess Significance

Age (yrs) 37.6 (10.6) 39.5 (11.7) 36.1 (9.1) 0.19

Gender (% female) 41 47 36 0.35
Duration of symptoms (days) 41.7 (54.7) 23.5 (32.5) 56.7 (64.4) 0.007
Mode of onset (%)
Gradual 31 26 33 0.64
Sudden (minimal/no perturbation) 39 48 33 0.24
Traumatic (fall, lifting, pulling, etc.) 30 26 33 0.44
Prior history of low back pain (%) 83 88 79 0.37
Episodes of low back pain becoming more frequent (%) 35 25 44 0.10
Sitting ranked as best position (%) 31 31 31 0.97
Standing ranked as best position (%) 20 25 15 0.31
Walking ranked as best position (%) 23 16 28 0.21
Sitting ranked as worst position (%) 41 47 36 0.35
Standing ranked as worst position (%) 27 19 36 0.055
Walking ranked as worst position (%) 13 16 10 0.50
Values are means (SD) unless otherwise indicated.
 Clinical Prediction Rule for Spinal Manipulation • Flynn et al 2839

Table 3. Variables From the Baseline Clinical Examination Used in This Study
Manipulation Manipulation
Variable All Subjects Success Nonsuccess Significance

Nonorganic signs* 0.70 (1.0) 0.48 (0.77) 0.74 (1.1) 0.25

Total flexion (°) 75.7 (30.0) 75.3 (31.2) 81.1 (25.7) 0.40
Pelvic flexion (°) 38.0 (20.7) 37.4 (19.9) 41.1 (20.5) 0.45
Lumbar flexion (°) 38.3 (15.4) 37.6 (18.4) 40.0 (12.5) 0.58
Total extension (°) 21.5 (9.5) 21.4 (8.2) 21.4 (9.8) 0.98
Left side bending (°) 27.5 (7.8) 27.6 (7.7) 28.1 (8.0) 0.80
Right side bending (°) 27.3 (7.2) 27.0 (6.4) 28.1 (7.5) 0.53
Left and right side bending discrepancy† (°) 5.2 (4.7) 6.0 (4.4) 4.7 (5.0) 0.26
Left straight leg raise (°) 67.3 (16.8) 69.2 (14.9) 68.5 (16.3) 0.85
Right straight leg raise (°) 67.5 (16.7) 68.6 (14.3) 69.2 (16.7) 0.87
Left and right straight-leg raise discrepancy† (°) 6.0 (6.9) 5.6 (5.6) 6.5 (8.1) 0.59
Left hip internal rotation (°) 29.6 (9.8) 33.0 (9.7) 27.2 (9.3) 0.012
Right hip internal rotation (°) 30.3 (12.5) 29.6 (10.4) 25.9 (10.7) 0.15
Left and right hip internal rotation discrepancy† (°) 6.4 (5.7) 6.8 (6.5) 6.0 (5.3) 0.58
Left hip external rotation (°) 30.1 (11.9) 31.3 (13.3) 29.5 (12.9) 0.48
Right hip external rotation (°) 30.3 (12.5) 31.8 (12.6) 29.3 (11.1) 0.45
Left and right hip external rotation discrepancy† (°) 6.1 (5.3) 5.8 (5.3) 6.6 (5.4) 0.53
Hypomobility at one or more lumbar levels with spring testing (%) 86 97 77 0.016
Pain at one or more lumbar levels with spring testing (%) 92 97 87 0.14
Lateral shift present (%) 13 13 13 0.97
Peripheralizes with lumbar single movement testing (%) 25 16 33 0.088
Centralizes with lumbar single movement testing (%) 6 6 5 0.84
Values represent means (standard deviation) unless otherwise indicated.
* Mann-Whitney test used due to non-normal data distribution.
† Discrepancy values were calculated as the absolute value of the left minus the right.

his or her probability of success with manipulation from Discussion

45% to 95%. If the criteria were changed to three or
more variables present, the probability of success was Clinicians who routinely use spinal manipulation have
only increased to 68%. If two or fewer variables were encountered patients who experience a rapid, even dra-
present, the probability of success was virtually matic improvement as a result of one or two treatments,
unchanged. whereas others change very little. The ability to accu-

