British Journal of Neurosurgery, August 2009; 23(4): 401–405

ORIGINAL ARTICLE

The Short Form 36 health survey in spine disease—validation against

condition-specific measures

MATHEW R. GUILFOYLE, HELEN SEELEY & RODNEY J. LAING

Department of Neurosurgery, Addenbrooke’s Hospital, Cambridge, CB2 2QQ, UK

Br J Neurosurg Downloaded from informahealthcare.com by University of California Irvine on 11/02/14

Abstract
Health-Related Quality of Life (HRQoL) assessments in spinal disease offer the potential of outcome measures that better
represent patients’ disability and response to treatment. The Short Form 36 Health Survey (SF-36) is a generic HRQoL
questionnaire that has been extensively used in primary and secondary care, but before it can be routinely applied in patients
with spinal disease must be validated against traditional specific measures of physical and mental morbidity. Patients with
lumbar disc prolapse, lumbar canal stenosis, and cervical spondylotic radiculomyelopathy were identified from a
prospectively maintained database. Visual Analogue Scales (VAS) and condition-specific questionnaires including the
Roland Morris Disability Score (RMDS), Myelopathy Disability Index (MDI), and Hospital Anxiety and Depression Scales
(HADS), were completed alongside the SF-36 survey at baseline and following surgery. Convergent, discriminant, and
predictive validity were assessed by computing correlations between the specific and generic scores. In addition,
responsiveness (Standardised Response Mean, SRM) and floor and ceiling effects were examined. Data from 1623
assessments of 620 patients were available. Convergent validity was shown by strong correlations between condition-specific
physical scores (MDI or RMDS) and the Physical Function and Bodily Pain domains of SF-36 (r ¼ -0.52 to -0.76, all
For personal use only.

p 5 0.01). VAS for leg or arm pain were also strongly correlated with Bodily Pain domain scores (r ¼ -0.54 to -0.77, all
p 5 0.01). Discriminant validity was confirmed by non-significant partial correlations between Physical Function and Mental
Health SF-36 domains when controlled for HADS scores (r ¼ 70.01 to 0.02, p 4 0.05). Predictive validity was
demonstrated by similar correlations between pre- and post-operative scores for specific and generic instruments. Physical
Function, Bodily Pain, and Mental Health domains were all free of significant floor or ceiling effects and showed moderate to
good responsiveness (SRM 0.54-1.72). SF-36 domain scores are valid for measuring morbidity and surgical outcomes in
common spinal disorders.

Key words: Cervical Spine, lumbar Spine, outcome, SF-36.

that is sufficiently general to be used in any disease

Introduction
state whilst retaining accurate measurement of
There is growing awareness that traditional clinical morbidity due to the particular condition con-
outcome assessments do not adequately represent cerned. Eight domain scales reflecting all aspects
patients’ experience of disease or perception of of health are calculated from the questionnaire
treatment, and are often insensitive to important items, and two summary measures of physical and
changes in health status. Particularly in the context of mental health can be further derived from these
chronic diseases, outcome measures that do not take scales.
account of patients’ quality of life are likely to be SF-36 has been favourably compared to standard
inadequate for judging the real efficacy of health measures of physical disability in patients with
interventions. Furthermore, the vast number of cervical and lumbar spine disease, including the
different specific outcome scales in use for spinal Neck Disability Index,2 Myelopathy Index,3 Oswes-
disorders are largely incompatible when comparing try Back Pain Questionnaire,4 and Roland-Morris
results across studies. Disability Scale.5 Separately, the mental health
Many of the problems associated with condition- domains of SF-36 have been validated against
specific measures can be mitigated by generic specific assessments of depression and anxiety.6
health instruments, of which the Medical Outcomes However, there has not been a combined analysis
Study 36-item Short Form Health Survey (SF-36) to determine if SF-36 accurately reflects both
has proven the most popular.1 SF-36 is designed physical and mental morbidity in patients with spinal
to be a standardised patient-reported assessment disease.

Correspondence: R. J. Laing, Box 166 Addenbrooke’s Hospital, Hills Road, Cambridge CB2 2QQ, UK. Tel.: þ44(0)1223245151.
E-mail: [email protected]

Received for publication 17 December 2008. Accepted 5 January 2009.

ISSN 0268-8697 print/ISSN 1360-046X online ª The Neurosurgical Foundation
DOI: 10.1080/02688690902730731
 402 M. R. Guilfoyle et al.

