Knee osteoarthritis and obesity
International Journal of Obesity
Knee osteoarthritis and obesity
D Coggon 2 3
I Reading 2 3
P Croft 1 2
M McLaren 0 2
D Barrett 2 4
C Cooper 2 3
0 Department of Orthopaedic Surgery, Queen Alexandra Hospital , Cosham, Portsmouth , UK
1 Industrial and Community Health Research Centre, North Staffordshire Medical Institute , Stoke-on-Trent , UK
2 Southampton General Hospital , Southampton , UK
3 MRC Environmental Epidemiology Unit, University of Southampton, Southampton General Hospital , Southampton , UK
4 Department of Orthopaedic Surgery, Southampton General Hospital , Southampton , UK
OBJECTIVES: To assess the risk of knee osteoarthritis (OA) attributable to obesity, and the interactions between obesity and other established causes of the disorder. METHODS: We performed a population-based case ± control study in three health districts of England (Southampton, Portsmouth and North Staffordshire). A total of 525 men and women aged 45 y and over, consecutively listed for surgical treatment of primary knee OA, were compared with 525 controls matched by age, sex and family practitioner. RESULTS: Relative to a body mass index (BMI) of 24.0 ± 24.9 kg=m2, the risk of knee OA increased progressively from 0.1 (95% CI 0.0 ± 0.5) for a BMI < 20 kg=m2 to 13.6 (95% CI 5.1 ± 36.2) for a BMI of 36 kg=m2 or higher. If all overweight and obese people reduced their weight by 5 kg or until their BMI was within the recommended normal range, 24% of surgical cases of knee OA (95% CI 19 ± 27%) might be avoided. As a risk factor for knee OA obesity interacted more than additively with each of Heberden's nodes, earlier knee injury and meniscectomy. In comparison with subjects of normal weight, without Heberden's nodes, and with no history of knee injury, people with a combination of obesity, de®nite Heberden's nodes and previous knee injury had a relative risk of 78 (95% CI 17 ± 354). CONCLUSIONS: Our ®ndings give strong support to public health initiatives aimed at reducing the burden of knee OA by controlling obesity. People undergoing meniscectomy or with a history of knee injury might be a focus for targeted advice. International Journal of Obesity (2001) 25, 622 ± 627
osteoarthritis; epidemiology; prevention; knee
Osteoarthritis of the knee is a common cause of pain and
disability, especially in the elderly.1 In Britain, hospital
activity data2 suggest that 1.5% of people will undergo
surgical treatment for the disorder (usually total knee
arthroplasty) at some stage in their lives. Case ± control studies
have consistently demonstrated a strong association
between knee osteoarthritis and obesity,3 and in the
Framingham longitudinal study high body mass index (BMI)
predicted development of the disease in later life.4
Furthermore, analysis of women in the same cohort has shown that
incidence is lower in obese women who lose weight than in
those who do not;5 these data suggest that controlling
obesity can reduce risk.
Other established causes of knee osteoarthritis include
constitutional predisposition to osteoarthritis in multiple
joints (generalized osteoarthritis) and injury or surgery to
the knee.6 It is possible that when obesity is present in
combination with one or more of these risk factors, risk is
increased to the extent that targeted advice on weight
control would be appropriate. We have explored this question
using data from a case ± control study of knee osteoarthritis
requiring surgical treatment.
