General practice

Predicting who develops chronic low back pain in primary

care: a prospective study
Elaine Thomas, Alan J Silman, Peter R Croft, Ann C Papageorgiou, Malcolm I V Jayson,
Gary J Macfarlane

Arthritis Research Abstract tion those patients at high risk of persistent disabling
Campaign
Epidemiology Unit,
symptoms. Prediction of outcome has been examined
Objectives To quantify the relative contribution of previously with two main sources of prognostic factors:
School of
Epidemiology and premorbid and episode specific factors in clinical data directly related to the pain episode and
Health Sciences, determining the long term persistence of disabling health, lifestyle, and individual factors. In all studies to
Medical School,
University of
symptoms of low back pain. date, however, health, lifestyle, and individual factors
Manchester, Design Prospective cohort study. have been recorded after the onset of symptoms and
Manchester Setting Two general practices in the south
M13 9PT may therefore have been influenced by the pain
Manchester area. episode.
Elaine Thomas, Participants 180 patients, who previously participated
research statistician We used a prospective design to quantify the contri-
Alan J Silman,
in a cross sectional population survey, who consulted bution, firstly, of health, lifestyle, and individual factors
professor and director because of low back pain during the study period. measured before the onset of the episode of low back
Ann C They were followed at 1 week and 3 and 12 months pain leading to consultation and, secondly, of factors
Papageorgiou, after consultation.
studies coordinator specific to the episode, in predicting the long term per-
Gary J Macfarlane,
Main outcome measure Persistent disabling low back sistence of symptoms. We also determined whether sub-
senior lecturer pain in the 12 months after the consultation. jects at high risk of persistent disabling symptoms can be
University of Keele, Results Disabling low back pain persisted in one third identified early on the basis of such factors.
School of of participants after consultation and was more
Postgraduate common with increasing age, among those with a
Medicine, Industrial
and Community history of low back pain, and in women. Persistence of Participants and methods
Health Research symptoms was associated with “premorbid” factors
Centre, Hartshill, Cohort recruitment
(high levels of psychological distress (odds ratio 3.3;
Stoke on Trent Eligible participants were those registered with two
ST4 7QB 95% confidence interval 1.5 to 7.2), poor self rated
general practices in the south Manchester area who
Peter R Croft, health (3.6; 1.9 to 6.8), low levels of physical activity (2.8;
professor had participated in a previous cross sectional study3
1.4 to 5.6), smoking (2.1; 1.0 to 4.3), dissatisfaction with
and had consulted because of low back pain during the
Rheumatic Diseases employment (2.4; 1.3 to 4.5)) and factors related to the
Centre, University subsequent 18 month study period (fig). Ethical
episode of low back pain (duration of symptoms, pain
of Manchester, approval was obtained from the local health authority.
Clinical Sciences radiating to the leg (2.6; 1.3 to 5.1), widespread pain
Building, Hope
The survey had included information on demographic
(6.4; 2.7 to 15), and restriction in spinal mobility). A
Hospital, Salford data, lifestyle factors including levels of physical
M6 8HD multivariate model based on six factors identified
activity, smoking status, a single question on self rated
Malcolm I V Jayson, groups whose likelihood of persistent symptoms
general health,4 and the 12 item general health
professor ranged from 6% to 70%.
Correspondence to: Conclusions The presence of persistent low back
GJ Macfarlane pain is determined not only by clinical factors
G.Macfarlane@
man.ac.uk associated with pain but also by the premorbid state. Interview / examination

BMJ 1999;318:1662–7 Introduction 1 week 3 months 12 months

Consultation
Each year in the United Kingdom 7% of the adult popu-
lation present to their general practitioner with low back
Consulting episodes
pain1 at a cost in excess of £500 million to the NHS.2
Episodes of acute back pain are perceived to resolve
rapidly with only a small proportion of sufferers experi-
encing persistent or recurrent symptoms leading to dis- 0 18 months
ability. Most of the costs linked to the treatment of back Recruitment
survey
pain apply to this small proportion.
Clearly, it would be advantageous with respect to Summary of design of follow up study of patients with low back
clinical management to be able to identify at presenta- pain presenting to primary care