The purpose of the present study was to examine 36 domains that should measure different concepts.
the relationship between SF-36 domains and condi- Predictive validity was examined by comparing the
tion-specific measures of physical and mental health ability of baseline condition-specific and generic
in a prospectively observed cohort of patients with scores to predicted post-operative outcomes. Re-
lumbar disc prolapse, lumbar canal stenosis, or sponsiveness of SF-36 domains was analysed using
cervical spondylosis requiring surgery. standardised response means (SRM ¼ [mean change
in score] / [standard deviation of change in score])
and considered good if 40.8, moderate if 40.5, and
Materials and methods small if 50.5.13 Floor and ceiling effects of the SF-36
domains were considered to be present if greater than
Patients
15% of responses were at either extreme of the scale.
The prospective database of all (unselected) patients Independent groups and repeated measures were
under the care of the senior author (RJCL) since compared with the Mann-Whitney U-test and
1998 was queried for the data. Patients undergoing Wilcoxon Signed Ranks test, respectively. Correla-
anterior cervical discectomy for spondylotic radicu- tions were evaluated with Pearson’s r and Spear-
Br J Neurosurg Downloaded from informahealthcare.com by University of California Irvine on 11/02/14

lomyelopathy, lumbar discectomy for radiculopathy, man’s r coefficients. Because higher condition-
or decompressive laminectomy for lumbar stenosis specific and visual analogue scores indicated greater
were included. Clinical diagnosis was made by the morbidity whereas higher SF-36 scores represent
senior author and confirmed by findings on magnetic greater health, ‘positive’ correlations will be negative.
resonance imaging. All procedures were performed Results were considered significant at the 5% level
by or under the supervision of the senior author. (p 5 0.05). All analyses were done with SPSS 15.0.
Operative details have been reported elsewhere and
outcomes are comparable to those reported from
other centres.7–9 Results
Patient population
For personal use only.

Disease-specific and generic assessments

Six hundred and twenty patients were included in the
Patients were assessed prior to surgery and at follow- analysis comprising 203 with lumbar disc prolapse
up appointments at 3 months and 12–60 months (LD), 177 with lumbar stenosis (LS), and 240 with
post-operatively. At each point patients answered a cervical spondylotic radiculomyelopathy (CS). All
questionnaire comprising the SF-36 survey, visual patients provided complete pre-operative question-
analogue scales for back/neck pain and leg/arm pain, naires. Data were available for 88.1% and 73.7% of
and condition-specific disability instruments. The patients at early and late follow-up, respectively, giving
specific psychological morbidity assessment was the 1623 combined condition-specific and generic assess-
Hospital Anxiety and Depression Scale (HADS),10 ments. Demographic data are shown in Table I.
completed by all patients. The specific physical
morbidity measurement in the lumbar surgery group
Convergent validity
was the Roland-Morris Disability Score (RMDS),11
while the cervical surgery group completed the For all patient subgroups there were strong correla-
Myelopathy Disability Index (MDI).12 From the tions between the specific disability scale (RMDS or
SF-36 questionnaire 8 domain scores were calculated MDI) and the Physical Function and Bodily Pain
as per the published instructions1—Physical Func- domains of the SF-36 (r ¼ 70.52 to 70.76, all
tioning (PF), Role Physical (RP), Bodily Pain (BP), p 5 0.01; Table II). The Bodily Pain domain also
General Health (GH), Vitality (VI), Social Function- showed strong correlation with the visual analogue
ing (SF), Role Emotional (RE), and Mental Health scales for leg or arm pain (r ¼ 70.54 to 70.77, all
(MH). p 5 0.01; Table II). Similarly, there were strong
correlations between the HADS components and the
Mental Health domain of SF-36 (r ¼ 70.61 to
Analysis
70.76, all p 5 0.01; Table II).
To determine the construct validity of SF-36 the
statistical relationship with visual analogue scales and
condition-specific scores was investigated. Conver- TABLE I. Study sample characteristics
gent validity was assessed by comparing the physical
Age Sex
and mental condition-specific scores with SF-36 Group N Mean (SD) M:F
domains that purport to measure the same aspect
of health, i.e., Physical Functioning, Bodily Pain, and LD 203 41.3 (12.8) 1.11:1
Mental Health domains. In addition, Bodily Pain LS 177 66.7 (11.5) 1.13:1
CS 240 54.0 (13.9) 1.61:1
scores were compared with the visual analogue
scales. By contrast, discriminant validity was tested Number of patients, age, and sex for each group. LD, Lumbar
by comparing the condition-specific scores with SF- Discectomy; LS, Lumbar Stenosis; CS, Cervical Spondylosis.
 SF-36 and Spine Disease 403