Table 4. Special Tests for SI Dysfunction Used in This Study

Manipulation Manipulation
Reliability All Subjects (%) Success (%) Nonsuccess (%) Significance

Provocation tests
Posterior shear test 0.70 56 56 56 0.99
Sacral sulcus test 0.64 67 75 62 0.23
Patrick test 0.60 44 47 41 0.62
Gaenslen test 0.54 44 47 44 0.78
Resisted hip abduction 0.41 36 34 38 0.72
Sacral thrust test 0.41 54 63 49 0.25
Compression–distraction test 0.26 23 16 31 0.14
Motion tests
Gillet test 0.59 60 66 56 0.43
Seated flexion test 0.25 66 72 59 0.26
Long-sitting test 0.21 46 39 51 0.35
Prone knee bend test 0.21 56 47 62 0.22
Standing flexion test ⫺0.08 54 47 62 0.22
Symmetry tests
ASIS symmetry in standing 0.31 36 41 33 0.53
Iliac crest symmetry in standing 0.23 33 34 33 0.93
PSIS symmetry in sitting 0.23 31 31 31 0.97
PSIS symmetry in standing 0.13 41 44 38 0.65
Ischial tuberosity heights in prone 0.03 41 35 46 0.43
Pubic tubercle symmetry in supine ⫺0.04 49 41 56 0.18
ASIS ⫽ anterior superior iliac spine; PSIS ⫽ posterior superior iliac spine.
Reliability measured with kappa and weighted kappa statistics. Percentage values represent the percentage of positive tests in all patients and the success and
nonsuccess groups.
2840 Spine • Volume 27 • Number 24 • 2002

Table 5. Accuracy Statistics (With 95% Confidence Intervals) for Individual Variables for Predicting Success
Variable Associated With Success Sensitivity Specificity Positive Likelihood Ratio

Duration of symptoms ⱕ15 days 0.56 (0.39, 0.72) 0.87 (0.73, 0.94) 4.39 (1.83, 10.51)
Episodes not becoming more frequent 0.75 (0.58, 0.87) 0.44 (0.29, 0.59) 1.33 (0.95, 1.87)
Standing not ranked as worst position 0.84 (0.67, 0.93) 0.36 (0.23, 0.52) 1.31 (1.0, 1.74)
FABQ work subscale ⱕ18 0.84 (0.68, 0.93) 0.49 (0.34, 0.64) 1.65 (1.17, 2.31)
Symptoms in the low back only 0.41 (0.26, 0.58) 0.77 (0.62, 0.87) 1.76 (0.87, 3.58)
Symptoms not distal to the knee 0.88 (0.72, 0.95) 0.36 (0.23, 0.52) 1.36 (1.04, 1.79)
At least one hip internal rotation range of motion ⬎35° 0.50 (0.34, 0.66) 0.85 (0.70, 0.93) 3.25 (1.44, 7.33)
Hypomobility at one or more lumbar levels with spring testing 0.97 (0.84, 0.99) 0.23 (0.13, 0.38) 1.26 (1.05, 1.51)
Pain at one or more lumbar levels with spring testing 0.97 (0.84, 0.99) 0.13 (0.056, 0.27) 1.11 (0.97, 1.27)
Does not peripheralize with lumbar single movement testing 0.84 (0.68, 0.93) 0.33 (0.21, 0.49) 1.27 (0.97, 1.65)
Negative compression/distraction test 0.84 (0.68, 0.93) 0.31 (0.19, 0.46) 1.22 (0.94, 1.58)
FABQ ⫽ Fear-Avoidance Beliefs Questionnaire.