1662 BMJ VOLUME 318 19 JUNE 1999 www.bmj.com

 General practice

questionnaire5 to measure participants’ levels of Univariate analysis assessed the individual associa-
“psychological distress” and current (during the past tions of these putative risk factors and are presented
month) and past low back pain. The lower back was as odds ratios with 95% confidence intervals. When we
defined, on a manikin, as the area bordered above by considered premorbid and episode specific factors
the 12th rib and below by the gluteal folds. together, those variables significantly associated with
Current work status was recorded and social class persistent symptoms were used in a backwards
was derived from job title by using the classification stepwise logistic regression procedure to determine
method of the Office of Population Censuses and Sur- whether a small group of factors could accurately pre-
veys.6 Participants were also asked to rate their level of dict those patients who would have persistent
satisfaction either with their current job or their disabling symptoms.
current work status (retired, seeking work, working in A “jack knife” technique was used to test the
the home) with a 5 point scale from “very satisfied” to accuracy of the resulting model. This method involves
“severely dissatisfied.” the removal of data from each participant in turn, and
a model is then derived by using the data from the
Cohort follow up remaining participants. The predicted outcome for the
During the 18 month study period all consultations for removed participant, on the basis of this model, is then
low back pain were identified weekly by using the gen- compared with their observed outcome status. This
eral practices’ computerised records systems. Partici- process is repeated for each participant.
pants had a home interview and examination carried All analyses were conducted with the stata
out by a research nurse. The median time between statistical software package.9
consultation and the nurse visit was 1 week (interquar-
tile range 1-2 weeks).
The interview after consultation provided infor-
Table 1 Number (percentage) of patients with disabling low back
mation on factors directly related to the episode of low pain who visited their general practitioner with new episode of pain
back pain leading to that consultation: duration of cur-
Interview Men Women Total
rent episode, mode of onset (sudden or gradual), radia-
1 week 47 (64) 84 (79) 131 (73)
tion of pain to leg (either above or below the knee), and
3 month 28 (38) 59 (56) 87 (48)
pain elsewhere in the body. The physical examination
12 month 21 (28) 55 (52) 76 (42)
consisted of five measures of spinal mobility: standing Each time point 18 (24) 43 (41) 61 (34)
extension, lateral flexion, finger to floor distance, knee
extension, and a modified Schober’s measurement.
Details of the methods used to measure each of these Table 2 Association between demographic and premorbid
movements have been described previously.7 Each spi- factors and persistent low back pain (univariate analysis)
nal movement was considered “restricted” at the point Persistent low back pain
Odds ratio
that best discriminated between participants consult- Characteristic No (n=119) Yes (n=61) (95% CI)
ing because of low back pain and a separate group of Men 56 18 1
people who had never experienced low back pain.7 Women 63 43 2.12 (1.1 to 4.1)
Participants were interviewed again at 3 and 12 Age (years)*:
months after the initial consultation with the main 18-29 16 2 1
objective of determining current symptoms (fig). 30-44 41 16 3.12 (0.6 to 15)
45-59 38 23 4.84 (1.0 to 23)
Outcome 60-75 24 20 6.67 (1.4 to 33)
At each of the three interviews participants were asked General health questionnaire score†:
12-21 49 10 1
about the presence of low back pain on that day and
22-48 69 47 3.34 (1.5 to 7.2)
asked to mark the severity of any pain on a visual
Self rated health†:
analogue scale from 0 to 10; a score of 0 or 1 was defined
Excellent/good 83 24 1
as no pain. Disability was measured at each of the three
Fair/poor 36 37 3.55 (1.9 to 6.8)
interviews with the Hanover back pain activity schedule.8 Physical activity compared with peers†:
This schedule was developed for use in subjects with More/same 97 38 1
back pain and has been found to compare well with Less 21 23 2.80 (1.4 to 5.6)
other similar instruments. It inquires about the ease of First episode of low back pain†:
carrying out 12 items of daily activity in the previous 2 Yes 24 4 1
weeks, with scores for each item being summed to a per- No 95 57 3.60 (1.2 to 11)
centage value where 100% represents total ability. Ever smoked†:
The primary outcome considered in this study was Never 40 12 1
“persistent disabling low back pain” defined as the Ever 79 49 2.07 (1.0 to 4.3)