The cases were residents of North Staffordshire, Portsmouth
and Southampton Health Districts who were consecutively
placed on a waiting list for knee surgery because of
osteoarthritis during a 2 y period. In Portsmouth and North
Staffordshire we included private as well as National Health
Service (NHS) patients but for logistic reasons that was not
possible in Southampton. Each patient's case ± notes and
radiographs were reviewed to con®rm the diagnosis of
osteoarthritis, and to exclude those with underlying
rheumatoid arthritis, ankylosing spondylitis, or injury to the
knee within the past 12 months. The radiographic severity
of osteoarthritis in the knee requiring surgery was graded
according to the Kellgren and Lawrence classi®cation.7
Eligible cases who agreed to take part in the study were
visited at home by a research nurse who administered a
structured questionnaire. Among other things this asked
about earlier surgery to the knees and whether they had
ever suffered a knee injury bad enough to impair weight
bearing for a week or longer. In addition, the nurse examined
their hands for the presence of Heberden's nodes (as a marker
for osteoarthritis at a different site) and measured their
height (using a portable stadiometer) and weight (using
electronic scales). A Heberden's node was de®ned as a
palpable, tender or non-tender, bony swelling adjacent to the
distal interphalangeal joint of a ®nger and was assessed using
a three-point scale (none, possible or de®nite). If a clear
swelling was palpable and felt to be bony, the node was
assigned as de®nite; if the research nurse was uncertain as to
bony texture, it was assigned as possible. Although the
research nurse was not the principal member of the research
staff responsible for the selection of cases and controls, the
inclusion of questions pertaining to knee pain, disability and
surgery made it impossible for her to be blinded as to case ±
For each case who was interviewed we sought a control of
the same sex and matched as closely as possible for date of
birth, who was registered with the same general practitioner
as the case, and who had not undergone earlier surgery for
osteoarthritis. Following an introductory letter, those who
agreed to take part were visited at home and interviewed and
examined according to the same protocol as the cases.
Where possible, controls who could not be contacted or
declined to participate were replaced.
Associations of knee osteoarthritis with obesity, alone and
in combination with other risk factors, were examined by
conditional logistic regression.
We identi®ed a total of 729 cases who were eligible for
inclusion in the study, of whom 675 (93%) agreed to be
interviewed. The response rate from controls was rather
lower, interviews being obtained from 57% of the 1171
men and women whom we attempted to recruit. The
incomplete response occurred because the general practitioner did
not want the person to be approached (152 controls) or
because the control could not be contacted or declined to
take part (354). This left 665 matched pairs in which both
the case and control had been interviewed; 140 of these were
excluded from analysis because the control had previously
undergone knee surgery for osteoarthritis (eight pairs) or
information was missing on one or more of weight, height,
Heberden's nodes and history of knee injury or surgery (132
pairs). Missing information on these risk factors was more
frequent among cases (72%) than among controls (25%)
among these 132 pairs (in the remaining 3%, information
was missing in both case and control of a given pair).
However, the relative proportions of subjects with missing
information on each speci®c risk factor were similar among
cases and controls (BMI 31% of cases, 36% of controls;
previous injury 44% of cases, 58% of controls; Heberden's
nodes no cases, 3% of controls).
Of the 525 cases who were included in the analysis, 205
were men and 320 were women. Their ages ranged from 47
to 93 y with a median of 72 y, and the ages of all but six
controls were matched to within 2 y. The largest age
difference between a case and matched control was 4.1 y. In all,
78% of cases had a Kellgren and Lawrence grade of 3 or 4 in
the knee listed for surgery. Of the remainder, 17% had
Kellgren and Lawrence grade 2 osteoarthritis, 4% grade 1,
and 1% grade 0.
The median BMI in the cases was 28.1 kg=m2
(inter-quartile range (IQR) 25.6 ± 31.3) as compared with 25.3 kg=m2
(IQR 23.1 ± 28.0) in the controls. Table 1 shows the
distribution of cases and controls according to whether they were
underweight (BMI < 20.0 kg=m2), normal weight (BMI 20.0 ±
24.9 kg=m2), overweight (BMI 25.0 ± 29.9 kg=m2) or obese
(BMI 30.0 kg=m2), and the associated risks of knee
osteoarthritis. Odds ratios increased progressively across these
categories, more than a third of cases being classi®ed as obese.
To assess the relation of obesity to risk of knee
osteoarthritis in more detail, BMI was partitioned into 12 categories
with roughly equal numbers of subjects in each. Figure 1
shows odds ratios (OR) for each category relative to a BMI of
24.0 ± 24.9. There was a progressive increase in risk
throughout the range from 0.1 (95% CI 0.0 ± 0.5) for a BMI less than
20 to 13.6 (95% CI 5.1 ± 36.2) for a BMI of 36 or higher, and
within the range of the data the trend was well approximated
by an exponential function as illustrated. This took the form:
OR (BMI=24)i, where the point estimate for i was 5.2 with
95% CI 4.2 ± 6.3.