rately predict which patients will have which response a Manipulation is thought to be indicated in the presence
priori would be immensely beneficial for clinical deci- of hypomobility. Interestingly, although the technique
sion-making. Similar to other studies,21,40,41,44 we were used in this study is described as affecting the SI region, it
unable to show acceptable accuracy for any individual was lumbar hypomobility that entered the prediction
tests proposed to identify SI dysfunction. Furthermore, model. This finding reinforces the idea that the manipu-
we found that the reliability of these tests in a population lation technique is not specific to the SI region but im-
of individuals with LBP is less than optimal. As noted by pacts the lumbar spine as well.7,17,29 Manipulation is
previous researchers, provocation tests as a whole are generally thought to be contraindicated in patients with
more reliable tests than motion or symmetry tests.9,37 radiculopathy.5 We excluded patients with signs of nerve
However, by considering other variables and combining root compression. However, some patients with symp-
findings, we were able to develop a clinical prediction toms distal to the knee were included, and these patients
rule that may be useful for assisting clinicians in classi- tended not to succeed. Finally, the FABQ quantifies a
fying patients as likely to respond to this manipulation patient’s fear of pain and subsequent avoidance of activ-
technique. ity.69 The FABQ work subscale has been previously cor-
The developed clinical prediction rule contains five related with work loss and disability in patients with
variables: duration of symptoms ⬍16 days, at least one chronic and acute LBP.24,35,69 Our results suggest that
hip with ⬎35° of internal rotation, lumbar hypomobil- patients with high levels of fear-avoidance beliefs about
ity, no symptoms distal to the knee, and an FABQ work work activities are unlikely to respond to manipulation.
score ⬍19. These findings are generally consistent with These individuals likely require an alternative treatment
previous theories and research. Randomized trials have approach.15
suggested that patients with more acute symptoms re- The usefulness of a clinical prediction rule for classi-
spond better to manipulation.30,42 Our results support
fying patients is best expressed using likelihood ratio
this hypothesis. Hip rotation range of motion discrepan-
statistics. The PLR expresses the change in odds favoring
cies have been reported in patients with LBP.2,14,51 Pre-
the outcome when the patient meets the prediction rule’s
vious studies in patients with “nonspecific” LBP have
criteria.57 In our sample, 45% of subjects were successful
found greater external rotation than internal rota-
without any attempt at prediction. In other words, ran-
tion.2,11,51 As a whole, patients in this study had greater
domly manipulating individuals with nonradicular LBP
external than internal rotation; however, increased inter-
may result in success about 45% of the time. Using a
nal rotation was associated with manipulation success.
criterion of at least 4 of 5 variables present at baseline
(PLR ⫽ 24.38), the probability of success is raised to
Table 6. Number of Subjects in the Success and 95%; therefore, these individuals should be manipu-
Nonsuccess Groups at Each Level of the Clinical lated. If only three variables are present, the probability
Prediction Rule increased to 68%, which is likely sufficient to warrant an
attempt at manipulation in these patients. When two or
No. of Predictor No. of Subjects in No. of Subjects in fewer variables are present, the probability of success
Variables the Manipulation the Manipulation
Present Success Group Nonsuccess Group changes little, and clinicians should consider alternative
treatments if such can be identified that may have a prob-
5 6 0 ability of success ⬎45%.
4 14 1
3 10 13 An important consideration in the examination of di-
2 2 19 agnostic tests is the reference standard against which
1 0 5 tests are judged. Previous studies of tests for SI dysfunc-
0 0 1
tion have generally used immediate pain relief with SI
 Clinical Prediction Rule for Spinal Manipulation • Flynn et al 2841

Table 7. Clinical Prediction Rule

No. of Predictor Probability of
Variables Present Sensitivity Specificity Positive Likelihood Ratio Success* (%)

5 0.19 (0.09, 0.35) 1.00 (0.91, 1.00) infinite (2.02, infinite) —

4⫹ 0.63 (0.45, 0.77) 0.97 (0.87, 1.0) 24.38 (4.63, 139.41) 95
3⫹ 0.94 (0.80, 0.98) 0.64 (0.48, 0.77) 2.61 (1.78, 4.15) 68
2⫹ 1.00 (0.89, 1.0) 0.15 (0.07, 0.30) 1.18 (1.09, 1.42) 49
1⫹ 1.00 (0.89, 1.0) 0.03 (0.005, 0.13) 1.03 (1.01, 1.15) 46
* The probability of success is calculated using the positive likelihood ratio and assumes a pretest probability of success of 45%. Accuracy statistics with 95%
confidence intervals for individual variables for predicting success.