presence of both low back pain and disability (Hanover Alcohol drinker†:
Weekly 70 23 1
score < 75%) at each follow up interview (1 week and
Never/hardly ever 49 38 2.36 (1.3 to 4.4)
3 and 12 months).
Employment status†:
Working 74 25 1
Statistical analysis Not working 45 36 2.37 (1.3 to 4.4)
The analysis examined the relation between persistent Satisfied with employment situation†:
disabling low back pain and, firstly, premorbid factors Yes 66 21 1
and, secondly, episode specific factors, among those No 53 40 2.37 (1.3 to 4.5)
participants who had consulted with an episode of low *Odds ratios adjusted for sex.
back pain which started after recruitment to the study. †Odds ratios adjusted for sex and age (four groups).

BMJ VOLUME 318 19 JUNE 1999 www.bmj.com 1663

General practice

Results Table 4 Predictors of presence of persistent disabling low back

During the 18 month study period consultations were pain. Demographic, premorbid, and episode specific factors
recorded for 442 of the 4501 participants (9.8%) from included in stepwise logistic regression model
the cross sectional study conducted in the study Predictive factor Odds ratio (95% CI)
general practices. Interviews 1 week after consultation Men 1
were completed for 294. We defined a new episode to Women 2.26 (1.0 to 5.1)
be where pain that caused the consultation started First episode of low back pain:

after the cross sectional survey was carried out, which Yes 1
No 2.76 (0.8 to 9.9)
resulted in the exclusion of 48 participants whose epi-
Satisfied with employment situation:
sode of back pain had started before this. Hence 246
Yes 1
subjects were eligible for follow up, of whom 180 pro-
No 2.62 (1.2 to 5.8)
vided complete information at 1 week and 3 and 12 Radiating leg pain:
months after consultation and are the subjects used in No 1
further analyses. Yes 1.89 (0.8 to 4.4)
Outcome of low back pain—The percentage of partici- Widespread pain:
pants who reported disabling low back pain was 73% at No 1
1 week and 48% at 3 months, with only a slight subse- Yes 3.44 (1.3 to 9.3)
quent decrease to 42% at 12 months after consultation. Spinal restrictions:
About a third (61; 34%) of participants were classified None or 1 1
as having persistent disabling low back pain—that is, 2–5 3.08 (1.3 to 7.3)
low back pain with related disability at each of the three
follow up interviews (table 1). smoking, a low alcohol intake, not being employed,
Premorbid factors—Sex and age predicted persistent and dissatisfaction with current employment or work
disabling low back pain: a doubling in odds of a poor status. Each of the factors was associated with a twofold
outcome was seen for women compared with men to fivefold increase in odds of persistent symptoms
(odds ratio 2.1; 95% confidence interval 1.1 to 4.1) and (table 2).
the likelihood of a poor outcome increased with older Episode specific factors—The strongest episode spe-
age. Premorbid factors (that is, measured before the cific predictor of a poor outcome was the presence of
onset of the consulting episode) associated with a poor widespread pain (axial skeletal pain in addition to pain
outcome were high levels of psychological distress, above and below the waist and on the right and left side
below average self rated health, low levels of physical of the body); it was associated with a sixfold increase in
activity, a history of low back pain, current or previous odds of a poor outcome (6.4; 2.7 to 15.0). A long
duration of symptoms before consultation, the reporting
Table 3 Association between factors specific to episode and of radiating leg pain, and restriction in spinal movement
persistent low back pain (univariate analysis) were also significantly associated with a twofold to
Persistent low back pain fivefold increase in odds of a poor outcome (table 3).
Odds ratio
Characteristic No (n=119) Yes (n=61) (95% CI) Predicting outcome: multivariate model—All premor-
Duration of pain before current visit (weeks)*: bid and episode specific factors that were significantly
0-3 95 40 1 associated with outcome were candidate variables for a
4-12 20 12 1.43 (0.6 to 3.2) stepwise logistic regression model. The final model
>13 4 9 5.34 (1.6 to 18) consisted of six factors: two premorbid (history of low
Radiating leg pain*: back pain, dissatisfaction with current employment or
No 54 15 1 work status), three episode specific (widespread pain,
Yes 65 46 2.55 (1.3 to 5.1) radiating leg pain, restriction in two or more spinal
Widespread pain*: movements), and sex (table 4). When we used the jack
No 110 40 1
knife procedure the outcome of 74% of participants
Yes 9 21 6.42 (2.7 to 15)
was correctly classified with a higher negative
Usual duration of low back pain (days)*:
predictive value (77%) than positive predictive value
<7 31 17 1
7-30 34 21 1.01 (0.4 to 2.4)
(60%). The likelihood of persistent disabling low back
>30 15 16 1.65 (0.6 to 6.7)
pain increased with the number of factors reported:
Standing extension*: only 6% of the participants who reported fewer than
Not restricted 37 9 1 three factors had a poor outcome compared with 70%
Restricted 71 39 2.26 (0.9 to 5.2) of participants who reported more than four (table 5).
Finger to floor*: Non-participants—Among patients who consulted
Not restricted 67 12 1 with low back pain and who were eligible to participate
Restricted 46 37 4.49 (2.1 to 9.5) in this follow up study, 148 (33%) refused to participate
Lateral flexion*: in the initial interview or were not contacted for
Not restricted 82 28 1 logistical reasons. Information on premorbid factors
Restricted 33 23 2.04 (0.9 to 4.0)
found to predict poor outcome is shown for these non-
Modified Schober’s*:
participants compared with participants in table 6. Non-
Not restricted 95 33 1
participants were more likely to be women and aged
Restricted 18 17 2.72 (1.3 to 5.9)
under 45 years but otherwise did not differ. Sixty six par-
Knee extension*:
Not restricted 51 16 1
ticipants initially interviewed were subsequently lost to
Restricted 36 27 2.39 (1.1 to 5.1) follow up. In table 7 they are compared with those who
*Odds ratios adjusted for sex and age (four groups).
completed follow up with respect to premorbid and