Discriminant validity Predictive validity

In all three patient groups the condition-specific In the LD subgroup, baseline RMDS was not
physical scales and HADS components displayed correlated with RMDS at 3 months (r ¼ 0.16,
significant correlation (r ¼ 0.32 to 0.56, all p 4 0.05) or at later follow-up (r ¼ 0.17, p 4 0.05).
p 5 0.01). As these instruments are specific and well Baseline SF-36 Physical Functioning mirrored this
validated, this finding implies a true association pattern and was not correlated RMDS at either
between severity of spinal disease and mental follow-up assessment (r ¼ 70.14 to 70.15,
morbidity, as opposed to a failing of the scales to p 4 0.05). By contrast, for LS patients, baseline
distinguish aspects of health. Similarly, SF-36 RMDS was correlated with RMDS score at 3 months
Physical Functioning and Mental Health domains (r ¼ 0.54, p 5 0.01) and later (r ¼ 0.49, p 5 0.01)
were highly correlated (r ¼ 0.40 to 0.56, p 5 0.01). and the pre-operative SF-36 Physical Function score
Partial correlations between SF-36 Physical Func- was also predictive of RMDS at follow-up
tioning and Mental Health domains, controlling for (r ¼ 70.28 to 70.30, p 5 0.01). Baseline MDI was
both HAD components, were not significant observed to predict MDI at 3 months (r ¼ 0.65,
Br J Neurosurg Downloaded from informahealthcare.com by University of California Irvine on 11/02/14

(r ¼ 70.01 to 0.02, p 4 0.05; Table III), confirming p 5 0.01) and late follow-up (r ¼ 0.45, p 5 0.01) in
that the SF-36 domains discriminate physical and the CS subgroup and pre-operative SF-36 Physical
mental health to the same degree as the condition- Function was similarly predictive of MDI post-
specific scales. operatively (r ¼ 70.48 to 70.60, p 5 0.01).

Floor and ceiling effects

TABLE II. Correlation of condition-specific and visual analogue
scales against selected SF-36 domains SF-36 Physical Function, Bodily Pain, General
SF-36 Domains
Health, Vitality, and Mental Health domains were
all free of significant floor and ceiling effects, but the
Group Scale PF BP MH remaining SF-36 domains had either significant floor
For personal use only.

LD RMDS 70.74** 70.75** 70.50**

or ceiling effects, or both (Table IV).
HADS-A 70.52** 70.57** 70.66**
HADS-D 70.70** 70.67** 70.68**
VAS – 70.77** –
Responsiveness
LS RMDS 70.66** 70.66** 70.37** Standardised response means (SRM, Table V) for
HADS-A 70.43** 70.49** 70.66**
HADS-D 70.59** 70.58** 70.61**
the SF-36 Physical Function domain at late follow-
VAS – 70.71** – up for each of the subgroups were 0.86–1.72, thus
CS MI 70.76** 70.56** 70.39** satisfying Cohen’s criteria for good responsiveness.
HADS-A 70.46** 70.58** 70.76** SRM for the Bodily Pain domain was good in LD
HADS-D 70.63** 70.63** 70.67** and LS groups and moderate in the CS patients.
VAS – 70.57** –

Values are negative because higher scores indicate greater

Discussion
morbidity on the specific scales whereas SF-36 domains are
constructed so that higher scores reflect better health. Abbrevia- This study has compared self-reported condition-
tions as in Table I and in addition RMDS, Roland Morris
specific and generic outcome measures in a large
Disability Score; HADS, Hospital Anxiety and Depression Scale;
MI, Myelopathy Index; NDI, Neck Disability Index; PF, Physical number of patients with common spinal diseases and
Function; BP, Bodily Pain; MH, Mental Health; VAS, Visual the findings support the use of SF-36 to measure
Analogue Scale for leg pain in LD and LS groups and arm pain in morbidity and response to surgical intervention.
CS group. **p 5 0.01. Strong correlations between SF-36 Physical Func-
tion and Bodily Pain domains and the condition-
specific physical disability scores were observed in all
TABLE III. Partial correlations between condition-specific scales patient groups. Similarly, SF-36 Mental Health and
and SF-36 domains Hospital Anxiety and Depression Scale components
were strongly correlated. When controlled for the
SF-36 Domains
underlying association between physical and mental
Group Scale PF BP MH morbidity, condition-specific disability scores and
LD RMDS 70.51** 70.54** 70.05{
the SF-36 Mental Health domain were not signifi-
LS RMDS 70.50** 70.50** 0.01{ cantly correlated. These results fulfil the basic
CS MI 70.52** 70.22** 0.05{ psychometric requirements for construct validity,
namely that scales purporting to measure the same
Correlations between variables controlled for HADS score.
underlying concept are correlated (convergent valid-
Significant association between specific and generic physical scores
remains, but correlation between physical and mental scores is no ity) and those that measure different concepts are not
longer apparent. Abbreviations as in Table II. { p 4 0.05 (discriminant validity). Further analysis showed the
**p 5 0.01. SF-36 domains have similar predictive power to their
 404 M. R. Guilfoyle et al.