joint anesthetic injection. In our opinion, clinicians per- prove outcomes and clinical decision-making before it can
forming these tests are not as interested in pathoana- be advocated for widespread use.38,48
tomic speculations (i.e., is the SI joint generating the
pain?) as they are in determining if the patient will re-
Key Points
spond to a particular intervention. We therefore used a
reference standard representative of the desired outcome ● Special tests purported to identify patients with
of the tests (i.e., responding to manipulation). The use of low back pain who will respond to manipulation
50% improvement on the OSW as the reference standard were largely unsuccessful in doing so.
was based on previous research involving the interven- ● The best univariate predictor of success with ma-
tion used in this study. In three previous studies, patients nipulation was the duration of the current symp-
thought to be matched to this intervention experienced toms; more acute symptoms were more likely to
mean improvements in OSW scores from 57% to 83%, respond favorably.
whereas patients receiving unmatched interventions ex- ● Five variables were identified to form a clinical pre-
perienced mean improvements ranging from 20% to diction rule for patients with low back pain likely to
38% over a 1– 4-week period.16,22,23 We therefore respond favorably to spinal manipulation: duration
thought that requiring 50% improvement in the OSW of symptoms ⬍16 days, FABQ work subscale score
over a 2– 4-day period would provide adequate distinc- ⬍19, at least one hip with ⬎35° of internal rotation
range of motion, hypomobility in the lumbar spine,
tion between patients responding to the intervention and
and no symptoms distal to the knee.
those simply benefiting from the favorable natural his-
● The presence of four of five variables in the pre-
tory of LBP.
diction rule increased the likelihood of success with
The patients participating in this study should be rep-
manipulation from 45% to 95%.
resentative of patients seeking physical therapy services
in large metropolitan areas. The eight physical therapists
involved in the study had varying degrees of skills in References
spinal manipulative therapy. However, the manipulation 1. Clinical research agenda for physical therapy. Phys Ther 2000;80:499 –513.
technique employed is a standard technique used in 2. Barbee-Ellison JB, Rose SJ, Sahrmann SA. Patterns of hip rotation range of
motion: comparisons between healthy subjects and patients with low back
physical therapist education programs. Therefore, the pain. Phys Ther 1990;70:537– 41.
results should be generalizable to outpatient clinics treat- 3. Barnsley L, Lord S, Wallis B, et al. False-positive rates of cervical zygapoph-
ysial joint blocks. Clin J Pain 1993;9:124 –30.
ing individuals with LBP. 4. Bemis T, Daniel M. Validation of the long sitting test on subjects with ilio-
A three-step process for developing and testing a clin- sacral dysfunction. J Orthop Sports Phys Ther 1987;8:336 – 45.
ical prediction rule is recommended.48 The first step is 5. Bigos S, Bowyer O, Braen G, et al. Acute low back problems in adults
[AHCPR Publication No. 95-0642]. Rockville, MD: Agency for Health Care
developing the rule, the second step is validation, and the Policy and Research, Public Health Service, U.S. Department of Health and
third step is an assessment of the impact of the rule on Human Services, 1994.
6. Borkan JM, Koes B, Reis S, et al. A report from the second international
clinical behavior. The purpose of the present study was
forum for primary care research on low back pain: reexamining priorities.
to develop a clinical prediction rule that would identify Spine 1998;23:1992– 6.
individuals with LBP who respond favorably to a specific 7. Bourdillon JF, Day EA, Bookhout MR. Spinal Manipulation. 5th ed. Oxford:
Butterworth-Heinemann, 1992;156 – 8.
spinal manipulation. In the present study, only one ma- 8. Bouter LM, van Tulder MW, Koes BW. Methodologic issues in low back
nipulation technique was used, and it is unknown pain research in primary care. Spine 1998;23:2014 –20.
whether other techniques would provide similar results. 9. Broadhurst NA, Bond MJ. Pain provocation for the assessment of sacroiliac
joint dysfunction. J Spinal Dis 1998;11:341–5.
Validation of the proposed clinical prediction rule is the 10. Cherkin DC, Deyo RA, Battie M, et al. A comparison of physical therapy,
purpose of an ongoing randomized controlled trial where chiropractic manipulation, and provision of an educational booklet for the
subjects meeting the prediction criteria receive either the treatment of patients with low back pain. N Engl J Med 1998;339:1021–9.
11. Chesworth BM, Padfield BJ, Helewa A, et al. A comparison of hip mobility
spinal manipulation technique or a competing therapy. Ul- in patients with nonspecific low back pain. Physiother Can 1994;46:267–74.
timately, any clinical prediction rule must be shown to im- 12. Cibulka MT, Delitto A, Koldehoff R. Changes in innominate tilt after ma-
2842 Spine • Volume 27 • Number 24 • 2002