1664 BMJ VOLUME 318 19 JUNE 1999 www.bmj.com

 General practice

Comparisons with other studies

Table 5 Likelihood of persistent disabling low back pain
Studies examining outcome in a cross sectional fashion
according to number of risk factors present
at 1 year have found a poor outcome in 30-50% of
No with Observed percentage subjects,11–14 which is consistent with our finding that
No of factors No of persistent with persistent
present* subjects† symptoms symptoms 34% of participants have persistent disabling pain.
Five/six 30 21 70 Results from a recent study from the Netherlands found
Four 43 15 35 that at 12 months after the initial consultation only 10%
Three 45 12 27 still suffered from the same episode of pain which had
None/one/two 49 3 6 originally led to consultation and that 75% had reported
*Factors predicting persistence: female sex, dissatisfaction with employment at least one recurrence in the 12 month period.15 It is
situation, history of low back pain, radiating leg pain, widespread pain, two or therefore likely that the group in the current study with
more restrictions in spinal movement.
†Does not total 180 as subjects with missing data for any of these six risk “persistent” pain are a combination of subjects with con-
factors were not included in multivariate model. tinuous pain and those who had one or multiple recur-
rences during the follow up period.
Table 6 Differences in premorbid factors between subjects who Although women had a poorer outcome than men,
were and were not interviewed after consultation predictors of outcome were similar in both sexes. Pre-
vious results from this study have shown that high
Interviewed after consultation
levels of psychological distress, as measured by the
Characteristic Yes (n=294) No (n=148) P value*
general health questionnaire, and dissatisfaction with
Men 123 51
Women 171 97
0.134 employment increase the risk of a future episode of
Age (years):
low back pain in patients presenting to primary care
18-29 37 37 among those initially pain free.16 17 The present study
30-44 96 44 indicates that these factors also influence the
0.002
45-59 92 29 persistence of symptoms. Adverse psychological
60-75 69 38 factors have previously been reported to be associated
General health questionnaire score: with a poor outcome.13 14 18–20 In those studies, however,
12-21 84 40
0.730
psychological data were collected after the onset of
22-48 200 103 symptoms, making it impossible to determine whether
Self rated health: adverse psychological factors predated or were a
Excellent/good 166 80 consequence of pain. By using a prospective design we
0.769
Fair/poor 127 65
have uniquely collected psychological information
Physical activity of peers:
before the onset of pain; and the results suggest that
More/same 199 109
0.191 such factors do have an early influence on onset of
Less 91 37
Ever smoked:
symptoms and outcome.
Never 81 44
In common with other studies of low back pain13 14
0.631 we found that a history of symptoms was highly predic-
Ever 213 104
First episode of low back pain: tive of persistent symptoms, although, interestingly, the
No 60 26 participant’s assessment of duration of symptoms in a
0.538
Yes 230 117
Alcohol drinker:
Never/hardly 149 72 Table 7 Differences in premorbid and episode specific factors
0.687
Weekly 145 76 between subjects who did and did not completed follow up
Satisfied with employment situation:
Completed 12 month follow up
Yes 131 55
0.286 Characteristic Yes (n=180) No (n=66) P value*
No 152 80
Men 74 25
*Groups were compared with ÷2 test. 0.647
Women 106 41
Age (years):
episode specific factors. Participants lost to follow up 18-29 18 15
were younger but did not differ with respect to any other 30-44 57 25
0.021
predictors of outcome. 45-59 61 14
60-75 44 12
First episode of low back pain:
Discussion Yes 28 12
0.621
We have previously shown in this population that No 152 54
Satisfied with employment situation:
chronic symptoms are common after consultation for a
Yes 87 29
new episode of low back pain.10 By using a small group 0.607
No 93 36
of factors, both premorbid and specific to the episode,
Radiating leg pain:
we have now shown that it is possible to define groups at No 69 29
high risk of a poor outcome. In comparison with the 0.426
Yes 111 37
population reported in our previous paper on Widespread pain:
outcome,10 the current study population includes only No 150 59
0.239
those who participated in the cross sectional survey, Yes 30 7
patients consulting with back pain over the subsequent Spinal restrictions:
18 month period (instead of 12 months), and those <1 71 28
0.386
whose episode of pain started after the population >2 96 29
survey. *Groups were compared with ÷ test.2