TABLE IV. Floor and ceiling effects

SF-36 Domains

Group Score PF RP BP GH VI SF RE MH

LD 0 5.0 40.7 8.1 0.6 1.0 6.3 16.7 0.2

100 9.7 27.4 7.3 7.4 4.4 29.5 60.5 7.3
LS 0 7.4 43.9 3.7 0.2 1.3 2.6 20.4 0.2
100 4.3 18.0 7.6 2.9 2.8 24.3 55.9 7.0
CS 0 7.7 50.9 3.7 0.6 2.4 8.1 29.0 0.5
100 6.9 20.4 11.3 2.6 1.4 22.9 52.1 3.7

Percentage of responses at the extremes of domain scales (i.e., 0 or 100). Bold text highlights domains where there are significant floor and/or
ceiling effects (415%).
Abbreviations as in Table II and in addition RP, Role Physical; GH, General Health; VI, Vitality; SF, Social Functioning; RE, Role
Emotional.
Br J Neurosurg Downloaded from informahealthcare.com by University of California Irvine on 11/02/14

TABLE V. Responsiveness measures in these groups.14,16 This study shows

that SF-36 Physical Functioning, Bodily Pain, and
SF-36 Domains
Mental Health domains are in themselves adequate to
Group PF BP MH assess morbidity in spinal disease, obviating the need
for any further distillation of scores into summary
LD 1.72 1.57 0.82
LS 1.04 1.07 0.56
measures.
CS 0.86 0.65 0.54 This study has demonstrated that SF-36 is suitable
for measuring both physical and mental health in
Standardised response means ([Mean change] / [Standard devia- routine spinal surgery practice. Practically, the ques-
tion of changes]) for the three important SF-36 domains.
tionnaire takes 5–10 minutes to complete and is
Abbreviations as in Table II.
routinely given to patients waiting in the clinic without
For personal use only.

requiring additional time to be set aside or introducing

condition-specific counterparts. For all patients the delays. The SF-36 has also been validated as an
SF-36 Physical Functioning domain was free of any outcome measure in primary care19 and the current
floor or ceiling effects and was responsive to change analysis supports its use by general practitioners in
following surgery, and in Lumbar Disc and Lumbar assessment of spinal disorders. Employing a generic
Stenosis groups the Bodily Pain domain displayed score makes results transparently comparable to other
similar properties. In the Cervical Spondylosis group, interventions and across diseases, and has additional
Bodily Pain showed moderate responsiveness, a advantages including the availability of accurate
reflection of the fact that pain is not the predominant normative data acquired from the general population.
symptom in this condition. Further, calculating Quality-Adjusted Life Years for
SF-36 has become a familiar outcome measure in cost-benefit analyses is simple using instruments such
both medical and surgical neurology practice. Vali- as SF-6D, a preference-based health utility index
dation analyses have been reported for conditions derived from SF-36.20
including stroke,14 Parkinson’s disease,15 multiple
sclerosis,16 motor neuron disease,17 and headache.18
Conclusion
Though the SF-36 has in each case been recom-
mended for use, these studies have highlighted SF-36 is valid for assessing morbidity and health-
deficiencies with specific domains in respect to related quality of life due to cervical and lumbar
internal consistency, scaling, and floor and ceiling spine disease, and for measuring outcomes following
effects. Similarly, the present study found that Role- surgery. Given its additional benefits over condition-
Physical, Role-Emotional, and Social Functioning specific instruments it merits inclusion as a routine
domains are likely to be insensitive to significant outcome measure in these patients.
changes in health status, since a large proportion of
patients scored at the scale extremes. For this reason, Declaration of interest: The authors report no
these domains should not be used as primary conflicts of interest. The authors alone are respon-
outcome measures. sible for the content and writing of the paper.
The calculation of physical and mental health
summary measures using weighted sums of the eight
References
domain scores continues to be a controversial aspect
of SF-36. Originally established from principal com- 1 Ware JE, Snow KK, Kosinski M, et al. SF-36 Health Survey:
ponents analysis of normative data sets, similar factor Manual and Interpretation Guide. Boston: Nimrod Press; 1993.
2 McCarthy MJH, Grevitt MP, Silcocks P, et al. The reliability of
structures have typically not been reproducible in the Vernon and Mior neck disability index, and its validity
patients with neurological disorders, raising doubts compared with the short form-36 health survey questionnaire.
about the validity and interpretation of the summary Euro Spine J 2007;16(12):2111–17.
 SF-36 and Spine Disease 405