nipulation of the sacroiliac joint in patients with low back pain. Phys Ther 41. Levangie PK. The association between static pelvic asymmetry and low back
1988;68:1359 – 63. pain. Spine 1999;24:1234 – 42.
13. Cibulka MT, Koldehoff R. Clinical usefulness of a cluster of sacroiliac joint 42. MacDonald RS, Bell CMJ. An open controlled assessment of osteopathic
tests in patients with and without low back pain. J Orthop Sports Phys Ther manipulation in nonspecific low-back pain. Spine 1990;15:364 –70.
1999;29:83–92. 43. Maher CG, Latimer J, Adams R. An investigation of the reliability and
14. Cibulka MT, Sinacore DR, Cromer GS, et al. Unilateral hip rotation range of validity of posteroanterior spinal stiffness judgments made using a reference-
motion asymmetry in patients with sacroiliac joint regional pain. Spine 1998; based protocol. Phys Ther 1998;78:829 –37.
23:1009 –15. 44. Maigne J-Y, Aivaliklis A, Pfefer F. Results of sacroiliac joint double block
15. Crombez G, Vlaeyen JW, Heuts PH, et al. Pain-related fear is more disabling and value of sacroiliac pain provocation tests in 54 patients with low back
than fear itself: evidence on the role of pain-related fear in chronic back pain pain. Spine 1996;21:1889 –92.
disability. Pain 1999;80:329 –39. 45. Maigne J-Y, Boulahdour H, Chatellier G. Value of quantitative radionuclide
16. Delitto A, Cibulka MT, Erhard RE, et al. Evidence for use of an extension- bone scanning in the diagnosis of sacroiliac joint syndrome in 32 patients
mobilization category in acute low back syndrome: a prescriptive validation with low back pain. Eur Spine J 1998;7:328 –31.
pilot study. Phys Ther 1993;73:216 –28. 46. Mann NH, Brown MD, Hertz DB, et al. Initial-impression diagnosis using
17. Delitto A, Erhard RE, Bowling RW. A treatment-based classification ap- low-back pain patient pain drawings. Spine 1993;18:41–53.
proach to low back syndrome: identifying and staging patients for conser-
47. Mathews W, Morkel M, Mathews J. Manipulation and traction for sciatica
vative management. Phys Ther 1995;75:470 – 89.
and lumbago: physiotherapeutic techniques in two controlled trials. Phys-
18. Deyo RA, Centor RM. Assessing the responsiveness of functional scales to
iother Pract 1987;4:201– 6.
clinical change: an analogy to diagnostic test performance. J Chronic Dis
48. McGinn TG, Guyatt GH, Wyer PC, et al. Users’ guide to the medical litera-
1986;11:897–906.
ture: XXII. How to use articles about clinical decision rules. JAMA 2000;
19. DonTigny R. Dysfunction of the sacroiliac joint and its treatment. J Orthop
284:79 – 84.
Sports Phys Ther 1979;1:23–9.
49. McKenzie RA. The Lumbar Spine: Mechanical Diagnosis and Therapy.
20. Dreyfuss P, Dreyer S, Griffin J, et al. Positive sacroiliac screening tests in
Waikanae, New Zealand: Spinal Publications, 1989.
asymptomatic adults. Spine 1994;19:1138 – 43.
21. Dreyfuss P, Michaelsen M, Pauza K, et al. The value of medical history and 50. Meijne W, van Neerbos K, Aufdemkampe G, et al. Intraexaminer and interex-
physical examination in diagnosing sacroiliac joint pain. Spine 1996;21: aminer reliability of the Gillet test. J Manipulative Physiol Ther 1999;22:4 –9.