BMJ VOLUME 318 19 JUNE 1999 www.bmj.com 1665

General practice

previous episode(s) did not predict outcome. This

Key messages
association of outcome with history may indicate that in
some people episodes of low back pain after the first + Although many episodes of acute low back pain
become increasingly longer in duration. Alternatively it will resolve rapidly, around 30% result in
may simply reflect that people with a previous episode(s) persistent disabling symptoms
are likely to have future multiple episodes and have been
classified with persistent symptoms in this study. + The presence of persistent low back pain is not
In addition to the premorbid state, several factors only determined by clinical factors at the time of
specific to the episode were important in predicting onset but also by the “premorbid” state
outcome. The strongest adverse prognostic factor was + Patients with these adverse prognostic factors
when low back pain was part of a more widespread represent a group in which effective
pain syndrome. Chronic widespread pain, the cardinal interventions may result in substantially
feature of fibromyalgia, is known to have a poor reduced morbidity and healthcare costs
outcome, particularly when it is associated with high
levels of distress.21 It commonly occurs together with
other physical symptoms and may be a manifestation that the presence of persistent low back pain is not only
of somatisation. It is therefore not surprising that this determined by clinical factors associated with pain but
subgroup of patients presenting with “low back pain” also by the premorbid state. On the basis of five
have a notably poorer outcome. A further “pain distri- variables which are easily collected at the time of con-
bution” prognostic factor indicative of a poor outcome sultation and an examination of spinal movement, we
was the presence of leg pain. This, however, is likely to were able to define a group who had a high risk of per-
be mechanical in origin, indicating possible compres- sistent symptoms of low back pain during the follow up
sion of a nerve root or irritation of the spinal canal.13 14 year. Furthermore, patients with these adverse
In the literature on low back pain much attention prognostic factors are likely to represent those in
has been given to the predictive ability of spinal whom potentially effective interventions would have
movements13 18 22 23; our study has confirmed that restric- the greatest impact in terms of morbidity and
tion predicts poor outcome. This relation with the healthcare costs.
persistence of pain could be a direct consequence of spi-
nal restriction—that is, patients with a less mobile spine We thank the partners, staff, and patients of Brooklands and
Bowland General Practices, Manchester, for help with conduct-
have more severe symptoms. Alternatively, as suggested ing the study. Margaret Carrington, Jane Barnett, and Hannah
in a cross sectional study examining the relation Chambers carried out the study interviews.
between behavioural and biomechanical factors, restric- Contributors: PRC, AJS, and ACP designed the study proto-
tion may be associated with pain tolerance that is, in col, discussed core ideas, coordinated the study, and participated
in analysis and interpretation of data and writing the paper. ET
turn, subject to psychological influences.24 A further
and GJM discussed core ideas and participated in data analysis
analysis in the present study, that stratified participants and interpretation and took the lead in the writing of the paper.
according to high and low levels of psychological MIVJ participated in the design and development of the study
distress, showed that restriction in spinal movements and the writing of the paper. All the authors are guarantors for
predicted poor outcome in both groups, suggesting that the paper.
Funding: Arthritis Research Campaign and the National
the former may be the more likely explanation. Back Pain Association.
Competing interests: None declared.
Potential bias
1 McCormick A, Fleming D, Charlton J. Morbidity statistics from general prac-
A concern in any follow up study is the possibility of tice. Fourth national study 1991-92. Office of Population Censuses and Surveys.
bias occurring due to loss to follow up at various stages. London: HMSO, 1995. (Series MB5 No 3.)
2 United Kingdom Department of Health, Clinical Standards Advisory
These losses are to some extent inevitable in a study Group. Epidemiology review: the epidemiology and cost of back pain. London:
from an urban population. We have investigated the HMSO, 1994.
3 Papageorgiou AC, Croft PR, Ferry S, Jayson MIV, Silman AJ. Estimating
potential for such biases on our results. With the the prevalence of low back pain in the general population. Evidence from
exception of age and sex, participants who were not the South Manchester back pain survey. Spine 1995;20:1889-94.
4 Stewart AL, Hays RD, Ware JE. Health perceptions, energy/fatigue, and
followed up or who were lost to follow up did not dif- health distress measures. In: Stewart AL, Ware JE, eds. Measuring function-
fer from those who were recruited and studied for the ing and well-being: the medical outcomes study approach. Durham, NC: Duke
University Press, 1992:143-72.
entire follow up period in terms of factors that 5 Goldberg D, Williams P. A user’s guide to the general health questionnaire.
predicted outcome. This increased propensity for loss Windsor: NEFR-Nelson, 1988.
6 Office of Population Censuses and Surveys. Standard occupational classifi-
in the younger participants may be due, in part, to their cation. London: HMSO, 1991.
moving out of the area. These differences would bias 7 Thomas E, Silman AJ, Papageorgiou AC, Macfarlane GJ, Croft PR. Asso-
our results, however, only if their relation with outcome ciations between measures of spinal mobility and low back pain: an
analysis of new attenders in primary care. Spine 1998;23:343-7.
was different in participants and non-participants. 8 Kohlmann T, Rapse H. Der Funktionsfragebogen Hannover zur
Overall, our data do not suggest any important biases alltagsnahen Diagnostik der Funktionsbeeinträchtigung durch Rücken-
schmerzen (FFbH-R). Rehabilitation 1996;35:I-VIII.
that would prevent our results being extrapolated to all 9 Statacorp. Stata statistical software. Release 5.0. College Station, TX: Stata
those eligible to be followed up. Corporation, 1997.
10 Croft PR. Macfarlane GJ, Papageorgiou AC, Thomas E, Silman AJ. Out-
come of low back pain in general practice: a prospective study. BMJ
Conclusions 1998;316:1356-9.
11 Philips HC, Grant L. The evolution of chronic back pain problems: a lon-
In conclusion, we have shown that a substantial gitudinal study. Behav Res Ther 1991;29:435-41.
proportion of patients who present to primary care 12 Von Korff M, Deyo RA, Cherkin D, Barlow W. Back pain in primary care.
Outcomes at 1 year. Spine 1993;18:855-62.
with low back pain will have persistent symptoms over 13 Burton AK, Tillotson KM, Main CJ, Hollis S. Psychosocial predictors of
the 12 months after consultation. By uniquely outcome in acute and subchronic low back trouble. Spine 1995;20:722-8.
14 Cherkin DC, Deyo RA, Street JH, Barlow W. Predicting poor outcome for
collecting data on risk factors before the onset of the back pain seen in primary care using patients’ own criteria. Spine
pain that resulted in the consultation, we have shown 1996;21:2900-7.