3 Singh A, Crockard HA. Comparison of seven different scales 12 Casey AT, Bland JM, Crockard HA. Development of a
used to quantify severity of cervical spondylotic myelopathy functional scoring system for rheumatoid arthritis patients with
and post-operative improvement. J Outc Measure 2001;5(1): cervical myelopathy. Ann Rheumat Dis 1996;55(12):901–6.
798–818. 13 Cohen J. Statistical power analysis for the behavioural sciences.
4 Walsh TL, Hanscom B, Lurie JD, et al. Is a condition-specific Hillsdale, NJ: Lawrence Erlbaum Associates, 1988.
instrument for patients with low back pain/leg symptoms really 14 Hobart JC, Williams LS, Moran K, et al. Quality of life
necessary? The responsiveness of the Oswestry Disability measurement after stroke: uses and abuses of the SF-36 Stroke
Index, MODEMS, and the SF-36. Spine 2003;28(6):607–15. 2002;33(5):1348–56.
5 Turner JA, Fulton-Kehoe D, Franklin G, et al. Comparison of 15 Jenkinson C, Peto V, Fitzpatrick R, et al. Self-reported
the Roland-Morris Disability Questionnaire and generic health functioning and well-being in patients with Parkinson’s
status measures: a population-based study of workers’ com- disease: comparison of the short-form health survey (SF-36)
pensation back injury claimants. Spine 2003;28(10):1061–7; and the Parkinson’s Disease Questionnaire (PDQ-39). Age
discussion 1067. Ageing 1995;24(6):505–9.
6 Stoll T, Kauer Y, Buchi S, et al. Prediction of depression in 16 Hobart J, Freeman J, Lamping D, et al. The SF-36 in multiple
systemic lupus erythematosus patients using SF-36 Mental sclerosis: why basic assumptions must be tested. J Neurol
Health scores. Rheumatology 2001;40(6):695–8. Neurosurg Psych 2001;71(3):363–70.
7 Latimer M, Haden N, Seeley HM, et al. Measurement of 17 Dallmeijer AJ, de Groot V, Roorda LD, et al. Cross-
outcome in patients with cervical spondylotic myelopathy diagnostic validity of the SF-36 physical functioning scale in
Br J Neurosurg Downloaded from informahealthcare.com by University of California Irvine on 11/02/14

treated surgically. Br J Neurosurg 2002;16(6):545–9. patients with stroke, multiple sclerosis and amyotrophic lateral
8 Wilby MJ, Seeley H, Laing RJ. Laminectomy for lumbar canal sclerosis: a study using Rasch analysis. J Rehabil Med 2007;
stenosis: a safe and effective treatment. Br J Neurosurg 2006; 39(2):163–9.
20(6):391–5. 18 Wang SJ, Fuh JL, Lu SR, et al. Quality of life differs among
9 Guilfoyle MR, Ganesan D, Seeley H, et al. Prospective study of headache diagnoses: analysis of SF-36 survey in 901 headache
outcomes in lumbar discectomy. Br J Neurosurg 2007;21(4): patients. Pain 2001;89(2–3):285–92.
389–95. 19 Brazier J, Harper R, Jones N, et al. Validating the SF-36 health
10 Zigmond A, Snaith R. The hospital anxiety and depression survey questionnaire: new outcome measure for primary care.
scale. Acta Psychiatr Scand 1983;67(6):361–70. BMJ 1992;305(6846):160–4.
11 Roland M, Morris R. A study of the natural history of back 20 Brazier J, Usherwood T, Harper R, et al. Deriving a preference-
pain. Part I: development of a reliable and sensitive measure of based single index from the UK SF-36 Health Survey. J Clin
disability in low-back pain. Spine 1983;8(2):141–4. Epidemiol 1998;51(11):1115–28.
For personal use only.