2594 – 602. 51. Mellin G. Correlation of hip mobility with degree of back pain and lumbar
22. Erhard RE, Delitto A, Cibulka MT. Relative effectiveness of an extension spinal mobility in chronic low back pain patients. Spine 1988;13:668 –70.
program and a combined program of manipulation and flexion and exten- 52. Postacchini F, Facchini M, Palieri P. Efficacy of various forms of conservative
sion exercises in patients with acute low back syndrome. Phys Ther 1994; treatments in low back pain: a comparative study. Neuro Orthop 1988;6:
74:1093–100. 28 –35.
23. Fritz JM, George S. The use of a classification approach to identify subgroups 53. Potter NA, Rothstein JM. Intertester reliability of selected clinical tests of the
of patients with acute low back pain: inter-rater reliability and short-term sacroiliac joint. Phys Ther 1985;65:1671–5.
treatment outcomes. Spine 2000;25:106 –14. 54. Rasmussen GG. Manipulation in treatment of low-back pain: a randomized
24. Fritz JM, George SZ, Delitto A. The role of fear avoidance beliefs in acute clinical trial. Manual Med 1979;1:8 –10.
low back pain: relationships with current and future disability and work 55. Reid MC, Lachs MS, Feinstein AR. Use of methodological standards in
status. Pain 2001;94:7–15. diagnostic test research: getting better but still not good. JAMA 1995;274:
25. Fritz JM, Irrgang JJ. A Comparison of a Modified Oswestry Disability Question- 645–51.
naire and the Quebec Back Pain Disability Scale. Phys Ther 2001;81:776–88. 56. Rose SJ. Physical therapy diagnosis: role and function. Phys Ther 1989;69:
26. Gibson T, Grahame R, Harkness J, et al. Controlled comparison of short- 535–7.
wave diathermy treatment with osteopathic treatment in non-specific low- 57. Sackett DL. A primer on the precision and accuracy of the clinical examina-
back pain. Lancet 1985;1:1258 – 60. tion. JAMA 1992;267:2638 – 44.
27. Glover JR, Morris JG, Kholsa T. Back pain: a randomized clinical trial of 58. Sackett DL, Haynes RB, Guyatt GH, et al. Clinical Epidemiology: A Basic Sci-
rotational manipulation of the trunk. Br J Ind Med 1974;31:59 – 64. ence for Clinical Medicine. 2nd ed. Boston, MA: Little Brown, 1991:119 –39.
28. Godfrey CM, Morgan PP, Schatzker J. A randomized trial of manipulation 59. Schwarzer AC, Aprill CN, Derby R, et al. The false positive rate of uncon-
for low back pain in a medical setting. Spine 1984;9:301– 4. trolled diagnostic blocks of the lumbar zygapophyseal joints. Pain 1994;58:
29. Greenman P. Osteopathic manipulation of the lumbar spine and pelvis. In: 195–200.
White A, Anderson A, eds. Conservative Care of Low Back Pain. Baltimore, 60. Sikorski JM. A rationalized approach to physiotherapy for low back pain.
MD: Williams & Wilkins, 1991:210 –5. Spine 1985;10:571–9.
30. Hadler NM, Curtis P, Gillings DB, et al. A benefit of spinal manipulation as 61. Slipman CW, Sterenfeld EB, Chou LH, et al. The predictive value of provoc-
an adjunctive therapy for acute low-back pain: a stratified controlled study. ative sacroiliac joint stress maneuvers in the diagnosis of sacroiliac joint
Spine 1987;12:703–5. syndrome. Arch Phys Med Rehabil 1998;79:288 –92.