We used this formula to estimate the proportion of
surgical cases of knee osteoarthritis in the community that might
be avoided by controlling obesity. The complete derivation
of this attributable proportion is provided in Appendix 1. We
calculated the fraction of cases in the study sample that
Odds ratios derived from matched, unadjusted analyses. Each risk factor was
analysed in a separate regression model. aFor de®nition see text.
would have been eliminated if the risk among people who
were overweight or obese were reduced to that of people
with lower BMIs, assuming three different degrees of weight
loss (Table 2). Lowering their risk to the level associated with
a BMI in the normal range would have prevented more than
half of all cases, and even a more modest reduction
(equivalent to reducing weight by no more than 5 kg) would have
eliminated about a quarter of cases.
Table 3 shows the associations of knee osteoarthritis with
Heberden's nodes, earlier knee injury and meniscectomy
when each of these variables was examined independently.
For the purposes of this analysis, we only counted injuries
and meniscectomies in a case and his or her matched control
where they occurred on the side for which the case's surgical
treatment of osteoarthritis was planned and clearly preceded
the onset of the case's osteoarthritic symptoms on that side.
As expected, each risk factor was signi®cantly associated with
Table 4 shows associations with the same risk factors
when subjects were classi®ed according to their BMI. BMI
interacted with the other risk factors more than additively so
that, for example, the OR for obesity when combined with a
history of knee injury was 21.6 as compared with 8.2 in the
absence of knee injury and 6.1 for knee injury in subjects of
normal weight. A combination of obesity, de®nite
Heberden's nodes and earlier knee injury carried an even higher
risk (OR 78, 95% CI 17 ± 354 in comparison with normal
weight, no Heberden's nodes and no knee injury). All of the
above analyses were repeated separately in men and women.
The patterns of interaction between risk factors were similar
in the two sexes, but the risks associated with obesity tended
to be rather higher in women than in men. The table also
Proportion of cases (95% CI)
that might be eliminated
10.9% (8.8 ± 12.8%)
23.6% (19.4 ± 27.4%)
57.1% (46.7 ± 64.2%)
Previous knee injurya
illustrates that the risk of knee osteoarthritis associated with
nodes, injury and meniscectomy does in fact vary by
category of BMI.
With the assumptions that the overall lifetime risk of
surgery for knee osteoarthritis is 1.5% and that the
distribution of risk factors in the general population is similar to that
in our control group, Table 5 gives estimates of the lifetime
risk of surgically treated knee osteoarthritis according to BMI
and the presence or absence of other risk factors. In people
with no Heberden's nodes and no history of knee injury the
excess lifetime risk associated with obesity was estimated as
2.0%, but this rose to an excess of 9.5% for people who have
both de®nite Heberden's nodes and a past history of knee
Our ®ndings suggest that a large proportion of severe knee
osteoarthritis is attributable to obesity, and that risk is
particularly high among obese people who also have other
risk factors for the disease. The risk estimates are broadly in
accord with other population-based studies of
osteoarthritis.6,8 ± 11 In particular, the ®rst National Health and Nutrition
Examination survey conducted throughout the United States
in 1971 ± 1975 showed that women with a BMI between 30
and 35 kg=m2 had almost four times the risk of knee
osteoarthritis of women with a BMI under 25.8 Our study explored
the relation of obesity to knee osteoarthritis and its
interaction with other risk factors in greater detail than previous
The cases we studied came from geographically de®ned
populations in two regions of England (the South and the
Midlands), and although we were not able to recruit all
privately treated patients, only a small minority of those
eligible for entry to the study are likely to have been missed.
Thus, our study sample should be fairly representative of
cases that occur nationally.
The response rate from potential controls was lower than
for cases, but this is not surprising in a community-based
study of elderly people. In general, we would expect those
too frail or ill to participate to be underweight more often
than overweight, which if anything would cause the risks
associated with obesity to be under-estimated rather than
Another potential source of bias was differences in recall
between cases and controls. In particular, cases may have
remembered earlier injuries more completely than controls.