1666 BMJ VOLUME 318 19 JUNE 1999 www.bmj.com

 General practice

15 Van den Hoogen HJM, Koes BW, van Eijk JT, Bouter LM, Devillé W. On 20 Dionne CE, Koepsell TD, Von Korff M, Deyo RA, Barlow WE, Checkoway
the course of low back pain in general practice: a one year follow up H. Predicting long-term functional limitations among back pain patients
study. Ann Rheum Dis 1998;57:13-9. in primary care. J Clin Epidemiol 1997;50:31-43.
16 Croft PR, Papageorgiou AC, Ferry S, Thomas E, Jayson MIV, Silman AJ. 21 Macfarlane GJ, Thomas E, Papageorgiou AC, Schollum J, Croft PR. The
Psychological distress and low back pain: Evidence from a prospective natural history of chronic pain in the community: a better prognosis than
study in the general population. Spine 1996;20:2731-7. in the clinic? J Rheumatol 1996;23:1617-20.
17 Papageorgiou AC, Macfarlane GJ, Thomas E, Croft PR, Jayson MIV, 22 Troup JDG, Martin JW, Lloyd DCEF. Back pain in industry. A prospective
Silman AJ. Psychosocial factors in the work place—do they predict new survey. Spine 1981;6:61-9.
episodes of low back pain? Spine 1997;22:1137-42. 23 Burton AK, Tillotson KM. Prediction of the clinical course of low-back
18 Main CJ, Wood PL, Hollis S, Spanswick CC, Waddell G. The distress and trouble using multivariable models. Spine 1991;16:7-14.
risk assessment method. A simple patient classification to identify distress 24 Pope MH, Rosen JC, Wilder DG, Frymoyer JW. The relation between bio-
and evaluate the risk of poor outcome. Spine 1992;17:42-52. mechanical and psychological factors in patients with low-back pain.
19 Coste J, Delecoeuillerie G, Cohen de Lara A, Le Parc JM, Paolaggi JB. Spine 1980;5:173-8.
Clinical course and prognostic factors in acute low back pain: an incep-
tion cohort study in primary care practice. BMJ 1994;308:577-80. (Accepted 31 March 1999)