31. Hagen MD. Test characteristics: how good is that test? Med Decis Making 62. Sutton SE. Postural imbalances: examination and treatment utilizing flexion
1995;22:213–33. tests. J Am Osteopath Assoc 1978;77:456 – 65.
32. Hogan QH, Abram SE. Neural blockade for diagnosis and prognosis: a 63. Triano JJ, McGregor M, Hondras MA, et al. Manipulative therapy versus
review. Anesthesiology 1997;86:216 – 41. education programs in chronic low back pain. Spine 1995;20:948 –55.
33. Jaeschke R, Guyatt G, Sackett DL. Users’ guides to the medical literature: III. 64. Van Dillen LR, Sahrmann SA, Norton BJ, et al. Reliability of physical exam-
How to use an article about a diagnostic test. A. Are the results of the study ination items used for classification of patients with low back pain. Phys Ther
valid? JAMA 1994;271:389 –91. 1998;78:979 – 88.
34. Jensen MP, Turner JA, Romano JM. What is the maximum number of levels 65. van Tulder MW, Koes BW, Bouter LM. Conservative treatment of acute and
needed in pain intensity measurement? Pain 1994;58:387–92. chronic nonspecific low back pain: a systematic review of randomized con-
35. Klenerman L, Slade PD, Stanley IM, et al. The prediction of chronicity in trolled trials of the most common interventions. Spine 1997;22:2128 –56.
patients with an acute attack of low back pain in a general practice setting. 66. van Tulder MW, Malmivaara A, Esmail R, et al. Exercise therapy for low
Spine 1995;20:478 – 84. back pain: a systematic review within the framework of the Cochrane Col-
36. Koes BW, Bouter LM, van Mameren H, et al. The effectiveness of manual laboration Back Review Group. Spine 2000;25:2784 –96.
therapy, physiotherapy and treatment by the general practitioner for back 67. Waddell G. Low back pain: a twentieth century health care enigma. Spine
and neck complaints: a randomized clinical trial. Spine 1992;17:28 –35. 1996;21:2820 –5.
37. Laslett M, Williams M. The reliability of selected provocation tests for sac- 68. Waddell G, McCulloch JA, Kummel E, et al. Nonorganic signs in low-back
roiliac joint pathology. Spine 1994;19:1245–9. pain. Spine 1980;5:117–25.
38. Laupacis A, Sekar N, Stiell IG. Clinical prediction rules: a review and suggested 69. Waddell G, Newton M, Henderson I, et al. A Fear-Avoidance Beliefs Ques-
modifications of methodological standards. JAMA 1997;277:488 –94. tionnaire (FABQ), the role of fear-avoidance beliefs in chronic low back pain
39. Leboeuf-Yde C, Lauritsen JM, Lauritzen T. Why has the search for causes of and disability. Pain 1993;52:157– 68.
low back pain largely been nonconclusive? Spine 1997;22:877– 81. 70. Werneke M, Hart DL, Cook D. A descriptive study of the centralization
40. Levangie PK. Four clinical tests of sacroiliac joint dysfunction: the associa- phenomenon: a prospective analysis. Spine 1999;24:676 – 83.
tion of test results with innominate torsion among patients with and without 71. Wreje U, Nordgren B, Aberg H. Treatment of pelvic joint dysfunction in
low back pain. Phys Ther 1999;79:1043–57. primary care: a controlled study. Scand J Prim Care 1992;10:310 –5.
 Clinical Prediction Rule for Spinal Manipulation • Flynn et al 2843