This possibility cannot be excluded, but by restricting our
de®nition to more severe injuries that interfered with weight
The calculations assume that the overall lifetime risk of surgery for knee osteoarthritis is 1.5% and that the prevalence of risk factors in the general population is
similar to that in the control group.
bearing for at least a week, we should have reduced the
chance of such errors. We also evaluated the extent to
which information on the key risk factors might have been
differentially missed between cases and controls. Although
missing information was more frequent among cases than
among controls, there were similar proportions of those in
whom data on BMI, Heberden's nodes and knee injury were
A more important limitation was our ability to measure
obesity only after the cases' osteoarthritis had reached an
advanced stage. This meant that the association of obesity
with knee osteoarthritis could have been exaggerated if some
patients had put on weight as a consequence of immobility
caused by their disease. Against a major contribution of this
type, however, is the fact that similar associations have been
found with the incidence, progression and later prevalence
of knee osteoarthritis in prospective studies where height
and weight were measured at the start of follow-up.5,9,12 ± 15
Furthermore, in one such study, subjects who lost weight
during follow-up had a lower risk than those who did not.5
When set alongside these ®ndings and those of other
retrospective investigations, our observations indicate that
control of obesity could have a major impact on the occurrence
of severe knee osteoarthritis.
Of course, knee osteoarthritis is only one of many adverse
consequences of obesity, but persuading people to reduce
weight and then maintain the loss is dif®cult.16 One factor
which may in¯uence their motivation is the perceived
personal gain in relation to the sacri®ces that must be
made. Table 4 indicates that the risks associated with obesity
are particularly high in people who have other risk factors
for knee osteoarthritis, and Table 5 attempts to translate the
observed associations into estimates of the potential gain to
individuals from controlling weight, with risks quanti®ed by
an absolute rather than a relative measure. These
calculations involve several assumptions and simpli®cations. The
lifetime risk of surgery for knee osteoarthritis is taken as
1.5%, and it is assumed that the distribution of risk factors in
the general population is similar to that in our control
group. Furthermore, no account is taken of the age at
which people ®rst develop Heberden's nodes or suffer knee
injuries. Nevertheless, the ®ndings suggest that the
reduction in risk from controlling weight could be substantial in
obese people with de®nite Heberden's nodes and a past
history of signi®cant knee injury. This information might
provide an added incentive to weight loss in such
Another potential target for advice about weight control
is patients undergoing meniscectomy. Most of the cases in
our study who reported meniscectomy would have been
treated by open surgery, and we do not yet know whether
similar risks of osteoarthritis are associated with
arthroscopic meniscectomy. However, it would be prudent to
advise patients who have a menisectomy about the need
to avoid obesity in order to minimize their risk of further
Ultimately, con®rmation of the bene®ts of weight control
in reducing the occurrence and severity of knee
osteoarthritis would best come from prospective intervention studies.
Meanwhile, our ®ndings give encouragement to public
health initiatives aimed at reducing obesity, and identify
certain groups of people who might be a focus for targeted
advice. They also highlight a further adverse effect that can
be expected if the prevalence of obesity in the general
population continues to rise.17 Projections of the impact
of this trend on future demands for health care should take
into account the likely increase in requirements for knee
This research was supported by a grant from the Arthritis
Research Campaign and IR's post was funded by the Colt
Foundation. We thank Syd Anstee, Trish Byng, Leslie
Campbell, Gillian Latham and Gill Smith who carried out the
®eldwork; Vanessa Cox and Graham Wield who supported
the data handling and analysis; and Sue McIntosh who
prepared the manuscript.
The attributable proportion of cases due to an exposure is
RRe ÿ 1
where RRe is the relative risk of the exposed to the unexposed
group (estimated by an odds ratio, ORe), Ce is the number of
cases exposed and Ct is the total number of cases.
By using the function OR (BMI=24)i where i 5.2 (4.2 ±
6.3), we can calculate the individual OR for every subject for
their actual weight vs their hypothetical reduced weight as:
where BMIak is the actual BMI for subject k and BMIrk is the
reduced BMI for subject k made by reducing their weight.