Statistics notes
Variables and parameters
Douglas G Altman, J Martin Bland

Like all specialist areas, statistics has developed its own ICRF Medical
80 Statistics Group,
language. As we have noted before,1 much confusion Frequency Centre for Statistics
may arise when a word in common use is also given a in Medicine,
technical meaning. Statistics abounds in such terms, Institute of Health
60 Sciences, Oxford
including normal, random, variance, significant, etc. OX3 7LF
Two commonly confused terms are variable and Douglas G Altman,
parameter; here we explain and contrast them. 40 professor of statistics
in medicine
Information recorded about a sample of individu-
als (often patients) comprises measurements such as Department of
20 Public Health
blood pressure, age, or weight and attributes such as Sciences, St
blood group, stage of disease, and diabetes. Values of George’s Hospital
these will vary among the subjects; in this context Medical School,
0 London SW17 0RE
35 40 45 50 55
blood pressure, weight, blood group and so on are J Martin Bland,
Albumin (g/l)
variables. Variables are quantities which vary from professor of medical
individual to individual. Measurements of serum albumin in 481 white men aged over 20 statistics
By contrast, parameters do not relate to actual (data from Dr W G Miller) Correspondence to:
Professor Altman.
measurements or attributes but to quantities defining a
theoretical model. The figure shows the distribution of (more generally known as regression coefficients) are
BMJ 1999;318:1667
measurements of serum albumin in 481 white men the parameters defining the model. They have no
aged over 20 with mean 46.14 and standard deviation meaning for individuals, although they can be used to
3.08 g/l. For the empirical data the mean and SD are predict an individual’s lung function from their height.
called sample estimates. They are properties of the col- In some contexts parameters are values that can be
lection of individuals. Also shown is the normal1 distri- altered to see what happens to the performance of
bution which fits the data most closely. It too has mean some system. For example, the performance of a
46.14 and SD 3.08 g/l. For the theoretical distribution screening programme (such as positive predictive
the mean and SD are called parameters. There is not value or cost effectiveness) will depend on aspects such
one normal distribution but many, called a family of as the sensitivity and specificity of the screening test. If
distributions. Each member of the family is defined by we look to see how the performance would change if,
its mean and SD, the parameters1 which specify the say, sensitivity and specificity were improved, then we
particular theoretical normal distribution with which are treating these as parameters rather than using the
we are dealing. In this case, they give the best estimate values observed in a real set of data.
of the population distribution of serum albumin if we Parameter is a technical term which has only
can assume that in the population serum albumin has recently found its way into general use, unfortunately
a normal distribution. without keeping its correct meaning. It is common in
Most statistical methods, such as t tests, are called medical journals to find variables incorrectly called
parametric because they estimate parameters of some parameters (but not in the BMJ we hope2). Another
underlying theoretical distribution. Non-parametric common misuse of parameter is as a limit or boundary,
methods, such as the Mann-Whitney U test and the log as in “within certain parameters.” This misuse seems to
rank test for survival data, do not assume any particu- have arisen from confusion between parameter and
lar family for the distribution of the data and so do not perimeter.
estimate any parameters for such a distribution. Misuse of medical terms is rightly deprecated. Like
Another use of the word parameter relates to its other language errors it leads to confusion and the loss
original mathematical meaning as the value(s) defining of valuable distinction. Misuse of non-medical terms
one of a family of curves. If we fit a regression model, should be viewed likewise.
such as that describing the relation between lung func- 1 Altman DG, Bland JM. The normal distribution. BMJ 1995;310:298.
tion and height, the slope and intercept of this line 2 Endpiece: What’s a parameter? BMJ 1998;316:1877.

BMJ VOLUME 318 19 JUNE 1999 www.bmj.com 1667