Appendix
Special Tests for SI Dysfunction

Procedure Criteria for Positive

Tests for symmetry

PSIS symmetry standing17 Palpation of right and left PSIS with the patient standing One PSIS judged to be higher than the other
ASIS symmetry standing17 Palpation of right and left ASIS with the patient standing One ASIS judged to be higher than the other
Iliac crest symmetry Palpation of right and left iliac crest with the patient standing One iliac crest judged to be higher than the other
standing17
Landmark symmetry Comparison of PSIS, ASIS, and iliac crest findings in standing All landmarks are not level and all landmarks are
standing17 not high on the same side
PSIS symmetry sitting13 Palpation of right and left PSIS with the patient sitting One PSIS judged to be higher than the other
Pubic tubercle symmetry Palpation of the right and left tubercle with the patient supine One pubic tubercle judged to be higher than the
supine29 other
Ischial tuberosity symmetry Palpation of the right and left ischial tuberosity with the One ischial tuberosity judged to be higher than
in prone29 patient prone the other
Motion tests
Standing flexion test13,62 The patient is standing and the relative heights of the PSIS A change in the relative relationship of the PSIS
are assessed. The patient is asked to flex forward as far as is found in the fully flexed position.
possible, with the examiner continuing to palpate the PSIS.
Seated flexion test17,62 The patient is seated and the relative heights of the PSIS are A change in the relative relationship of the PSIS
judged. The patient is asked to bend forward as far as is found in the fully flexed position.
possible, with the examiner continuing to palpate the PSIS.
Long-sitting test4,13 The patient is supine with hips and knees extended. The A change in the relative position of medial
examiner grasps around each ankle with the thumbs below malleoli occurs.
the medial malleoli. A visual estimation of leg length is
made. The patient is assisted to a long-sitting position, and
the examiner reexamines the relative leg lengths.
Prone knee bend test12, 17
The patient is prone. The relative leg lengths are assessed by A change in relative lengths occurs between the
looking at the heels. The examiner passively flexes the two positions.
patient’s knees to approximately 90°. The relative leg
lengths are assessed a gain in this position
Gillet test50 The patient is standing. The examiner places one thumb The PSIS fails to move posterior and inferior with
under the PSIS on the side being tested, with the other respect to S2.
thumb over the S2 spinous process. The patient is
instructed to stand on one leg and flex the other hip and
knee, bringing the leg toward the chest.
Patrick test—range of The patient is placed in the test position by flexing, abducting, A difference in the range of motion exists
motion9,50 and externally rotating the hip of the tested leg, placing the between the two sides.
lateral malleolus on the knee of the opposite leg.
Overpressure is applied to the medial aspect of the knee.
The amount of motion available in the tested extremity is
compared with the opposite side.
Provocation tests
Gaenslen test26,45,50 The patient is supine with both legs extended. The leg being Pain is reproduced in either SI joint region with
tested is passively brought into full hip and knee flexion, performance of the test.
while the opposite hip is maintained in an extended
position. Overpressure is applied to the flexed extremity.
Posterior shear test9,26,50 The patient is supine. The hip is flexed to 90° and adducted. Buttock pain is produced.
The examiner applies an axial force through the femur at
different angles of hip adduction–abduction.
Compression/distraction The patient is supine. Pressure is applied first in a posterior Pain is reproduced in the SI joint region with
test26,50 and lateral direction (compression) on the ASIS either maneuver.
simultaneously. Pressure is then applied in an anterior and
medial direction on the ASIS (distraction).
Patrick test—Buttock pain26 The patient’s hip is flexed, abducted, and externally rotated Buttock or low back pain is produced.
by placing the lateral malleolus on the knee of the opposite
leg. Overpressure is applied to the medial aspect of the
knee while the pelvis is stabilized.
Patrick test—Groin pain26 Same as above. Groin pain is produced.
Resisted hip abduction9 The patient is supine with the hip in about 30° of abduction. Buttock pain is produced.
The examiner pushes the leg medially to cause an
isometric contraction of the hip abductors.
Sacral sulcus test21 The patient is prone. The examiner palpates with firm Pain is reproduced in the SI region.
pressure in the region directly medial to the PSIS.
Sacral thrust test26 The patient is prone. The examiner delivers an anteriorly Pain is reproduced in the SI region.
directed thrust directly over the sacrum.
ASIS ⫽ anterior superior iliac spine; PSIS ⫽ posterior superior iliac spine; SI ⫽ sacroiliac.