Then the attributable proportion of cases that might be
eliminated by weight loss, using the above two equations is
Ce ORk ÿ 1
That is the summation over all exposed (overweight or
1 Cooper C. Epidemiology of Osteoarthritis . In: Klippel JH , Dieppe PA (eds). Rheumatology (2nd edn) . Mosby: London; 1998. pp 8.2.1 ± 8.2.8.
2 Williams M , Frankel S , Nanchahal K , Coast J , Donovan J . Epidemiologically based needs assessment: total knee replacement . Health Care Evaluation Unit , University of Bristol: Bristol; 1992 .
3 Spector TD . The fat on the joint: osteoarthritis and obesity . J Rheumatol 1990 ; 17 : 283 ± 284 .
4 Felson DT , Anderson JJ , Naimark A , Walker AM , Meenan RF . Obesity and knee osteoarthritis . The Framingham Study. Ann Intern Med 1988 ; 109 : 18 ± 24 .
5 Felson DT , Zhang Y , Anthony JM , Naimark A , Anderson JJ . Weight loss reduces the risk for symptomatic knee OA in women: the Framingham Study . Ann Intern Med 1992 ; 116 : 535 ± 539 .
6 Cooper C , McAlindon T , Snow S , Vines K , Young P , Kirwan J , Dieppe P . Mechanical and constitutional risk factors for symptomatic knee osteoarthritis: differences between medial tibiofemoral and patellofemoral disease . J Rheumatol 1994 ; 21 : 307 ± 313 .
7 Kellgren JH , Lawrence JS . Atlas of Standard Radiographs: The Epidemiology of Chronic Rheumatism , Vol 2 . Oxford: Blackwell; 1963 .
8 Anderson JJ , Felson DT . Factors associated with osteoarthritis of the knee in the ®rst national health and nutrition survey (HANES 1): evidence for an association with overweight, race and physical demands of work . Am J Epidemiol 1988 ; 128 : 179 ± 189 .
9 Schouten JSAG , Van den Ouweland FA , Valkenburg HA . A twelve year follow-up study in the general population on prognostic factors of cartilage loss in osteoarthritis of the knee . Ann Rheum Dis 1992 ; 51 : 932 ± 937 .
10 Davis MA , Ettinger WH , Neuhaus JM , Hauck WW . Sex differences in osteoarthritis of the knee; the role of obesity . Am J Epidemiol 1988 ; 127 : 1019 ± 1030 .
11 Hart DJ , Spector TD . The relationship of obesity, fat distribution and osteoarthritis in women in the general population: the Chingford study . J Rheumatol 1993 ; 20 : 331 ± 335 .
12 Spector TD , Hart DJ , Doyle DV . Incidence and progression of osteoarthritis in women with unilateral knee disease in the general population: the effect of obesity . Ann Rheum Dis 1994 ; 53 : 565 ± 568 .
13 Felson DT , Zhang Y , Hannan MT , Naimark A , Weissman B , Aliabadi P , Levy D . Risk factors for incident radiographic knee osteoarthritis in the elderly: the Framingham Study . Arthritis Rheum 1997 ; 40 : 728 ± 733 .
14 Oliveria SA , Felson DT , Cirillo PA , Reed JI , Walker AM . Body weight, body mass index and incident symptomatic osteoarthritis of the hand, hip and knee . Epidemiology 1999 ; 10 : 161 ± 166 .
15 Hart DJ , Doyle DV , Spector TD . Incidence and risk factors for radiographic knee osteoarthritis in middle aged women: the Chingford Study . Arthritis Rheum 1999 ; 42 : 17 ± 24 .
16 Summerbell CD , Watts C , Higgins JPT , Garrow JS . Randomised controlled trial of novel simple and well supervised weight reducing diets in outpatients . Br Med J 1998 ; 317 : 1487 ± 1489 .
17 Bost L , Primatesta P , Dong W. Anthropometric measures and children's iron status . In: Prescott-Clarke P , Primatesta P (eds). Health survey for England' 96 , Vol 1 . The Stationery Of®ce: London; 1998 . pp 282 ± 304 .